log.cc 168 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
    : at_least_one_stmt_committed(0), incident(FALSE), m_pending(0),
157
    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));
185 186 187 188
    if (pending())
    {
      delete pending();
    }
189 190
    set_pending(0);
    reinit_io_cache(&trans_log, WRITE_CACHE, pos, 0, 0);
191
    trans_log.end_of_file= max_binlog_cache_size;
192 193
    if (pos < before_stmt_pos)
      before_stmt_pos= MY_OFF_T_UNDEF;
194 195 196 197

    /*
      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
198
      to set the at_least_one_stmt_committed flag correctly. In other word, if
199 200 201 202
      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.
     */
203
    at_least_one_stmt_committed= (pos > 0);
204 205 206 207 208 209 210 211 212
  }

  /*
    Reset the entire contents of the transaction cache, emptying it
    completely.
   */
  void reset() {
    if (!empty())
      truncate(0);
213
    before_stmt_pos= MY_OFF_T_UNDEF;
214
    incident= FALSE;
215
    trans_log.end_of_file= max_binlog_cache_size;
216
    DBUG_ASSERT(empty());
217 218 219 220 221 222 223 224 225 226 227 228
  }

  Rows_log_event *pending() const
  {
    return m_pending;
  }

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

229
  IO_CACHE trans_log;                         // The transaction cache
230

231 232 233 234 235 236 237 238 239 240
  void set_incident(void)
  {
    incident= TRUE;
  }
  
  bool has_incident(void)
  {
    return(incident);
  }

241 242 243 244
  /**
    Boolean that is true if there is at least one statement in the
    transaction cache.
  */
245
  bool at_least_one_stmt_committed;
246
  bool incident;
247

248 249 250 251 252 253 254 255 256 257 258 259
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;
260 261
};

262
handlerton *binlog_hton;
263

264 265 266 267 268 269 270 271 272 273 274 275 276
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 */
  }
}

277

278 279 280 281 282 283 284 285 286 287 288 289
/* 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") :
290 291 292 293 294 295 296 297 298 299 300 301
                                  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))
302
        return QUERY_LOG_SLOW;
303 304
      return 0;
    }
305 306 307 308
  }
  return 0;
}

309

310 311 312 313 314 315 316 317 318
Log_to_csv_event_handler::Log_to_csv_event_handler()
{
}


Log_to_csv_event_handler::~Log_to_csv_event_handler()
{
}

319

320 321 322 323 324 325 326
void Log_to_csv_event_handler::cleanup()
{
  logger.is_log_tables_initialized= FALSE;
}

/* log event handlers */

327
/**
328 329
  Log command to the general log table

330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354
  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
355 356 357
*/

bool Log_to_csv_event_handler::
358
  log_general(THD *thd, time_t event_time, const char *user_host,
359 360
              uint user_host_len, int thread_id,
              const char *command_type, uint command_type_len,
361 362
              const char *sql_text, uint sql_text_len,
              CHARSET_INFO *client_cs)
363
{
364 365 366 367 368 369
  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
370
  uint field_index;
371 372 373 374 375
  Silence_log_table_errors error_handler;
  Open_tables_state open_tables_backup;
  ulonglong save_thd_options;
  bool save_time_zone_used;

376 377 378 379 380
  /*
    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;
381 382 383 384 385 386 387 388 389 390 391 392 393

  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;

394 395 396 397 398 399 400 401 402 403 404 405 406 407
  /*
    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;

408 409 410 411
  if (!(table= open_performance_schema_table(thd, & table_list,
                                             & open_tables_backup)))
    goto err;

412 413
  need_close= TRUE;

414
  if (table->file->extra(HA_EXTRA_MARK_AS_LOG_TABLE) ||
415 416 417 418 419 420 421
      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;
422

423 424 425 426 427 428 429 430 431
  /*
    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).
  */
432

433
  /* check that all columns exist */
434
  if (table->s->fields < 6)
435 436
    goto err;

437 438
  DBUG_ASSERT(table->field[0]->type() == MYSQL_TYPE_TIMESTAMP);

439 440
  ((Field_timestamp*) table->field[0])->store_timestamp((my_time_t)
                                                        event_time);
441

442 443 444 445
  /* 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) ||
446
      table->field[4]->store(command_type, command_type_len, client_cs))
447 448
    goto err;

449 450 451 452
  /*
    A positive return value in store() means truncation.
    Still logging a message in the log in this case.
  */
453
  table->field[5]->flags|= FIELDFLAG_HEX_ESCAPE;
454 455
  if (table->field[5]->store(sql_text, sql_text_len, client_cs) < 0)
    goto err;
456 457

  /* mark all fields as not null */
458 459 460 461 462
  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();
463

464 465 466 467 468 469
  /* 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();
  }

470
  /* log table entries are not replicated */
471
  if (table->file->ha_write_row(table->record[0]))
472
    goto err;
473

474
  result= FALSE;
475

476
err:
477
  if (result && !thd->killed)
478 479
    sql_print_error("Failed to write to mysql.general_log: %s",
                    error_handler.message());
480

481 482 483 484 485 486 487 488 489 490 491 492 493
  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;
494 495 496 497 498 499 500
}


/*
  Log a query to the slow log table

  SYNOPSIS
501
    log_slow()
502 503 504 505 506 507
    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
508 509
    query_time        Amount of time the query took to execute (in microseconds)
    lock_time         Amount of time the query was locked (in microseconds)
510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528
    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,
529
           ulonglong query_utime, ulonglong lock_utime, bool is_command,
530 531
           const char *sql_text, uint sql_text_len)
{
532 533 534 535 536
  TABLE_LIST table_list;
  TABLE *table;
  bool result= TRUE;
  bool need_close= FALSE;
  bool need_rnd_end= FALSE;
537
  Silence_log_table_errors error_handler;
538
  Open_tables_state open_tables_backup;
539
  CHARSET_INFO *client_cs= thd->variables.character_set_client;
540
  bool save_time_zone_used;
541
  DBUG_ENTER("Log_to_csv_event_handler::log_slow");
542

543
  thd->push_internal_handler(& error_handler);
544 545 546 547 548 549
  /*
    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;

550 551 552 553 554 555 556 557 558
  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;

559 560 561
  if (!(table= open_performance_schema_table(thd, & table_list,
                                             & open_tables_backup)))
    goto err;
562 563 564

  need_close= TRUE;

565
  if (table->file->extra(HA_EXTRA_MARK_AS_LOG_TABLE) ||
566 567 568 569 570 571 572
      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;
573

574
  restore_record(table, s->default_values);    // Get empty record
575

576 577 578 579
  /* check that all columns exist */
  if (table->s->fields < 11)
    goto err;

580 581
  /* store the time and user values */
  DBUG_ASSERT(table->field[0]->type() == MYSQL_TYPE_TIMESTAMP);
582 583
  ((Field_timestamp*) table->field[0])->store_timestamp((my_time_t)
                                                        current_time);
584 585
  if (table->field[1]->store(user_host, user_host_len, client_cs))
    goto err;
586 587 588

  if (query_start_arg)
  {
589 590
    longlong query_time= (longlong) (query_utime/1000000);
    longlong lock_time=  (longlong) (lock_utime/1000000);
591 592 593 594 595
    /*
      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)
    */
596
    MYSQL_TIME t;
597 598
    t.neg= 0;

599
    /* fill in query_time field */
600 601
    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))
602
      goto err;
603
    /* lock_time */
604 605
    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))
606
      goto err;
607
    /* rows_sent */
608 609
    if (table->field[4]->store((longlong) thd->sent_row_count, TRUE))
      goto err;
610
    /* rows_examined */
611 612
    if (table->field[5]->store((longlong) thd->examined_row_count, TRUE))
      goto err;
613 614 615 616 617 618 619 620
  }
  else
  {
    table->field[2]->set_null();
    table->field[3]->set_null();
    table->field[4]->set_null();
    table->field[5]->set_null();
  }
621
  /* fill database field */
622
  if (thd->db)
623
  {
624 625
    if (table->field[6]->store(thd->db, thd->db_length, client_cs))
      goto err;
626 627
    table->field[6]->set_notnull();
  }
628

629
  if (thd->stmt_depends_on_first_successful_insert_id_in_prev_stmt)
630
  {
631 632 633 634 635
    if (table->
        field[7]->store((longlong)
                        thd->first_successful_insert_id_in_prev_stmt_for_binlog,
                        TRUE))
      goto err;
636 637
    table->field[7]->set_notnull();
  }
638

639 640 641 642 643 644 645
  /*
    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)
646
  {
647 648 649 650
    if (table->
        field[8]->store((longlong)
          thd->auto_inc_intervals_in_cur_stmt_for_binlog.minimum(), TRUE))
      goto err;
651 652
    table->field[8]->set_notnull();
  }
653

654 655
  if (table->field[9]->store((longlong) server_id, TRUE))
    goto err;
656
  table->field[9]->set_notnull();
657

658 659 660 661 662 663
  /*
    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)
664 665
    goto err;

666
  /* log table entries are not replicated */
667
  if (table->file->ha_write_row(table->record[0]))
668
    goto err;
669

670
  result= FALSE;
671

672
err:
673 674
  thd->pop_internal_handler();

675
  if (result && !thd->killed)
676 677
    sql_print_error("Failed to write to mysql.slow_log: %s",
                    error_handler.message());
678

679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720
  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);
721 722 723 724 725 726 727 728
  if (table)
  {
    result= 0;
    close_performance_schema_table(thd, & open_tables_backup);
  }
  else
    result= 1;

729
  DBUG_RETURN(result);
730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753
}

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();
}


754
/** Wrapper around MYSQL_LOG::write() for slow log. */
755 756 757 758

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,
759
           ulonglong query_utime, ulonglong lock_utime, bool is_command,
760 761
           const char *sql_text, uint sql_text_len)
{
762 763 764 765 766 767 768 769
  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;
770 771 772
}


773
/**
774 775 776 777 778
   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::
779
  log_general(THD *thd, time_t event_time, const char *user_host,
780 781
              uint user_host_len, int thread_id,
              const char *command_type, uint command_type_len,
782 783
              const char *sql_text, uint sql_text_len,
              CHARSET_INFO *client_cs)
784
{
785 786 787 788 789 790 791
  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;
792 793 794 795 796 797 798 799
}


bool Log_to_file_event_handler::init()
{
  if (!is_initialized)
  {
    if (opt_slow_log)
800
      mysql_slow_log.open_slow_log(sys_var_slow_log_path.value);
801 802

    if (opt_log)
803
      mysql_log.open_query_log(sys_var_general_log_path.value);
804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820

    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 */
821
  if (opt_log)
822
    mysql_log.reopen_file();
823
  if (opt_slow_log)
824
    mysql_slow_log.reopen_file();
825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846
}

/*
  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;
847
  Log_event_handler **current_handler;
848 849

  /* currently we don't need locking here as there is no error_log table */
850
  for (current_handler= error_log_handler_list ; *current_handler ;)
851 852 853 854 855 856
    error= (*current_handler++)->log_error(level, format, args) || error;

  return error;
}


petr@mysql.com's avatar
petr@mysql.com committed
857
void LOGGER::cleanup_base()
858 859
{
  DBUG_ASSERT(inited == 1);
860
  rwlock_destroy(&LOCK_logger);
861
  if (table_log_handler)
petr@mysql.com's avatar
petr@mysql.com committed
862
  {
863
    table_log_handler->cleanup();
petr@mysql.com's avatar
petr@mysql.com committed
864
    delete table_log_handler;
865
    table_log_handler= NULL;
petr@mysql.com's avatar
petr@mysql.com committed
866
  }
867 868 869 870 871
  if (file_log_handler)
    file_log_handler->cleanup();
}


petr@mysql.com's avatar
petr@mysql.com committed
872 873 874 875
void LOGGER::cleanup_end()
{
  DBUG_ASSERT(inited == 1);
  if (file_log_handler)
876
  {
petr@mysql.com's avatar
petr@mysql.com committed
877
    delete file_log_handler;
878 879 880
    file_log_handler=NULL;
  }
  inited= 0;
petr@mysql.com's avatar
petr@mysql.com committed
881 882 883
}


884
/**
885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901
  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 */
902
  init_error_log(LOG_FILE);
903 904

  file_log_handler->init_pthread_objects();
905
  my_rwlock_init(&LOCK_logger, NULL);
906 907 908 909 910 911 912 913 914 915 916 917 918 919
}


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;
}


920 921 922
bool LOGGER::flush_logs(THD *thd)
{
  int rc= 0;
923

924
  /*
925 926
    Now we lock logger, as nobody should be able to use logging routines while
    log tables are closed
927
  */
928
  logger.lock_exclusive();
929 930 931

  /* reopen log files */
  file_log_handler->flush();
932

933 934
  /* end of log flush */
  logger.unlock();
935
  return rc;
936 937 938 939 940 941 942 943 944
}


/*
  Log slow query with all enabled log event handlers

  SYNOPSIS
    slow_log_print()

945 946 947 948
    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)
949 950

  RETURN
951 952
    FALSE   OK
    TRUE    error occured
953 954 955
*/

bool LOGGER::slow_log_print(THD *thd, const char *query, uint query_length,
956 957
                            ulonglong current_utime)

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

967
  DBUG_ASSERT(thd->enable_slow_log);
968 969 970 971 972 973
  /*
    Print the message to the buffer if we have slow log enabled
  */

  if (*slow_log_handler_list)
  {
974
    time_t current_time;
975

976
    /* do not log slow queries from replication threads */
977
    if (thd->slave_thread && !opt_log_slow_slave_statements)
978 979
      return 0;

980
    lock_shared();
981 982 983 984 985
    if (!opt_slow_log)
    {
      unlock();
      return 0;
    }
986 987

    /* fill in user_host value: the format is "%s[%s] @ %s [%s]" */
988 989 990 991 992 993 994 995 996
    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)
997
    {
998 999 1000 1001
      query_utime= (current_utime - thd->start_utime);
      lock_utime=  (thd->utime_after_lock - thd->start_utime);
    }
    else
1002
    {
1003
      query_utime= lock_utime= 0;
1004 1005 1006 1007 1008 1009 1010 1011 1012
    }

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

1013 1014
    for (current_handler= slow_log_handler_list; *current_handler ;)
      error= (*current_handler++)->log_slow(thd, current_time, thd->start_time,
1015
                                            user_host_buff, user_host_len,
1016
                                            query_utime, lock_utime, is_command,
1017 1018 1019 1020 1021 1022 1023
                                            query, query_length) || error;

    unlock();
  }
  return error;
}

1024 1025
bool LOGGER::general_log_write(THD *thd, enum enum_server_command command,
                               const char *query, uint query_length)
1026 1027 1028
{
  bool error= FALSE;
  Log_event_handler **current_handler= general_log_handler_list;
1029
  char user_host_buff[MAX_USER_HOST_SIZE + 1];
1030 1031 1032
  Security_context *sctx= thd->security_ctx;
  uint user_host_len= 0;
  time_t current_time;
1033

1034
  DBUG_ASSERT(thd);
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
      log_general(thd, current_time, user_host_buff,
1053
                  user_host_len, thd->thread_id,
1054 1055 1056 1057 1058 1059
                  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

    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;
}

1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278
/** 
    This function checks if a transactional talbe was updated by the
    current statement.

