log.cc 159 KB
Newer Older
1
/* Copyright 2000-2008 MySQL AB, 2008 Sun Microsystems, Inc.
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 bool test_if_number(const char *str,
			   long *res, bool allow_wildcards);
57 58 59 60 61 62 63
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);
64

65 66 67 68 69 70 71 72 73
/**
  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
{
74
  char m_message[MYSQL_ERRMSG_SIZE];
75 76
public:
  Silence_log_table_errors()
77 78 79
  {
    m_message[0]= '\0';
  }
80 81 82

  virtual ~Silence_log_table_errors() {}

83
  virtual bool handle_error(uint sql_errno, const char *message,
84 85
                            MYSQL_ERROR::enum_warning_level level,
                            THD *thd);
86
  const char *message() const { return m_message; }
87 88 89 90
};

bool
Silence_log_table_errors::handle_error(uint /* sql_errno */,
91
                                       const char *message_arg,
92 93 94
                                       MYSQL_ERROR::enum_warning_level /* level */,
                                       THD * /* thd */)
{
95
  strmake(m_message, message_arg, sizeof(m_message)-1);
96 97 98 99
  return TRUE;
}


100 101 102 103 104 105 106 107
sql_print_message_func sql_print_message_handlers[3] =
{
  sql_print_information,
  sql_print_warning,
  sql_print_error
};


108 109
char *make_default_log_name(char *buff,const char* log_ext)
{
110
  strmake(buff, pidfile_name, FN_REFLEN-5);
111
  return fn_format(buff, buff, mysql_data_home, log_ext,
112
                   MYF(MY_UNPACK_FILENAME|MY_REPLACE_EXT));
113 114
}

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
/*
  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&);
};

150
/*
151
  Helper class to store binary log transaction data.
152
*/
153 154 155
class binlog_trx_data {
public:
  binlog_trx_data()
156 157
    : at_least_one_stmt(0), incident(FALSE), m_pending(0),
    before_stmt_pos(MY_OFF_T_UNDEF)
158 159 160 161 162 163 164 165 166 167 168 169 170 171
  {
    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);
  }

172 173
  bool empty() const
  {
174
    return pending() == NULL && my_b_tell(&trans_log) == 0;
175
  }
176 177 178 179 180 181 182

  /*
    Truncate the transaction cache to a certain position. This
    includes deleting the pending event.
   */
  void truncate(my_off_t pos)
  {
183 184
    DBUG_PRINT("info", ("truncating to position %lu", (ulong) pos));
    DBUG_PRINT("info", ("before_stmt_pos=%lu", (ulong) pos));
185 186 187
    delete pending();
    set_pending(0);
    reinit_io_cache(&trans_log, WRITE_CACHE, pos, 0, 0);
188
    trans_log.end_of_file= max_binlog_cache_size;
189 190
    if (pos < before_stmt_pos)
      before_stmt_pos= MY_OFF_T_UNDEF;
191 192 193 194 195 196 197 198 199 200

    /*
      The only valid positions that can be truncated to are at the
      beginning of a statement. We are relying on this fact to be able
      to set the at_least_one_stmt flag correctly. In other word, if
      we are truncating to the beginning of the transaction cache,
      there will be no statements in the cache, otherwhise, we will
      have at least one statement in the transaction cache.
     */
    at_least_one_stmt= (pos > 0);
201 202 203 204 205 206 207 208 209
  }

  /*
    Reset the entire contents of the transaction cache, emptying it
    completely.
   */
  void reset() {
    if (!empty())
      truncate(0);
210
    before_stmt_pos= MY_OFF_T_UNDEF;
211
    incident= FALSE;
212
    trans_log.end_of_file= max_binlog_cache_size;
213
    DBUG_ASSERT(empty());
214 215 216 217 218 219 220 221 222 223 224 225
  }

  Rows_log_event *pending() const
  {
    return m_pending;
  }

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

226
  IO_CACHE trans_log;                         // The transaction cache
227

228 229 230 231 232 233 234 235 236 237
  void set_incident(void)
  {
    incident= TRUE;
  }
  
  bool has_incident(void)
  {
    return(incident);
  }

238 239 240 241 242
  /**
    Boolean that is true if there is at least one statement in the
    transaction cache.
  */
  bool at_least_one_stmt;
243
  bool incident;
244

245 246 247 248 249 250 251 252 253 254 255 256
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;
257 258
};

259
handlerton *binlog_hton;
260

261 262 263 264 265 266 267 268 269 270 271 272 273
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 */
  }
}

274

275 276 277 278 279 280 281 282 283 284 285 286
/* 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") :
287 288 289 290 291 292 293 294 295 296 297 298
                                  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))
299
        return QUERY_LOG_SLOW;
300 301
      return 0;
    }
302 303 304 305
  }
  return 0;
}

306

307 308 309 310 311 312 313 314 315
Log_to_csv_event_handler::Log_to_csv_event_handler()
{
}


Log_to_csv_event_handler::~Log_to_csv_event_handler()
{
}

316

317 318 319 320 321 322 323
void Log_to_csv_event_handler::cleanup()
{
  logger.is_log_tables_initialized= FALSE;
}

/* log event handlers */

324
/**
325 326
  Log command to the general log table

327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351
  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
352 353 354
*/

bool Log_to_csv_event_handler::
355
  log_general(THD *thd, time_t event_time, const char *user_host,
356 357
              uint user_host_len, int thread_id,
              const char *command_type, uint command_type_len,
358 359
              const char *sql_text, uint sql_text_len,
              CHARSET_INFO *client_cs)
360
{
361 362 363 364 365 366
  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
367
  uint field_index;
368 369 370 371 372
  Silence_log_table_errors error_handler;
  Open_tables_state open_tables_backup;
  ulonglong save_thd_options;
  bool save_time_zone_used;

373 374 375 376 377
  /*
    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;
378 379 380 381 382 383 384 385 386 387 388 389 390

  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;

391 392 393 394 395 396 397 398 399 400 401 402 403 404
  /*
    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;

405 406 407 408
  if (!(table= open_performance_schema_table(thd, & table_list,
                                             & open_tables_backup)))
    goto err;

409 410
  need_close= TRUE;

411
  if (table->file->extra(HA_EXTRA_MARK_AS_LOG_TABLE) ||
412 413 414 415 416 417 418
      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;
419

420 421 422 423 424 425 426 427 428
  /*
    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).
  */
429

430
  /* check that all columns exist */
431
  if (table->s->fields < 6)
432 433
    goto err;

434 435
  DBUG_ASSERT(table->field[0]->type() == MYSQL_TYPE_TIMESTAMP);

436 437
  ((Field_timestamp*) table->field[0])->store_timestamp((my_time_t)
                                                        event_time);
438

439 440 441 442
  /* 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) ||
443
      table->field[4]->store(command_type, command_type_len, client_cs))
444 445
    goto err;

446 447 448 449
  /*
    A positive return value in store() means truncation.
    Still logging a message in the log in this case.
  */
450
  table->field[5]->flags|= FIELDFLAG_HEX_ESCAPE;
451 452
  if (table->field[5]->store(sql_text, sql_text_len, client_cs) < 0)
    goto err;
453 454

  /* mark all fields as not null */
455 456 457 458 459
  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();
460

461 462 463 464 465 466
  /* 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();
  }

467
  /* log table entries are not replicated */
468
  if (table->file->ha_write_row(table->record[0]))
469
    goto err;
470

471
  result= FALSE;
472

473
err:
474
  if (result && !thd->killed)
475 476
    sql_print_error("Failed to write to mysql.general_log: %s",
                    error_handler.message());
477

478 479 480 481 482 483 484 485 486 487 488 489 490
  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;
491 492 493 494 495 496 497
}


/*
  Log a query to the slow log table

  SYNOPSIS
498
    log_slow()
499 500 501 502 503 504
    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
505 506
    query_time        Amount of time the query took to execute (in microseconds)
    lock_time         Amount of time the query was locked (in microseconds)
507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525
    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,
526
           ulonglong query_utime, ulonglong lock_utime, bool is_command,
527 528
           const char *sql_text, uint sql_text_len)
{
529 530 531 532 533
  TABLE_LIST table_list;
  TABLE *table;
  bool result= TRUE;
  bool need_close= FALSE;
  bool need_rnd_end= FALSE;
534
  Silence_log_table_errors error_handler;
535
  Open_tables_state open_tables_backup;
536
  CHARSET_INFO *client_cs= thd->variables.character_set_client;
537
  bool save_time_zone_used;
538
  DBUG_ENTER("Log_to_csv_event_handler::log_slow");
539

540
  thd->push_internal_handler(& error_handler);
541 542 543 544 545 546
  /*
    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;

547 548 549 550 551 552 553 554 555
  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;

556 557 558
  if (!(table= open_performance_schema_table(thd, & table_list,
                                             & open_tables_backup)))
    goto err;
559 560 561

  need_close= TRUE;

562
  if (table->file->extra(HA_EXTRA_MARK_AS_LOG_TABLE) ||
563 564 565 566 567 568 569
      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;
570

571
  restore_record(table, s->default_values);    // Get empty record
572

573 574 575 576
  /* check that all columns exist */
  if (table->s->fields < 11)
    goto err;

577 578
  /* store the time and user values */
  DBUG_ASSERT(table->field[0]->type() == MYSQL_TYPE_TIMESTAMP);
579 580
  ((Field_timestamp*) table->field[0])->store_timestamp((my_time_t)
                                                        current_time);
581 582
  if (table->field[1]->store(user_host, user_host_len, client_cs))
    goto err;
583 584 585

  if (query_start_arg)
  {
586 587
    longlong query_time= (longlong) (query_utime/1000000);
    longlong lock_time=  (longlong) (lock_utime/1000000);
588 589 590 591 592
    /*
      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)
    */
593
    MYSQL_TIME t;
594 595
    t.neg= 0;

596
    /* fill in query_time field */
597 598
    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))
599
      goto err;
600
    /* lock_time */
601 602
    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))
603
      goto err;
604
    /* rows_sent */
605 606
    if (table->field[4]->store((longlong) thd->sent_row_count, TRUE))
      goto err;
607
    /* rows_examined */
608 609
    if (table->field[5]->store((longlong) thd->examined_row_count, TRUE))
      goto err;
610 611 612 613 614 615 616 617
  }
  else
  {
    table->field[2]->set_null();
    table->field[3]->set_null();
    table->field[4]->set_null();
    table->field[5]->set_null();
  }
618
  /* fill database field */
619
  if (thd->db)
620
  {
621 622
    if (table->field[6]->store(thd->db, thd->db_length, client_cs))
      goto err;
623 624
    table->field[6]->set_notnull();
  }
625

626
  if (thd->stmt_depends_on_first_successful_insert_id_in_prev_stmt)
627
  {
628 629 630 631 632
    if (table->
        field[7]->store((longlong)
                        thd->first_successful_insert_id_in_prev_stmt_for_binlog,
                        TRUE))
      goto err;
633 634
    table->field[7]->set_notnull();
  }
635

636 637 638 639 640 641 642
  /*
    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)
643
  {
644 645 646 647
    if (table->
        field[8]->store((longlong)
          thd->auto_inc_intervals_in_cur_stmt_for_binlog.minimum(), TRUE))
      goto err;
648 649
    table->field[8]->set_notnull();
  }
650

651 652
  if (table->field[9]->store((longlong) server_id, TRUE))
    goto err;
653
  table->field[9]->set_notnull();
654

655 656 657 658 659 660
  /*
    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)
661 662
    goto err;

663
  /* log table entries are not replicated */
664
  if (table->file->ha_write_row(table->record[0]))
665
    goto err;
666

667
  result= FALSE;
668

669
err:
670 671
  thd->pop_internal_handler();

672
  if (result && !thd->killed)
673 674
    sql_print_error("Failed to write to mysql.slow_log: %s",
                    error_handler.message());
675

676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717
  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);
718 719 720 721 722 723 724 725
  if (table)
  {
    result= 0;
    close_performance_schema_table(thd, & open_tables_backup);
  }
  else
    result= 1;

726
  DBUG_RETURN(result);
727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750
}

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();
}


751
/** Wrapper around MYSQL_LOG::write() for slow log. */
752 753 754 755

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,
756
           ulonglong query_utime, ulonglong lock_utime, bool is_command,
757 758
           const char *sql_text, uint sql_text_len)
{
759 760 761 762 763 764 765 766
  Silence_log_table_errors error_handler;
  thd->push_internal_handler(&error_handler);
  bool retval= mysql_slow_log.write(thd, current_time, query_start_arg,
                                    user_host, user_host_len,
                                    query_utime, lock_utime, is_command,
                                    sql_text, sql_text_len);
  thd->pop_internal_handler();
  return retval;
767 768 769
}


770
/**
771 772 773 774 775
   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::
776
  log_general(THD *thd, time_t event_time, const char *user_host,
777 778
              uint user_host_len, int thread_id,
              const char *command_type, uint command_type_len,
779 780
              const char *sql_text, uint sql_text_len,
              CHARSET_INFO *client_cs)
781
{
782 783 784 785 786 787 788
  Silence_log_table_errors error_handler;
  thd->push_internal_handler(&error_handler);
  bool retval= mysql_log.write(event_time, user_host, user_host_len,
                               thread_id, command_type, command_type_len,
                               sql_text, sql_text_len);
  thd->pop_internal_handler();
  return retval;
789 790 791 792 793 794 795 796
}


bool Log_to_file_event_handler::init()
{
  if (!is_initialized)
  {
    if (opt_slow_log)
797
      mysql_slow_log.open_slow_log(sys_var_slow_log_path.value);
798 799

    if (opt_log)
800
      mysql_log.open_query_log(sys_var_general_log_path.value);
801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817

    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 */
818
  if (opt_log)
819
    mysql_log.reopen_file();
820
  if (opt_slow_log)
821
    mysql_slow_log.reopen_file();
822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843
}

/*
  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;
844
  Log_event_handler **current_handler;
845 846

  /* currently we don't need locking here as there is no error_log table */
847
  for (current_handler= error_log_handler_list ; *current_handler ;)
848 849 850 851 852 853
    error= (*current_handler++)->log_error(level, format, args) || error;

  return error;
}


petr@mysql.com's avatar
petr@mysql.com committed
854
void LOGGER::cleanup_base()
855 856
{
  DBUG_ASSERT(inited == 1);
857
  rwlock_destroy(&LOCK_logger);
858
  if (table_log_handler)
petr@mysql.com's avatar
petr@mysql.com committed
859
  {
860
    table_log_handler->cleanup();
petr@mysql.com's avatar
petr@mysql.com committed
861
    delete table_log_handler;
862
    table_log_handler= NULL;
petr@mysql.com's avatar
petr@mysql.com committed
863
  }
864 865 866 867 868
  if (file_log_handler)
    file_log_handler->cleanup();
}


petr@mysql.com's avatar
petr@mysql.com committed
869 870 871 872
void LOGGER::cleanup_end()
{
  DBUG_ASSERT(inited == 1);
  if (file_log_handler)
873
  {
petr@mysql.com's avatar
petr@mysql.com committed
874
    delete file_log_handler;
875 876 877
    file_log_handler=NULL;
  }
  inited= 0;
petr@mysql.com's avatar
petr@mysql.com committed
878 879 880
}


881
/**
882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898
  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 */
899
  init_error_log(LOG_FILE);
900 901

  file_log_handler->init_pthread_objects();
902
  my_rwlock_init(&LOCK_logger, NULL);
903 904 905 906 907 908 909 910 911 912 913 914 915 916
}


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;
}


917 918 919
bool LOGGER::flush_logs(THD *thd)
{
  int rc= 0;
920

921
  /*
922 923
    Now we lock logger, as nobody should be able to use logging routines while
    log tables are closed
924
  */
925
  logger.lock_exclusive();
926 927 928

  /* reopen log files */
  file_log_handler->flush();
929

930 931
  /* end of log flush */
  logger.unlock();
932
  return rc;
933 934 935 936 937 938 939 940 941
}


/*
  Log slow query with all enabled log event handlers

  SYNOPSIS
    slow_log_print()

942 943 944 945
    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)
946 947

  RETURN
948 949
    FALSE   OK
    TRUE    error occured
950 951 952
*/

bool LOGGER::slow_log_print(THD *thd, const char *query, uint query_length,
953 954
                            ulonglong current_utime)

955 956
{
  bool error= FALSE;
957
  Log_event_handler **current_handler;
958
  bool is_command= FALSE;
959
  char user_host_buff[MAX_USER_HOST_SIZE + 1];
960
  Security_context *sctx= thd->security_ctx;
961
  uint user_host_len= 0;
962
  ulonglong query_utime, lock_utime;
963 964 965 966 967 968 969

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

  if (*slow_log_handler_list)
  {
970
    time_t current_time;
971

972
    /* do not log slow queries from replication threads */
973
    if (thd->slave_thread && !opt_log_slow_slave_statements)
974 975
      return 0;

976
    lock_shared();
977 978 979 980 981
    if (!opt_slow_log)
    {
      unlock();
      return 0;
    }
982 983

    /* fill in user_host value: the format is "%s[%s] @ %s [%s]" */
984 985 986 987 988 989 990 991 992
    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)
993
    {
994 995 996 997
      query_utime= (current_utime - thd->start_utime);
      lock_utime=  (thd->utime_after_lock - thd->start_utime);
    }
    else
998
    {
999
      query_utime= lock_utime= 0;
1000 1001 1002 1003 1004 1005 1006 1007 1008
    }