    @param thd The client thread that executed the current statement.
    @return
      @c true if a transactional table was updated, @false otherwise.
*/
static bool stmt_has_updated_trans_table(THD *thd)
{
  Ha_trx_info *ha_info;

1279
  for (ha_info= thd->transaction.stmt.ha_list; ha_info && ha_info->is_started(); ha_info= ha_info->next())
1280 1281 1282 1283 1284 1285
  {
    if (ha_info->is_trx_read_write() && ha_info->ht() != binlog_hton)
      return (TRUE);
  }
  return (FALSE);
}
1286

1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306
 /*
  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);
1307
  if (thd_get_ha_data(thd, binlog_hton) == NULL)
1308 1309
    thd->binlog_setup_trx_data();
  binlog_trx_data *const trx_data=
1310
    (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
1311 1312
  DBUG_ASSERT(mysql_bin_log.is_open());
  *pos= trx_data->position();
1313
  DBUG_PRINT("return", ("*pos: %lu", (ulong) *pos));
1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336
  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");
1337
  DBUG_PRINT("enter", ("pos: %lu", (ulong) pos));
1338

1339
  DBUG_ASSERT(thd_get_ha_data(thd, binlog_hton) != NULL);
1340 1341 1342 1343
  /* Only true if binlog_trans_log_savepos() wasn't called before */
  DBUG_ASSERT(pos != ~(my_off_t) 0);

  binlog_trx_data *const trx_data=
1344
    (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
1345 1346 1347 1348 1349
  trx_data->truncate(pos);
  DBUG_VOID_RETURN;
}


1350 1351
/*
  this function is mostly a placeholder.
1352
  conceptually, binlog initialization (now mostly done in MYSQL_BIN_LOG::open)
1353 1354 1355
  should be moved here.
*/

1356
int binlog_init(void *p)
1357
{
1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368
  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;
1369
  return 0;
1370 1371
}

1372
static int binlog_close_connection(handlerton *hton, THD *thd)
1373
{
1374
  binlog_trx_data *const trx_data=
1375
    (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
1376
  DBUG_ASSERT(trx_data->empty());
1377
  thd_set_ha_data(thd, binlog_hton, NULL);
1378
  trx_data->~binlog_trx_data();
1379
  my_free((uchar*)trx_data, MYF(0));
1380 1381 1382
  return 0;
}

1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405
/*
  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).
 */
1406
static int
1407 1408
binlog_end_trans(THD *thd, binlog_trx_data *trx_data,
                 Log_event *end_ev, bool all)
1409
{
1410
  DBUG_ENTER("binlog_end_trans");
1411 1412
  int error=0;
  IO_CACHE *trans_log= &trx_data->trans_log;
1413 1414
  DBUG_PRINT("enter", ("transaction: %s  end_ev: 0x%lx",
                       all ? "all" : "stmt", (long) end_ev));
1415 1416 1417
  DBUG_PRINT("info", ("thd->options={ %s%s}",
                      FLAGSTR(thd->options, OPTION_NOT_AUTOCOMMIT),
                      FLAGSTR(thd->options, OPTION_BEGIN)));
1418

1419 1420 1421 1422 1423 1424
  /*
    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.
  */
1425
  if (end_ev != NULL)
1426
  {
1427 1428
    if (thd->binlog_flush_pending_rows_event(TRUE))
      DBUG_RETURN(1);
1429
    /*
1430 1431 1432 1433
      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.

1434 1435 1436 1437 1438
      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.
     */
1439 1440
    error= mysql_bin_log.write(thd, &trx_data->trans_log, end_ev,
                               trx_data->has_incident());
1441 1442 1443 1444
    trx_data->reset();

    statistic_increment(binlog_cache_use, &LOCK_status);
    if (trans_log->disk_writes != 0)
1445
    {
1446 1447
      statistic_increment(binlog_cache_disk_use, &LOCK_status);
      trans_log->disk_writes= 0;
1448
    }
1449 1450 1451
  }
  else
  {
1452 1453 1454
    /*
      If rolling back an entire transaction or a single statement not
      inside a transaction, we reset the transaction cache.
1455

1456 1457 1458
      If rolling back a statement in a transaction, we truncate the
      transaction cache to remove the statement.
     */
1459
    thd->binlog_remove_pending_rows_event(TRUE);
1460
    if (all || !(thd->options & (OPTION_BEGIN | OPTION_NOT_AUTOCOMMIT)))
1461 1462
    {
      if (trx_data->has_incident())
He Zhenxing's avatar
He Zhenxing committed
1463
        error= mysql_bin_log.write_incident(thd, TRUE);
1464
      trx_data->reset();
1465
    }
1466 1467
    else                                        // ...statement
      trx_data->truncate(trx_data->before_stmt_pos);
1468
  }
1469

1470
  DBUG_ASSERT(thd->binlog_get_pending_rows_event() == NULL);
1471
  DBUG_RETURN(error);
1472 1473
}

1474
static int binlog_prepare(handlerton *hton, THD *thd, bool all)
1475 1476 1477 1478 1479
{
  /*
    do nothing.
    just pretend we can do 2pc, so that MySQL won't
    switch to 1pc.
1480
    real work will be done in MYSQL_BIN_LOG::log_xid()
1481 1482 1483 1484
  */
  return 0;
}

1485 1486 1487 1488 1489 1490 1491 1492
/**
  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.
1493 1494
  @param all   This is @c true if this is a real transaction commit, and
               @false otherwise.
1495 1496 1497

  @see handlerton::commit
*/
1498
static int binlog_commit(handlerton *hton, THD *thd, bool all)
1499
{
1500
  int error= 0;
1501
  DBUG_ENTER("binlog_commit");
1502
  binlog_trx_data *const trx_data=
1503
    (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
1504

1505
  if (trx_data->empty())
1506
  {
1507
    // we're here because trans_log was flushed in MYSQL_BIN_LOG::log_xid()
1508
    trx_data->reset();
1509 1510
    DBUG_RETURN(0);
  }
1511

1512
  /*
1513
    We flush the cache if:
1514

1515 1516
     - we are committing a transaction or;
     - no statement was committed before and just non-transactional
1517
       tables were updated.
1518

1519
    Otherwise, we collect the changes.
1520
  */
1521
  DBUG_PRINT("debug",
1522
             ("all: %d, empty: %s, all.modified_non_trans_table: %s, stmt.modified_non_trans_table: %s",
1523 1524 1525 1526
              all,
              YESNO(trx_data->empty()),
              YESNO(thd->transaction.all.modified_non_trans_table),
              YESNO(thd->transaction.stmt.modified_non_trans_table)));
1527 1528 1529
  if (ending_trans(thd, all) ||
      (trans_has_no_stmt_committed(thd, all) &&
       !stmt_has_updated_trans_table(thd) && stmt_has_updated_non_trans_table(thd)))
1530
  {
He Zhenxing's avatar
He Zhenxing committed
1531
    Query_log_event qev(thd, STRING_WITH_LEN("COMMIT"), TRUE, TRUE, 0);
1532
    error= binlog_end_trans(thd, trx_data, &qev, all);
1533
  }
1534

1535 1536
  trx_data->at_least_one_stmt_committed = my_b_tell(&trx_data->trans_log) > 0;

1537 1538 1539
  if (!all)
    trx_data->before_stmt_pos = MY_OFF_T_UNDEF; // part of the stmt commit
  DBUG_RETURN(error);
1540 1541
}

1542 1543 1544 1545 1546 1547 1548 1549 1550 1551
/**
  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.
1552 1553
  @param all   This is @c true if this is a real transaction rollback, and
               @false otherwise.
1554 1555 1556

  @see handlerton::rollback
*/
1557
static int binlog_rollback(handlerton *hton, THD *thd, bool all)
1558 1559
{
  DBUG_ENTER("binlog_rollback");
1560 1561
  int error=0;
  binlog_trx_data *const trx_data=
1562
    (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
1563 1564 1565 1566 1567 1568

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

1569 1570 1571 1572
  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)));
1573
  if (mysql_bin_log.check_write_error(thd))
1574
  {
1575
    /*
1576 1577 1578 1579
      "all == true" means that a "rollback statement" triggered the error and
      this function was called. However, this must not happen as a rollback
      is written directly to the binary log. And in auto-commit mode, a single
      statement that is rolled back has the flag all == false.
1580
    */
1581
    DBUG_ASSERT(!all);
1582
    /*
1583 1584 1585 1586 1587 1588 1589 1590
      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.
1591
    */
1592
    if ((stmt_has_updated_non_trans_table(thd) ||
1593 1594 1595
        (thd->options & OPTION_KEEP_LOG)) &&
        mysql_bin_log.check_write_error(thd))
      trx_data->set_incident();
1596
    error= binlog_end_trans(thd, trx_data, 0, all);
1597
  }
1598 1599 1600 1601
  else
  {
   /*
      We flush the cache with a rollback, wrapped in a beging/rollback if:
1602 1603
        . aborting a transaction that modified a non-transactional table or
          the OPTION_KEEP_LOG is activate.
1604
        . aborting a statement that modified both transactional and
1605 1606
          non-transactional tables but which is not in the boundaries of any
          transaction or there was no early change;
1607
    */
1608 1609 1610 1611 1612 1613
    if ((ending_trans(thd, all) &&
        (trans_has_updated_non_trans_table(thd) ||
         (thd->options & OPTION_KEEP_LOG))) ||
        (trans_has_no_stmt_committed(thd, all) &&
         stmt_has_updated_non_trans_table(thd) &&
         thd->current_stmt_binlog_row_based))
1614 1615 1616 1617 1618 1619 1620 1621
    {
      Query_log_event qev(thd, STRING_WITH_LEN("ROLLBACK"), TRUE, TRUE, 0);
      error= binlog_end_trans(thd, trx_data, &qev, all);
    }
    /*
      Otherwise, we simply truncate the cache as there is no change on
      non-transactional tables as follows.
    */
1622 1623
    else if (ending_trans(thd, all) ||
             (!(thd->options & OPTION_KEEP_LOG) && !stmt_has_updated_non_trans_table(thd)))
1624 1625
      error= binlog_end_trans(thd, trx_data, 0, all);
  }
1626 1627
  if (!all)
    trx_data->before_stmt_pos = MY_OFF_T_UNDEF; // part of the stmt rollback
1628 1629 1630
  DBUG_RETURN(error);
}

1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643
/**
  Cleanup the cache.

  @param thd   The client thread that wants to clean up the cache.
*/
void MYSQL_BIN_LOG::reset_gathered_updates(THD *thd)
{
  binlog_trx_data *const trx_data=
    (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);

  trx_data->reset();
}

1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681
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);
}

1682 1683 1684 1685 1686 1687
/**
  @note
  How do we handle this (unlikely but legal) case:
  @verbatim
    [transaction] + [update to non-trans table] + [rollback to savepoint] ?
  @endverbatim
1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705
  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).
*/

1706
static int binlog_savepoint_set(handlerton *hton, THD *thd, void *sv)
1707 1708 1709
{
  DBUG_ENTER("binlog_savepoint_set");

1710
  binlog_trans_log_savepos(thd, (my_off_t*) sv);
1711
  /* Write it to the binary log */
1712

1713 1714 1715 1716
  String log_query;
  if (log_query.append(STRING_WITH_LEN("SAVEPOINT ")) ||
      log_query.append(thd->lex->ident.str, thd->lex->ident.length))
    DBUG_RETURN(1);
1717
  int errcode= query_error_code(thd, thd->killed == THD::NOT_KILLED);
1718 1719 1720
  Query_log_event qinfo(thd, log_query.c_ptr_safe(), log_query.length(),
                        TRUE, TRUE, errcode);
  DBUG_RETURN(mysql_bin_log.write(&qinfo));
1721 1722
}

1723
static int binlog_savepoint_rollback(handlerton *hton, THD *thd, void *sv)
1724 1725 1726 1727 1728 1729 1730 1731
{
  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.
  */
1732
  if (unlikely(trans_has_updated_non_trans_table(thd) || 
1733
               (thd->options & OPTION_KEEP_LOG)))
1734
  {
1735 1736 1737 1738
    String log_query;
    if (log_query.append(STRING_WITH_LEN("ROLLBACK TO ")) ||
        log_query.append(thd->lex->ident.str, thd->lex->ident.length))
      DBUG_RETURN(1);
1739
    int errcode= query_error_code(thd, thd->killed == THD::NOT_KILLED);
1740 1741 1742
    Query_log_event qinfo(thd, log_query.c_ptr_safe(), log_query.length(),
                          TRUE, TRUE, errcode);
    DBUG_RETURN(mysql_bin_log.write(&qinfo));
1743
  }
1744
  binlog_trans_log_truncate(thd, *(my_off_t*)sv);
1745 1746 1747
  DBUG_RETURN(0);
}

1748

1749 1750 1751 1752 1753
int check_binlog_magic(IO_CACHE* log, const char** errmsg)
{
  char magic[4];
  DBUG_ASSERT(my_b_tell(log) == 0);

1754
  if (my_b_read(log, (uchar*) magic, sizeof(magic)))
1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768
  {
    *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;
}

1769

1770 1771 1772 1773 1774
File open_binlog(IO_CACHE *log, const char *log_file_name, const char **errmsg)
{
  File file;
  DBUG_ENTER("open_binlog");

1775 1776
  if ((file = my_open(log_file_name, O_RDONLY | O_BINARY | O_SHARE, 
                      MYF(MY_WME))) < 0)
1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802
  {
    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
1803

rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1804 1805
#ifdef __NT__
static int eventSource = 0;
serg@serg.mylan's avatar
serg@serg.mylan committed
1806

1807
static void setup_windows_event_source()
monty@mysql.com's avatar
monty@mysql.com committed
1808
{
1809
  HKEY    hRegKey= NULL;
monty@mysql.com's avatar
monty@mysql.com committed
1810 1811 1812
  DWORD   dwError= 0;
  TCHAR   szPath[MAX_PATH];
  DWORD dwTypes;
1813

monty@mysql.com's avatar
monty@mysql.com committed
1814 1815 1816
  if (eventSource)               // Ensure that we are only called once
    return;
  eventSource= 1;
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1817

monty@mysql.com's avatar
monty@mysql.com committed
1818
  // Create the event source registry key
1819
  dwError= RegCreateKey(HKEY_LOCAL_MACHINE,
monty@mysql.com's avatar
monty@mysql.com committed
1820 1821
                          "SYSTEM\\CurrentControlSet\\Services\\EventLog\\Application\\MySQL", 
                          &hRegKey);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1822

monty@mysql.com's avatar
monty@mysql.com committed
1823 1824
  /* Name of the PE module that contains the message resource */
  GetModuleFileName(NULL, szPath, MAX_PATH);
serg@serg.mylan's avatar
serg@serg.mylan committed
1825

rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1826
  /* Register EventMessageFile */
1827
  dwError = RegSetValueEx(hRegKey, "EventMessageFile", 0, REG_EXPAND_SZ,
1828
                          (PBYTE) szPath, (DWORD) (strlen(szPath) + 1));
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1829

monty@mysql.com's avatar
monty@mysql.com committed
1830 1831 1832 1833 1834
  /* 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
1835

monty@mysql.com's avatar
monty@mysql.com committed
1836
  RegCloseKey(hRegKey);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1837 1838
}

monty@mysql.com's avatar
monty@mysql.com committed
1839
#endif /* __NT__ */
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1840 1841


1842 1843 1844 1845 1846 1847 1848 1849
/**
  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
1850 1851 1852

static int find_uniq_filename(char *name)
{
1853 1854 1855 1856
  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
1857
  reg1 struct fileinfo *file_info;
1858
  ulong                 max_found=0;
1859 1860
  size_t		buf_length, length;
  char			*start, *end;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1861 1862
  DBUG_ENTER("find_uniq_filename");

1863 1864 1865
  length= dirname_part(buff, name, &buf_length);
  start=  name + length;
  end=    strend(start);
1866

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1867
  *end='.';
1868
  length= (size_t) (end-start+1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1869 1870 1871 1872 1873 1874 1875 1876 1877

  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++)
  {
1878
    if (memcmp(file_info->name, start, length) == 0 &&
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1879 1880 1881 1882 1883 1884 1885 1886
	test_if_number(file_info->name+length, &number,0))
    {
      set_if_bigger(max_found,(ulong) number);
    }
  }
  my_dirend(dir_info);

  *end++='.';
1887
  sprintf(end,"%06ld",max_found+1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1888 1889 1890
  DBUG_RETURN(0);
}

1891

1892
void MYSQL_LOG::init(enum_log_type log_type_arg,
1893
                     enum cache_type io_cache_type_arg)
1894 1895 1896 1897 1898 1899 1900 1901 1902
{
  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;
}


1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918
bool MYSQL_LOG::init_and_set_log_file_name(const char *log_name,
                                           const char *new_name,
                                           enum_log_type log_type_arg,
                                           enum cache_type io_cache_type_arg)
{
  init(log_type_arg, io_cache_type_arg);

  if (new_name && !strmov(log_file_name, new_name))
    return TRUE;
  else if (!new_name && generate_new_name(log_file_name, log_name))
    return TRUE;

  return FALSE;
}


1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956
/*
  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;

  if (!(name= my_strdup(log_name, MYF(MY_WME))))
  {
    name= (char *)log_name; // for the error message
    goto err;
  }

1957 1958
  if (init_and_set_log_file_name(name, new_name,
                                 log_type_arg, io_cache_type_arg))
1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975
    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;

1976
  if (log_type == LOG_NORMAL)
1977 1978
  {
    char *end;
1979
    int len=my_snprintf(buff, sizeof(buff), "%s, Version: %s (%s). "
1980
#ifdef EMBEDDED_LIBRARY
1981 1982
                        "embedded library\n",
                        my_progname, server_version, MYSQL_COMPILATION_COMMENT
1983 1984
#elif __NT__
			"started with:\nTCP Port: %d, Named Pipe: %s\n",
1985 1986
                        my_progname, server_version, MYSQL_COMPILATION_COMMENT,
                        mysqld_port, mysqld_unix_port
1987 1988
#else
			"started with:\nTcp port: %d  Unix socket: %s\n",
1989 1990
                        my_progname, server_version, MYSQL_COMPILATION_COMMENT,
                        mysqld_port, mysqld_unix_port
1991 1992 1993 1994
#endif
                       );
    end= strnmov(buff + len, "Time                 Id Command    Argument\n",
                 sizeof(buff) - len);
1995
    if (my_b_write(&log_file, (uchar*) buff, (uint) (end-buff)) ||
1996 1997 1998 1999
	flush_io_cache(&log_file))
      goto err;
  }

2000
  log_state= LOG_OPENED;
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
  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);
2012
  log_state= LOG_CLOSED;
2013 2014 2015
  DBUG_RETURN(1);
}

2016
MYSQL_LOG::MYSQL_LOG()
serg@janus.mylan's avatar
serg@janus.mylan committed
2017 2018
  : 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
2019 2020
{
  /*
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2021 2022
    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
2023
    called only in main(). Doing initialization here would make it happen
2024
    before main().
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2025
  */
2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053
  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));
2054
  if (log_state == LOG_OPENED)
2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070
  {
    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);
    }
  }

2071
  log_state= (exiting & LOG_CLOSE_TO_BE_OPENED) ? LOG_TO_BE_OPENED : LOG_CLOSED;
2072 2073
  safeFree(name);
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2074 2075
}

2076
/** This is called only once. */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2077

monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
2078
void MYSQL_LOG::cleanup()
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2079
{
2080
  DBUG_ENTER("cleanup");
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2081
  if (inited)
2082
  {
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
2083
    inited= 0;
2084
    (void) pthread_mutex_destroy(&LOCK_log);
2085
    close(0);
2086
  }
2087
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2088 2089
}

2090

bk@work.mysql.com's avatar
bk@work.mysql.com committed
2091
int MYSQL_LOG::generate_new_name(char *new_name, const char *log_name)
2092
{
2093 2094
  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
2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107
  {
    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;
}

2108

2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120
/*
  Reopen the log file

  SYNOPSIS
    reopen_file()

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


2121
void MYSQL_QUERY_LOG::reopen_file()
2122
{
2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138
  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);

  /*
2139
     Note that at this point, log_state != LOG_CLOSED (important for is_open()).
2140 2141
  */

2142
  open(save_name, log_type, 0, io_cache_type);
2143 2144 2145 2146
  my_free(save_name, MYF(0));

  pthread_mutex_unlock(&LOCK_log);

2147
  DBUG_VOID_RETURN;
2148 2149
}

2150

2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175
/*
  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
*/

2176 2177 2178 2179
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
2180
{
2181 2182
  char buff[32];
  uint length= 0;
2183
  char local_time_buff[MAX_TIME_SIZE];
2184 2185 2186
  struct tm start;
  uint time_buff_len= 0;

2187 2188
  (void) pthread_mutex_lock(&LOCK_log);

2189 2190 2191
  /* Test if someone closed between the is_open test and lock */
  if (is_open())
  {
2192 2193 2194
    /* for testing output of timestamp and thread id */
    DBUG_EXECUTE_IF("reset_log_last_time", last_time= 0;);

2195 2196 2197 2198 2199 2200 2201
    /* 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);

2202
        time_buff_len= my_snprintf(local_time_buff, MAX_TIME_SIZE,
2203
                                   "%02d%02d%02d %2d:%02d:%02d\t",
2204 2205 2206 2207
                                   start.tm_year % 100, start.tm_mon + 1,
                                   start.tm_mday, start.tm_hour,
                                   start.tm_min, start.tm_sec);

2208
        if (my_b_write(&log_file, (uchar*) local_time_buff, time_buff_len))
2209 2210 2211
          goto err;
      }
      else
2212
        if (my_b_write(&log_file, (uchar*) "\t\t" ,2) < 0)
2213 2214
          goto err;

2215
      /* command_type, thread_id */
2216
      length= my_snprintf(buff, 32, "%5ld ", (long) thread_id);
2217

2218
    if (my_b_write(&log_file, (uchar*) buff, length))
2219 2220
      goto err;

2221
    if (my_b_write(&log_file, (uchar*) command_type, command_type_len))
2222 2223
      goto err;

2224
    if (my_b_write(&log_file, (uchar*) "\t", 1))
2225 2226 2227
      goto err;

    /* sql_text */
2228
    if (my_b_write(&log_file, (uchar*) sql_text, sql_text_len))
2229 2230
      goto err;

2231
    if (my_b_write(&log_file, (uchar*) "\n", 1) ||
2232 2233 2234 2235
        flush_io_cache(&log_file))
      goto err;
  }

2236
  (void) pthread_mutex_unlock(&LOCK_log);
2237 2238 2239 2240 2241 2242 2243 2244
  return FALSE;
err:

  if (!write_error)
  {
    write_error= 1;
    sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
  }
2245
  (void) pthread_mutex_unlock(&LOCK_log);
2246
  return TRUE;
2247 2248
}

2249

2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261
/*
  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
2262 2263
    query_utime       Amount of time the query took to execute (in microseconds)
    lock_utime        Amount of time the query was locked (in microseconds)
2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278
    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
*/

2279 2280
bool MYSQL_QUERY_LOG::write(THD *thd, time_t current_time,
                            time_t query_start_arg, const char *user_host,
2281 2282
                            uint user_host_len, ulonglong query_utime,
                            ulonglong lock_utime, bool is_command,
2283
                            const char *sql_text, uint sql_text_len)
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2284
{
2285
  bool error= 0;
2286
  DBUG_ENTER("MYSQL_QUERY_LOG::write");
2287

2288 2289
  (void) pthread_mutex_lock(&LOCK_log);

2290
  if (!is_open())
2291 2292
  {
    (void) pthread_mutex_unlock(&LOCK_log);
2293
    DBUG_RETURN(0);
2294
  }
2295 2296 2297 2298 2299

  if (is_open())
  {						// Safety agains reopen
    int tmp_errno= 0;
    char buff[80], *end;
2300
    char query_time_buff[22+7], lock_time_buff[22+7];
2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318
    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 */
2319
        if (my_b_write(&log_file, (uchar*) buff, buff_len))
2320 2321
          tmp_errno= errno;
      }
2322 2323
      const uchar uh[]= "# User@Host: ";
      if (my_b_write(&log_file, uh, sizeof(uh) - 1))
2324
        tmp_errno= errno;
2325
      if (my_b_write(&log_file, (uchar*) user_host, user_host_len))
2326
        tmp_errno= errno;
2327
      if (my_b_write(&log_file, (uchar*) "\n", 1))
2328 2329 2330
        tmp_errno= errno;
    }
    /* For slow query log */
2331 2332
    sprintf(query_time_buff, "%.6f", ulonglong2double(query_utime)/1000000.0);
    sprintf(lock_time_buff,  "%.6f", ulonglong2double(lock_utime)/1000000.0);
2333
    if (my_b_printf(&log_file,
2334
                    "# Query_time: %s  Lock_time: %s"
2335
                    " Rows_sent: %lu  Rows_examined: %lu\n",
2336
                    query_time_buff, lock_time_buff,
2337 2338 2339 2340 2341 2342 2343 2344 2345
                    (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);
    }
2346
    if (thd->stmt_depends_on_first_successful_insert_id_in_prev_stmt)
2347 2348
    {
      end=strmov(end, ",last_insert_id=");
2349 2350 2351
      end=longlong10_to_str((longlong)
                            thd->first_successful_insert_id_in_prev_stmt_for_binlog,
                            end, -10);
2352 2353
    }
    // Save value if we do an insert.
2354
    if (thd->auto_inc_intervals_in_cur_stmt_for_binlog.nb_elements() > 0)
2355 2356 2357 2358
    {
      if (!(specialflag & SPECIAL_SHORT_LOG_FORMAT))
      {
        end=strmov(end,",insert_id=");
2359 2360 2361
        end=longlong10_to_str((longlong)
                              thd->auto_inc_intervals_in_cur_stmt_for_binlog.minimum(),
                              end, -10);
2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376
      }
    }

    /*
      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';
2377 2378
      if (my_b_write(&log_file, (uchar*) "SET ", 4) ||
          my_b_write(&log_file, (uchar*) buff + 1, (uint) (end-buff)))
2379 2380 2381 2382 2383 2384
        tmp_errno= errno;
    }
    if (is_command)
    {
      end= strxmov(buff, "# administrator command: ", NullS);
      buff_len= (ulong) (end - buff);
2385
      my_b_write(&log_file, (uchar*) buff, buff_len);
2386
    }
2387 2388
    if (my_b_write(&log_file, (uchar*) sql_text, sql_text_len) ||
        my_b_write(&log_file, (uchar*) ";\n",2) ||
2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400
        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);
      }
    }
  }
2401
  (void) pthread_mutex_unlock(&LOCK_log);
2402
  DBUG_RETURN(error);
2403 2404
}

2405

2406 2407 2408 2409 2410
/**
  @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.
*/
2411
const char *MYSQL_LOG::generate_name(const char *log_name,
2412 2413
                                      const char *suffix,
                                      bool strip_ext, char *buff)
2414 2415 2416
{
  if (!log_name || !log_name[0])
  {
2417 2418 2419
    strmake(buff, pidfile_name, FN_REFLEN - strlen(suffix) - 1);
    return (const char *)
      fn_format(buff, buff, "", suffix, MYF(MY_REPLACE_EXT|MY_REPLACE_DIR));
2420 2421 2422 2423
  }
  // get rid of extension if the log is binary to avoid problems
  if (strip_ext)
  {
2424 2425
    char *p= fn_ext(log_name);
    uint length= (uint) (p - log_name);
Davi Arnaut's avatar
Davi Arnaut committed
2426
    strmake(buff, log_name, min(length, FN_REFLEN-1));
2427 2428 2429 2430 2431
    return (const char*)buff;
  }
  return log_name;
}

2432

2433

2434 2435
MYSQL_BIN_LOG::MYSQL_BIN_LOG()
  :bytes_written(0), prepared_xids(0), file_id(1), open_count(1),
2436
   need_start_event(TRUE),
2437
   is_relay_log(0),
2438 2439 2440 2441 2442 2443 2444 2445 2446 2447
   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));
2448
  bzero((char*) &purge_index_file, sizeof(purge_index_file));
2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491
}

/* 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,
2492
                                    const char *log_name, bool need_mutex)
2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516
{
  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)),
2517 2518
			0, MYF(MY_WME | MY_WAIT_IF_FULL)) ||
      DBUG_EVALUATE_IF("fault_injection_openning_index", 1, 0))
2519
  {
2520 2521 2522 2523 2524
    /*
      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().
    */
2525 2526 2527 2528
    if (index_file_nr >= 0)
      my_close(index_file_nr,MYF(0));
    return TRUE;
  }
2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550

#ifdef HAVE_REPLICATION
  /*
    Sync the index by purging any binary log file that is not registered.
    In other words, either purge binary log files that were removed from
    the index but not purged from the file system due to a crash or purge
    any binary log file that was created but not register in the index
    due to a crash.
  */

  if (set_purge_index_file_name(index_file_name_arg) ||
      open_purge_index_file(FALSE) ||
      purge_index_entry(NULL, NULL, need_mutex) ||
      close_purge_index_file() ||
      DBUG_EVALUATE_IF("fault_injection_recovering_index", 1, 0))
  {
    sql_print_error("MYSQL_BIN_LOG::open_index_file failed to sync the index "
                    "file.");
    return TRUE;
  }
#endif

2551 2552 2553
  return FALSE;
}

bk@work.mysql.com's avatar
bk@work.mysql.com committed
2554

2555
/**
2556
  Open a (new) binlog file.
2557

2558
  - Open the log file and the index file. Register the new
2559
  file name in it
2560
  - When calling this when the file is in use, you must have a locks
2561
  on LOCK_log and LOCK_index.
2562

2563
  @retval
2564
    0	ok
2565
  @retval
2566 2567 2568
    1	error
*/

2569 2570 2571 2572 2573 2574
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,
2575 2576
                         bool null_created_arg,
                         bool need_mutex)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2577
{
2578
  File file= -1;
2579

2580
  DBUG_ENTER("MYSQL_BIN_LOG::open");
2581
  DBUG_PRINT("enter",("log_type: %d",(int) log_type_arg));
2582

2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602
  if (init_and_set_log_file_name(log_name, new_name, log_type_arg,
                                 io_cache_type_arg))
  {
    sql_print_error("MSYQL_BIN_LOG::open failed to generate new file name.");
    DBUG_RETURN(1);
  }

#ifdef HAVE_REPLICATION
  if (open_purge_index_file(TRUE) ||
      register_create_index_entry(log_file_name) ||
      sync_purge_index_file() ||
      DBUG_EVALUATE_IF("fault_injection_registering_index", 1, 0))
  {
    sql_print_error("MSYQL_BIN_LOG::open failed to sync the index file.");
    DBUG_RETURN(1);
  }
  DBUG_EXECUTE_IF("crash_create_non_critical_before_update_index", abort(););
#endif

  write_error= 0;
2603

2604
  /* open the main log file */
2605 2606 2607 2608 2609 2610
  if (MYSQL_LOG::open(log_name, log_type_arg, new_name,
                      io_cache_type_arg))
  {
#ifdef HAVE_REPLICATION
    close_purge_index_file();
#endif
2611
    DBUG_RETURN(1);                            /* all warnings issued */
2612
  }
2613

2614
  init(no_auto_events_arg, max_size_arg);
2615

2616
  open_count++;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2617

2618
  DBUG_ASSERT(log_type == LOG_BIN);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2619 2620

  {
2621 2622 2623 2624 2625 2626 2627 2628 2629 2630
    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.
      */
2631
      if (my_b_safe_write(&log_file, (uchar*) BINLOG_MAGIC,
2632
			  BIN_LOG_HEADER_SIZE))
2633
        goto err;
2634 2635
      bytes_written+= BIN_LOG_HEADER_SIZE;
      write_file_name_to_index_file= 1;
2636
    }
2637