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

1009 1010
    for (current_handler= slow_log_handler_list; *current_handler ;)
      error= (*current_handler++)->log_slow(thd, current_time, thd->start_time,
1011
                                            user_host_buff, user_host_len,
1012
                                            query_utime, lock_utime, is_command,
1013 1014 1015 1016 1017 1018 1019
                                            query, query_length) || error;

    unlock();
  }
  return error;
}

1020 1021
bool LOGGER::general_log_write(THD *thd, enum enum_server_command command,
                               const char *query, uint query_length)
1022 1023 1024
{
  bool error= FALSE;
  Log_event_handler **current_handler= general_log_handler_list;
1025
  char user_host_buff[MAX_USER_HOST_SIZE + 1];
1026 1027 1028 1029
  Security_context *sctx= thd->security_ctx;
  ulong id;
  uint user_host_len= 0;
  time_t current_time;
1030

1031 1032 1033 1034
  if (thd)
    id= thd->thread_id;                 /* Normal thread */
  else
    id= 0;                              /* Log from connect handler */
1035

1036 1037 1038
  lock_shared();
  if (!opt_log)
  {
1039
    unlock();
1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050
    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)
1051
    error|= (*current_handler++)->
1052 1053 1054 1055 1056 1057 1058 1059
      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();

1060 1061 1062
  return error;
}

1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078
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);
}

1079
void LOGGER::init_error_log(uint error_log_printer)
1080
{
1081 1082
  if (error_log_printer & LOG_NONE)
  {
1083
    error_log_handler_list[0]= 0;
1084 1085 1086 1087 1088
    return;
  }

  switch (error_log_printer) {
  case LOG_FILE:
1089 1090 1091 1092
    error_log_handler_list[0]= file_log_handler;
    error_log_handler_list[1]= 0;
    break;
    /* these two are disabled for now */
1093
  case LOG_TABLE:
1094 1095
    DBUG_ASSERT(0);
    break;
1096
  case LOG_TABLE|LOG_FILE:
1097 1098 1099 1100 1101
    DBUG_ASSERT(0);
    break;
  }
}

1102
void LOGGER::init_slow_log(uint slow_log_printer)
1103
{
1104 1105
  if (slow_log_printer & LOG_NONE)
  {
1106
    slow_log_handler_list[0]= 0;
1107 1108 1109 1110 1111
    return;
  }

  switch (slow_log_printer) {
  case LOG_FILE:
1112 1113 1114
    slow_log_handler_list[0]= file_log_handler;
    slow_log_handler_list[1]= 0;
    break;
1115
  case LOG_TABLE:
1116 1117 1118
    slow_log_handler_list[0]= table_log_handler;
    slow_log_handler_list[1]= 0;
    break;
1119
  case LOG_TABLE|LOG_FILE:
1120 1121 1122 1123 1124 1125 1126
    slow_log_handler_list[0]= file_log_handler;
    slow_log_handler_list[1]= table_log_handler;
    slow_log_handler_list[2]= 0;
    break;
  }
}

1127
void LOGGER::init_general_log(uint general_log_printer)
1128
{
1129 1130
  if (general_log_printer & LOG_NONE)
  {
1131
    general_log_handler_list[0]= 0;
1132 1133 1134 1135 1136
    return;
  }

  switch (general_log_printer) {
  case LOG_FILE:
1137 1138 1139
    general_log_handler_list[0]= file_log_handler;
    general_log_handler_list[1]= 0;
    break;
1140
  case LOG_TABLE:
1141 1142 1143
    general_log_handler_list[0]= table_log_handler;
    general_log_handler_list[1]= 0;
    break;
1144
  case LOG_TABLE|LOG_FILE:
1145 1146 1147 1148 1149 1150 1151 1152
    general_log_handler_list[0]= file_log_handler;
    general_log_handler_list[1]= table_log_handler;
    general_log_handler_list[2]= 0;
    break;
  }
}


1153 1154
bool LOGGER::activate_log_handler(THD* thd, uint log_type)
{
1155
  MYSQL_QUERY_LOG *file_log;
1156 1157
  bool res= FALSE;
  lock_exclusive();
1158 1159 1160 1161
  switch (log_type) {
  case QUERY_LOG_SLOW:
    if (!opt_slow_log)
    {
1162 1163 1164
      file_log= file_log_handler->get_mysql_slow_log();

      file_log->open_slow_log(sys_var_slow_log_path.value);
1165 1166 1167 1168
      if (table_log_handler->activate_log(thd, QUERY_LOG_SLOW))
      {
        /* Error printed by open table in activate_log() */
        res= TRUE;
1169
        file_log->close(0);
1170 1171
      }
      else
1172 1173
      {
        init_slow_log(log_output_options);
1174
        opt_slow_log= TRUE;
1175
      }
1176 1177 1178 1179 1180
    }
    break;
  case QUERY_LOG_GENERAL:
    if (!opt_log)
    {
1181 1182 1183
      file_log= file_log_handler->get_mysql_log();

      file_log->open_query_log(sys_var_general_log_path.value);
1184 1185 1186 1187
      if (table_log_handler->activate_log(thd, QUERY_LOG_GENERAL))
      {
        /* Error printed by open table in activate_log() */
        res= TRUE;
1188
        file_log->close(0);
1189 1190
      }
      else
1191 1192
      {
        init_general_log(log_output_options);
1193
        opt_log= TRUE;
1194
      }
1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219
    }
    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:
1220
    assert(0);                                  // Impossible
1221 1222 1223 1224 1225
  }

  if (!(*tmp_opt))
    return;

1226
  lock_exclusive();
1227 1228 1229 1230 1231 1232
  file_log->close(0);
  *tmp_opt= FALSE;
  unlock();
}


1233 1234 1235
/* the parameters are unused for the log tables */
bool Log_to_csv_event_handler::init()
{
1236
  return 0;
1237 1238
}

1239 1240 1241
int LOGGER::set_handlers(uint error_log_printer,
                         uint slow_log_printer,
                         uint general_log_printer)
1242 1243
{
  /* error log table is not supported yet */
1244
  DBUG_ASSERT(error_log_printer < LOG_TABLE);
1245

1246
  lock_exclusive();
1247

1248
  if ((slow_log_printer & LOG_TABLE || general_log_printer & LOG_TABLE) &&
1249 1250
      !is_log_tables_initialized)
  {
1251 1252
    slow_log_printer= (slow_log_printer & ~LOG_TABLE) | LOG_FILE;
    general_log_printer= (general_log_printer & ~LOG_TABLE) | LOG_FILE;
1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267

    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;
}


1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287
 /*
  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);
1288
  if (thd_get_ha_data(thd, binlog_hton) == NULL)
1289 1290
    thd->binlog_setup_trx_data();
  binlog_trx_data *const trx_data=
1291
    (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
1292 1293
  DBUG_ASSERT(mysql_bin_log.is_open());
  *pos= trx_data->position();
1294
  DBUG_PRINT("return", ("*pos: %lu", (ulong) *pos));
1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317
  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");
1318
  DBUG_PRINT("enter", ("pos: %lu", (ulong) pos));
1319

1320
  DBUG_ASSERT(thd_get_ha_data(thd, binlog_hton) != NULL);
1321 1322 1323 1324
  /* Only true if binlog_trans_log_savepos() wasn't called before */
  DBUG_ASSERT(pos != ~(my_off_t) 0);

  binlog_trx_data *const trx_data=
1325
    (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
1326 1327 1328 1329 1330
  trx_data->truncate(pos);
  DBUG_VOID_RETURN;
}


1331 1332
/*
  this function is mostly a placeholder.
1333
  conceptually, binlog initialization (now mostly done in MYSQL_BIN_LOG::open)
1334 1335 1336
  should be moved here.
*/

1337
int binlog_init(void *p)
1338
{
1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349
  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;
1350
  return 0;
1351 1352
}

1353
static int binlog_close_connection(handlerton *hton, THD *thd)
1354
{
1355
  binlog_trx_data *const trx_data=
1356
    (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
1357
  DBUG_ASSERT(trx_data->empty());
1358
  thd_set_ha_data(thd, binlog_hton, NULL);
1359
  trx_data->~binlog_trx_data();
1360
  my_free((uchar*)trx_data, MYF(0));
1361 1362 1363
  return 0;
}

1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386
/*
  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).
 */
1387
static int
1388 1389
binlog_end_trans(THD *thd, binlog_trx_data *trx_data,
                 Log_event *end_ev, bool all)
1390
{
1391
  DBUG_ENTER("binlog_end_trans");
1392 1393
  int error=0;
  IO_CACHE *trans_log= &trx_data->trans_log;
1394 1395
  DBUG_PRINT("enter", ("transaction: %s  end_ev: 0x%lx",
                       all ? "all" : "stmt", (long) end_ev));
1396 1397 1398
  DBUG_PRINT("info", ("thd->options={ %s%s}",
                      FLAGSTR(thd->options, OPTION_NOT_AUTOCOMMIT),
                      FLAGSTR(thd->options, OPTION_BEGIN)));
1399

1400 1401 1402 1403 1404 1405
  /*
    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.
  */
1406
  if (end_ev != NULL)
1407
  {
1408 1409
    if (thd->binlog_flush_pending_rows_event(TRUE))
      DBUG_RETURN(1);
1410
    /*
1411 1412 1413 1414
      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.

1415 1416 1417 1418 1419
      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.
     */
1420 1421
    error= mysql_bin_log.write(thd, &trx_data->trans_log, end_ev,
                               trx_data->has_incident());
1422 1423
    trx_data->reset();

1424
    /*
1425 1426 1427 1428 1429 1430
      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)
1431
    {
1432 1433
      statistic_increment(binlog_cache_disk_use, &LOCK_status);
      trans_log->disk_writes= 0;
1434
    }
1435 1436 1437
  }
  else
  {
1438 1439 1440
    /*
      If rolling back an entire transaction or a single statement not
      inside a transaction, we reset the transaction cache.
1441

1442 1443 1444
      If rolling back a statement in a transaction, we truncate the
      transaction cache to remove the statement.
     */
1445
    thd->binlog_remove_pending_rows_event(TRUE);
1446
    if (all || !(thd->options & (OPTION_BEGIN | OPTION_NOT_AUTOCOMMIT)))
1447 1448 1449
    {
      if (trx_data->has_incident())
        mysql_bin_log.write_incident(thd, TRUE);
1450
      trx_data->reset();
1451
    }
1452 1453
    else                                        // ...statement
      trx_data->truncate(trx_data->before_stmt_pos);
1454 1455 1456 1457 1458 1459 1460

    /*
      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();
1461
  }
1462

1463
  DBUG_ASSERT(thd->binlog_get_pending_rows_event() == NULL);
1464
  DBUG_RETURN(error);
1465 1466
}

1467
static int binlog_prepare(handlerton *hton, THD *thd, bool all)
1468 1469 1470 1471 1472
{
  /*
    do nothing.
    just pretend we can do 2pc, so that MySQL won't
    switch to 1pc.
1473
    real work will be done in MYSQL_BIN_LOG::log_xid()
1474 1475 1476 1477
  */
  return 0;
}

1478 1479 1480 1481 1482 1483 1484 1485
/**
  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.
1486 1487
  @param all   This is @c true if this is a real transaction commit, and
               @false otherwise.
1488 1489 1490

  @see handlerton::commit
*/
1491
static int binlog_commit(handlerton *hton, THD *thd, bool all)
1492
{
1493
  int error= 0;
1494
  DBUG_ENTER("binlog_commit");
1495
  binlog_trx_data *const trx_data=
1496
    (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
1497

1498
  if (trx_data->empty())
1499
  {
1500
    // we're here because trans_log was flushed in MYSQL_BIN_LOG::log_xid()
1501
    trx_data->reset();
1502 1503
    DBUG_RETURN(0);
  }
1504

1505
  /*
1506
    We commit the transaction if:
1507

1508
     - We are not in a transaction and committing a statement, or
1509

1510
     - We are in a transaction and a full transaction is committed
1511 1512

    Otherwise, we accumulate the statement
1513
  */
1514 1515 1516 1517 1518 1519 1520 1521 1522
  ulonglong const in_transaction=
    thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN);
  DBUG_PRINT("debug",
             ("all: %d, empty: %s, in_transaction: %s, all.modified_non_trans_table: %s, stmt.modified_non_trans_table: %s",
              all,
              YESNO(trx_data->empty()),
              YESNO(in_transaction),
              YESNO(thd->transaction.all.modified_non_trans_table),
              YESNO(thd->transaction.stmt.modified_non_trans_table)));
1523
  if (!in_transaction || all)
1524
  {
He Zhenxing's avatar
He Zhenxing committed
1525
    Query_log_event qev(thd, STRING_WITH_LEN("COMMIT"), TRUE, TRUE, 0);
1526 1527
    error= binlog_end_trans(thd, trx_data, &qev, all);
    goto end;
1528
  }
1529 1530 1531 1532 1533

end:
  if (!all)
    trx_data->before_stmt_pos = MY_OFF_T_UNDEF; // part of the stmt commit
  DBUG_RETURN(error);
1534 1535
}

1536 1537 1538 1539 1540 1541 1542 1543 1544 1545
/**
  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.
1546 1547
  @param all   This is @c true if this is a real transaction rollback, and
               @false otherwise.
1548 1549 1550

  @see handlerton::rollback
*/
1551
static int binlog_rollback(handlerton *hton, THD *thd, bool all)
1552 1553
{
  DBUG_ENTER("binlog_rollback");
1554 1555
  int error=0;
  binlog_trx_data *const trx_data=
1556
    (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
1557 1558 1559 1560 1561 1562

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

1563 1564 1565 1566
  DBUG_PRINT("debug", ("all: %s, all.modified_non_trans_table: %s, stmt.modified_non_trans_table: %s",
                       YESNO(all),
                       YESNO(thd->transaction.all.modified_non_trans_table),
                       YESNO(thd->transaction.stmt.modified_non_trans_table)));
1567 1568 1569 1570 1571
  if ((all && thd->transaction.all.modified_non_trans_table) ||
      (!all && thd->transaction.stmt.modified_non_trans_table &&
       !mysql_bin_log.check_write_error(thd)) ||
      ((thd->options & OPTION_KEEP_LOG) &&
        !mysql_bin_log.check_write_error(thd)))
1572
  {
1573 1574 1575 1576 1577 1578 1579 1580
    /*
      We write the transaction cache with a rollback last if we have
      modified any non-transactional table. We do this even if we are
      committing a single statement that has modified a
      non-transactional table since it can have modified a
      transactional table in that statement as well, which needs to be
      rolled back on the slave.
    */
He Zhenxing's avatar
He Zhenxing committed
1581
    Query_log_event qev(thd, STRING_WITH_LEN("ROLLBACK"), TRUE, TRUE, 0);
1582
    error= binlog_end_trans(thd, trx_data, &qev, all);
1583
  }
1584
  else
1585 1586
  {
    /*
1587 1588 1589 1590 1591 1592 1593 1594
      We reach this point if either only transactional tables were modified or
      the effect of a statement that did not get into the binlog needs to be
      rolled back. In the latter case, if a statement changed non-transactional
      tables or had the OPTION_KEEP_LOG associated, we write an incident event
      to the binlog in order to stop slaves and notify users that some changes
      on the master did not get into the binlog and slaves will be inconsistent.
      On the other hand, if a statement is transactional, we just safely roll it
      back.
1595
     */
1596 1597 1598 1599
    if ((thd->transaction.stmt.modified_non_trans_table ||
        (thd->options & OPTION_KEEP_LOG)) &&
        mysql_bin_log.check_write_error(thd))
      trx_data->set_incident();
1600
    error= binlog_end_trans(thd, trx_data, 0, all);
1601
  }
1602 1603
  if (!all)
    trx_data->before_stmt_pos = MY_OFF_T_UNDEF; // part of the stmt rollback
1604 1605 1606
  DBUG_RETURN(error);
}

1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 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 1643 1644
void MYSQL_BIN_LOG::set_write_error(THD *thd)
{
  DBUG_ENTER("MYSQL_BIN_LOG::set_write_error");

  write_error= 1;

  if (check_write_error(thd))
    DBUG_VOID_RETURN;

  if (my_errno == EFBIG)
    my_message(ER_TRANS_CACHE_FULL, ER(ER_TRANS_CACHE_FULL), MYF(MY_WME));
  else
    my_error(ER_ERROR_ON_WRITE, MYF(MY_WME), name, errno);

  DBUG_VOID_RETURN;
}

bool MYSQL_BIN_LOG::check_write_error(THD *thd)
{
  DBUG_ENTER("MYSQL_BIN_LOG::check_write_error");

  bool checked= FALSE;

  if (!thd->is_error())
    DBUG_RETURN(checked);

  switch (thd->main_da.sql_errno())
  {
    case ER_TRANS_CACHE_FULL:
    case ER_ERROR_ON_WRITE:
    case ER_BINLOG_LOGGING_IMPOSSIBLE:
      checked= TRUE;
    break;
  }

  DBUG_RETURN(checked);
}

1645 1646 1647 1648 1649 1650
/**
  @note
  How do we handle this (unlikely but legal) case:
  @verbatim
    [transaction] + [update to non-trans table] + [rollback to savepoint] ?
  @endverbatim
1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668
  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).
*/

1669
static int binlog_savepoint_set(handlerton *hton, THD *thd, void *sv)
1670 1671 1672
{
  DBUG_ENTER("binlog_savepoint_set");