2638
    if (need_start_event && !no_auto_events)
2639
    {
2640 2641 2642 2643 2644
      /*
        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
2645 2646 2647 2648 2649 2650
      /*
        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;
2651 2652
      if (!s.is_valid())
        goto err;
2653
      s.dont_set_created= null_created_arg;
2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668
      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
2669
        Format_description_log_event::Format_description_log_event(char* buf,).
2670 2671 2672 2673 2674 2675 2676 2677 2678
        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
2679
        Format_description_log_event::apply_event_impl().
2680 2681 2682
      */
      description_event_for_queue->created= 0;
      /* Don't set log_pos in event header */
2683
      description_event_for_queue->set_artificial_event();
2684

2685 2686 2687
      if (description_event_for_queue->write(&log_file))
        goto err;
      bytes_written+= description_event_for_queue->data_written;
2688
    }
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2689 2690
    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
2691
      goto err;
2692 2693 2694

    if (write_file_name_to_index_file)
    {
2695 2696 2697 2698 2699 2700 2701
#ifdef HAVE_REPLICATION
      DBUG_EXECUTE_IF("crash_create_critical_before_update_index", abort(););
#endif

      DBUG_ASSERT(my_b_inited(&index_file) != 0);
      reinit_io_cache(&index_file, WRITE_CACHE,
                      my_b_filelength(&index_file), 0, 0);
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2702 2703 2704 2705
      /*
        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.
      */
2706 2707 2708 2709 2710
      if (DBUG_EVALUATE_IF("fault_injection_updating_index", 1, 0) ||
          my_b_write(&index_file, (uchar*) log_file_name,
                     strlen(log_file_name)) ||
          my_b_write(&index_file, (uchar*) "\n", 1) ||
          flush_io_cache(&index_file) ||
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2711
          my_sync(index_file.file, MYF(MY_WME)))
2712 2713 2714 2715 2716
        goto err;

#ifdef HAVE_REPLICATION
      DBUG_EXECUTE_IF("crash_create_after_update_index", abort(););
#endif
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
2717
    }
2718
  }
2719 2720
  log_state= LOG_OPENED;

2721 2722 2723 2724
#ifdef HAVE_REPLICATION
  close_purge_index_file();
#endif

2725
  DBUG_RETURN(0);
2726 2727

err:
2728 2729 2730 2731 2732
#ifdef HAVE_REPLICATION
  if (is_inited_purge_index_file())
    purge_index_entry(NULL, NULL, need_mutex);
  close_purge_index_file();
#endif
2733 2734 2735
  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, \
2736
shutdown the MySQL server and restart it.", name, errno);
2737 2738 2739
  if (file >= 0)
    my_close(file,MYF(0));
  end_io_cache(&log_file);
2740 2741
  end_io_cache(&index_file);
  safeFree(name);
2742
  log_state= LOG_CLOSED;
2743
  DBUG_RETURN(1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2744 2745
}

2746

2747
int MYSQL_BIN_LOG::get_current_log(LOG_INFO* linfo)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2748 2749
{
  pthread_mutex_lock(&LOCK_log);
2750 2751 2752 2753 2754
  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
2755
int MYSQL_BIN_LOG::raw_get_current_log(LOG_INFO* linfo)
2756
{
2757 2758
  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
2759 2760 2761
  return 0;
}

2762 2763
/**
  Move all data up in a file in an filename index file.
2764 2765 2766 2767 2768

    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.

2769 2770 2771 2772 2773 2774 2775
  @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
2776 2777 2778
    0	ok
*/

2779 2780
#ifdef HAVE_REPLICATION

2781 2782 2783 2784 2785
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;
2786
  uchar io_buf[IO_SIZE*2];
2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797
  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));
2798
    if (my_write(file, io_buf, bytes_read, MYF(MY_WME | MY_NABP)))
2799 2800 2801
      goto err;
  }
  /* The following will either truncate the file or fill the end with \n' */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2802 2803
  if (my_chsize(file, offset - init_offset, '\n', MYF(MY_WME)) ||
      my_sync(file, MYF(MY_WME)))
2804 2805 2806 2807 2808 2809 2810 2811 2812 2813
    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);
}

2814
#endif /* HAVE_REPLICATION */
2815

2816 2817
/**
  Find the position in the log-index-file for the given log name.
2818

2819 2820 2821 2822 2823 2824
  @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
2825

2826
  @note
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
2827 2828
    On systems without the truncate function the file will end with one or
    more empty lines.  These will be ignored when reading the file.
2829

2830
  @retval
2831
    0			ok
2832 2833 2834
  @retval
    LOG_INFO_EOF	        End of log-index-file found
  @retval
2835 2836 2837
    LOG_INFO_IO		Got IO error while reading file
*/

2838
int MYSQL_BIN_LOG::find_log_pos(LOG_INFO *linfo, const char *log_name,
2839
			    bool need_lock)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2840
{
2841 2842 2843 2844 2845
  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
2846

2847
  /*
2848 2849
    Mutex needed because we need to make sure the file pointer does not
    move from under our feet
2850
  */
2851
  if (need_lock)
2852
    pthread_mutex_lock(&LOCK_index);
2853 2854 2855 2856 2857
  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);

2858
  for (;;)
2859
  {
2860
    uint length;
2861 2862 2863 2864
    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
2865
    {
2866 2867 2868
      /* 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
2869 2870
    }

2871 2872
    // if the log entry matches, null string matching anything
    if (!log_name ||
sasha@mysql.sashanet.com's avatar
sasha@mysql.sashanet.com committed
2873
	(log_name_len == length-1 && fname[log_name_len] == '\n' &&
2874
	 !memcmp(fname, log_name, log_name_len)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2875
    {
2876
      DBUG_PRINT("info",("Found log file entry"));
2877
      fname[length-1]=0;			// remove last \n
2878 2879
      linfo->index_file_start_offset= offset;
      linfo->index_file_offset = my_b_tell(&index_file);
2880
      break;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2881
    }
2882
  }
2883

2884
  if (need_lock)
2885
    pthread_mutex_unlock(&LOCK_index);
2886
  DBUG_RETURN(error);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2887
}
2888

2889

2890 2891
/**
  Find the position in the log-index-file for the given log name.
2892

2893
  @param
2894 2895
    linfo		Store here the next log file name and position to
			the file name after that.
2896
  @param
2897 2898 2899
    need_lock		Set this to 1 if the parent doesn't already have a
			lock on LOCK_index

2900
  @note
2901
    - Before calling this function, one has to call find_log_pos()
2902
    to set up 'linfo'
2903
    - Mutex needed because we need to make sure the file pointer does not move
2904
    from under our feet
2905

2906
  @retval
2907
    0			ok
2908 2909 2910
  @retval
    LOG_INFO_EOF	        End of log-index-file found
  @retval
2911 2912 2913
    LOG_INFO_IO		Got IO error while reading file
*/

2914
int MYSQL_BIN_LOG::find_next_log(LOG_INFO* linfo, bool need_lock)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2915
{
2916
  int error= 0;
2917
  uint length;
2918 2919
  char *fname= linfo->log_file_name;

2920 2921
  if (need_lock)
    pthread_mutex_lock(&LOCK_index);
2922 2923 2924 2925 2926 2927 2928 2929
  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)
2930
  {
2931
    error = !index_file.error ? LOG_INFO_EOF : LOG_INFO_IO;
2932 2933
    goto err;
  }
2934
  fname[length-1]=0;				// kill \n
2935
  linfo->index_file_offset = my_b_tell(&index_file);
2936

bk@work.mysql.com's avatar
bk@work.mysql.com committed
2937
err:
2938 2939
  if (need_lock)
    pthread_mutex_unlock(&LOCK_index);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2940 2941 2942
  return error;
}

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2943

2944 2945 2946
/**
  Delete all logs refered to in the index file.
  Start writing to a new log file.
2947

2948
  The new index file will only contain this file.
2949

2950
  @param thd		Thread
2951

2952 2953
  @note
    If not called from slave thread, write start event to new log
2954

2955
  @retval
2956
    0	ok
2957
  @retval
2958 2959 2960
    1   error
*/

2961
bool MYSQL_BIN_LOG::reset_logs(THD* thd)
2962 2963
{
  LOG_INFO linfo;
2964
  bool error=0;
2965
  const char* save_name;
2966
  DBUG_ENTER("reset_logs");
2967

tomas@poseidon.ndb.mysql.com's avatar
tomas@poseidon.ndb.mysql.com committed
2968
  ha_reset_logs(thd);
2969 2970 2971 2972
  /*
    We need to get both locks to be sure that no one is trying to
    write to the index log file.
  */
2973
  pthread_mutex_lock(&LOCK_log);
2974
  pthread_mutex_lock(&LOCK_index);
2975

2976 2977 2978 2979 2980 2981
  /*
    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.
  */
2982
  VOID(pthread_mutex_lock(&LOCK_thread_count));
2983 2984 2985 2986

  /* Save variables so that we can reopen the log */
  save_name=name;
  name=0;					// Protect against free
2987
  close(LOG_CLOSE_TO_BE_OPENED);
2988

2989 2990 2991 2992 2993
  /*
    First delete all old log files and then update the index file.
    As we first delete the log files and do not use sort of logging,
    a crash may lead to an inconsistent state where the index has
    references to non-existent files.
2994

2995 2996 2997
    We need to invert the steps and use the purge_index_file methods
    in order to make the operation safe.
  */
2998
  if (find_log_pos(&linfo, NullS, 0))
2999 3000 3001 3002
  {
    error=1;
    goto err;
  }
3003

3004
  for (;;)
3005
  {
3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019
    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
      {
3020
        push_warning_printf(current_thd, MYSQL_ERROR::WARN_LEVEL_WARN,
3021 3022 3023 3024 3025 3026 3027 3028 3029 3030
                            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;
      }
    }
3031
    if (find_next_log(&linfo, 0))
3032 3033
      break;
  }
3034 3035

  /* Start logging with a new file */
3036
  close(LOG_CLOSE_INDEX);
3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050
  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
    {
3051
      push_warning_printf(current_thd, MYSQL_ERROR::WARN_LEVEL_WARN,
3052 3053 3054 3055 3056 3057 3058 3059 3060 3061
                          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;
    }
  }
3062
  if (!thd->slave_thread)
3063
    need_start_event=1;
3064 3065
  if (!open_index_file(index_file_name, 0, FALSE))
    open(save_name, log_type, 0, io_cache_type, no_auto_events, max_size, 0, FALSE);
3066
  my_free((uchar*) save_name, MYF(0));
3067

3068
err:
3069
  VOID(pthread_mutex_unlock(&LOCK_thread_count));
3070
  pthread_mutex_unlock(&LOCK_index);
3071
  pthread_mutex_unlock(&LOCK_log);
3072
  DBUG_RETURN(error);
3073 3074
}

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
3075

3076
/**
3077 3078
  Delete relay log files prior to rli->group_relay_log_name
  (i.e. all logs which are not involved in a non-finished group
3079 3080
  (transaction)), remove them from the index file and start on next
  relay log.
3081

3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095
  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
3096
    - This is only called from the slave-execute thread when it has read
3097
    all commands from a relay log and want to switch to a new relay log.
3098
    - When this happens, we can be in an active transaction as
3099 3100 3101
    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).
3102

3103
  @retval
3104
    0			ok
3105 3106 3107
  @retval
    LOG_INFO_EOF	        End of log-index-file found
  @retval
3108
    LOG_INFO_SEEK	Could not allocate IO cache
3109
  @retval
3110
    LOG_INFO_IO		Got IO error while reading file
3111 3112
*/

hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
3113
#ifdef HAVE_REPLICATION
3114

3115
int MYSQL_BIN_LOG::purge_first_log(Relay_log_info* rli, bool included)
3116
{
3117
  int error;
3118
  char *to_purge_if_included= NULL;
3119 3120
  DBUG_ENTER("purge_first_log");

3121 3122
  DBUG_ASSERT(is_open());
  DBUG_ASSERT(rli->slave_running == 1);
3123
  DBUG_ASSERT(!strcmp(rli->linfo.log_file_name,rli->event_relay_log_name));
3124

3125
  pthread_mutex_lock(&LOCK_index);
3126
  to_purge_if_included= my_strdup(rli->group_relay_log_name, MYF(0));
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
3127

3128 3129
  /*
    Read the next log file name from the index file and pass it back to
3130
    the caller.
3131
  */
3132 3133
  if((error=find_log_pos(&rli->linfo, rli->event_relay_log_name, 0)) || 
     (error=find_next_log(&rli->linfo, 0)))
3134 3135
  {
    char buff[22];
3136 3137 3138
    sql_print_error("next log error: %d  offset: %s  log: %s included: %d",
                    error,
                    llstr(rli->linfo.index_file_offset,buff),
3139
                    rli->event_relay_log_name,
3140
                    included);
3141 3142
    goto err;
  }
3143

3144
  /*
3145
    Reset rli's coordinates to the current log.
3146
  */
3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160
  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);
3161
    rli->notify_group_relay_log_name_update();
3162
  }
3163 3164 3165

  /* 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
3166

3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186
  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.
   */
3187
  if((error=find_log_pos(&rli->linfo, rli->event_relay_log_name, 0)))
3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200
  {
    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);

3201
err:
3202
  my_free(to_purge_if_included, MYF(0));
3203
  pthread_mutex_unlock(&LOCK_index);
3204
  DBUG_RETURN(error);
3205 3206
}

3207 3208
/**
  Update log index_file.
3209 3210
*/

3211
int MYSQL_BIN_LOG::update_log_index(LOG_INFO* log_info, bool need_update_threads)
3212 3213 3214 3215 3216
{
  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
3217 3218
  if (need_update_threads)
    adjust_linfo_offsets(log_info->index_file_start_offset);
3219 3220
  return 0;
}
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
3221

3222
/**
3223 3224
  Remove all logs before the given log from disk and from the index file.

3225 3226 3227 3228 3229 3230 3231
  @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
3232

3233
  @note
3234 3235 3236
    If any of the logs before the deleted one is in use,
    only purge logs up to this one.