1673
  binlog_trans_log_savepos(thd, (my_off_t*) sv);
1674
  /* Write it to the binary log */
1675 1676

  int errcode= query_error_code(thd, thd->killed == THD::NOT_KILLED);
1677 1678
  int const error=
    thd->binlog_query(THD::STMT_QUERY_TYPE,
1679
                      thd->query, thd->query_length, TRUE, FALSE, errcode);
1680
  DBUG_RETURN(error);
1681 1682
}

1683
static int binlog_savepoint_rollback(handlerton *hton, THD *thd, void *sv)
1684 1685 1686 1687 1688 1689 1690 1691
{
  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.
  */
1692 1693
  if (unlikely(thd->transaction.all.modified_non_trans_table || 
               (thd->options & OPTION_KEEP_LOG)))
1694
  {
1695
    int errcode= query_error_code(thd, thd->killed == THD::NOT_KILLED);
1696
    int error=
1697
      thd->binlog_query(THD::STMT_QUERY_TYPE,
1698
                        thd->query, thd->query_length, TRUE, FALSE, errcode);
1699
    DBUG_RETURN(error);
1700
  }
1701
  binlog_trans_log_truncate(thd, *(my_off_t*)sv);
1702 1703 1704
  DBUG_RETURN(0);
}

1705

1706 1707 1708 1709 1710
int check_binlog_magic(IO_CACHE* log, const char** errmsg)
{
  char magic[4];
  DBUG_ASSERT(my_b_tell(log) == 0);

1711
  if (my_b_read(log, (uchar*) magic, sizeof(magic)))
1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725
  {
    *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;
}

1726

1727 1728 1729 1730 1731
File open_binlog(IO_CACHE *log, const char *log_file_name, const char **errmsg)
{
  File file;
  DBUG_ENTER("open_binlog");

1732 1733
  if ((file = my_open(log_file_name, O_RDONLY | O_BINARY | O_SHARE, 
                      MYF(MY_WME))) < 0)
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
  {
    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
1760

rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1761 1762
#ifdef __NT__
static int eventSource = 0;
serg@serg.mylan's avatar
serg@serg.mylan committed
1763

1764
static void setup_windows_event_source()
monty@mysql.com's avatar
monty@mysql.com committed
1765
{
1766
  HKEY    hRegKey= NULL;
monty@mysql.com's avatar
monty@mysql.com committed
1767 1768 1769
  DWORD   dwError= 0;
  TCHAR   szPath[MAX_PATH];
  DWORD dwTypes;
1770

monty@mysql.com's avatar
monty@mysql.com committed
1771 1772 1773
  if (eventSource)               // Ensure that we are only called once
    return;
  eventSource= 1;
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1774

monty@mysql.com's avatar
monty@mysql.com committed
1775
  // Create the event source registry key
1776
  dwError= RegCreateKey(HKEY_LOCAL_MACHINE,
monty@mysql.com's avatar
monty@mysql.com committed
1777 1778
                          "SYSTEM\\CurrentControlSet\\Services\\EventLog\\Application\\MySQL", 
                          &hRegKey);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1779

monty@mysql.com's avatar
monty@mysql.com committed
1780 1781
  /* Name of the PE module that contains the message resource */
  GetModuleFileName(NULL, szPath, MAX_PATH);
serg@serg.mylan's avatar
serg@serg.mylan committed
1782

rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1783
  /* Register EventMessageFile */
1784
  dwError = RegSetValueEx(hRegKey, "EventMessageFile", 0, REG_EXPAND_SZ,
1785
                          (PBYTE) szPath, (DWORD) (strlen(szPath) + 1));
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1786

monty@mysql.com's avatar
monty@mysql.com committed
1787 1788 1789 1790 1791
  /* 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
1792

monty@mysql.com's avatar
monty@mysql.com committed
1793
  RegCloseKey(hRegKey);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1794 1795
}

monty@mysql.com's avatar
monty@mysql.com committed
1796
#endif /* __NT__ */
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1797 1798


1799 1800 1801 1802 1803 1804 1805 1806
/**
  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
1807 1808 1809

static int find_uniq_filename(char *name)
{
1810 1811 1812 1813
  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
1814
  reg1 struct fileinfo *file_info;
1815
  ulong                 max_found=0;
1816 1817
  size_t		buf_length, length;
  char			*start, *end;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1818 1819
  DBUG_ENTER("find_uniq_filename");

1820 1821 1822
  length= dirname_part(buff, name, &buf_length);
  start=  name + length;
  end=    strend(start);
1823

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1824
  *end='.';
1825
  length= (size_t) (end-start+1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1826 1827 1828 1829 1830 1831 1832 1833 1834

  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++)
  {
1835
    if (bcmp((uchar*) file_info->name, (uchar*) start, length) == 0 &&
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1836 1837 1838 1839 1840 1841 1842 1843
	test_if_number(file_info->name+length, &number,0))
    {
      set_if_bigger(max_found,(ulong) number);
    }
  }
  my_dirend(dir_info);

  *end++='.';
1844
  sprintf(end,"%06ld",max_found+1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1845 1846 1847
  DBUG_RETURN(0);
}

1848

1849
void MYSQL_LOG::init(enum_log_type log_type_arg,
1850
                     enum cache_type io_cache_type_arg)
1851 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 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919
{
  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;

1920
  if (log_type == LOG_NORMAL)
1921 1922
  {
    char *end;
1923
    int len=my_snprintf(buff, sizeof(buff), "%s, Version: %s (%s). "
1924
#ifdef EMBEDDED_LIBRARY
1925 1926
                        "embedded library\n",
                        my_progname, server_version, MYSQL_COMPILATION_COMMENT
1927 1928
#elif __NT__
			"started with:\nTCP Port: %d, Named Pipe: %s\n",
1929 1930
                        my_progname, server_version, MYSQL_COMPILATION_COMMENT,
                        mysqld_port, mysqld_unix_port
1931 1932
#else
			"started with:\nTcp port: %d  Unix socket: %s\n",
1933 1934
                        my_progname, server_version, MYSQL_COMPILATION_COMMENT,
                        mysqld_port, mysqld_unix_port
1935 1936 1937 1938
#endif
                       );
    end= strnmov(buff + len, "Time                 Id Command    Argument\n",
                 sizeof(buff) - len);
1939
    if (my_b_write(&log_file, (uchar*) buff, (uint) (end-buff)) ||
1940 1941 1942 1943
	flush_io_cache(&log_file))
      goto err;
  }

1944
  log_state= LOG_OPENED;
1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955
  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);
1956
  log_state= LOG_CLOSED;
1957 1958 1959
  DBUG_RETURN(1);
}

1960
MYSQL_LOG::MYSQL_LOG()
serg@janus.mylan's avatar
serg@janus.mylan committed
1961 1962
  : 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
1963 1964
{
  /*
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1965 1966
    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
1967
    called only in main(). Doing initialization here would make it happen
1968
    before main().
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1969
  */
1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997
  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));
1998
  if (log_state == LOG_OPENED)
1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
  {
    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);
    }
  }

2015
  log_state= (exiting & LOG_CLOSE_TO_BE_OPENED) ? LOG_TO_BE_OPENED : LOG_CLOSED;
2016 2017
  safeFree(name);
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2018 2019
}

2020
/** This is called only once. */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2021

monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
2022
void MYSQL_LOG::cleanup()
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2023
{
2024
  DBUG_ENTER("cleanup");
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2025
  if (inited)
2026
  {
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
2027
    inited= 0;
2028
    (void) pthread_mutex_destroy(&LOCK_log);
2029
    close(0);
2030
  }
2031
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2032 2033
}

2034

bk@work.mysql.com's avatar
bk@work.mysql.com committed
2035
int MYSQL_LOG::generate_new_name(char *new_name, const char *log_name)
2036
{
2037 2038
  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
2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051
  {
    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;
}

2052

2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064
/*
  Reopen the log file

  SYNOPSIS
    reopen_file()

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


2065
void MYSQL_QUERY_LOG::reopen_file()
2066
{
2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082
  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);

  /*
2083
     Note that at this point, log_state != LOG_CLOSED (important for is_open()).
2084 2085
  */

2086
  open(save_name, log_type, 0, io_cache_type);
2087 2088 2089 2090
  my_free(save_name, MYF(0));

  pthread_mutex_unlock(&LOCK_log);

2091
  DBUG_VOID_RETURN;
2092 2093
}

2094

2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119
/*
  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
*/

2120 2121 2122 2123
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
2124
{
2125 2126
  char buff[32];
  uint length= 0;
2127
  char local_time_buff[MAX_TIME_SIZE];
2128 2129 2130
  struct tm start;
  uint time_buff_len= 0;

2131 2132
  (void) pthread_mutex_lock(&LOCK_log);

2133 2134 2135
  /* Test if someone closed between the is_open test and lock */
  if (is_open())
  {
2136 2137 2138
    /* for testing output of timestamp and thread id */
    DBUG_EXECUTE_IF("reset_log_last_time", last_time= 0;);

2139 2140 2141 2142 2143 2144 2145
    /* 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);

2146
        time_buff_len= my_snprintf(local_time_buff, MAX_TIME_SIZE,
2147
                                   "%02d%02d%02d %2d:%02d:%02d\t",
2148 2149 2150 2151
                                   start.tm_year % 100, start.tm_mon + 1,
                                   start.tm_mday, start.tm_hour,
                                   start.tm_min, start.tm_sec);

2152
        if (my_b_write(&log_file, (uchar*) local_time_buff, time_buff_len))
2153 2154 2155
          goto err;
      }
      else
2156
        if (my_b_write(&log_file, (uchar*) "\t\t" ,2) < 0)
2157 2158
          goto err;

2159
      /* command_type, thread_id */
2160
      length= my_snprintf(buff, 32, "%5ld ", (long) thread_id);
2161

2162
    if (my_b_write(&log_file, (uchar*) buff, length))
2163 2164
      goto err;

2165
    if (my_b_write(&log_file, (uchar*) command_type, command_type_len))
2166 2167
      goto err;

2168
    if (my_b_write(&log_file, (uchar*) "\t", 1))
2169 2170 2171
      goto err;

    /* sql_text */
2172
    if (my_b_write(&log_file, (uchar*) sql_text, sql_text_len))
2173 2174
      goto err;

2175
    if (my_b_write(&log_file, (uchar*) "\n", 1) ||
2176 2177 2178 2179
        flush_io_cache(&log_file))
      goto err;
  }

2180
  (void) pthread_mutex_unlock(&LOCK_log);
2181 2182 2183 2184 2185 2186 2187 2188
  return FALSE;
err:

  if (!write_error)
  {
    write_error= 1;
    sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
  }
2189
  (void) pthread_mutex_unlock(&LOCK_log);
2190
  return TRUE;
2191 2192
}

2193

2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205
/*
  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
2206 2207
    query_utime       Amount of time the query took to execute (in microseconds)
    lock_utime        Amount of time the query was locked (in microseconds)
2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222
    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
*/

2223 2224
bool MYSQL_QUERY_LOG::write(THD *thd, time_t current_time,
                            time_t query_start_arg, const char *user_host,
2225 2226
                            uint user_host_len, ulonglong query_utime,
                            ulonglong lock_utime, bool is_command,
2227
                            const char *sql_text, uint sql_text_len)
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2228
{
2229
  bool error= 0;
2230
  DBUG_ENTER("MYSQL_QUERY_LOG::write");
2231

2232 2233
  (void) pthread_mutex_lock(&LOCK_log);

2234
  if (!is_open())
2235 2236
  {
    (void) pthread_mutex_unlock(&LOCK_log);
2237
    DBUG_RETURN(0);
2238
  }
2239 2240 2241 2242 2243

  if (is_open())
  {						// Safety agains reopen
    int tmp_errno= 0;
    char buff[80], *end;
2244
    char query_time_buff[22+7], lock_time_buff[22+7];
2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262
    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 */
2263
        if (my_b_write(&log_file, (uchar*) buff, buff_len))
2264 2265
          tmp_errno= errno;
      }
2266 2267
      const uchar uh[]= "# User@Host: ";
      if (my_b_write(&log_file, uh, sizeof(uh) - 1))
2268
        tmp_errno= errno;
2269
      if (my_b_write(&log_file, (uchar*) user_host, user_host_len))
2270
        tmp_errno= errno;
2271
      if (my_b_write(&log_file, (uchar*) "\n", 1))
2272 2273 2274
        tmp_errno= errno;
    }
    /* For slow query log */
2275 2276
    sprintf(query_time_buff, "%.6f", ulonglong2double(query_utime)/1000000.0);
    sprintf(lock_time_buff,  "%.6f", ulonglong2double(lock_utime)/1000000.0);
2277
    if (my_b_printf(&log_file,
2278
                    "# Query_time: %s  Lock_time: %s"
2279
                    " Rows_sent: %lu  Rows_examined: %lu\n",
2280
                    query_time_buff, lock_time_buff,
2281 2282 2283 2284 2285 2286 2287 2288 2289
                    (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);
    }
2290
    if (thd->stmt_depends_on_first_successful_insert_id_in_prev_stmt)
2291 2292
    {
      end=strmov(end, ",last_insert_id=");
2293 2294 2295
      end=longlong10_to_str((longlong)
                            thd->first_successful_insert_id_in_prev_stmt_for_binlog,
                            end, -10);
2296 2297
    }
    // Save value if we do an insert.
2298
    if (thd->auto_inc_intervals_in_cur_stmt_for_binlog.nb_elements() > 0)
2299 2300 2301 2302
    {
      if (!(specialflag & SPECIAL_SHORT_LOG_FORMAT))
      {
        end=strmov(end,",insert_id=");
2303 2304 2305
        end=longlong10_to_str((longlong)
                              thd->auto_inc_intervals_in_cur_stmt_for_binlog.minimum(),
                              end, -10);
2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320
      }
    }

    /*
      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';
2321 2322
      if (my_b_write(&log_file, (uchar*) "SET ", 4) ||
          my_b_write(&log_file, (uchar*) buff + 1, (uint) (end-buff)))
2323 2324 2325 2326 2327 2328
        tmp_errno= errno;
    }
    if (is_command)
    {
      end= strxmov(buff, "# administrator command: ", NullS);
      buff_len= (ulong) (end - buff);
2329
      my_b_write(&log_file, (uchar*) buff, buff_len);
2330
    }
2331 2332
    if (my_b_write(&log_file, (uchar*) sql_text, sql_text_len) ||
        my_b_write(&log_file, (uchar*) ";\n",2) ||
2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344
        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);
      }
    }
  }
2345
  (void) pthread_mutex_unlock(&LOCK_log);
2346
  DBUG_RETURN(error);
2347 2348
}

2349

2350 2351 2352 2353 2354
/**
  @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.
*/
2355
const char *MYSQL_LOG::generate_name(const char *log_name,
2356 2357
                                      const char *suffix,
                                      bool strip_ext, char *buff)
2358 2359 2360
{
  if (!log_name || !log_name[0])
  {
2361 2362 2363
    strmake(buff, pidfile_name, FN_REFLEN - strlen(suffix) - 1);
    return (const char *)
      fn_format(buff, buff, "", suffix, MYF(MY_REPLACE_EXT|MY_REPLACE_DIR));
2364 2365 2366 2367
  }
  // get rid of extension if the log is binary to avoid problems
  if (strip_ext)
  {
2368 2369 2370
    char *p= fn_ext(log_name);
    uint length= (uint) (p - log_name);
    strmake(buff, log_name, min(length, FN_REFLEN));
2371 2372 2373 2374 2375
    return (const char*)buff;
  }
  return log_name;
}

2376

2377

2378 2379 2380
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),
2381
   is_relay_log(0),
2382 2383 2384 2385 2386 2387 2388 2389 2390 2391
   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));
2392
  bzero((char*) &purge_temp, sizeof(purge_temp));
2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435
}

/* 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,
2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460
                                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
2461
			0, MYF(MY_WME | MY_WAIT_IF_FULL)))
2462
  {
2463 2464 2465 2466 2467
    /*
      TODO: all operations creating/deleting the index file or a log, should
      call my_sync_dir() or my_sync_dir_by_file() to be durable.
      TODO: file creation should be done with my_create() not my_open().
    */
2468 2469 2470 2471 2472 2473 2474
    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
2475

2476
/**
2477
  Open a (new) binlog file.
2478

2479
  - Open the log file and the index file. Register the new
2480
  file name in it
2481
  - When calling this when the file is in use, you must have a locks
2482
  on LOCK_log and LOCK_index.
2483

2484
  @retval
2485
    0	ok
2486
  @retval
2487 2488 2489
    1	error
*/

2490 2491 2492 2493 2494 2495 2496
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
2497
{
2498
  File file= -1;
2499
  DBUG_ENTER("MYSQL_BIN_LOG::open");
2500
  DBUG_PRINT("enter",("log_type: %d",(int) log_type_arg));
2501 2502

  write_error=0;
2503

2504 2505 2506
  /* 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 */
2507

2508
  init(no_auto_events_arg, max_size_arg);
2509

2510
  open_count++;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2511

2512
  DBUG_ASSERT(log_type == LOG_BIN);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2513 2514

  {
2515 2516 2517 2518 2519 2520 2521 2522 2523 2524
    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.
      */
2525
      if (my_b_safe_write(&log_file, (uchar*) BINLOG_MAGIC,
2526
			  BIN_LOG_HEADER_SIZE))
2527
        goto err;
2528 2529
      bytes_written+= BIN_LOG_HEADER_SIZE;
      write_file_name_to_index_file= 1;
2530
    }
2531

2532
    DBUG_ASSERT(my_b_inited(&index_file) != 0);
2533 2534
    reinit_io_cache(&index_file, WRITE_CACHE,
                    my_b_filelength(&index_file), 0, 0);
2535
    if (need_start_event && !no_auto_events)
2536
    {
2537 2538 2539 2540 2541
      /*
        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
2542 2543 2544 2545 2546 2547
      /*
        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;
2548 2549
      if (!s.is_valid())
        goto err;
2550
      s.dont_set_created= null_created_arg;
2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565
      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
2566
        Format_description_log_event::Format_description_log_event(char* buf,).
2567 2568 2569 2570 2571 2572 2573 2574 2575
        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
2576
        Format_description_log_event::apply_event_impl().
2577 2578 2579
      */
      description_event_for_queue->created= 0;
      /* Don't set log_pos in event header */
2580
      description_event_for_queue->set_artificial_event();
2581

2582 2583 2584
      if (description_event_for_queue->write(&log_file))
        goto err;
      bytes_written+= description_event_for_queue->data_written;
2585
    }
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2586 2587
    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
2588
      goto err;
2589 2590 2591

    if (write_file_name_to_index_file)
    {
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2592 2593 2594 2595
      /*
        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.
      */
2596
      if (my_b_write(&index_file, (uchar*) log_file_name,
2597
		     strlen(log_file_name)) ||
2598
	  my_b_write(&index_file, (uchar*) "\n", 1) ||
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2599 2600
	  flush_io_cache(&index_file) ||
          my_sync(index_file.file, MYF(MY_WME)))
2601
	goto err;
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
2602
    }
2603
  }
2604 2605
  log_state= LOG_OPENED;

2606
  DBUG_RETURN(0);
2607 2608

err:
2609 2610 2611
  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, \
2612
shutdown the MySQL server and restart it.", name, errno);
2613 2614 2615
  if (file >= 0)
    my_close(file,MYF(0));
  end_io_cache(&log_file);
2616 2617
  end_io_cache(&index_file);
  safeFree(name);
2618
  log_state= LOG_CLOSED;
2619
  DBUG_RETURN(1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2620 2621
}

2622

2623
int MYSQL_BIN_LOG::get_current_log(LOG_INFO* linfo)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2624 2625
{
  pthread_mutex_lock(&LOCK_log);
2626 2627 2628 2629 2630
  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
2631
int MYSQL_BIN_LOG::raw_get_current_log(LOG_INFO* linfo)
2632
{
2633 2634
  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
2635 2636 2637
  return 0;
}

2638 2639
/**
  Move all data up in a file in an filename index file.
2640 2641 2642 2643 2644

    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.

2645 2646 2647 2648 2649 2650 2651
  @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
2652 2653 2654
    0	ok
*/

2655 2656
#ifdef HAVE_REPLICATION

2657 2658 2659 2660 2661
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;
2662
  uchar io_buf[IO_SIZE*2];
2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673
  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));
2674
    if (my_write(file, io_buf, bytes_read, MYF(MY_WME | MY_NABP)))
2675 2676 2677
      goto err;
  }
  /* The following will either truncate the file or fill the end with \n' */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2678 2679
  if (my_chsize(file, offset - init_offset, '\n', MYF(MY_WME)) ||
      my_sync(file, MYF(MY_WME)))
2680 2681 2682 2683 2684 2685 2686 2687 2688 2689
    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);
}

2690
#endif /* HAVE_REPLICATION */
2691

2692 2693
/**
  Find the position in the log-index-file for the given log name.
2694

2695 2696 2697 2698 2699 2700
  @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
2701

2702
  @note
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
2703 2704
    On systems without the truncate function the file will end with one or
    more empty lines.  These will be ignored when reading the file.
2705

2706
  @retval
2707
    0			ok
2708 2709 2710
  @retval
    LOG_INFO_EOF	        End of log-index-file found
  @retval
2711 2712 2713
    LOG_INFO_IO		Got IO error while reading file
*/

2714
int MYSQL_BIN_LOG::find_log_pos(LOG_INFO *linfo, const char *log_name,
2715
			    bool need_lock)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2716
{
2717 2718 2719 2720 2721
  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
2722

2723
  /*
2724 2725
    Mutex needed because we need to make sure the file pointer does not
    move from under our feet
2726
  */
2727
  if (need_lock)
2728
    pthread_mutex_lock(&LOCK_index);
2729 2730 2731 2732 2733
  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);

2734
  for (;;)
2735
  {
2736
    uint length;
2737 2738 2739 2740
    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
2741
    {
2742 2743 2744
      /* 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
2745 2746
    }

2747 2748
    // if the log entry matches, null string matching anything
    if (!log_name ||
sasha@mysql.sashanet.com's avatar
sasha@mysql.sashanet.com committed
2749
	(log_name_len == length-1 && fname[log_name_len] == '\n' &&
2750
	 !memcmp(fname, log_name, log_name_len)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2751
    {
2752
      DBUG_PRINT("info",("Found log file entry"));
2753
      fname[length-1]=0;			// remove last \n
2754 2755
      linfo->index_file_start_offset= offset;
      linfo->index_file_offset = my_b_tell(&index_file);
2756
      break;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2757
    }
2758
  }
2759

2760
  if (need_lock)
2761
    pthread_mutex_unlock(&LOCK_index);
2762
  DBUG_RETURN(error);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2763
}
2764

2765

2766 2767
/**
  Find the position in the log-index-file for the given log name.
2768

2769
  @param
2770 2771
    linfo		Store here the next log file name and position to
			the file name after that.
2772
  @param
2773 2774 2775
    need_lock		Set this to 1 if the parent doesn't already have a
			lock on LOCK_index

2776
  @note
2777
    - Before calling this function, one has to call find_log_pos()
2778
    to set up 'linfo'
2779
    - Mutex needed because we need to make sure the file pointer does not move
2780
    from under our feet
2781

2782
  @retval
2783
    0			ok
2784 2785 2786
  @retval
    LOG_INFO_EOF	        End of log-index-file found
  @retval
2787 2788 2789
    LOG_INFO_IO		Got IO error while reading file
*/

2790
int MYSQL_BIN_LOG::find_next_log(LOG_INFO* linfo, bool need_lock)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2791
{
2792
  int error= 0;
2793
  uint length;
2794 2795
  char *fname= linfo->log_file_name;

2796 2797
  if (need_lock)
    pthread_mutex_lock(&LOCK_index);
2798 2799 2800 2801 2802 2803 2804 2805
  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)
2806
  {
2807
    error = !index_file.error ? LOG_INFO_EOF : LOG_INFO_IO;
2808 2809
    goto err;
  }
2810
  fname[length-1]=0;				// kill \n
2811
  linfo->index_file_offset = my_b_tell(&index_file);
2812

bk@work.mysql.com's avatar
bk@work.mysql.com committed
2813
err:
2814 2815
  if (need_lock)
    pthread_mutex_unlock(&LOCK_index);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2816 2817 2818
  return error;
}

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2819

2820 2821 2822
/**
  Delete all logs refered to in the index file.
  Start writing to a new log file.
2823

2824
  The new index file will only contain this file.
2825

2826
  @param thd		Thread
2827

2828 2829
  @note
    If not called from slave thread, write start event to new log
2830

2831
  @retval
2832
    0	ok
2833
  @retval
2834 2835 2836
    1   error
*/

2837
bool MYSQL_BIN_LOG::reset_logs(THD* thd)
2838 2839
{
  LOG_INFO linfo;
2840
  bool error=0;
2841
  const char* save_name;
2842
  DBUG_ENTER("reset_logs");
2843

tomas@poseidon.ndb.mysql.com's avatar
tomas@poseidon.ndb.mysql.com committed
2844
  ha_reset_logs(thd);
2845 2846 2847 2848
  /*
    We need to get both locks to be sure that no one is trying to
    write to the index log file.
  */
2849
  pthread_mutex_lock(&LOCK_log);
2850
  pthread_mutex_lock(&LOCK_index);
2851

2852 2853 2854 2855 2856 2857
  /*
    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.
  */
2858
  VOID(pthread_mutex_lock(&LOCK_thread_count));
2859 2860 2861 2862

  /* Save variables so that we can reopen the log */
  save_name=name;
  name=0;					// Protect against free
2863
  close(LOG_CLOSE_TO_BE_OPENED);
2864 2865 2866 2867

  /* First delete all old log files */

  if (find_log_pos(&linfo, NullS, 0))
2868 2869 2870 2871
  {
    error=1;
    goto err;
  }
2872

2873
  for (;;)
2874
  {
2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899
    if ((error= my_delete_allow_opened(linfo.log_file_name, MYF(0))) != 0)
    {
      if (my_errno == ENOENT) 
      {
        push_warning_printf(current_thd, MYSQL_ERROR::WARN_LEVEL_WARN,
                            ER_LOG_PURGE_NO_FILE, ER(ER_LOG_PURGE_NO_FILE),
                            linfo.log_file_name);
        sql_print_information("Failed to delete file '%s'",
                              linfo.log_file_name);
        my_errno= 0;
        error= 0;
      }
      else
      {
        push_warning_printf(current_thd, MYSQL_ERROR::WARN_LEVEL_ERROR,
                            ER_BINLOG_PURGE_FATAL_ERR,
                            "a problem with deleting %s; "
                            "consider examining correspondence "
                            "of your binlog index file "
                            "to the actual binlog files",
                            linfo.log_file_name);
        error= 1;
        goto err;
      }
    }
2900
    if (find_next_log(&linfo, 0))
2901 2902
      break;
  }
2903 2904

  /* Start logging with a new file */
2905
  close(LOG_CLOSE_INDEX);
2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930
  if ((error= my_delete_allow_opened(index_file_name, MYF(0))))	// Reset (open will update)
  {
    if (my_errno == ENOENT) 
    {
      push_warning_printf(current_thd, MYSQL_ERROR::WARN_LEVEL_WARN,
                          ER_LOG_PURGE_NO_FILE, ER(ER_LOG_PURGE_NO_FILE),
                          index_file_name);
      sql_print_information("Failed to delete file '%s'",
                            index_file_name);
      my_errno= 0;
      error= 0;
    }
    else
    {
      push_warning_printf(current_thd, MYSQL_ERROR::WARN_LEVEL_ERROR,
                          ER_BINLOG_PURGE_FATAL_ERR,
                          "a problem with deleting %s; "
                          "consider examining correspondence "
                          "of your binlog index file "
                          "to the actual binlog files",
                          index_file_name);
      error= 1;
      goto err;
    }
  }
2931
  if (!thd->slave_thread)
2932
    need_start_event=1;
2933
  if (!open_index_file(index_file_name, 0))
2934
    open(save_name, log_type, 0, io_cache_type, no_auto_events, max_size, 0);
2935
  my_free((uchar*) save_name, MYF(0));
2936

2937
err:
2938
  VOID(pthread_mutex_unlock(&LOCK_thread_count));
2939
  pthread_mutex_unlock(&LOCK_index);
2940
  pthread_mutex_unlock(&LOCK_log);
2941
  DBUG_RETURN(error);
2942 2943
}

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2944

2945
/**
2946 2947
  Delete relay log files prior to rli->group_relay_log_name
  (i.e. all logs which are not involved in a non-finished group
2948 2949
  (transaction)), remove them from the index file and start on next
  relay log.
2950

2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964
  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
2965
    - This is only called from the slave-execute thread when it has read
2966
    all commands from a relay log and want to switch to a new relay log.
2967
    - When this happens, we can be in an active transaction as
2968 2969 2970
    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).
2971

2972
  @retval
2973
    0			ok
2974 2975 2976
  @retval
    LOG_INFO_EOF	        End of log-index-file found
  @retval
2977
    LOG_INFO_SEEK	Could not allocate IO cache
2978
  @retval
2979
    LOG_INFO_IO		Got IO error while reading file
2980 2981
*/

hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
2982
#ifdef HAVE_REPLICATION
2983

2984
int MYSQL_BIN_LOG::purge_first_log(Relay_log_info* rli, bool included)
2985
{
2986
  int error;
2987
  char *to_purge_if_included= NULL;
2988 2989
  DBUG_ENTER("purge_first_log");

2990 2991
  DBUG_ASSERT(is_open());
  DBUG_ASSERT(rli->slave_running == 1);
2992
  DBUG_ASSERT(!strcmp(rli->linfo.log_file_name,rli->event_relay_log_name));
2993

2994
  pthread_mutex_lock(&LOCK_index);
2995
  to_purge_if_included= my_strdup(rli->group_relay_log_name, MYF(0));
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2996

2997 2998
  /*
    Read the next log file name from the index file and pass it back to
2999
    the caller.
3000
  */
3001 3002
  if((error=find_log_pos(&rli->linfo, rli->event_relay_log_name, 0)) || 
     (error=find_next_log(&rli->linfo, 0)))
3003 3004
  {
    char buff[22];
3005 3006 3007
    sql_print_error("next log error: %d  offset: %s  log: %s included: %d",
                    error,
                    llstr(rli->linfo.index_file_offset,buff),
3008
                    rli->event_relay_log_name,
3009
                    included);
3010 3011
    goto err;
  }
3012

3013
  /*
3014
    Reset rli's coordinates to the current log.
3015
  */
3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029
  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);
3030
    rli->notify_group_relay_log_name_update();
3031
  }
3032 3033 3034

  /* 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
3035

3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055
  DBUG_EXECUTE_IF("crash_before_purge_logs", abort(););

  pthread_mutex_lock(&rli->log_space_lock);
  rli->relay_log.purge_logs(to_purge_if_included, 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);

  /*
    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
  */
  pthread_cond_broadcast(&rli->log_space_cond);

  /*
   * Need to update the log pos because purge logs has been called 
   * after fetching initially the log pos at the begining of the method.
   */
3056
  if((error=find_log_pos(&rli->linfo, rli->event_relay_log_name, 0)))
3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069
  {
    char buff[22];
    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);
    goto err;
  }

  /* If included was passed, rli->linfo should be the first entry. */
  DBUG_ASSERT(!included || rli->linfo.index_file_start_offset == 0);

3070
err:
3071
  my_free(to_purge_if_included, MYF(0));
3072
  pthread_mutex_unlock(&LOCK_index);
3073
  DBUG_RETURN(error);
3074 3075
}

3076 3077
/**
  Update log index_file.
3078 3079
*/

3080
int MYSQL_BIN_LOG::update_log_index(LOG_INFO* log_info, bool need_update_threads)
3081 3082 3083 3084 3085
{
  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
3086 3087
  if (need_update_threads)
    adjust_linfo_offsets(log_info->index_file_start_offset);
3088 3089
  return 0;
}
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
3090

3091
/**
3092 3093
  Remove all logs before the given log from disk and from the index file.

3094 3095 3096 3097 3098 3099 3100
  @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
3101

3102
  @note
3103 3104 3105
    If any of the logs before the deleted one is in use,
    only purge logs up to this one.