3237 3238 3239
  @retval
    0			ok
  @retval
3240
    LOG_INFO_EOF		to_log not found
3241
    LOG_INFO_EMFILE             too many files opened
3242 3243
    LOG_INFO_FATAL              if any other than ENOENT error from
                                my_stat() or my_delete()
3244 3245
*/

3246
int MYSQL_BIN_LOG::purge_logs(const char *to_log, 
3247 3248 3249 3250
                          bool included,
                          bool need_mutex, 
                          bool need_update_threads, 
                          ulonglong *decrease_log_space)
3251
{
3252
  int error= 0;
3253
  bool exit_loop= 0;
3254
  LOG_INFO log_info;
3255
  THD *thd= current_thd;
3256
  DBUG_ENTER("purge_logs");
3257
  DBUG_PRINT("info",("to_log= %s",to_log));
3258

3259 3260
  if (need_mutex)
    pthread_mutex_lock(&LOCK_index);
3261 3262
  if ((error=find_log_pos(&log_info, to_log, 0 /*no mutex*/))) 
  {
3263
    sql_print_error("MYSQL_BIN_LOG::purge_logs was called with file %s not "
3264
                    "listed in the index.", to_log);
3265
    goto err;
3266 3267
  }

3268
  if ((error= open_purge_index_file(TRUE)))
3269
  {
3270 3271
    sql_print_error("MYSQL_BIN_LOG::purge_logs failed to sync the index file.");
    goto err;
3272
  }
3273

3274
  /*
3275
    File name exists in index file; delete until we find this file
3276 3277 3278
    or a file that is used.
  */
  if ((error=find_log_pos(&log_info, NullS, 0 /*no mutex*/)))
3279
    goto err;
3280
  while ((strcmp(to_log,log_info.log_file_name) || (exit_loop=included)) &&
3281
         !is_active(log_info.log_file_name) &&
3282
         !log_in_use(log_info.log_file_name))
3283
  {
3284
    if ((error= register_purge_index_entry(log_info.log_file_name)))
3285
    {
3286
      sql_print_error("MYSQL_BIN_LOG::purge_logs failed to copy %s to register file.",
3287 3288 3289 3290 3291 3292
                      log_info.log_file_name);
      goto err;
    }

    if (find_next_log(&log_info, 0) || exit_loop)
      break;
3293 3294 3295 3296 3297 3298 3299 3300 3301
  }

  DBUG_EXECUTE_IF("crash_purge_before_update_index", abort(););

  if ((error= sync_purge_index_file()))
  {
    sql_print_error("MSYQL_BIN_LOG::purge_logs failed to flush register file.");
    goto err;
  }
3302 3303

  /* We know how many files to delete. Update index file. */
3304
  if ((error=update_log_index(&log_info, need_update_threads)))
3305
  {
3306
    sql_print_error("MSYQL_BIN_LOG::purge_logs failed to update the index file");
3307 3308 3309
    goto err;
  }

3310 3311 3312 3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325
  DBUG_EXECUTE_IF("crash_purge_critical_after_update_index", abort(););

err:
  /* Read each entry from purge_index_file and delete the file. */
  if (is_inited_purge_index_file() &&
      (error= purge_index_entry(thd, decrease_log_space, FALSE)))
    sql_print_error("MSYQL_BIN_LOG::purge_logs failed to process registered files"
                    " that would be purged.");
  close_purge_index_file();

  DBUG_EXECUTE_IF("crash_purge_non_critical_after_update_index", abort(););

  if (need_mutex)
    pthread_mutex_unlock(&LOCK_index);
  DBUG_RETURN(error);
}
3326

3327 3328 3329 3330 3331 3332 3333
int MYSQL_BIN_LOG::set_purge_index_file_name(const char *base_file_name)
{
  int error= 0;
  DBUG_ENTER("MYSQL_BIN_LOG::set_purge_index_file_name");
  if (fn_format(purge_index_file_name, base_file_name, mysql_data_home,
                ".~rec~", MYF(MY_UNPACK_FILENAME | MY_SAFE_PATH |
                              MY_REPLACE_EXT)) == NULL)
3334
  {
3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435
    error= 1;
    sql_print_error("MYSQL_BIN_LOG::set_purge_index_file_name failed to set "
                      "file name.");
  }
  DBUG_RETURN(error);
}

int MYSQL_BIN_LOG::open_purge_index_file(bool destroy)
{
  int error= 0;
  File file= -1;

  DBUG_ENTER("MYSQL_BIN_LOG::open_purge_index_file");

  if (destroy)
    close_purge_index_file();

  if (!my_b_inited(&purge_index_file))
  {
    if ((file= my_open(purge_index_file_name, O_RDWR | O_CREAT | O_BINARY,
                       MYF(MY_WME | ME_WAITTANG))) < 0  ||
        init_io_cache(&purge_index_file, file, IO_SIZE,
                      (destroy ? WRITE_CACHE : READ_CACHE),
                      0, 0, MYF(MY_WME | MY_NABP | MY_WAIT_IF_FULL)))
    {
      error= 1;
      sql_print_error("MYSQL_BIN_LOG::open_purge_index_file failed to open register "
                      " file.");
    }
  }
  DBUG_RETURN(error);
}

int MYSQL_BIN_LOG::close_purge_index_file()
{
  int error= 0;

  DBUG_ENTER("MYSQL_BIN_LOG::close_purge_index_file");

  if (my_b_inited(&purge_index_file))
  {
    end_io_cache(&purge_index_file);
    error= my_close(purge_index_file.file, MYF(0));
  }
  my_delete(purge_index_file_name, MYF(0));
  bzero((char*) &purge_index_file, sizeof(purge_index_file));

  DBUG_RETURN(error);
}

bool MYSQL_BIN_LOG::is_inited_purge_index_file()
{
  DBUG_ENTER("MYSQL_BIN_LOG::is_inited_purge_index_file");
  DBUG_RETURN (my_b_inited(&purge_index_file));
}

int MYSQL_BIN_LOG::sync_purge_index_file()
{
  int error= 0;
  DBUG_ENTER("MYSQL_BIN_LOG::sync_purge_index_file");

  if ((error= flush_io_cache(&purge_index_file)) ||
      (error= my_sync(purge_index_file.file, MYF(MY_WME))))
    DBUG_RETURN(error);

  DBUG_RETURN(error);
}

int MYSQL_BIN_LOG::register_purge_index_entry(const char *entry)
{
  int error= 0;
  DBUG_ENTER("MYSQL_BIN_LOG::register_purge_index_entry");

  if ((error=my_b_write(&purge_index_file, (const uchar*)entry, strlen(entry))) ||
      (error=my_b_write(&purge_index_file, (const uchar*)"\n", 1)))
    DBUG_RETURN (error);

  DBUG_RETURN(error);
}

int MYSQL_BIN_LOG::register_create_index_entry(const char *entry)
{
  DBUG_ENTER("MYSQL_BIN_LOG::register_create_index_entry");
  DBUG_RETURN(register_purge_index_entry(entry));
}

int MYSQL_BIN_LOG::purge_index_entry(THD *thd, ulonglong *decrease_log_space,
                                     bool need_mutex)
{
  MY_STAT s;
  int error= 0;
  LOG_INFO log_info;
  LOG_INFO check_log_info;

  DBUG_ENTER("MYSQL_BIN_LOG:purge_index_entry");

  DBUG_ASSERT(my_b_inited(&purge_index_file));

  if ((error=reinit_io_cache(&purge_index_file, READ_CACHE, 0, 0, 0)))
  {
    sql_print_error("MSYQL_BIN_LOG::purge_index_entry failed to reinit register file "
3436 3437 3438 3439 3440 3441 3442 3443
                    "for read");
    goto err;
  }

  for (;;)
  {
    uint length;

3444
    if ((length=my_b_gets(&purge_index_file, log_info.log_file_name,
3445 3446
                          FN_REFLEN)) <= 1)
    {
3447
      if (purge_index_file.error)
3448
      {
3449 3450 3451
        error= purge_index_file.error;
        sql_print_error("MSYQL_BIN_LOG::purge_index_entry error %d reading from "
                        "register file.", error);
3452 3453 3454 3455 3456 3457 3458 3459 3460 3461
        goto err;
      }

      /* Reached EOF */
      break;
    }

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

3462
    if (!my_stat(log_info.log_file_name, &s, MYF(0)))
3463
    {
3464 3465 3466 3467 3468
      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.
3469
        */
3470 3471 3472 3473 3474 3475
        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);
        }
3476
        sql_print_information("Failed to execute my_stat on file '%s'",
3477
			      log_info.log_file_name);
3478 3479 3480 3481 3482 3483 3484
        my_errno= 0;
      }
      else
      {
        /*
          Other than ENOENT are fatal
        */
3485 3486
        if (thd)
        {
3487
          push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_WARN,
3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502
                              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);
        }
3503 3504 3505 3506 3507 3508
        error= LOG_INFO_FATAL;
        goto err;
      }
    }
    else
    {
3509
      if ((error= find_log_pos(&check_log_info, log_info.log_file_name, need_mutex)))
3510
      {
3511
        if (error != LOG_INFO_EOF)
3512
        {
3513 3514
          if (thd)
          {
3515
            push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_WARN,
3516 3517 3518
                                ER_BINLOG_PURGE_FATAL_ERR,
                                "a problem with deleting %s and "
                                "reading the binlog index file",
3519 3520
                                log_info.log_file_name);
          }
3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543
          else
          {
            sql_print_information("Failed to delete file '%s' and "
                                  "read the binlog index file",
                                  log_info.log_file_name);
          }
          goto err;
        }
           
        error= 0;
        if (!need_mutex)
        {
          /*
            This is to avoid triggering an error in NDB.
          */
          ha_binlog_index_purge_file(current_thd, log_info.log_file_name);
        }

        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;
3544 3545 3546
        }
        else
        {
3547
          if (my_errno == ENOENT)
3548
          {
3549 3550 3551 3552 3553 3554 3555 3556 3557
            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);
            }
            sql_print_information("Failed to delete file '%s'",
                                  log_info.log_file_name);
            my_errno= 0;
3558 3559 3560
          }
          else
          {
3561 3562
            if (thd)
            {
3563
              push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_WARN,
3564 3565
                                  ER_BINLOG_PURGE_FATAL_ERR,
                                  "a problem with deleting %s; "
3566 3567 3568 3569
                                  "consider examining correspondence "
                                  "of your binlog index file "
                                  "to the actual binlog files",
                                  log_info.log_file_name);
3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586
            }
            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);
            }
            if (my_errno == EMFILE)
            {
              DBUG_PRINT("info",
                         ("my_errno: %d, set ret = LOG_INFO_EMFILE", my_errno));
              error= LOG_INFO_EMFILE;
              goto err;
            }
            error= LOG_INFO_FATAL;
Kristofer Pettersson's avatar
Kristofer Pettersson committed
3587
            goto err;
3588
          }
3589 3590
        }
      }
3591
    }
3592
  }
3593

3594
err:
3595
  DBUG_RETURN(error);
3596 3597
}

3598
/**
3599 3600 3601
  Remove all logs before the given file date from disk and from the
  index file.

3602
  @param thd		Thread pointer
3603
  @param purge_time	Delete all log files before given date.
3604

3605
  @note
3606 3607 3608
    If any of the logs before the deleted one is in use,
    only purge logs up to this one.

3609
  @retval
3610
    0				ok
3611
  @retval
3612
    LOG_INFO_PURGE_NO_ROTATE	Binary file that can't be rotated
3613 3614
    LOG_INFO_FATAL              if any other than ENOENT error from
                                my_stat() or my_delete()
3615 3616
*/

3617
int MYSQL_BIN_LOG::purge_logs_before_date(time_t purge_time)
3618 3619
{
  int error;
3620
  char to_log[FN_REFLEN];
3621 3622
  LOG_INFO log_info;
  MY_STAT stat_area;
3623
  THD *thd= current_thd;
3624
  
3625 3626 3627
  DBUG_ENTER("purge_logs_before_date");

  pthread_mutex_lock(&LOCK_index);
3628
  to_log[0]= 0;
3629 3630

  if ((error=find_log_pos(&log_info, NullS, 0 /*no mutex*/)))
3631
    goto err;
3632 3633

  while (strcmp(log_file_name, log_info.log_file_name) &&
3634 3635
	 !is_active(log_info.log_file_name) &&
         !log_in_use(log_info.log_file_name))
3636
  {
3637 3638 3639 3640 3641 3642
    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.
3643
        */
3644 3645 3646 3647 3648 3649 3650
        my_errno= 0;
      }
      else
      {
        /*
          Other than ENOENT are fatal
        */
3651 3652
        if (thd)
        {
3653
          push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_WARN,
3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665
                              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);
        }
3666 3667 3668 3669 3670 3671
        error= LOG_INFO_FATAL;
        goto err;
      }
    }
    else
    {
3672 3673 3674
      if (stat_area.st_mtime < purge_time) 
        strmake(to_log, 
                log_info.log_file_name, 
3675
                sizeof(log_info.log_file_name) - 1);
3676
      else
3677 3678
        break;
    }
3679 3680
    if (find_next_log(&log_info, 0))
      break;
3681
  }
3682

3683
  error= (to_log[0] ? purge_logs(to_log, 1, 0, 1, (ulonglong *) 0) : 0);
3684

3685 3686
err:
  pthread_mutex_unlock(&LOCK_index);
3687
  DBUG_RETURN(error);
3688
}
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
3689
#endif /* HAVE_REPLICATION */
3690

3691

3692 3693
/**
  Create a new log file name.
3694

3695
  @param buf		buf of at least FN_REFLEN where new name is stored
3696

3697
  @note
3698 3699 3700
    If file name will be longer then FN_REFLEN it will be truncated
*/

3701
void MYSQL_BIN_LOG::make_log_name(char* buf, const char* log_ident)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3702
{
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3703
  uint dir_len = dirname_length(log_file_name); 
3704
  if (dir_len >= FN_REFLEN)
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3705 3706
    dir_len=FN_REFLEN-1;
  strnmov(buf, log_file_name, dir_len);
3707
  strmake(buf+dir_len, log_ident, FN_REFLEN - dir_len -1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3708 3709
}

3710

3711 3712
/**
  Check if we are writing/reading to the given log file.
3713 3714
*/

3715
bool MYSQL_BIN_LOG::is_active(const char *log_file_name_arg)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3716
{
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3717
  return !strcmp(log_file_name, log_file_name_arg);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3718 3719
}

3720

3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740
/*
  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);
}


3741 3742
/**
  Start writing to a new log file or reopen the old file.
3743

3744
  @param need_lock		Set to 1 if caller has not locked LOCK_log
3745

3746
  @note
3747 3748 3749
    The new file name is stored last in the index file
*/

3750
void MYSQL_BIN_LOG::new_file_impl(bool need_lock)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3751
{
3752 3753
  char new_name[FN_REFLEN], *new_name_ptr, *old_name;

3754
  DBUG_ENTER("MYSQL_BIN_LOG::new_file_impl");
3755
  if (!is_open())
3756 3757 3758 3759
  {
    DBUG_PRINT("info",("log is closed"));
    DBUG_VOID_RETURN;
  }
3760 3761 3762

  if (need_lock)
    pthread_mutex_lock(&LOCK_log);
3763 3764
  pthread_mutex_lock(&LOCK_index);

3765 3766 3767
  safe_mutex_assert_owner(&LOCK_log);
  safe_mutex_assert_owner(&LOCK_index);

3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780
  /*
    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);
3781 3782
    while (prepared_xids) {
      DBUG_PRINT("info", ("prepared_xids=%lu", prepared_xids));
3783
      pthread_cond_wait(&COND_prep_xids, &LOCK_prep_xids);
3784
    }
3785 3786 3787
    pthread_mutex_unlock(&LOCK_prep_xids);
  }

3788
  /* Reuse old name if not binlog and not update log */
3789
  new_name_ptr= name;
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
3790

3791
  /*
3792 3793 3794
    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.
3795
  */
3796 3797 3798
  if (generate_new_name(new_name, name))
    goto end;
  new_name_ptr=new_name;
3799

3800
  if (log_type == LOG_BIN)
3801
  {
3802
    if (!no_auto_events)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3803
    {
3804
      /*
3805 3806
        We log the whole file name for log file as the user may decide
        to change base names at some point.
3807
      */
3808
      Rotate_log_event r(new_name+dirname_length(new_name),
3809
                         0, LOG_EVENT_OFFSET, is_relay_log ? Rotate_log_event::RELAY_LOG : 0);
3810
      r.write(&log_file);
3811
      bytes_written += r.data_written;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3812
    }
3813 3814 3815 3816 3817
    /*
      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.
    */
3818
    signal_update();
3819 3820 3821
  }
  old_name=name;
  name=0;				// Don't free name
3822
  close(LOG_CLOSE_TO_BE_OPENED | LOG_CLOSE_INDEX);
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3823

3824
  /*
3825
     Note that at this point, log_state != LOG_CLOSED (important for is_open()).
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3826
  */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3827

3828
  /*
3829
     new_file() is only used for rotation (in FLUSH LOGS or because size >
3830
     max_binlog_size or max_relay_log_size).
3831 3832 3833 3834
     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.
3835
  */
3836

3837
  /* reopen index binlog file, BUG#34582 */
3838 3839 3840
  if (!open_index_file(index_file_name, 0, FALSE))
    open(old_name, log_type, new_name_ptr,
         io_cache_type, no_auto_events, max_size, 1, FALSE);
3841 3842
  my_free(old_name,MYF(0));