3106 3107 3108
  @retval
    0			ok
  @retval
3109
    LOG_INFO_EOF		to_log not found
3110
    LOG_INFO_EMFILE             too many files opened
3111 3112
    LOG_INFO_FATAL              if any other than ENOENT error from
                                my_stat() or my_delete()
3113 3114
*/

3115
int MYSQL_BIN_LOG::purge_logs(const char *to_log, 
3116 3117 3118 3119
                          bool included,
                          bool need_mutex, 
                          bool need_update_threads, 
                          ulonglong *decrease_log_space)
3120 3121
{
  int error;
3122
  bool exit_loop= 0;
3123
  LOG_INFO log_info;
3124
  THD *thd= current_thd;
3125
  DBUG_ENTER("purge_logs");
3126
  DBUG_PRINT("info",("to_log= %s",to_log));
3127

3128 3129
  if (need_mutex)
    pthread_mutex_lock(&LOCK_index);
3130 3131 3132 3133
  if ((error=find_log_pos(&log_info, to_log, 0 /*no mutex*/))) 
  {
    sql_print_error("MYSQL_LOG::purge_logs was called with file %s not "
                    "listed in the index.", to_log);
3134
    goto err;
3135 3136 3137 3138 3139 3140 3141 3142 3143
  }

  /*
    For crash recovery reasons the index needs to be updated before
    any files are deleted. Move files to be deleted into a temp file
    to be processed after the index is updated.
  */
  if (!my_b_inited(&purge_temp))
  {
3144 3145
    if ((error=open_cached_file(&purge_temp, mysql_tmpdir, TEMP_PREFIX,
                                DISK_BUFFER_SIZE, MYF(MY_WME))))
3146 3147 3148 3149 3150 3151 3152
    {
      sql_print_error("MYSQL_LOG::purge_logs failed to open purge_temp");
      goto err;
    }
  }
  else
  {
3153
    if ((error=reinit_io_cache(&purge_temp, WRITE_CACHE, 0, 0, 1)))
3154 3155 3156 3157 3158 3159
    {
      sql_print_error("MYSQL_LOG::purge_logs failed to reinit purge_temp "
                      "for write");
      goto err;
    }
  }
3160

3161
  /*
3162
    File name exists in index file; delete until we find this file
3163 3164 3165
    or a file that is used.
  */
  if ((error=find_log_pos(&log_info, NullS, 0 /*no mutex*/)))
3166
    goto err;
3167 3168
  while ((strcmp(to_log,log_info.log_file_name) || (exit_loop=included)) &&
         !log_in_use(log_info.log_file_name))
3169
  {
3170
    if ((error=my_b_write(&purge_temp, (const uchar*)log_info.log_file_name,
3171
                          strlen(log_info.log_file_name))) ||
3172
        (error=my_b_write(&purge_temp, (const uchar*)"\n", 1)))
3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183
    {
      sql_print_error("MYSQL_LOG::purge_logs failed to copy %s to purge_temp",
                      log_info.log_file_name);
      goto err;
    }

    if (find_next_log(&log_info, 0) || exit_loop)
      break;
 }

  /* We know how many files to delete. Update index file. */
3184
  if ((error=update_log_index(&log_info, need_update_threads)))
3185 3186 3187 3188 3189 3190 3191 3192
  {
    sql_print_error("MSYQL_LOG::purge_logs failed to update the index file");
    goto err;
  }

  DBUG_EXECUTE_IF("crash_after_update_index", abort(););

  /* Switch purge_temp for read. */
3193
  if ((error=reinit_io_cache(&purge_temp, READ_CACHE, 0, 0, 0)))
3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221 3222
  {
    sql_print_error("MSYQL_LOG::purge_logs failed to reinit purge_temp "
                    "for read");
    goto err;
  }

  /* Read each entry from purge_temp and delete the file. */
  for (;;)
  {
    uint length;

    if ((length=my_b_gets(&purge_temp, log_info.log_file_name,
                          FN_REFLEN)) <= 1)
    {
      if (purge_temp.error)
      {
        error= purge_temp.error;
        sql_print_error("MSYQL_LOG::purge_logs error %d reading from "
                        "purge_temp", error);
        goto err;
      }

      /* Reached EOF */
      break;
    }

    /* Get rid of the trailing '\n' */
    log_info.log_file_name[length-1]= 0;

3223 3224
    ha_binlog_index_purge_file(current_thd, log_info.log_file_name);

3225 3226
    MY_STAT s;
    if (!my_stat(log_info.log_file_name, &s, MYF(0)))
3227
    {
3228 3229 3230 3231 3232
      if (my_errno == ENOENT) 
      {
        /*
          It's not fatal if we can't stat a log file that does not exist;
          If we could not stat, we won't delete.
3233
        */
3234 3235 3236 3237 3238 3239
        if (thd)
        {
          push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_WARN,
                              ER_LOG_PURGE_NO_FILE, ER(ER_LOG_PURGE_NO_FILE),
                              log_info.log_file_name);
        }
3240
        sql_print_information("Failed to execute my_stat on file '%s'",
3241
			      log_info.log_file_name);
3242 3243 3244 3245 3246 3247 3248
        my_errno= 0;
      }
      else
      {
        /*
          Other than ENOENT are fatal
        */
3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266
        if (thd)
        {
          push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_ERROR,
                              ER_BINLOG_PURGE_FATAL_ERR,
                              "a problem with getting info on being purged %s; "
                              "consider examining correspondence "
                              "of your binlog index file "
                              "to the actual binlog files",
                              log_info.log_file_name);
        }
        else
        {
          sql_print_information("Failed to delete log file '%s'; "
                                "consider examining correspondence "
                                "of your binlog index file "
                                "to the actual binlog files",
                                log_info.log_file_name);
        }
3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282
        error= LOG_INFO_FATAL;
        goto err;
      }
    }
    else
    {
      DBUG_PRINT("info",("purging %s",log_info.log_file_name));
      if (!my_delete(log_info.log_file_name, MYF(0)))
      {
        if (decrease_log_space)
          *decrease_log_space-= s.st_size;
      }
      else
      {
        if (my_errno == ENOENT) 
        {
3283 3284 3285 3286 3287 3288
          if (thd)
          {
            push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_WARN,
                                ER_LOG_PURGE_NO_FILE, ER(ER_LOG_PURGE_NO_FILE),
                                log_info.log_file_name);
          }
3289 3290 3291 3292 3293 3294
          sql_print_information("Failed to delete file '%s'",
                                log_info.log_file_name);
          my_errno= 0;
        }
        else
        {
3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312
          if (thd)
          {
            push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_ERROR,
                                ER_BINLOG_PURGE_FATAL_ERR,
                                "a problem with deleting %s; "
                                "consider examining correspondence "
                                "of your binlog index file "
                                "to the actual binlog files",
                                log_info.log_file_name);
          }
          else
          {
            sql_print_information("Failed to delete file '%s'; "
                                  "consider examining correspondence "
                                  "of your binlog index file "
                                  "to the actual binlog files",
                                  log_info.log_file_name);
          }
3313 3314 3315 3316 3317
          if (my_errno == EMFILE)
          {
            DBUG_PRINT("info",
                       ("my_errno: %d, set ret = LOG_INFO_EMFILE", my_errno));
            error= LOG_INFO_EMFILE;
Kristofer Pettersson's avatar
Kristofer Pettersson committed
3318
            goto err;
3319
          }
3320 3321 3322 3323
          error= LOG_INFO_FATAL;
          goto err;
        }
      }
3324
    }
3325
  }
3326

3327
err:
3328
  close_cached_file(&purge_temp);
3329 3330
  if (need_mutex)
    pthread_mutex_unlock(&LOCK_index);
3331
  DBUG_RETURN(error);
3332 3333
}

3334
/**
3335 3336 3337
  Remove all logs before the given file date from disk and from the
  index file.

3338
  @param thd		Thread pointer
3339
  @param purge_time	Delete all log files before given date.
3340

3341
  @note
3342 3343 3344
    If any of the logs before the deleted one is in use,
    only purge logs up to this one.

3345
  @retval
3346
    0				ok
3347
  @retval
3348
    LOG_INFO_PURGE_NO_ROTATE	Binary file that can't be rotated
3349 3350
    LOG_INFO_FATAL              if any other than ENOENT error from
                                my_stat() or my_delete()
3351 3352
*/

3353
int MYSQL_BIN_LOG::purge_logs_before_date(time_t purge_time)
3354 3355
{
  int error;
3356
  char to_log[FN_REFLEN];
3357 3358
  LOG_INFO log_info;
  MY_STAT stat_area;
3359
  THD *thd= current_thd;
3360
  
3361 3362 3363
  DBUG_ENTER("purge_logs_before_date");

  pthread_mutex_lock(&LOCK_index);
3364
  to_log[0]= 0;
3365 3366

  if ((error=find_log_pos(&log_info, NullS, 0 /*no mutex*/)))
3367
    goto err;
3368 3369 3370 3371

  while (strcmp(log_file_name, log_info.log_file_name) &&
	 !log_in_use(log_info.log_file_name))
  {
3372 3373 3374 3375 3376 3377
    if (!my_stat(log_info.log_file_name, &stat_area, MYF(0)))
    {
      if (my_errno == ENOENT) 
      {
        /*
          It's not fatal if we can't stat a log file that does not exist.
3378 3379 3380 3381 3382 3383 3384
        */
        if (thd)
        {
          push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_WARN,
                              ER_LOG_PURGE_NO_FILE, ER(ER_LOG_PURGE_NO_FILE),
                              log_info.log_file_name);
        }
3385 3386
        sql_print_information("Failed to execute my_stat on file '%s'",
                              log_info.log_file_name);
3387 3388 3389 3390 3391 3392 3393
        my_errno= 0;
      }
      else
      {
        /*
          Other than ENOENT are fatal
        */
3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408
        if (thd)
        {
          push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_ERROR,
                              ER_BINLOG_PURGE_FATAL_ERR,
                              "a problem with getting info on being purged %s; "
                              "consider examining correspondence "
                              "of your binlog index file "
                              "to the actual binlog files",
                              log_info.log_file_name);
        }
        else
        {
          sql_print_information("Failed to delete log file '%s'",
                                log_info.log_file_name);
        }
3409 3410 3411 3412 3413 3414
        error= LOG_INFO_FATAL;
        goto err;
      }
    }
    else
    {
3415 3416 3417 3418 3419
      if (stat_area.st_mtime < purge_time) 
        strmake(to_log, 
                log_info.log_file_name, 
                sizeof(log_info.log_file_name));
      else
3420 3421
        break;
    }
3422 3423
    if (find_next_log(&log_info, 0))
      break;
3424
  }
3425

3426
  error= (to_log[0] ? purge_logs(to_log, 1, 0, 1, (ulonglong *) 0) : 0);
3427

3428 3429
err:
  pthread_mutex_unlock(&LOCK_index);
3430
  DBUG_RETURN(error);
3431
}
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
3432
#endif /* HAVE_REPLICATION */
3433

3434

3435 3436
/**
  Create a new log file name.
3437

3438
  @param buf		buf of at least FN_REFLEN where new name is stored
3439

3440
  @note
3441 3442 3443
    If file name will be longer then FN_REFLEN it will be truncated
*/

3444
void MYSQL_BIN_LOG::make_log_name(char* buf, const char* log_ident)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3445
{
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3446
  uint dir_len = dirname_length(log_file_name); 
3447
  if (dir_len >= FN_REFLEN)
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3448 3449
    dir_len=FN_REFLEN-1;
  strnmov(buf, log_file_name, dir_len);
3450
  strmake(buf+dir_len, log_ident, FN_REFLEN - dir_len -1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3451 3452
}

3453

3454 3455
/**
  Check if we are writing/reading to the given log file.
3456 3457
*/

3458
bool MYSQL_BIN_LOG::is_active(const char *log_file_name_arg)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3459
{
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3460
  return !strcmp(log_file_name, log_file_name_arg);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3461 3462
}

3463

3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483
/*
  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);
}


3484 3485
/**
  Start writing to a new log file or reopen the old file.
3486

3487
  @param need_lock		Set to 1 if caller has not locked LOCK_log
3488

3489
  @note
3490 3491 3492
    The new file name is stored last in the index file
*/

3493
void MYSQL_BIN_LOG::new_file_impl(bool need_lock)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3494
{
3495 3496
  char new_name[FN_REFLEN], *new_name_ptr, *old_name;

3497
  DBUG_ENTER("MYSQL_BIN_LOG::new_file_impl");
3498
  if (!is_open())
3499 3500 3501 3502
  {
    DBUG_PRINT("info",("log is closed"));
    DBUG_VOID_RETURN;
  }
3503 3504 3505

  if (need_lock)
    pthread_mutex_lock(&LOCK_log);
3506 3507
  pthread_mutex_lock(&LOCK_index);

3508 3509 3510
  safe_mutex_assert_owner(&LOCK_log);
  safe_mutex_assert_owner(&LOCK_index);

3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523
  /*
    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);
3524 3525
    while (prepared_xids) {
      DBUG_PRINT("info", ("prepared_xids=%lu", prepared_xids));
3526
      pthread_cond_wait(&COND_prep_xids, &LOCK_prep_xids);
3527
    }
3528 3529 3530
    pthread_mutex_unlock(&LOCK_prep_xids);
  }

3531
  /* Reuse old name if not binlog and not update log */
3532
  new_name_ptr= name;
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
3533

3534
  /*
3535 3536 3537
    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.
3538
  */
3539 3540 3541
  if (generate_new_name(new_name, name))
    goto end;
  new_name_ptr=new_name;
3542

3543
  if (log_type == LOG_BIN)
3544
  {
3545
    if (!no_auto_events)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3546
    {
3547
      /*
3548 3549
        We log the whole file name for log file as the user may decide
        to change base names at some point.
3550
      */
3551
      Rotate_log_event r(new_name+dirname_length(new_name),
3552
                         0, LOG_EVENT_OFFSET, is_relay_log ? Rotate_log_event::RELAY_LOG : 0);
3553
      r.write(&log_file);
3554
      bytes_written += r.data_written;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3555
    }
3556 3557 3558 3559 3560
    /*
      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.
    */
3561
    signal_update();
3562 3563 3564
  }
  old_name=name;
  name=0;				// Don't free name
3565
  close(LOG_CLOSE_TO_BE_OPENED);
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3566

3567
  /*
3568
     Note that at this point, log_state != LOG_CLOSED (important for is_open()).
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3569
  */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3570

3571
  /*
3572
     new_file() is only used for rotation (in FLUSH LOGS or because size >
3573
     max_binlog_size or max_relay_log_size).
3574 3575 3576 3577
     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.
3578
  */
3579

3580
  open(old_name, log_type, new_name_ptr,
3581
       io_cache_type, no_auto_events, max_size, 1);
3582 3583
  my_free(old_name,MYF(0));

3584
end:
3585 3586
  if (need_lock)
    pthread_mutex_unlock(&LOCK_log);
3587 3588
  pthread_mutex_unlock(&LOCK_index);

3589
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3590 3591
}

3592

3593
bool MYSQL_BIN_LOG::append(Log_event* ev)
3594 3595 3596
{
  bool error = 0;
  pthread_mutex_lock(&LOCK_log);
3597
  DBUG_ENTER("MYSQL_BIN_LOG::append");
3598

3599
  DBUG_ASSERT(log_file.type == SEQ_READ_APPEND);
3600 3601 3602 3603
  /*
    Log_event::write() is smart enough to use my_b_write() or
    my_b_append() depending on the kind of cache we have.
  */
3604 3605 3606 3607 3608
  if (ev->write(&log_file))
  {
    error=1;
    goto err;
  }
3609
  bytes_written+= ev->data_written;
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
3610
  DBUG_PRINT("info",("max_size: %lu",max_size));
3611
  if ((uint) my_b_append_tell(&log_file) > max_size)
3612
    new_file_without_locking();
3613

3614
err:
3615
  pthread_mutex_unlock(&LOCK_log);
3616
  signal_update();				// Safe as we don't call close
3617
  DBUG_RETURN(error);
3618 3619
}

3620

3621
bool MYSQL_BIN_LOG::appendv(const char* buf, uint len,...)
3622
{
3623
  bool error= 0;
3624
  DBUG_ENTER("MYSQL_BIN_LOG::appendv");
3625 3626
  va_list(args);
  va_start(args,len);
3627

3628
  DBUG_ASSERT(log_file.type == SEQ_READ_APPEND);
3629

3630
  safe_mutex_assert_owner(&LOCK_log);
3631 3632
  do
  {
3633
    if (my_b_append(&log_file,(uchar*) buf,len))
3634
    {
3635 3636
      error= 1;
      goto err;
3637
    }
3638
    bytes_written += len;
3639
  } while ((buf=va_arg(args,const char*)) && (len=va_arg(args,uint)));
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
3640
  DBUG_PRINT("info",("max_size: %lu",max_size));
3641
  if ((uint) my_b_append_tell(&log_file) > max_size)
3642
    new_file_without_locking();
3643 3644

err:
3645 3646
  if (!error)
    signal_update();
3647
  DBUG_RETURN(error);
3648
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3649

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
3650

3651
bool MYSQL_BIN_LOG::flush_and_sync()
3652
{
3653 3654 3655 3656 3657
  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)
3658 3659
  {
    sync_binlog_counter= 0;
3660
    err=my_sync(fd, MYF(MY_WME));
3661
  }
3662
  return err;
3663 3664
}

3665
void MYSQL_BIN_LOG::start_union_events(THD *thd, query_id_t query_id_param)
3666 3667 3668 3669 3670
{
  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;
3671
  thd->binlog_evt_union.first_query_id= query_id_param;
3672 3673
}

3674
void MYSQL_BIN_LOG::stop_union_events(THD *thd)
3675 3676 3677 3678 3679
{
  DBUG_ASSERT(thd->binlog_evt_union.do_union);
  thd->binlog_evt_union.do_union= FALSE;
}

3680
bool MYSQL_BIN_LOG::is_query_in_union(THD *thd, query_id_t query_id_param)
3681 3682 3683 3684 3685
{
  return (thd->binlog_evt_union.do_union && 
          query_id_param >= thd->binlog_evt_union.first_query_id);
}

3686 3687 3688 3689 3690 3691 3692 3693 3694 3695

/*
  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=
3696
    (binlog_trx_data*) thd_get_ha_data(this, binlog_hton);
3697 3698 3699 3700

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

3701
  trx_data= (binlog_trx_data*) my_malloc(sizeof(binlog_trx_data), MYF(MY_ZEROFILL));
3702 3703 3704 3705
  if (!trx_data ||
      open_cached_file(&trx_data->trans_log, mysql_tmpdir,
                       LOG_PREFIX, binlog_cache_size, MYF(MY_WME)))
  {
3706
    my_free((uchar*)trx_data, MYF(MY_ALLOW_ZERO_PTR));
3707 3708
    DBUG_RETURN(1);                      // Didn't manage to set it up
  }
3709
  thd_set_ha_data(this, binlog_hton, trx_data);
3710

3711
  trx_data= new (thd_get_ha_data(this, binlog_hton)) binlog_trx_data;
3712

3713 3714 3715
  DBUG_RETURN(0);
}

3716
/*
3717 3718
  Function to start a statement and optionally a transaction for the
  binary log.
3719

3720 3721
  SYNOPSIS
    binlog_start_trans_and_stmt()
3722

3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743 3744 3745 3746
  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()
{
3747
  binlog_trx_data *trx_data= (binlog_trx_data*) thd_get_ha_data(this, binlog_hton);
3748 3749 3750 3751 3752 3753
  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)));

3754
  if (trx_data == NULL ||
3755
      trx_data->before_stmt_pos == MY_OFF_T_UNDEF)
3756
  {
3757
    this->binlog_set_stmt_begin();
3758
    if (options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN))
mats@romeo.(none)'s avatar
mats@romeo.(none) committed
3759 3760
      trans_register_ha(this, TRUE, binlog_hton);
    trans_register_ha(this, FALSE, binlog_hton);
3761 3762 3763 3764 3765 3766 3767 3768 3769 3770
    /*
      Mark statement transaction as read/write. We never start
      a binary log transaction and keep it read-only,
      therefore it's best to mark the transaction read/write just
      at the same time we start it.
      Not necessary to mark the normal transaction read/write
      since the statement-level flag will be propagated automatically
      inside ha_commit_trans.
    */
    ha_data[binlog_hton->slot].ha_info[0].set_trx_read_write();
3771 3772 3773 3774
  }
  DBUG_VOID_RETURN;
}

3775 3776
void THD::binlog_set_stmt_begin() {
  binlog_trx_data *trx_data=
3777
    (binlog_trx_data*) thd_get_ha_data(this, binlog_hton);
3778 3779 3780 3781 3782 3783 3784 3785 3786

  /*
    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);
3787
  trx_data= (binlog_trx_data*) thd_get_ha_data(this, binlog_hton);
3788 3789 3790 3791
  trx_data->before_stmt_pos= pos;
}


3792 3793
/*
  Write a table map to the binary log.
3794 3795
 */

3796 3797
int THD::binlog_write_table_map(TABLE *table, bool is_trans)
{
3798
  int error;
3799
  DBUG_ENTER("THD::binlog_write_table_map");
3800 3801
  DBUG_PRINT("enter", ("table: 0x%lx  (%s: #%lu)",
                       (long) table, table->s->table_name.str,
3802
                       table->s->table_map_id));
3803 3804

  /* Pre-conditions */
3805
  DBUG_ASSERT(is_current_stmt_binlog_format_row() && mysql_bin_log.is_open());
3806 3807 3808 3809 3810 3811 3812 3813
  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);

3814 3815
  if (is_trans && binlog_table_maps == 0)
    binlog_start_trans_and_stmt();
3816

3817
  if ((error= mysql_bin_log.write(&the_event)))
3818 3819
    DBUG_RETURN(error);

3820
  binlog_table_maps++;
3821 3822 3823 3824
  table->s->table_map_version= mysql_bin_log.table_map_version();
  DBUG_RETURN(0);
}

3825 3826 3827 3828
Rows_log_event*
THD::binlog_get_pending_rows_event() const
{
  binlog_trx_data *const trx_data=
3829
    (binlog_trx_data*) thd_get_ha_data(this, binlog_hton);
3830 3831 3832 3833 3834 3835
  /*
    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.
   */
3836
  return trx_data ? trx_data->pending() : NULL;
3837 3838 3839 3840 3841
}

void
THD::binlog_set_pending_rows_event(Rows_log_event* ev)
{
3842
  if (thd_get_ha_data(this, binlog_hton) == NULL)
3843 3844
    binlog_setup_trx_data();

3845
  binlog_trx_data *const trx_data=
3846
    (binlog_trx_data*) thd_get_ha_data(this, binlog_hton);
3847

3848
  DBUG_ASSERT(trx_data);
3849
  trx_data->set_pending(ev);
3850 3851 3852
}


3853 3854 3855 3856 3857 3858 3859 3860 3861
/**
  Remove the pending rows event, discarding any outstanding rows.

  If there is no pending rows event available, this is effectively a
  no-op.
 */
int
MYSQL_BIN_LOG::remove_pending_rows_event(THD *thd)
{
3862
  DBUG_ENTER("MYSQL_BIN_LOG::remove_pending_rows_event");
3863 3864 3865 3866 3867 3868 3869 3870 3871 3872 3873 3874 3875 3876 3877

  binlog_trx_data *const trx_data=
    (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);

  DBUG_ASSERT(trx_data);

  if (Rows_log_event* pending= trx_data->pending())
  {
    delete pending;
    trx_data->set_pending(NULL);
  }

  DBUG_RETURN(0);
}

3878 3879 3880 3881 3882
/*
  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.
*/
3883 3884 3885
int
MYSQL_BIN_LOG::flush_and_set_pending_rows_event(THD *thd,
                                                Rows_log_event* event)
3886
{
3887
  DBUG_ENTER("MYSQL_BIN_LOG::flush_and_set_pending_rows_event(event)");
3888
  DBUG_ASSERT(mysql_bin_log.is_open());
3889
  DBUG_PRINT("enter", ("event: 0x%lx", (long) event));
3890 3891 3892 3893

  int error= 0;

  binlog_trx_data *const trx_data=
3894
    (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
3895 3896 3897

  DBUG_ASSERT(trx_data);

3898
  DBUG_PRINT("info", ("trx_data->pending(): 0x%lx", (long) trx_data->pending()));
3899

3900
  if (Rows_log_event* pending= trx_data->pending())
3901 3902 3903 3904 3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924
  {
    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);
3925
      set_write_error(thd);
3926 3927 3928 3929 3930 3931 3932 3933 3934 3935 3936 3937 3938 3939 3940 3941 3942 3943 3944 3945 3946 3947 3948 3949 3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961
      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);
  }

3962
  thd->binlog_set_pending_rows_event(event);
3963 3964 3965 3966

  DBUG_RETURN(error);
}

3967 3968
/**
  Write an event to the binary log.
3969 3970
*/

3971
bool MYSQL_BIN_LOG::write(Log_event *event_info)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3972
{
3973 3974
  THD *thd= event_info->thd;
  bool error= 1;
3975
  DBUG_ENTER("MYSQL_BIN_LOG::write(Log_event *)");
3976

3977 3978
  if (thd->binlog_evt_union.do_union)
  {
3979 3980 3981 3982
    /*
      In Stored function; Remember that function call caused an update.
      We will log the function call to the binary log on function exit
    */
3983 3984 3985 3986
    thd->binlog_evt_union.unioned_events= TRUE;
    thd->binlog_evt_union.unioned_events_trans |= event_info->cache_stmt;
    DBUG_RETURN(0);
  }
3987 3988 3989 3990 3991 3992 3993

  /*
    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.

3994 3995 3996
    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.
3997
  */
3998 3999
  bool const end_stmt=
    thd->prelocked_mode && thd->lex->requires_prelocking();
4000 4001
  if (thd->binlog_flush_pending_rows_event(end_stmt))
    DBUG_RETURN(error);
4002

4003 4004
  pthread_mutex_lock(&LOCK_log);

4005
  /*
guilhem@mysql.com's avatar
guilhem@mysql.com committed
4006 4007 4008 4009
     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.
  */
4010
  if (likely(is_open()))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4011
  {
4012
    IO_CACHE *file= &log_file;
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
4013
#ifdef HAVE_REPLICATION
4014
    /*
4015 4016 4017
      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)"
4018
    */
4019
    const char *local_db= event_info->get_db();
4020
    if ((thd && !(thd->options & OPTION_BIN_LOG)) ||
lars@mysql.com's avatar
Merge  
lars@mysql.com committed
4021
	(!binlog_filter->db_ok(local_db)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4022
    {
4023
      VOID(pthread_mutex_unlock(&LOCK_log));
4024
      DBUG_RETURN(0);
4025
    }
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
4026
#endif /* HAVE_REPLICATION */
4027

4028
#if defined(USING_TRANSACTIONS) 
4029 4030 4031 4032 4033 4034 4035
    /*
      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)
4036
      - or if the event asks for it (cache_stmt == TRUE).
4037 4038 4039
    */
    if (opt_using_transactions && thd)
    {
4040 4041
      if (thd->binlog_setup_trx_data())
        goto err;
4042

4043
      binlog_trx_data *const trx_data=
4044
        (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
4045
      IO_CACHE *trans_log= &trx_data->trans_log;
4046 4047
      my_off_t trans_log_pos= my_b_tell(trans_log);
      if (event_info->get_cache_stmt() || trans_log_pos != 0)
4048
      {
4049
        DBUG_PRINT("info", ("Using trans_log: cache: %d, trans_log_pos: %lu",
4050
                            event_info->get_cache_stmt(),
4051
                            (ulong) trans_log_pos));
4052
        thd->binlog_start_trans_and_stmt();
4053
        file= trans_log;
4054
      }
4055
      /*
4056
        TODO as Mats suggested, for all the cases above where we write to
4057 4058
        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
4059
        LOCK_log.
4060
      */
4061
    }
4062
#endif /* USING_TRANSACTIONS */
4063
    DBUG_PRINT("info",("event type: %d",event_info->get_type_code()));
4064

4065 4066 4067 4068
    /*
      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
4069 4070

    /*
4071 4072
      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
4073 4074
    */

4075 4076 4077 4078
    /*
      If row-based binlogging, Insert_id, Rand and other kind of "setting
      context" events are not needed.
    */
4079
    if (thd)
4080
    {
4081
      if (!thd->is_current_stmt_binlog_format_row())
4082
      {
4083
        if (thd->stmt_depends_on_first_successful_insert_id_in_prev_stmt)
4084 4085
        {
          Intvar_log_event e(thd,(uchar) LAST_INSERT_ID_EVENT,
4086
                             thd->first_successful_insert_id_in_prev_stmt_for_binlog);
4087 4088 4089
          if (e.write(file))
            goto err;
        }
4090
        if (thd->auto_inc_intervals_in_cur_stmt_for_binlog.nb_elements() > 0)
4091
        {
4092 4093 4094 4095 4096 4097
          DBUG_PRINT("info",("number of auto_inc intervals: %u",
                             thd->auto_inc_intervals_in_cur_stmt_for_binlog.
                             nb_elements()));
          Intvar_log_event e(thd, (uchar) INSERT_ID_EVENT,
                             thd->auto_inc_intervals_in_cur_stmt_for_binlog.
                             minimum());
4098 4099 4100 4101 4102 4103 4104 4105 4106 4107 4108 4109 4110 4111
          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;
4112
            get_dynamic(&thd->user_var_events,(uchar*) &user_var_event, i);
4113 4114 4115 4116 4117 4118 4119 4120 4121 4122
            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;
          }
        }
4123
      }
4124
    }
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
4125

4126 4127
    /*
       Write the SQL command
4128
     */
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
4129

4130 4131
    if (event_info->write(file) || 
        DBUG_EVALUATE_IF("injecting_fault_writing", 1, 0))
4132
      goto err;
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
4133

4134
    if (file == &log_file) // we are writing to the real log (disk)
4135
    {
4136
      if (flush_and_sync())
monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
4137
	goto err;
4138 4139
      signal_update();
      rotate_and_purge(RP_LOCK_LOG_IS_ALREADY_LOCKED);
4140
    }
monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
4141
    error=0;
4142

4143 4144
err:
    if (error)
4145
      set_write_error(thd);
4146
  }
4147

4148 4149 4150
  if (event_info->flags & LOG_EVENT_UPDATE_TABLE_MAP_VERSION_F)
    ++m_table_map_version;

4151
  pthread_mutex_unlock(&LOCK_log);
4152
  DBUG_RETURN(error);
4153 4154
}

4155 4156 4157 4158 4159 4160 4161 4162 4163

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,
4164
                    ulonglong current_utime)
4165
{
4166
  return logger.slow_log_print(thd, query, query_length, current_utime);
4167 4168 4169
}


4170 4171 4172 4173 4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189 4190 4191 4192 4193 4194 4195 4196 4197
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;
}


4198 4199 4200 4201 4202 4203
bool general_log_print(THD *thd, enum enum_server_command command,
                       const char *format, ...)
{
  va_list args;
  uint error= 0;

4204 4205 4206 4207
  /* Print the message to the buffer if we want to log this king of commands */
  if (! logger.log_command(thd, command))
    return FALSE;

4208 4209 4210 4211 4212 4213 4214
  va_start(args, format);
  error= logger.general_log_print(thd, command, format, args);
  va_end(args);

  return error;
}

4215 4216 4217 4218 4219 4220 4221 4222 4223 4224
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;
}

4225
void MYSQL_BIN_LOG::rotate_and_purge(uint flags)
4226
{
4227 4228
  if (!(flags & RP_LOCK_LOG_IS_ALREADY_LOCKED))
    pthread_mutex_lock(&LOCK_log);
4229 4230
  if ((flags & RP_FORCE_ROTATE) ||
      (my_b_tell(&log_file) >= (my_off_t) max_size))
4231
  {
4232
    new_file_without_locking();
4233 4234 4235
#ifdef HAVE_REPLICATION
    if (expire_logs_days)
    {
4236
      time_t purge_time= my_time(0) - expire_logs_days*24*60*60;
4237 4238 4239 4240 4241
      if (purge_time >= 0)
        purge_logs_before_date(purge_time);
    }
#endif
  }
4242 4243
  if (!(flags & RP_LOCK_LOG_IS_ALREADY_LOCKED))
    pthread_mutex_unlock(&LOCK_log);
4244
}
4245

4246
uint MYSQL_BIN_LOG::next_file_id()
4247 4248 4249 4250 4251 4252 4253 4254
{
  uint res;
  pthread_mutex_lock(&LOCK_log);
  res = file_id++;
  pthread_mutex_unlock(&LOCK_log);
  return res;
}

4255

4256 4257 4258 4259 4260 4261 4262 4263 4264 4265 4266 4267 4268 4269 4270 4271 4272 4273 4274 4275
/*
  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;
4276
  uint length= my_b_bytes_in_cache(cache), group, carry, hdr_offs;
4277 4278 4279
  long val;
  uchar header[LOG_EVENT_HEADER_LEN];

4280 4281 4282 4283 4284 4285 4286 4287 4288 4289 4290 4291 4292 4293
  /*
    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.
  */

4294
  group= (uint)my_b_tell(&log_file);
4295 4296
  hdr_offs= carry= 0;

4297 4298
  do
  {
4299 4300 4301 4302 4303 4304

    /*
      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))
4305 4306 4307
    {
      DBUG_ASSERT(carry < LOG_EVENT_HEADER_LEN);

4308
      /* assemble both halves */
4309 4310
      memcpy(&header[carry], (char *)cache->read_pos, LOG_EVENT_HEADER_LEN - carry);

4311
      /* fix end_log_pos */
4312 4313 4314
      val= uint4korr(&header[LOG_POS_OFFSET]) + group;
      int4store(&header[LOG_POS_OFFSET], val);

4315
      /* write the first half of the split header */
4316 4317 4318
      if (my_b_write(&log_file, header, carry))
        return ER_ERROR_ON_WRITE;

4319 4320 4321 4322
      /*
        copy fixed second half of header to cache so the correct
        version will be written later.
      */
4323 4324
      memcpy((char *)cache->read_pos, &header[carry], LOG_EVENT_HEADER_LEN - carry);

4325
      /* next event header at ... */
4326
      hdr_offs = uint4korr(&header[EVENT_LEN_OFFSET]) - carry;
4327 4328 4329 4330

      carry= 0;
    }

4331 4332
    /* if there is anything to write, process it. */

4333
    if (likely(length > 0))
4334
    {
4335
      /*
4336 4337 4338 4339
        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").
4340
      */
4341

4342
      while (hdr_offs < length)
4343
      {
4344 4345 4346 4347
        /*
          partial header only? save what we can get, process once
          we get the rest.
        */
4348

4349
        if (hdr_offs + LOG_EVENT_HEADER_LEN > length)
4350
        {
4351
          carry= length - hdr_offs;
4352
          memcpy(header, (char *)cache->read_pos + hdr_offs, carry);
4353
          length= hdr_offs;
4354 4355 4356 4357
        }
        else
        {
          /* we've got a full event-header, and it came in one piece */
4358

4359
          uchar *log_pos= (uchar *)cache->read_pos + hdr_offs + LOG_POS_OFFSET;
4360

4361 4362 4363
          /* fix end_log_pos */
          val= uint4korr(log_pos) + group;
          int4store(log_pos, val);
4364

4365 4366 4367
          /* next event header at ... */
          log_pos= (uchar *)cache->read_pos + hdr_offs + EVENT_LEN_OFFSET;
          hdr_offs += uint4korr(log_pos);
4368

4369
        }
4370
      }
4371 4372

      /*
4373 4374 4375 4376 4377 4378
        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:
4379
      */
4380
      hdr_offs -= length;
4381 4382 4383
    }

    /* Write data to the binary log file */
4384
    if (my_b_write(&log_file, cache->read_pos, length))
4385
      return ER_ERROR_ON_WRITE;
4386
    cache->read_pos=cache->read_end;		// Mark buffer used up
4387
  } while ((length= my_b_fill(cache)));
4388

4389
  DBUG_ASSERT(carry == 0);
4390 4391 4392 4393 4394 4395 4396

  if (sync_log)
    flush_and_sync();

  return 0;                                     // All OK
}

4397 4398 4399 4400 4401 4402 4403 4404 4405 4406 4407 4408 4409 4410 4411 4412 4413 4414 4415 4416 4417 4418 4419 4420 4421 4422 4423 4424 4425
/*
  Helper function to get the error code of the query to be binlogged.
 */
int query_error_code(THD *thd, bool not_killed)
{
  int error;
  
  if (not_killed)
  {
    error= thd->is_error() ? thd->main_da.sql_errno() : 0;