3843
end:
3844 3845
  if (need_lock)
    pthread_mutex_unlock(&LOCK_log);
3846 3847
  pthread_mutex_unlock(&LOCK_index);

3848
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3849 3850
}

3851

3852
bool MYSQL_BIN_LOG::append(Log_event* ev)
3853 3854 3855
{
  bool error = 0;
  pthread_mutex_lock(&LOCK_log);
3856
  DBUG_ENTER("MYSQL_BIN_LOG::append");
3857

3858
  DBUG_ASSERT(log_file.type == SEQ_READ_APPEND);
3859 3860 3861 3862
  /*
    Log_event::write() is smart enough to use my_b_write() or
    my_b_append() depending on the kind of cache we have.
  */
3863 3864 3865 3866 3867
  if (ev->write(&log_file))
  {
    error=1;
    goto err;
  }
3868
  bytes_written+= ev->data_written;
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
3869
  DBUG_PRINT("info",("max_size: %lu",max_size));
3870
  if ((uint) my_b_append_tell(&log_file) > max_size)
3871
    new_file_without_locking();
3872

3873
err:
3874
  pthread_mutex_unlock(&LOCK_log);
3875
  signal_update();				// Safe as we don't call close
3876
  DBUG_RETURN(error);
3877 3878
}

3879

3880
bool MYSQL_BIN_LOG::appendv(const char* buf, uint len,...)
3881
{
3882
  bool error= 0;
3883
  DBUG_ENTER("MYSQL_BIN_LOG::appendv");
3884 3885
  va_list(args);
  va_start(args,len);
3886

3887
  DBUG_ASSERT(log_file.type == SEQ_READ_APPEND);
3888

3889
  safe_mutex_assert_owner(&LOCK_log);
3890 3891
  do
  {
3892
    if (my_b_append(&log_file,(uchar*) buf,len))
3893
    {
3894 3895
      error= 1;
      goto err;
3896
    }
3897
    bytes_written += len;
3898
  } while ((buf=va_arg(args,const char*)) && (len=va_arg(args,uint)));
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
3899
  DBUG_PRINT("info",("max_size: %lu",max_size));
3900
  if ((uint) my_b_append_tell(&log_file) > max_size)
3901
    new_file_without_locking();
3902 3903

err:
3904 3905
  if (!error)
    signal_update();
3906
  DBUG_RETURN(error);
3907
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3908

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
3909

3910
bool MYSQL_BIN_LOG::flush_and_sync()
3911
{
3912 3913 3914 3915 3916
  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)
3917 3918
  {
    sync_binlog_counter= 0;
3919
    err=my_sync(fd, MYF(MY_WME));
3920
  }
3921
  return err;
3922 3923
}

3924
void MYSQL_BIN_LOG::start_union_events(THD *thd, query_id_t query_id_param)
3925 3926 3927 3928 3929
{
  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;
3930
  thd->binlog_evt_union.first_query_id= query_id_param;
3931 3932
}

3933
void MYSQL_BIN_LOG::stop_union_events(THD *thd)
3934 3935 3936 3937 3938
{
  DBUG_ASSERT(thd->binlog_evt_union.do_union);
  thd->binlog_evt_union.do_union= FALSE;
}

3939
bool MYSQL_BIN_LOG::is_query_in_union(THD *thd, query_id_t query_id_param)
3940 3941 3942 3943 3944
{
  return (thd->binlog_evt_union.do_union && 
          query_id_param >= thd->binlog_evt_union.first_query_id);
}

3945 3946 3947 3948 3949 3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967 3968 3969 3970 3971 3972 3973 3974 3975 3976 3977 3978 3979 3980 3981 3982 3983 3984 3985 3986 3987 3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004 4005
/**
  This function checks if a transaction, either a multi-statement
  or a single statement transaction is about to commit or not.

  @param thd The client thread that executed the current statement.
  @param all Committing a transaction (i.e. TRUE) or a statement
             (i.e. FALSE).
  @return
    @c true if committing a transaction, otherwise @c false.
*/
bool ending_trans(const THD* thd, const bool all)
{
  return (all || (!all && !(thd->options & 
                  (OPTION_BEGIN | OPTION_NOT_AUTOCOMMIT))));
}

/**
  This function checks if a non-transactional table was updated by
  the current transaction.

  @param thd The client thread that executed the current statement.
  @return
    @c true if a non-transactional table was updated, @c false
    otherwise.
*/
bool trans_has_updated_non_trans_table(const THD* thd)
{
  return (thd->transaction.all.modified_non_trans_table ||
          thd->transaction.stmt.modified_non_trans_table);
}

/**
  This function checks if any statement was committed and cached.

  @param thd The client thread that executed the current statement.
  @param all Committing a transaction (i.e. TRUE) or a statement
             (i.e. FALSE).
  @return
    @c true if at a statement was committed and cached, @c false
    otherwise.
*/
bool trans_has_no_stmt_committed(const THD* thd, bool all)
{
  binlog_trx_data *const trx_data=
    (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);

  return (!all && !trx_data->at_least_one_stmt_committed);
}

/**
  This function checks if a non-transactional table was updated by the
  current statement.

  @param thd The client thread that executed the current statement.
  @return
    @c true if a non-transactional table was updated, @c false otherwise.
*/
bool stmt_has_updated_non_trans_table(const THD* thd)
{
  return (thd->transaction.stmt.modified_non_trans_table);
}
4006 4007 4008 4009 4010 4011 4012 4013 4014 4015

/*
  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=
4016
    (binlog_trx_data*) thd_get_ha_data(this, binlog_hton);
4017 4018 4019 4020

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

4021
  trx_data= (binlog_trx_data*) my_malloc(sizeof(binlog_trx_data), MYF(MY_ZEROFILL));
4022 4023 4024 4025
  if (!trx_data ||
      open_cached_file(&trx_data->trans_log, mysql_tmpdir,
                       LOG_PREFIX, binlog_cache_size, MYF(MY_WME)))
  {
4026
    my_free((uchar*)trx_data, MYF(MY_ALLOW_ZERO_PTR));
4027 4028
    DBUG_RETURN(1);                      // Didn't manage to set it up
  }
4029
  thd_set_ha_data(this, binlog_hton, trx_data);
4030

4031
  trx_data= new (thd_get_ha_data(this, binlog_hton)) binlog_trx_data;
4032

4033 4034 4035
  DBUG_RETURN(0);
}

4036
/*
4037 4038
  Function to start a statement and optionally a transaction for the
  binary log.
4039

4040 4041
  SYNOPSIS
    binlog_start_trans_and_stmt()
4042

4043 4044 4045 4046 4047 4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059 4060 4061 4062 4063 4064 4065 4066
  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()
{
4067
  binlog_trx_data *trx_data= (binlog_trx_data*) thd_get_ha_data(this, binlog_hton);
4068 4069 4070 4071 4072 4073
  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)));

4074
  if (trx_data == NULL ||
4075
      trx_data->before_stmt_pos == MY_OFF_T_UNDEF)
4076
  {
4077
    this->binlog_set_stmt_begin();
4078
    if (options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN))
mats@romeo.(none)'s avatar
mats@romeo.(none) committed
4079 4080
      trans_register_ha(this, TRUE, binlog_hton);
    trans_register_ha(this, FALSE, binlog_hton);
4081 4082 4083 4084 4085 4086 4087 4088 4089 4090
    /*
      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();
4091 4092 4093 4094
  }
  DBUG_VOID_RETURN;
}

4095 4096
void THD::binlog_set_stmt_begin() {
  binlog_trx_data *trx_data=
4097
    (binlog_trx_data*) thd_get_ha_data(this, binlog_hton);
4098 4099 4100 4101 4102 4103 4104 4105 4106

  /*
    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);
4107
  trx_data= (binlog_trx_data*) thd_get_ha_data(this, binlog_hton);
4108 4109 4110 4111
  trx_data->before_stmt_pos= pos;
}


4112 4113
/*
  Write a table map to the binary log.
4114 4115
 */

4116 4117
int THD::binlog_write_table_map(TABLE *table, bool is_trans)
{
4118
  int error;
4119
  DBUG_ENTER("THD::binlog_write_table_map");
4120 4121
  DBUG_PRINT("enter", ("table: 0x%lx  (%s: #%lu)",
                       (long) table, table->s->table_name.str,
4122
                       table->s->table_map_id));
4123 4124

  /* Pre-conditions */
4125
  DBUG_ASSERT(current_stmt_binlog_row_based && mysql_bin_log.is_open());
4126 4127 4128
  DBUG_ASSERT(table->s->table_map_id != ULONG_MAX);

  Table_map_log_event
4129
    the_event(this, table, table->s->table_map_id, is_trans);
4130

4131 4132
  if (is_trans && binlog_table_maps == 0)
    binlog_start_trans_and_stmt();
4133

4134
  if ((error= mysql_bin_log.write(&the_event)))
4135 4136
    DBUG_RETURN(error);

4137
  binlog_table_maps++;
4138 4139 4140
  DBUG_RETURN(0);
}

4141 4142 4143 4144
Rows_log_event*
THD::binlog_get_pending_rows_event() const
{
  binlog_trx_data *const trx_data=
4145
    (binlog_trx_data*) thd_get_ha_data(this, binlog_hton);
4146 4147 4148 4149 4150 4151
  /*
    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.
   */
4152
  return trx_data ? trx_data->pending() : NULL;
4153 4154 4155 4156 4157
}

void
THD::binlog_set_pending_rows_event(Rows_log_event* ev)
{
4158
  if (thd_get_ha_data(this, binlog_hton) == NULL)
4159 4160
    binlog_setup_trx_data();

4161
  binlog_trx_data *const trx_data=
4162
    (binlog_trx_data*) thd_get_ha_data(this, binlog_hton);
4163

4164
  DBUG_ASSERT(trx_data);
4165
  trx_data->set_pending(ev);
4166 4167 4168
}


4169 4170 4171 4172 4173 4174 4175 4176 4177
/**
  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)
{
4178
  DBUG_ENTER("MYSQL_BIN_LOG::remove_pending_rows_event");
4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189 4190 4191 4192 4193

  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);
}

4194 4195 4196 4197 4198
/*
  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.
*/
4199 4200 4201
int
MYSQL_BIN_LOG::flush_and_set_pending_rows_event(THD *thd,
                                                Rows_log_event* event)
4202
{
4203
  DBUG_ENTER("MYSQL_BIN_LOG::flush_and_set_pending_rows_event(event)");
4204
  DBUG_ASSERT(mysql_bin_log.is_open());
4205
  DBUG_PRINT("enter", ("event: 0x%lx", (long) event));
4206 4207 4208 4209

  int error= 0;

  binlog_trx_data *const trx_data=
4210
    (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
4211 4212 4213

  DBUG_ASSERT(trx_data);

4214
  DBUG_PRINT("info", ("trx_data->pending(): 0x%lx", (long) trx_data->pending()));
4215

4216
  if (Rows_log_event* pending= trx_data->pending())
4217 4218 4219 4220 4221 4222 4223 4224 4225 4226 4227
  {
    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;

    /*
4228 4229
      If we are not writing to the log file directly, we could avoid
      locking the log.
4230 4231 4232 4233 4234 4235 4236 4237 4238
    */
    pthread_mutex_lock(&LOCK_log);

    /*
      Write pending event to log file or transaction cache
    */
    if (pending->write(file))
    {
      pthread_mutex_unlock(&LOCK_log);
4239
      set_write_error(thd);
4240 4241 4242 4243 4244 4245 4246 4247 4248 4249 4250 4251 4252 4253 4254 4255 4256 4257
      DBUG_RETURN(1);
    }

    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);
  }

4258
  thd->binlog_set_pending_rows_event(event);
4259 4260 4261 4262

  DBUG_RETURN(error);
}

4263 4264
/**
  Write an event to the binary log.
4265 4266
*/

4267
bool MYSQL_BIN_LOG::write(Log_event *event_info)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4268
{
4269 4270
  THD *thd= event_info->thd;
  bool error= 1;
4271
  DBUG_ENTER("MYSQL_BIN_LOG::write(Log_event *)");
4272

4273 4274
  if (thd->binlog_evt_union.do_union)
  {
4275 4276 4277 4278
    /*
      In Stored function; Remember that function call caused an update.
      We will log the function call to the binary log on function exit
    */
4279 4280 4281 4282
    thd->binlog_evt_union.unioned_events= TRUE;
    thd->binlog_evt_union.unioned_events_trans |= event_info->cache_stmt;
    DBUG_RETURN(0);
  }
4283 4284 4285 4286 4287 4288 4289

  /*
    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.

4290 4291 4292
    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.
4293
  */
4294 4295
  bool const end_stmt=
    thd->prelocked_mode && thd->lex->requires_prelocking();
4296 4297
  if (thd->binlog_flush_pending_rows_event(end_stmt))
    DBUG_RETURN(error);
4298

4299 4300
  pthread_mutex_lock(&LOCK_log);

4301
  /*
guilhem@mysql.com's avatar
guilhem@mysql.com committed
4302 4303 4304 4305
     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.
  */
4306
  if (likely(is_open()))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4307
  {
4308
    IO_CACHE *file= &log_file;
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
4309
#ifdef HAVE_REPLICATION
4310
    /*
4311 4312 4313
      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)"
4314
    */
4315
    const char *local_db= event_info->get_db();
4316
    if ((thd && !(thd->options & OPTION_BIN_LOG)) ||
4317 4318 4319
	(thd->lex->sql_command != SQLCOM_ROLLBACK_TO_SAVEPOINT &&
         thd->lex->sql_command != SQLCOM_SAVEPOINT &&
         !binlog_filter->db_ok(local_db)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4320
    {
4321
      VOID(pthread_mutex_unlock(&LOCK_log));
4322
      DBUG_RETURN(0);
4323
    }
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
4324
#endif /* HAVE_REPLICATION */
4325

4326
#if defined(USING_TRANSACTIONS) 
4327 4328
    /*
      Should we write to the binlog cache or to the binlog on disk?
4329

4330
      Write to the binlog cache if:
4331 4332 4333 4334 4335 4336 4337 4338 4339 4340 4341 4342
      1 - a transactional engine/table is updated (stmt_has_updated_trans_table == TRUE);
      2 - or the event asks for it (cache_stmt == TRUE);
      3 - or the cache 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).
      
      Write to the binlog on disk if only a non-transactional engine is
      updated and:
      1 - the binlog cache is empty or;
      2 - --binlog-direct-non-transactional-updates is set and we are about to
      use the statement format. When using the row format (cache_stmt == TRUE).
4343 4344 4345
    */
    if (opt_using_transactions && thd)
    {
4346 4347
      if (thd->binlog_setup_trx_data())
        goto err;
4348

4349
      binlog_trx_data *const trx_data=
4350
        (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
4351
      IO_CACHE *trans_log= &trx_data->trans_log;
4352
      my_off_t trans_log_pos= my_b_tell(trans_log);
4353 4354 4355
      if (event_info->get_cache_stmt() || stmt_has_updated_trans_table(thd) ||
          (!thd->variables.binlog_direct_non_trans_update &&
            trans_log_pos != 0))
4356
      {
4357
        DBUG_PRINT("info", ("Using trans_log: cache: %d, trans_log_pos: %lu",
4358
                            event_info->get_cache_stmt(),
4359
                            (ulong) trans_log_pos));
4360
        thd->binlog_start_trans_and_stmt();
4361
        file= trans_log;
4362
      }
4363
      /*
4364
        TODO as Mats suggested, for all the cases above where we write to
4365 4366
        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
4367
        LOCK_log.
4368
      */
4369
    }
4370
#endif /* USING_TRANSACTIONS */
4371
    DBUG_PRINT("info",("event type: %d",event_info->get_type_code()));
4372

4373 4374 4375 4376
    /*
      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
4377 4378

    /*
4379 4380
      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
4381 4382
    */

4383 4384 4385 4386
    /*
      If row-based binlogging, Insert_id, Rand and other kind of "setting
      context" events are not needed.
    */
4387
    if (thd)
4388
    {
4389
      if (!thd->current_stmt_binlog_row_based)
4390
      {
4391
        if (thd->stmt_depends_on_first_successful_insert_id_in_prev_stmt)
4392 4393
        {
          Intvar_log_event e(thd,(uchar) LAST_INSERT_ID_EVENT,
4394
                             thd->first_successful_insert_id_in_prev_stmt_for_binlog);
4395 4396 4397
          if (e.write(file))
            goto err;
        }
4398
        if (thd->auto_inc_intervals_in_cur_stmt_for_binlog.nb_elements() > 0)
4399
        {
4400 4401 4402 4403 4404 4405
          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());
4406 4407 4408 4409 4410 4411 4412 4413 4414 4415 4416 4417 4418 4419
          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;
4420
            get_dynamic(&thd->user_var_events,(uchar*) &user_var_event, i);
4421 4422 4423 4424 4425 4426 4427 4428 4429 4430
            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;
          }
        }
4431
      }
4432
    }
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
4433

4434 4435
    /*
       Write the SQL command
4436
     */
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
4437

4438 4439
    if (event_info->write(file) || 
        DBUG_EVALUATE_IF("injecting_fault_writing", 1, 0))
4440
      goto err;
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
4441

4442
    if (file == &log_file) // we are writing to the real log (disk)
4443
    {
4444
      if (flush_and_sync())
monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
4445
	goto err;
4446 4447
      signal_update();
      rotate_and_purge(RP_LOCK_LOG_IS_ALREADY_LOCKED);
4448
    }
monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
4449
    error=0;
4450

4451 4452
err:
    if (error)
4453
      set_write_error(thd);
4454
  }
4455 4456

  pthread_mutex_unlock(&LOCK_log);
4457
  DBUG_RETURN(error);
4458 4459
}

4460 4461 4462 4463 4464 4465 4466 4467 4468

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,
4469
                    ulonglong current_utime)
4470
{
4471
  return logger.slow_log_print(thd, query, query_length, current_utime);
4472 4473 4474
}


4475 4476 4477 4478 4479 4480 4481 4482 4483 4484 4485 4486 4487 4488 4489 4490 4491 4492 4493 4494 4495 4496 4497 4498 4499 4500 4501 4502
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;
}


4503 4504 4505 4506 4507 4508
bool general_log_print(THD *thd, enum enum_server_command command,
                       const char *format, ...)
{
  va_list args;
  uint error= 0;

4509 4510 4511 4512
  /* Print the message to the buffer if we want to log this king of commands */
  if (! logger.log_command(thd, command))
    return FALSE;

4513 4514 4515 4516 4517 4518 4519
  va_start(args, format);
  error= logger.general_log_print(thd, command, format, args);
  va_end(args);

  return error;
}

4520 4521 4522 4523 4524 4525 4526 4527 4528 4529
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;
}

4530
void MYSQL_BIN_LOG::rotate_and_purge(uint flags)
4531
{
4532 4533 4534
#ifdef HAVE_REPLICATION
  bool check_purge= false;
#endif
4535 4536
  if (!(flags & RP_LOCK_LOG_IS_ALREADY_LOCKED))
    pthread_mutex_lock(&LOCK_log);
4537 4538
  if ((flags & RP_FORCE_ROTATE) ||
      (my_b_tell(&log_file) >= (my_off_t) max_size))
4539
  {
4540
    new_file_without_locking();
4541
#ifdef HAVE_REPLICATION
4542
    check_purge= true;
4543 4544
#endif
  }
4545 4546
  if (!(flags & RP_LOCK_LOG_IS_ALREADY_LOCKED))
    pthread_mutex_unlock(&LOCK_log);
4547 4548 4549 4550 4551 4552 4553 4554 4555 4556 4557 4558 4559

#ifdef HAVE_REPLICATION
  /*
    NOTE: Run purge_logs wo/ holding LOCK_log
          as it otherwise will deadlock in ndbcluster_binlog_index_purge_file
  */
  if (check_purge && expire_logs_days)
  {
    time_t purge_time= my_time(0) - expire_logs_days*24*60*60;
    if (purge_time >= 0)
      purge_logs_before_date(purge_time);
  }
#endif
4560
}
4561

4562
uint MYSQL_BIN_LOG::next_file_id()
4563 4564 4565 4566 4567 4568 4569 4570
{
  uint res;
  pthread_mutex_lock(&LOCK_log);
  res = file_id++;
  pthread_mutex_unlock(&LOCK_log);
  return res;
}

4571

4572 4573 4574 4575 4576 4577 4578 4579 4580 4581 4582 4583 4584 4585 4586 4587 4588 4589 4590 4591
/*
  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;
4592
  uint length= my_b_bytes_in_cache(cache), group, carry, hdr_offs;
4593 4594 4595
  long val;
  uchar header[LOG_EVENT_HEADER_LEN];

4596 4597 4598 4599 4600 4601 4602 4603 4604 4605 4606 4607 4608 4609
  /*
    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.
  */

4610
  group= (uint)my_b_tell(&log_file);
4611 4612
  hdr_offs= carry= 0;

4613 4614
  do
  {
4615 4616 4617 4618 4619 4620

    /*
      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))
4621 4622 4623
    {
      DBUG_ASSERT(carry < LOG_EVENT_HEADER_LEN);

4624
      /* assemble both halves */
4625 4626
      memcpy(&header[carry], (char *)cache->read_pos, LOG_EVENT_HEADER_LEN - carry);

4627
      /* fix end_log_pos */
4628 4629 4630
      val= uint4korr(&header[LOG_POS_OFFSET]) + group;
      int4store(&header[LOG_POS_OFFSET], val);

4631
      /* write the first half of the split header */
4632 4633 4634
      if (my_b_write(&log_file, header, carry))
        return ER_ERROR_ON_WRITE;

4635 4636 4637 4638
      /*
        copy fixed second half of header to cache so the correct
        version will be written later.
      */
4639 4640
      memcpy((char *)cache->read_pos, &header[carry], LOG_EVENT_HEADER_LEN - carry);

4641
      /* next event header at ... */
4642
      hdr_offs = uint4korr(&header[EVENT_LEN_OFFSET]) - carry;
4643 4644 4645 4646

      carry= 0;
    }

4647 4648
    /* if there is anything to write, process it. */

4649
    if (likely(length > 0))
4650
    {
4651
      /*
4652 4653 4654 4655
        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").
4656
      */
4657

4658
      while (hdr_offs < length)
4659
      {
4660 4661 4662 4663
        /*
          partial header only? save what we can get, process once
          we get the rest.
        */
4664

4665
        if (hdr_offs + LOG_EVENT_HEADER_LEN > length)
4666
        {
4667
          carry= length - hdr_offs;
4668
          memcpy(header, (char *)cache->read_pos + hdr_offs, carry);
4669
          length= hdr_offs;
4670 4671 4672 4673
        }
        else
        {
          /* we've got a full event-header, and it came in one piece */
4674

4675
          uchar *log_pos= (uchar *)cache->read_pos + hdr_offs + LOG_POS_OFFSET;
4676

4677 4678 4679
          /* fix end_log_pos */
          val= uint4korr(log_pos) + group;
          int4store(log_pos, val);
4680

4681 4682 4683
          /* next event header at ... */
          log_pos= (uchar *)cache->read_pos + hdr_offs + EVENT_LEN_OFFSET;
          hdr_offs += uint4korr(log_pos);
4684

4685
        }
4686
      }
4687 4688

      /*
4689 4690 4691 4692 4693 4694
        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:
4695
      */
4696
      hdr_offs -= length;
4697 4698 4699
    }

    /* Write data to the binary log file */
4700
    if (my_b_write(&log_file, cache->read_pos, length))
4701
      return ER_ERROR_ON_WRITE;
4702
    cache->read_pos=cache->read_end;		// Mark buffer used up
4703
  } while ((length= my_b_fill(cache)));
4704

4705
  DBUG_ASSERT(carry == 0);
4706 4707 4708 4709 4710 4711 4712

  if (sync_log)
    flush_and_sync();

  return 0;                                     // All OK
}

4713 4714 4715 4716 4717 4718 4719
/*
  Helper function to get the error code of the query to be binlogged.
 */
int query_error_code(THD *thd, bool not_killed)
{
  int error;
  
4720
  if (not_killed || (thd->killed == THD::KILL_BAD_DATA))
4721 4722 4723 4724 4725 4726 4727 4728 4729 4730 4731 4732 4733 4734 4735 4736 4737 4738 4739 4740 4741
  {
    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;
}

4742 4743 4744 4745 4746 4747 4748 4749 4750 4751
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);
He Zhenxing's avatar
He Zhenxing committed
4752
  error= ev.write(&log_file);
4753 4754 4755 4756 4757 4758 4759 4760 4761 4762 4763 4764
  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);
}

4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775
/**
  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
4776
                        contents of the cache.
4777 4778 4779
  @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
4780

4781 4782 4783 4784 4785 4786
  @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.
4787 4788
*/

4789 4790
bool MYSQL_BIN_LOG::write(THD *thd, IO_CACHE *cache, Log_event *commit_event,
                          bool incident)
4791
{
4792
  DBUG_ENTER("MYSQL_BIN_LOG::write(THD *, IO_CACHE *, Log_event *)");
4793
  VOID(pthread_mutex_lock(&LOCK_log));
4794

4795 4796 4797
  /* NULL would represent nothing to replicate after ROLLBACK */
  DBUG_ASSERT(commit_event != NULL);

4798
  DBUG_ASSERT(is_open());
4799
  if (likely(is_open()))                       // Should always be true
4800
  {
4801
    /*
4802 4803 4804 4805
      We only bother to write to the binary log if there is anything
      to write.
     */
    if (my_b_tell(cache) > 0)
4806
    {
4807
      /*
4808 4809 4810
        Log "BEGIN" at the beginning of every transaction.  Here, a
        transaction is either a BEGIN..COMMIT block or a single
        statement in autocommit mode.
4811
      */
He Zhenxing's avatar
He Zhenxing committed
4812
      Query_log_event qinfo(thd, STRING_WITH_LEN("BEGIN"), TRUE, TRUE, 0);
4813

4814 4815 4816 4817 4818 4819 4820 4821 4822 4823 4824 4825 4826 4827 4828 4829 4830 4831 4832
      /*
        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();
                      });
4833

4834 4835
      if ((write_error= write_cache(cache, false, false)))
        goto err;
4836

4837 4838
      if (commit_event && commit_event->write(&log_file))
        goto err;
4839 4840 4841 4842

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

4843 4844 4845 4846 4847 4848 4849 4850 4851 4852
      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
4853
    }
4854

4855 4856 4857 4858 4859 4860 4861 4862
    /*
      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.
    */
4863
    if (commit_event && commit_event->get_type_code() == XID_EVENT)
4864 4865 4866 4867 4868
    {
      pthread_mutex_lock(&LOCK_prep_xids);
      prepared_xids++;
      pthread_mutex_unlock(&LOCK_prep_xids);
    }
4869 4870
    else
      rotate_and_purge(RP_LOCK_LOG_IS_ALREADY_LOCKED);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4871
  }
4872
  VOID(pthread_mutex_unlock(&LOCK_log));
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
4873

4874
  DBUG_RETURN(0);
4875 4876

err:
4877 4878 4879 4880 4881
  if (!write_error)
  {
    write_error= 1;
    sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
  }
4882
  VOID(pthread_mutex_unlock(&LOCK_log));
4883
  DBUG_RETURN(1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4884 4885
}

4886

4887 4888
/**
  Wait until we get a signal that the binary log has been updated.
4889

4890 4891 4892 4893
  @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.
4894

4895
  @note
4896
    One must have a lock on LOCK_log before calling this function.
4897
    This lock will be released before return! That's required by
guilhem@mysql.com's avatar
guilhem@mysql.com committed
4898
    THD::enter_cond() (see NOTES in sql_class.h).
4899 4900
*/

4901
void MYSQL_BIN_LOG::wait_for_update(THD* thd, bool is_slave)
4902
{
monty@mysql.com's avatar
monty@mysql.com committed
4903 4904
  const char *old_msg;
  DBUG_ENTER("wait_for_update");
4905

monty@mysql.com's avatar
monty@mysql.com committed
4906
  old_msg= thd->enter_cond(&update_cond, &LOCK_log,
4907
                           is_slave ?
monty@mysql.com's avatar
monty@mysql.com committed
4908
                           "Has read all relay log; waiting for the slave I/O "
4909
                           "thread to update it" :
monty@mysql.com's avatar
monty@mysql.com committed
4910
                           "Has sent all binlog to slave; waiting for binlog "
4911
                           "to be updated");
4912 4913
  pthread_cond_wait(&update_cond, &LOCK_log);
  thd->exit_cond(old_msg);
monty@mysql.com's avatar
monty@mysql.com committed
4914
  DBUG_VOID_RETURN;
4915
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4916

4917

4918 4919
/**
  Close the log file.
4920

4921 4922 4923 4924 4925
  @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
4926

4927
  @note
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
4928 4929
    One can do an open on the object at once after doing a close.
    The internal structures are not freed until cleanup() is called
4930 4931
*/

4932
void MYSQL_BIN_LOG::close(uint exiting)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4933
{					// One can't set log_type here!
4934
  DBUG_ENTER("MYSQL_BIN_LOG::close");
4935
  DBUG_PRINT("enter",("exiting: %d", (int) exiting));
4936
  if (log_state == LOG_OPENED)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4937
  {
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
4938
#ifdef HAVE_REPLICATION
4939 4940
    if (log_type == LOG_BIN && !no_auto_events &&
	(exiting & LOG_CLOSE_STOP_EVENT))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4941 4942
    {
      Stop_log_event s;
4943
      s.write(&log_file);
4944
      bytes_written+= s.data_written;
4945
      signal_update();
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4946
    }
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
4947
#endif /* HAVE_REPLICATION */
4948 4949

    /* don't pwrite in a file opened with O_APPEND - it doesn't work */
4950
    if (log_file.type == WRITE_CACHE && log_type == LOG_BIN)
4951 4952
    {
      my_off_t offset= BIN_LOG_HEADER_SIZE + FLAGS_OFFSET;
4953 4954
      my_off_t org_position= my_tell(log_file.file, MYF(0));
      uchar flags= 0;            // clearing LOG_EVENT_BINLOG_IN_USE_F
4955
      my_pwrite(log_file.file, &flags, 1, offset, MYF(0));
4956 4957 4958 4959 4960 4961 4962
      /*
        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));
4963 4964
    }

4965 4966
    /* 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
4967
  }
4968 4969 4970 4971 4972 4973

  /*
    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.
  */

4974
  if ((exiting & LOG_CLOSE_INDEX) && my_b_inited(&index_file))
4975
  {
4976 4977
    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
4978
    {
4979 4980
      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
4981
    }
4982
  }
4983
  log_state= (exiting & LOG_CLOSE_TO_BE_OPENED) ? LOG_TO_BE_OPENED : LOG_CLOSED;
4984
  safeFree(name);
4985
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4986 4987 4988
}


4989
void MYSQL_BIN_LOG::set_max_size(ulong max_size_arg)
4990 4991 4992 4993 4994 4995 4996 4997
{
  /*
    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.
  */
4998
  DBUG_ENTER("MYSQL_BIN_LOG::set_max_size");
guilhem@mysql.com's avatar
guilhem@mysql.com committed
4999
  pthread_mutex_lock(&LOCK_log);
5000 5001
  if (is_open())
    max_size= max_size_arg;
guilhem@mysql.com's avatar
guilhem@mysql.com committed
5002 5003
  pthread_mutex_unlock(&LOCK_log);
  DBUG_VOID_RETURN;
5004
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5005

5006

5007 5008
/**
  Check if a string is a valid number.
5009

5010 5011 5012
  @param str			String to test
  @param res			Store value here
  @param allow_wildcards	Set to 1 if we should ignore '%' and '_'
5013

5014
  @note
5015 5016 5017
    For the moment the allow_wildcards argument is not used
    Should be move to some other file.