    /* thd->main_da.sql_errno() might be ER_SERVER_SHUTDOWN or
       ER_QUERY_INTERRUPTED, So here we need to make sure that error
       is not set to these errors when specified not_killed by the
       caller.
    */
    if (error == ER_SERVER_SHUTDOWN || error == ER_QUERY_INTERRUPTED)
      error= 0;
  }
  else
  {
    /* killed status for DELAYED INSERT thread should never be used */
    DBUG_ASSERT(!(thd->system_thread & SYSTEM_THREAD_DELAYED_INSERT));
    error= thd->killed_errno();
  }

  return error;
}

4426 4427 4428 4429 4430 4431 4432 4433 4434 4435 4436 4437 4438 4439 4440 4441 4442 4443 4444 4445 4446 4447 4448
bool MYSQL_BIN_LOG::write_incident(THD *thd, bool lock)
{
  uint error= 0;
  DBUG_ENTER("MYSQL_BIN_LOG::write_incident");
  LEX_STRING const write_error_msg=
    { C_STRING_WITH_LEN("error writing to the binary log") };
  Incident incident= INCIDENT_LOST_EVENTS;
  Incident_log_event ev(thd, incident, write_error_msg);
  if (lock)
    pthread_mutex_lock(&LOCK_log);
  ev.write(&log_file);
  if (lock)
  {
    if (!error && !(error= flush_and_sync()))
    {
      signal_update();
      rotate_and_purge(RP_LOCK_LOG_IS_ALREADY_LOCKED);
    }
    pthread_mutex_unlock(&LOCK_log);
  }
  DBUG_RETURN(error);
}

4449 4450 4451 4452 4453 4454 4455 4456 4457 4458 4459
/**
  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
4460
                        contents of the cache.
4461 4462 4463
  @param incident       Defines if an incident event should be created to
                        notify that some non-transactional changes did
                        not get into the binlog.
guilhem@mysql.com's avatar
guilhem@mysql.com committed
4464

4465 4466 4467 4468 4469 4470
  @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.
4471 4472
*/

4473 4474
bool MYSQL_BIN_LOG::write(THD *thd, IO_CACHE *cache, Log_event *commit_event,
                          bool incident)
4475
{
4476
  DBUG_ENTER("MYSQL_BIN_LOG::write(THD *, IO_CACHE *, Log_event *)");
4477
  VOID(pthread_mutex_lock(&LOCK_log));
4478

4479 4480 4481
  /* NULL would represent nothing to replicate after ROLLBACK */
  DBUG_ASSERT(commit_event != NULL);

4482
  DBUG_ASSERT(is_open());
4483
  if (likely(is_open()))                       // Should always be true
4484
  {
4485
    /*
4486 4487 4488 4489
      We only bother to write to the binary log if there is anything
      to write.
     */
    if (my_b_tell(cache) > 0)
4490
    {
4491
      /*
4492 4493 4494
        Log "BEGIN" at the beginning of every transaction.  Here, a
        transaction is either a BEGIN..COMMIT block or a single
        statement in autocommit mode.
4495
      */
He Zhenxing's avatar
He Zhenxing committed
4496
      Query_log_event qinfo(thd, STRING_WITH_LEN("BEGIN"), TRUE, TRUE, 0);
4497

4498 4499 4500 4501 4502 4503 4504 4505 4506 4507 4508 4509 4510 4511 4512 4513 4514 4515 4516
      /*
        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();
                      });
4517

4518 4519
      if ((write_error= write_cache(cache, false, false)))
        goto err;
4520

4521 4522
      if (commit_event && commit_event->write(&log_file))
        goto err;
4523 4524 4525 4526

      if (incident && write_incident(thd, FALSE))
        goto err;

4527 4528 4529 4530 4531 4532 4533 4534 4535 4536
      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
4537
    }
4538

4539 4540 4541 4542 4543 4544 4545 4546
    /*
      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.
    */
4547
    if (commit_event && commit_event->get_type_code() == XID_EVENT)
4548 4549 4550 4551 4552
    {
      pthread_mutex_lock(&LOCK_prep_xids);
      prepared_xids++;
      pthread_mutex_unlock(&LOCK_prep_xids);
    }
4553 4554
    else
      rotate_and_purge(RP_LOCK_LOG_IS_ALREADY_LOCKED);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4555
  }
4556
  VOID(pthread_mutex_unlock(&LOCK_log));
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
4557

4558
  DBUG_RETURN(0);
4559 4560

err:
4561 4562 4563 4564 4565
  if (!write_error)
  {
    write_error= 1;
    sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
  }
4566
  VOID(pthread_mutex_unlock(&LOCK_log));
4567
  DBUG_RETURN(1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4568 4569
}

4570

4571 4572
/**
  Wait until we get a signal that the binary log has been updated.
4573

4574 4575 4576 4577
  @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.
4578

4579
  @note
4580
    One must have a lock on LOCK_log before calling this function.
4581
    This lock will be released before return! That's required by
guilhem@mysql.com's avatar
guilhem@mysql.com committed
4582
    THD::enter_cond() (see NOTES in sql_class.h).
4583 4584
*/

4585
void MYSQL_BIN_LOG::wait_for_update(THD* thd, bool is_slave)
4586
{
monty@mysql.com's avatar
monty@mysql.com committed
4587 4588
  const char *old_msg;
  DBUG_ENTER("wait_for_update");
4589

monty@mysql.com's avatar
monty@mysql.com committed
4590
  old_msg= thd->enter_cond(&update_cond, &LOCK_log,
4591
                           is_slave ?
monty@mysql.com's avatar
monty@mysql.com committed
4592
                           "Has read all relay log; waiting for the slave I/O "
4593
                           "thread to update it" :
monty@mysql.com's avatar
monty@mysql.com committed
4594
                           "Has sent all binlog to slave; waiting for binlog "
4595
                           "to be updated");
4596 4597
  pthread_cond_wait(&update_cond, &LOCK_log);
  thd->exit_cond(old_msg);
monty@mysql.com's avatar
monty@mysql.com committed
4598
  DBUG_VOID_RETURN;
4599
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4600

4601

4602 4603
/**
  Close the log file.
4604

4605 4606 4607 4608 4609
  @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
4610

4611
  @note
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
4612 4613
    One can do an open on the object at once after doing a close.
    The internal structures are not freed until cleanup() is called
4614 4615
*/

4616
void MYSQL_BIN_LOG::close(uint exiting)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4617
{					// One can't set log_type here!
4618
  DBUG_ENTER("MYSQL_BIN_LOG::close");
4619
  DBUG_PRINT("enter",("exiting: %d", (int) exiting));
4620
  if (log_state == LOG_OPENED)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4621
  {
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
4622
#ifdef HAVE_REPLICATION
4623 4624
    if (log_type == LOG_BIN && !no_auto_events &&
	(exiting & LOG_CLOSE_STOP_EVENT))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4625 4626
    {
      Stop_log_event s;
4627
      s.write(&log_file);
4628
      bytes_written+= s.data_written;
4629
      signal_update();
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4630
    }
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
4631
#endif /* HAVE_REPLICATION */
4632 4633

    /* don't pwrite in a file opened with O_APPEND - it doesn't work */
4634
    if (log_file.type == WRITE_CACHE && log_type == LOG_BIN)
4635 4636
    {
      my_off_t offset= BIN_LOG_HEADER_SIZE + FLAGS_OFFSET;
4637 4638
      my_off_t org_position= my_tell(log_file.file, MYF(0));
      uchar flags= 0;            // clearing LOG_EVENT_BINLOG_IN_USE_F
4639
      my_pwrite(log_file.file, &flags, 1, offset, MYF(0));
4640 4641 4642 4643 4644 4645 4646
      /*
        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));
4647 4648
    }

4649 4650
    /* 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
4651
  }
4652 4653 4654 4655 4656 4657

  /*
    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.
  */

4658
  if ((exiting & LOG_CLOSE_INDEX) && my_b_inited(&index_file))
4659
  {
4660 4661
    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
4662
    {
4663 4664
      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
4665
    }
4666
  }
4667
  log_state= (exiting & LOG_CLOSE_TO_BE_OPENED) ? LOG_TO_BE_OPENED : LOG_CLOSED;
4668
  safeFree(name);
4669
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4670 4671 4672
}


4673
void MYSQL_BIN_LOG::set_max_size(ulong max_size_arg)
4674 4675 4676 4677 4678 4679 4680 4681
{
  /*
    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.
  */
4682
  DBUG_ENTER("MYSQL_BIN_LOG::set_max_size");
guilhem@mysql.com's avatar
guilhem@mysql.com committed
4683
  pthread_mutex_lock(&LOCK_log);
4684 4685
  if (is_open())
    max_size= max_size_arg;
guilhem@mysql.com's avatar
guilhem@mysql.com committed
4686 4687
  pthread_mutex_unlock(&LOCK_log);
  DBUG_VOID_RETURN;
4688
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4689

4690

4691 4692
/**
  Check if a string is a valid number.
4693

4694 4695 4696
  @param str			String to test
  @param res			Store value here
  @param allow_wildcards	Set to 1 if we should ignore '%' and '_'
4697

4698
  @note
4699 4700 4701
    For the moment the allow_wildcards argument is not used
    Should be move to some other file.

4702
  @retval
4703
    1	String is a number
4704
  @retval
4705
    0	Error
4706
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4707 4708 4709 4710 4711 4712 4713 4714 4715 4716 4717 4718

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++;
4719 4720
  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
4721 4722 4723 4724 4725 4726 4727
  {
    flag=1;
    str++;
  }
  if (*str == '.')
  {
    for (str++ ;
4728
	 my_isdigit(files_charset_info,*str) ||
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4729 4730 4731 4732 4733 4734 4735 4736 4737 4738 4739 4740 4741 4742 4743 4744 4745 4746 4747
	   (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
}
4748

monty@mysql.com's avatar
monty@mysql.com committed
4749

4750 4751 4752
bool flush_error_log()
{
  bool result=0;
4753
  if (opt_error_log)
4754 4755 4756 4757
  {
    char err_renamed[FN_REFLEN], *end;
    end= strmake(err_renamed,log_error_file,FN_REFLEN-4);
    strmov(end, "-old");
4758
    VOID(pthread_mutex_lock(&LOCK_error_log));
4759 4760 4761 4762 4763 4764
#ifdef __WIN__
    char err_temp[FN_REFLEN+4];
    /*
     On Windows is necessary a temporary file for to rename
     the current error file.
    */
4765
    strxmov(err_temp, err_renamed,"-tmp",NullS);
4766 4767 4768
    (void) my_delete(err_temp, MYF(0)); 
    if (freopen(err_temp,"a+",stdout))
    {
4769 4770 4771 4772
      int fd;
      size_t bytes;
      uchar buf[IO_SIZE];

4773
      freopen(err_temp,"a+",stderr);
4774
      setbuf(stderr, NULL);
4775 4776 4777
      (void) my_delete(err_renamed, MYF(0));
      my_rename(log_error_file,err_renamed,MYF(0));
      if (freopen(log_error_file,"a+",stdout))
4778
      {
4779
        freopen(log_error_file,"a+",stderr);
4780 4781
        setbuf(stderr, NULL);
      }
4782

4783 4784
      if ((fd = my_open(err_temp, O_RDONLY, MYF(0))) >= 0)
      {
4785 4786 4787
        while ((bytes= my_read(fd, buf, IO_SIZE, MYF(0))) &&
               bytes != MY_FILE_ERROR)
          my_fwrite(stderr, buf, bytes, MYF(0));
4788 4789 4790 4791 4792 4793 4794 4795 4796
        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))
4797
   {
4798
     freopen(log_error_file,"a+",stderr);
4799 4800
     setbuf(stderr, NULL);
   }
4801 4802 4803
   else
     result= 1;
#endif
4804
    VOID(pthread_mutex_unlock(&LOCK_error_log));
4805 4806 4807
  }
   return result;
}
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
4808

4809
void MYSQL_BIN_LOG::signal_update()
monty@mysql.com's avatar
monty@mysql.com committed
4810
{
4811
  DBUG_ENTER("MYSQL_BIN_LOG::signal_update");
monty@mysql.com's avatar
monty@mysql.com committed
4812 4813 4814 4815
  pthread_cond_broadcast(&update_cond);
  DBUG_VOID_RETURN;
}

4816
#ifdef __NT__
4817
static void print_buffer_to_nt_eventlog(enum loglevel level, char *buff,
4818
                                        size_t length, size_t buffLen)
4819 4820
{
  HANDLE event;
4821
  char   *buffptr= buff;
serg@serg.mylan's avatar
serg@serg.mylan committed
4822
  DBUG_ENTER("print_buffer_to_nt_eventlog");
4823

4824 4825
  /* Add ending CR/LF's to string, overwrite last chars if necessary */
  strmov(buffptr+min(length, buffLen-5), "\r\n\r\n");
4826

monty@mysql.com's avatar
monty@mysql.com committed
4827 4828
  setup_windows_event_source();
  if ((event= RegisterEventSource(NULL,"MySQL")))
4829
  {
monty@mysql.com's avatar
monty@mysql.com committed
4830
    switch (level) {
4831
      case ERROR_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
4832
        ReportEvent(event, EVENTLOG_ERROR_TYPE, 0, MSG_DEFAULT, NULL, 1, 0,
4833
                    (LPCSTR*)&buffptr, NULL);
4834 4835
        break;
      case WARNING_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
4836
        ReportEvent(event, EVENTLOG_WARNING_TYPE, 0, MSG_DEFAULT, NULL, 1, 0,
4837
                    (LPCSTR*) &buffptr, NULL);
4838 4839
        break;
      case INFORMATION_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
4840
        ReportEvent(event, EVENTLOG_INFORMATION_TYPE, 0, MSG_DEFAULT, NULL, 1,
4841
                    0, (LPCSTR*) &buffptr, NULL);
4842 4843 4844 4845 4846 4847 4848
        break;
    }
    DeregisterEventSource(event);
  }

  DBUG_VOID_RETURN;
}
monty@mysql.com's avatar
monty@mysql.com committed
4849 4850
#endif /* __NT__ */

4851

4852
/**
monty@mysql.com's avatar
monty@mysql.com committed
4853 4854
  Prints a printf style message to the error log and, under NT, to the
  Windows event log.
4855

4856 4857
  This function prints the message into a buffer and then sends that buffer
  to other functions to write that message to other logging sources.
4858

4859 4860 4861
  @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
4862

4863
  @returns
4864 4865 4866
    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
4867
*/
monty@mysql.com's avatar
monty@mysql.com committed
4868

4869
#ifndef EMBEDDED_LIBRARY
4870 4871 4872 4873 4874 4875 4876 4877 4878 4879
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));

4880
  skr= my_time(0);
4881 4882
  localtime_r(&skr, &tm_tmp);
  start=&tm_tmp;
4883

4884 4885 4886 4887 4888 4889 4890 4891 4892 4893 4894 4895 4896 4897 4898 4899 4900 4901
  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
4902
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
4903 4904
{
  char   buff[1024];
4905
  size_t length;
4906
  DBUG_ENTER("vprint_msg_to_log");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4907

4908
  length= my_vsnprintf(buff, sizeof(buff), format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
4909
  print_buffer_to_file(level, buff);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4910 4911

#ifdef __NT__
monty@mysql.com's avatar
monty@mysql.com committed
4912
  print_buffer_to_nt_eventlog(level, buff, length, sizeof(buff));
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4913
#endif
4914

4915
  DBUG_RETURN(0);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4916
}
4917
#endif /*EMBEDDED_LIBRARY*/
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4918

4919

monty@mysql.com's avatar
monty@mysql.com committed
4920
void sql_print_error(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4921 4922
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
4923
  DBUG_ENTER("sql_print_error");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4924

serg@serg.mylan's avatar
serg@serg.mylan committed
4925
  va_start(args, format);
4926
  error_log_print(ERROR_LEVEL, format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
4927
  va_end(args);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4928 4929 4930 4931 4932

  DBUG_VOID_RETURN;
}


monty@mysql.com's avatar
monty@mysql.com committed
4933
void sql_print_warning(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4934 4935
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
4936
  DBUG_ENTER("sql_print_warning");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4937

serg@serg.mylan's avatar
serg@serg.mylan committed
4938
  va_start(args, format);
4939
  error_log_print(WARNING_LEVEL, format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
4940
  va_end(args);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4941 4942 4943 4944 4945

  DBUG_VOID_RETURN;
}


monty@mysql.com's avatar
monty@mysql.com committed
4946
void sql_print_information(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4947 4948
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
4949
  DBUG_ENTER("sql_print_information");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4950

serg@serg.mylan's avatar
serg@serg.mylan committed
4951
  va_start(args, format);
4952
  error_log_print(INFORMATION_LEVEL, format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
4953
  va_end(args);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4954

guilhem@mysql.com's avatar
guilhem@mysql.com committed
4955 4956
  DBUG_VOID_RETURN;
}
4957

4958

4959 4960 4961 4962 4963 4964 4965 4966
/********* 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
4967 4968
  states - active, syncing, pool.
  there could be only one page in active or syncing states,
4969
  but many in pool - pool is fifo queue.
4970
  usual lifecycle of a page is pool->active->syncing->pool
4971
  "active" page - is a page where new xid's are logged.
4972 4973
  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.
4974
  when the sync is done the page is moved to a pool and an active page
4975 4976 4977 4978 4979 4980 4981 4982 4983 4984 4985 4986 4987
  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.
4988 4989 4990 4991
  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)

4992 4993
  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
4994 4995
  dirty - we don't sync removals to disk.
*/
4996 4997 4998 4999 5000

ulong tc_log_page_waits= 0;

#ifdef HAVE_MMAP

5001 5002
#define TC_LOG_HEADER_SIZE (sizeof(tc_log_magic)+1)

5003
static const char tc_log_magic[]={(char) 254, 0x23, 0x05, 0x74};
5004

5005
ulong opt_tc_log_size= TC_LOG_MIN_SIZE;
5006
ulong tc_log_max_pages_used=0, tc_log_page_size=0, tc_log_cur_pages_used=0;
5007 5008 5009 5010 5011 5012 5013

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

5014
  DBUG_ASSERT(total_ha_2pc > 1);
5015 5016
  DBUG_ASSERT(opt_name && opt_name[0]);

serg@serg.mylan's avatar
serg@serg.mylan committed
5017
  tc_log_page_size= my_getpagesize();
5018 5019 5020
  DBUG_ASSERT(TC_LOG_PAGE_SIZE % tc_log_page_size == 0);

  fn_format(logname,opt_name,mysql_data_home,"",MY_UNPACK_FILENAME);
5021
  if ((fd= my_open(logname, O_RDWR, MYF(0))) < 0)
5022
  {
5023 5024
    if (my_errno != ENOENT)
      goto err;
5025 5026
    if (using_heuristic_recover())
      return 1;
5027
    if ((fd= my_create(logname, CREATE_MODE, O_RDWR, MYF(MY_WME))) < 0)
5028 5029
      goto err;
    inited=1;
5030
    file_length= opt_tc_log_size;
5031 5032 5033 5034 5035
    if (my_chsize(fd, file_length, 0, MYF(MY_WME)))
      goto err;
  }
  else
  {
5036 5037
    inited= 1;
    crashed= TRUE;
serg@serg.mylan's avatar
serg@serg.mylan committed
5038
    sql_print_information("Recovering after a crash using %s", opt_name);
5039 5040 5041 5042 5043 5044
    if (tc_heuristic_recover)
    {
      sql_print_error("Cannot perform automatic crash recovery when "
                      "--tc-heuristic-recover is used");
      goto err;
    }
5045
    file_length= my_seek(fd, 0L, MY_SEEK_END, MYF(MY_WME+MY_FAE));
5046 5047 5048 5049
    if (file_length == MY_FILEPOS_ERROR || file_length % tc_log_page_size)
      goto err;
  }

5050
  data= (uchar *)my_mmap(0, (size_t)file_length, PROT_READ|PROT_WRITE,
5051 5052 5053 5054 5055 5056 5057 5058
                        MAP_NOSYNC|MAP_SHARED, fd, 0);
  if (data == MAP_FAILED)
  {
    my_errno=errno;
    goto err;
  }
  inited=2;

5059
  npages=(uint)file_length/tc_log_page_size;
5060 5061 5062 5063 5064 5065 5066 5067 5068 5069 5070 5071 5072 5073 5074 5075 5076
  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);
5077
  pages[0].start=pages[0].end-pages[0].size;
5078 5079 5080
  pages[npages-1].next=0;
  inited=4;

5081
  if (crashed && recover())
5082
      goto err;
5083

5084
  memcpy(data, tc_log_magic, sizeof(tc_log_magic));
5085
  data[sizeof(tc_log_magic)]= (uchar)total_ha_2pc;
5086 5087 5088 5089 5090 5091 5092 5093 5094 5095 5096
  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;

5097
  syncing= 0;
5098 5099 5100 5101 5102 5103 5104 5105 5106 5107 5108
  active=pages;
  pool=pages+1;
  pool_last=pages+npages-1;

  return 0;

err:
  close();
  return 1;
}

5109 5110
/**
  there is no active page, let's got one from the pool.
5111

5112 5113 5114
  Two strategies here:
    -# take the first from the pool
    -# if there're waiters - take the one with the most free space.
5115

5116 5117 5118
  @todo
    TODO page merging. try to allocate adjacent page first,
    so that they can be flushed both in one sync
5119
*/
5120

5121 5122 5123 5124 5125
void TC_LOG_MMAP::get_active_from_pool()
{
  PAGE **p, **best_p=0;
  int best_free;

5126 5127 5128
  if (syncing)
    pthread_mutex_lock(&LOCK_pool);

5129 5130 5131
  do
  {
    best_p= p= &pool;
5132 5133
    if ((*p)->waiters == 0) // can the first page be used ?
      break;                // yes - take it.
5134

5135
    best_free=0;            // no - trying second strategy
5136 5137 5138 5139 5140 5141 5142 5143 5144 5145 5146 5147
    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;
5148
  if (active->free == active->size) // we've chosen an empty page
5149 5150 5151 5152 5153
  {
    tc_log_cur_pages_used++;
    set_if_bigger(tc_log_max_pages_used, tc_log_cur_pages_used);
  }

5154
  if ((*best_p)->next)              // unlink the page from the pool
5155 5156 5157
    *best_p=(*best_p)->next;
  else
    pool_last=*best_p;
5158 5159 5160

  if (syncing)
    pthread_mutex_unlock(&LOCK_pool);
5161 5162
}

5163 5164 5165 5166
/**
  @todo
  perhaps, increase log size ?
*/
5167 5168 5169 5170 5171 5172 5173 5174 5175
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);
5176
  return 1; // always return 1
5177 5178
}

5179 5180
/**
  Record that transaction XID is committed on the persistent storage.
5181 5182 5183 5184 5185 5186 5187 5188 5189 5190 5191 5192 5193 5194 5195

    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.
5196

5197 5198 5199 5200 5201 5202 5203
  @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.
5204 5205
*/

5206
int TC_LOG_MMAP::log_xid(THD *thd, my_xid xid)
5207 5208 5209 5210 5211 5212 5213
{
  int err;
  PAGE *p;
  ulong cookie;

  pthread_mutex_lock(&LOCK_active);

5214 5215 5216 5217 5218 5219 5220
  /*
    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.
  */
5221 5222 5223
  while (unlikely(active && active->free == 0))
    pthread_cond_wait(&COND_active, &LOCK_active);

5224
  /* no active page ? take one from the pool */
5225 5226 5227 5228 5229 5230
  if (active == 0)
    get_active_from_pool();

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

5231
  /* searching for an empty slot */
5232 5233 5234 5235 5236 5237
  while (*p->ptr)
  {
    p->ptr++;
    DBUG_ASSERT(p->ptr < p->end);               // because p->free > 0
  }

5238 5239
  /* found! store xid there and mark the page dirty */
  cookie= (ulong)((uchar *)p->ptr - data);      // can never be zero
5240 5241 5242 5243 5244
  *p->ptr++= xid;
  p->free--;
  p->state= DIRTY;

  /* to sync or not to sync - this is the question */
5245
  pthread_mutex_unlock(&LOCK_active);
5246 5247 5248
  pthread_mutex_lock(&LOCK_sync);
  pthread_mutex_unlock(&p->lock);

5249
  if (syncing)
5250 5251
  {                                          // somebody's syncing. let's wait
    p->waiters++;
5252
    /*
monty@mishka.local's avatar
monty@mishka.local committed
5253
      note - it must be while (), not do ... while () here
5254 5255 5256
      as p->state may be not DIRTY when we come here
    */
    while (p->state == DIRTY && syncing)
5257 5258 5259 5260 5261 5262 5263 5264 5265 5266 5267
      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
5268 5269 5270 5271
  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
5272 5273 5274 5275 5276 5277 5278 5279 5280 5281 5282 5283 5284
  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;

5285
  DBUG_ASSERT(syncing != active);
5286 5287 5288 5289 5290

  /*
    sit down and relax - this can take a while...
    note - no locks are held at this point
  */
5291
  err= my_msync(fd, syncing->start, 1, MS_SYNC);
5292

5293
  /* page is synced. let's move it to the pool */
5294
  pthread_mutex_lock(&LOCK_pool);
5295 5296 5297 5298 5299
  pool_last->next=syncing;
  pool_last=syncing;
  syncing->next=0;
  syncing->state= err ? ERROR : POOL;
  pthread_cond_broadcast(&syncing->cond);    // signal "sync done"
5300 5301 5302
  pthread_cond_signal(&COND_pool);           // in case somebody's waiting
  pthread_mutex_unlock(&LOCK_pool);

5303
  /* marking 'syncing' slot free */
5304
  pthread_mutex_lock(&LOCK_sync);
5305
  syncing=0;
5306 5307 5308 5309 5310
  pthread_cond_signal(&active->cond);        // wake up a new syncer
  pthread_mutex_unlock(&LOCK_sync);
  return err;
}

5311
/**
5312
  erase xid from the page, update page free space counters/pointers.
5313
  cookie points directly to the memory where xid was logged.
5314
*/
5315

5316 5317 5318 5319 5320 5321 5322 5323 5324 5325 5326 5327 5328
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);
5329
  if (p->free == p->size)               // the page is completely empty
5330 5331
    statistic_decrement(tc_log_cur_pages_used, &LOCK_status);
  if (p->waiters == 0)                 // the page is in pool and ready to rock
5332
    pthread_cond_signal(&COND_pool);   // ping ... for overflow()
5333 5334 5335 5336 5337
  pthread_mutex_unlock(&p->lock);
}

void TC_LOG_MMAP::close()
{
serg@serg.mylan's avatar
serg@serg.mylan committed
5338
  uint i;
5339 5340 5341 5342 5343 5344 5345 5346 5347
  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
5348
    for (i=0; i < npages; i++)
5349 5350 5351 5352 5353 5354 5355
    {
      if (pages[i].ptr == 0)
        break;
      pthread_mutex_destroy(&pages[i].lock);
      pthread_cond_destroy(&pages[i].cond);
    }
  case 3:
5356
    my_free((uchar*)pages, MYF(0));
5357
  case 2:
5358
    my_munmap((char*)data, (size_t)file_length);
5359 5360 5361 5362 5363 5364 5365 5366 5367 5368 5369 5370 5371 5372 5373 5374 5375 5376 5377 5378 5379 5380 5381 5382 5383
  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)
  {
5384
    sql_print_error("Recovery failed! You must enable "
5385 5386 5387
                    "exactly %d storage engines that support "
                    "two-phase commit protocol",
                    data[sizeof(tc_log_magic)]);
5388 5389 5390 5391
    goto err1;
  }

  if (hash_init(&xids, &my_charset_bin, tc_log_page_size/3, 0,
5392
                sizeof(my_xid), 0, 0, MYF(0)))
5393 5394 5395 5396 5397
    goto err1;

  for ( ; p < end_p ; p++)
  {
    for (my_xid *x=p->start; x < p->end; x++)
5398
      if (*x && my_hash_insert(&xids, (uchar *)x))
5399 5400 5401 5402 5403 5404 5405
        goto err2; // OOM
  }

  if (ha_recover(&xids))
    goto err2;

  hash_free(&xids);
5406
  bzero(data, (size_t)file_length);
5407 5408 5409 5410 5411 5412 5413 5414 5415 5416 5417
  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
5418 5419 5420 5421 5422
#endif

TC_LOG *tc_log;
TC_LOG_DUMMY tc_log_dummy;
TC_LOG_MMAP  tc_log_mmap;
5423

5424 5425
/**
  Perform heuristic recovery, if --tc-heuristic-recover was used.
5426

5427
  @note
5428 5429
    no matter whether heuristic recovery was successful or not
    mysqld must exit. So, return value is the same in both cases.
5430 5431 5432 5433 5434

  @retval
    0	no heuristic recovery was requested
  @retval
    1   heuristic recovery was performed
5435 5436
*/

5437 5438 5439 5440 5441 5442 5443 5444 5445 5446 5447 5448 5449
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 ******/
5450
#define TC_LOG_BINLOG MYSQL_BIN_LOG
5451

5452 5453 5454
/**
  @todo
  keep in-memory list of prepared transactions
5455 5456 5457 5458 5459
  (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!
*/

5460
int TC_LOG_BINLOG::open(const char *opt_name)
5461
{
5462 5463
  LOG_INFO log_info;
  int      error= 1;
5464 5465

  DBUG_ASSERT(total_ha_2pc > 1);
5466
  DBUG_ASSERT(opt_name && opt_name[0]);
5467 5468 5469 5470

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

5471 5472 5473 5474 5475 5476 5477
  if (!my_b_inited(&index_file))
  {
    /* There was a failure to open the index file, can't open the binlog */
    cleanup();
    return 1;
  }

5478
  if (using_heuristic_recover())
5479 5480 5481 5482
  {
    /* generate a new binlog to mask a corrupted one */
    open(opt_name, LOG_BIN, 0, WRITE_CACHE, 0, max_binlog_size, 0);
    cleanup();
5483
    return 1;
5484
  }
5485

5486
  if ((error= find_log_pos(&log_info, NullS, 1)))
5487
  {
5488 5489 5490 5491 5492
    if (error != LOG_INFO_EOF)
      sql_print_error("find_log_pos() failed (error: %d)", error);
    else
      error= 0;
    goto err;
5493
  }
5494 5495 5496 5497 5498

  {
    const char *errmsg;
    IO_CACHE    log;
    File        file;
5499 5500
    Log_event  *ev=0;
    Format_description_log_event fdle(BINLOG_VERSION);
5501
    char        log_name[FN_REFLEN];
5502 5503 5504 5505

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

5506
    do
5507
    {
5508 5509 5510 5511 5512 5513 5514
      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;
5515
    }
5516

5517
    if ((file= open_binlog(&log, log_name, &errmsg)) < 0)
5518 5519 5520 5521 5522
    {
      sql_print_error("%s", errmsg);
      goto err;
    }

5523 5524 5525
    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
5526 5527
    {
      sql_print_information("Recovering after a crash using %s", opt_name);
5528
      error= recover(&log, (Format_description_log_event *)ev);
serg@serg.mylan's avatar
serg@serg.mylan committed
5529
    }
5530 5531
    else
      error=0;
5532 5533 5534 5535 5536 5537 5538 5539 5540 5541

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

    if (error)
      goto err;
  }

err:
5542
  return error;
5543 5544
}

5545
/** This is called on shutdown, after ha_panic. */
5546
void TC_LOG_BINLOG::close()
5547 5548 5549 5550 5551 5552
{
  DBUG_ASSERT(prepared_xids==0);
  pthread_mutex_destroy(&LOCK_prep_xids);
  pthread_cond_destroy (&COND_prep_xids);
}

5553 5554 5555
/**
  @todo
  group commit
5556

5557 5558 5559 5560
  @retval
    0    error
  @retval
    1    success
5561
*/
5562
int TC_LOG_BINLOG::log_xid(THD *thd, my_xid xid)
5563
{
5564
  DBUG_ENTER("TC_LOG_BINLOG::log");
5565
  Xid_log_event xle(thd, xid);
5566
  binlog_trx_data *trx_data=
5567
    (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
5568 5569 5570 5571 5572
  /*
    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));
5573 5574
}

5575
void TC_LOG_BINLOG::unlog(ulong cookie, my_xid xid)
5576
{
5577
  pthread_mutex_lock(&LOCK_prep_xids);
5578
  DBUG_ASSERT(prepared_xids > 0);
5579 5580
  if (--prepared_xids == 0) {
    DBUG_PRINT("info", ("prepared_xids=%lu", prepared_xids));
5581
    pthread_cond_signal(&COND_prep_xids);
5582
  }
5583
  pthread_mutex_unlock(&LOCK_prep_xids);
5584
  rotate_and_purge(0);     // as ::write() did not rotate
5585 5586
}

5587
int TC_LOG_BINLOG::recover(IO_CACHE *log, Format_description_log_event *fdle)
5588 5589 5590 5591 5592 5593
{
  Log_event  *ev;
  HASH xids;
  MEM_ROOT mem_root;

  if (! fdle->is_valid() ||
5594
      hash_init(&xids, &my_charset_bin, TC_LOG_PAGE_SIZE/3, 0,
5595
                sizeof(my_xid), 0, 0, MYF(0)))
5596 5597
    goto err1;

5598
  init_alloc_root(&mem_root, TC_LOG_PAGE_SIZE, TC_LOG_PAGE_SIZE);
5599

5600 5601
  fdle->flags&= ~LOG_EVENT_BINLOG_IN_USE_F; // abort on the first error

5602 5603 5604 5605 5606
  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;
5607 5608
      uchar *x= (uchar *) memdup_root(&mem_root, (uchar*) &xev->xid,
                                      sizeof(xev->xid));
5609 5610 5611 5612 5613 5614 5615 5616 5617 5618 5619 5620 5621 5622 5623 5624 5625 5626 5627 5628 5629 5630 5631 5632 5633
      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;
}

5634 5635 5636 5637 5638 5639 5640 5641 5642 5643 5644 5645 5646 5647 5648 5649 5650 5651 5652 5653 5654 5655 5656

#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 */


5657
struct st_mysql_storage_engine binlog_storage_engine=
5658
{ MYSQL_HANDLERTON_INTERFACE_VERSION };
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
5659 5660 5661 5662

mysql_declare_plugin(binlog)
{
  MYSQL_STORAGE_ENGINE_PLUGIN,
5663 5664
  &binlog_storage_engine,
  "binlog",
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
5665
  "MySQL AB",
5666
  "This is a pseudo storage engine to represent the binlog in a transaction",
5667
  PLUGIN_LICENSE_GPL,
5668
  binlog_init, /* Plugin Init */
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
5669 5670
  NULL, /* Plugin Deinit */
  0x0100 /* 1.0 */,
5671 5672 5673
  NULL,                       /* status variables                */
  NULL,                       /* system variables                */
  NULL                        /* config options                  */
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
5674 5675
}
mysql_declare_plugin_end;