5018
  @retval
5019
    1	String is a number
5020
  @retval
5021
    0	Error
5022
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5023 5024 5025 5026 5027 5028 5029 5030 5031 5032 5033 5034

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++;
5035 5036
  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
5037 5038 5039 5040 5041 5042 5043
  {
    flag=1;
    str++;
  }
  if (*str == '.')
  {
    for (str++ ;
5044
	 my_isdigit(files_charset_info,*str) ||
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5045 5046 5047 5048 5049 5050 5051 5052 5053 5054 5055 5056 5057 5058 5059 5060 5061 5062 5063
	   (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
}
5064

monty@mysql.com's avatar
monty@mysql.com committed
5065

5066 5067 5068 5069 5070 5071 5072 5073 5074 5075 5076 5077 5078 5079 5080 5081
/*
  Unfortunately, there seems to be no good way
  to restore the original streams upon failure.
*/
static bool redirect_std_streams(const char *file)
{
  if (freopen(file, "a+", stdout) && freopen(file, "a+", stderr))
  {
    setbuf(stderr, NULL);
    return FALSE;
  }

  return TRUE;
}


5082 5083 5084
bool flush_error_log()
{
  bool result=0;
5085
  if (opt_error_log)
5086 5087
  {
    char err_renamed[FN_REFLEN], *end;
5088
    end= strmake(err_renamed,log_error_file,FN_REFLEN-5);
5089
    strmov(end, "-old");
5090
    VOID(pthread_mutex_lock(&LOCK_error_log));
5091
#ifdef __WIN__
5092
    char err_temp[FN_REFLEN+5];
5093 5094 5095 5096
    /*
     On Windows is necessary a temporary file for to rename
     the current error file.
    */
5097
    strxmov(err_temp, err_renamed,"-tmp",NullS);
5098 5099 5100
    (void) my_delete(err_temp, MYF(0)); 
    if (freopen(err_temp,"a+",stdout))
    {
5101 5102 5103 5104
      int fd;
      size_t bytes;
      uchar buf[IO_SIZE];

5105
      freopen(err_temp,"a+",stderr);
5106
      setbuf(stderr, NULL);
5107 5108
      (void) my_delete(err_renamed, MYF(0));
      my_rename(log_error_file,err_renamed,MYF(0));
5109
      redirect_std_streams(log_error_file);
5110

5111 5112
      if ((fd = my_open(err_temp, O_RDONLY, MYF(0))) >= 0)
      {
5113 5114 5115
        while ((bytes= my_read(fd, buf, IO_SIZE, MYF(0))) &&
               bytes != MY_FILE_ERROR)
          my_fwrite(stderr, buf, bytes, MYF(0));
5116 5117 5118 5119 5120 5121 5122 5123
        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));
5124
   if (redirect_std_streams(log_error_file))
5125 5126
     result= 1;
#endif
5127
    VOID(pthread_mutex_unlock(&LOCK_error_log));
5128 5129 5130
  }
   return result;
}
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
5131

5132
void MYSQL_BIN_LOG::signal_update()
monty@mysql.com's avatar
monty@mysql.com committed
5133
{
5134
  DBUG_ENTER("MYSQL_BIN_LOG::signal_update");
monty@mysql.com's avatar
monty@mysql.com committed
5135 5136 5137 5138
  pthread_cond_broadcast(&update_cond);
  DBUG_VOID_RETURN;
}

5139
#ifdef __NT__
5140
static void print_buffer_to_nt_eventlog(enum loglevel level, char *buff,
5141
                                        size_t length, size_t buffLen)
5142 5143
{
  HANDLE event;
5144
  char   *buffptr= buff;
serg@serg.mylan's avatar
serg@serg.mylan committed
5145
  DBUG_ENTER("print_buffer_to_nt_eventlog");
5146

5147 5148
  /* Add ending CR/LF's to string, overwrite last chars if necessary */
  strmov(buffptr+min(length, buffLen-5), "\r\n\r\n");
5149

monty@mysql.com's avatar
monty@mysql.com committed
5150 5151
  setup_windows_event_source();
  if ((event= RegisterEventSource(NULL,"MySQL")))
5152
  {
monty@mysql.com's avatar
monty@mysql.com committed
5153
    switch (level) {
5154
      case ERROR_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
5155
        ReportEvent(event, EVENTLOG_ERROR_TYPE, 0, MSG_DEFAULT, NULL, 1, 0,
5156
                    (LPCSTR*)&buffptr, NULL);
5157 5158
        break;
      case WARNING_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
5159
        ReportEvent(event, EVENTLOG_WARNING_TYPE, 0, MSG_DEFAULT, NULL, 1, 0,
5160
                    (LPCSTR*) &buffptr, NULL);
5161 5162
        break;
      case INFORMATION_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
5163
        ReportEvent(event, EVENTLOG_INFORMATION_TYPE, 0, MSG_DEFAULT, NULL, 1,
5164
                    0, (LPCSTR*) &buffptr, NULL);
5165 5166 5167 5168 5169 5170 5171
        break;
    }
    DeregisterEventSource(event);
  }

  DBUG_VOID_RETURN;
}
monty@mysql.com's avatar
monty@mysql.com committed
5172 5173
#endif /* __NT__ */

5174

5175
#ifndef EMBEDDED_LIBRARY
5176 5177
static void print_buffer_to_file(enum loglevel level, const char *buffer,
                                 size_t length)
5178 5179 5180 5181 5182 5183 5184 5185 5186
{
  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));

5187
  skr= my_time(0);
5188 5189
  localtime_r(&skr, &tm_tmp);
  start=&tm_tmp;
5190

5191
  fprintf(stderr, "%02d%02d%02d %2d:%02d:%02d [%s] %.*s\n",
5192 5193 5194 5195 5196 5197 5198 5199
          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"),
5200
          (int) length, buffer);
5201 5202 5203 5204 5205 5206 5207

  fflush(stderr);

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

5208 5209 5210
/**
  Prints a printf style message to the error log and, under NT, to the
  Windows event log.
5211

5212 5213 5214 5215 5216 5217 5218 5219 5220 5221 5222 5223
  This function prints the message into a buffer and then sends that buffer
  to other functions to write that message to other logging sources.

  @param level          The level of the msg significance
  @param format         Printf style format of message
  @param args           va_list list of arguments for the message

  @returns
    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@production.mysql.com's avatar
rburnett@production.mysql.com committed
5224
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
5225 5226
{
  char   buff[1024];
5227
  size_t length;
5228
  DBUG_ENTER("vprint_msg_to_log");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
5229

5230
  length= my_vsnprintf(buff, sizeof(buff), format, args);
5231
  print_buffer_to_file(level, buff, length);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
5232 5233

#ifdef __NT__
monty@mysql.com's avatar
monty@mysql.com committed
5234
  print_buffer_to_nt_eventlog(level, buff, length, sizeof(buff));
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
5235
#endif
5236

5237
  DBUG_RETURN(0);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
5238
}
5239
#endif /* EMBEDDED_LIBRARY */
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
5240

5241

monty@mysql.com's avatar
monty@mysql.com committed
5242
void sql_print_error(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
5243 5244
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
5245
  DBUG_ENTER("sql_print_error");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
5246

serg@serg.mylan's avatar
serg@serg.mylan committed
5247
  va_start(args, format);
5248
  error_log_print(ERROR_LEVEL, format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
5249
  va_end(args);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
5250 5251 5252 5253 5254

  DBUG_VOID_RETURN;
}


monty@mysql.com's avatar
monty@mysql.com committed
5255
void sql_print_warning(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
5256 5257
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
5258
  DBUG_ENTER("sql_print_warning");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
5259

serg@serg.mylan's avatar
serg@serg.mylan committed
5260
  va_start(args, format);
5261
  error_log_print(WARNING_LEVEL, format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
5262
  va_end(args);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
5263 5264 5265 5266 5267

  DBUG_VOID_RETURN;
}


monty@mysql.com's avatar
monty@mysql.com committed
5268
void sql_print_information(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
5269 5270
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
5271
  DBUG_ENTER("sql_print_information");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
5272

serg@serg.mylan's avatar
serg@serg.mylan committed
5273
  va_start(args, format);
5274
  error_log_print(INFORMATION_LEVEL, format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
5275
  va_end(args);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
5276

guilhem@mysql.com's avatar
guilhem@mysql.com committed
5277 5278
  DBUG_VOID_RETURN;
}
5279

5280

5281 5282 5283 5284 5285 5286 5287 5288
/********* 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
5289 5290
  states - active, syncing, pool.
  there could be only one page in active or syncing states,
5291
  but many in pool - pool is fifo queue.
5292
  usual lifecycle of a page is pool->active->syncing->pool
5293
  "active" page - is a page where new xid's are logged.
5294 5295
  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.
5296
  when the sync is done the page is moved to a pool and an active page
5297 5298 5299 5300 5301 5302 5303 5304 5305 5306 5307 5308 5309
  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.
5310 5311 5312 5313
  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)

5314 5315
  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
5316 5317
  dirty - we don't sync removals to disk.
*/
5318 5319 5320 5321 5322

ulong tc_log_page_waits= 0;

#ifdef HAVE_MMAP

5323 5324
#define TC_LOG_HEADER_SIZE (sizeof(tc_log_magic)+1)

5325
static const char tc_log_magic[]={(char) 254, 0x23, 0x05, 0x74};
5326

5327
ulong opt_tc_log_size= TC_LOG_MIN_SIZE;
5328
ulong tc_log_max_pages_used=0, tc_log_page_size=0, tc_log_cur_pages_used=0;
5329 5330 5331 5332 5333 5334 5335

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

5336
  DBUG_ASSERT(total_ha_2pc > 1);
5337 5338
  DBUG_ASSERT(opt_name && opt_name[0]);

serg@serg.mylan's avatar
serg@serg.mylan committed
5339
  tc_log_page_size= my_getpagesize();
5340 5341 5342
  DBUG_ASSERT(TC_LOG_PAGE_SIZE % tc_log_page_size == 0);

  fn_format(logname,opt_name,mysql_data_home,"",MY_UNPACK_FILENAME);
5343
  if ((fd= my_open(logname, O_RDWR, MYF(0))) < 0)
5344
  {
5345 5346
    if (my_errno != ENOENT)
      goto err;
5347 5348
    if (using_heuristic_recover())
      return 1;
5349
    if ((fd= my_create(logname, CREATE_MODE, O_RDWR, MYF(MY_WME))) < 0)
5350 5351
      goto err;
    inited=1;
5352
    file_length= opt_tc_log_size;
5353 5354 5355 5356 5357
    if (my_chsize(fd, file_length, 0, MYF(MY_WME)))
      goto err;
  }
  else
  {
5358 5359
    inited= 1;
    crashed= TRUE;
serg@serg.mylan's avatar
serg@serg.mylan committed
5360
    sql_print_information("Recovering after a crash using %s", opt_name);
5361 5362 5363 5364 5365 5366
    if (tc_heuristic_recover)
    {
      sql_print_error("Cannot perform automatic crash recovery when "
                      "--tc-heuristic-recover is used");
      goto err;
    }
5367
    file_length= my_seek(fd, 0L, MY_SEEK_END, MYF(MY_WME+MY_FAE));
5368 5369 5370 5371
    if (file_length == MY_FILEPOS_ERROR || file_length % tc_log_page_size)
      goto err;
  }

5372
  data= (uchar *)my_mmap(0, (size_t)file_length, PROT_READ|PROT_WRITE,
5373 5374 5375 5376 5377 5378 5379 5380
                        MAP_NOSYNC|MAP_SHARED, fd, 0);
  if (data == MAP_FAILED)
  {
    my_errno=errno;
    goto err;
  }
  inited=2;

5381
  npages=(uint)file_length/tc_log_page_size;
5382 5383 5384 5385 5386 5387 5388 5389 5390 5391 5392 5393 5394 5395 5396 5397 5398
  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);
5399
  pages[0].start=pages[0].end-pages[0].size;
5400 5401 5402
  pages[npages-1].next=0;
  inited=4;

5403
  if (crashed && recover())
5404
      goto err;
5405

5406
  memcpy(data, tc_log_magic, sizeof(tc_log_magic));
5407
  data[sizeof(tc_log_magic)]= (uchar)total_ha_2pc;
5408 5409 5410 5411 5412 5413 5414 5415 5416 5417 5418
  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;

5419
  syncing= 0;
5420 5421 5422 5423 5424 5425 5426 5427 5428 5429 5430
  active=pages;
  pool=pages+1;
  pool_last=pages+npages-1;

  return 0;

err:
  close();
  return 1;
}

5431 5432
/**
  there is no active page, let's got one from the pool.
5433

5434 5435 5436
  Two strategies here:
    -# take the first from the pool
    -# if there're waiters - take the one with the most free space.
5437

5438 5439 5440
  @todo
    TODO page merging. try to allocate adjacent page first,
    so that they can be flushed both in one sync
5441
*/
5442

5443 5444 5445 5446 5447
void TC_LOG_MMAP::get_active_from_pool()
{
  PAGE **p, **best_p=0;
  int best_free;

5448 5449 5450
  if (syncing)
    pthread_mutex_lock(&LOCK_pool);

5451 5452 5453
  do
  {
    best_p= p= &pool;
5454 5455
    if ((*p)->waiters == 0) // can the first page be used ?
      break;                // yes - take it.
5456

5457
    best_free=0;            // no - trying second strategy
5458 5459 5460 5461 5462 5463 5464 5465 5466 5467 5468 5469
    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;
5470
  if (active->free == active->size) // we've chosen an empty page
5471 5472 5473 5474 5475
  {
    tc_log_cur_pages_used++;
    set_if_bigger(tc_log_max_pages_used, tc_log_cur_pages_used);
  }

5476
  if ((*best_p)->next)              // unlink the page from the pool
5477 5478 5479
    *best_p=(*best_p)->next;
  else
    pool_last=*best_p;
5480 5481 5482

  if (syncing)
    pthread_mutex_unlock(&LOCK_pool);
5483 5484
}

5485 5486 5487 5488
/**
  @todo
  perhaps, increase log size ?
*/
5489 5490 5491 5492 5493 5494 5495 5496 5497
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);
5498
  return 1; // always return 1
5499 5500
}

5501 5502
/**
  Record that transaction XID is committed on the persistent storage.
5503 5504 5505 5506 5507 5508 5509 5510 5511 5512 5513 5514 5515 5516 5517

    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.
5518

5519 5520 5521 5522 5523 5524 5525
  @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.
5526 5527
*/

5528
int TC_LOG_MMAP::log_xid(THD *thd, my_xid xid)
5529 5530 5531 5532 5533 5534 5535
{
  int err;
  PAGE *p;
  ulong cookie;

  pthread_mutex_lock(&LOCK_active);

5536 5537 5538 5539 5540 5541 5542
  /*
    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.
  */
5543 5544 5545
  while (unlikely(active && active->free == 0))
    pthread_cond_wait(&COND_active, &LOCK_active);

5546
  /* no active page ? take one from the pool */
5547 5548 5549 5550 5551 5552
  if (active == 0)
    get_active_from_pool();

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

5553
  /* searching for an empty slot */
5554 5555 5556 5557 5558 5559
  while (*p->ptr)
  {
    p->ptr++;
    DBUG_ASSERT(p->ptr < p->end);               // because p->free > 0
  }

5560 5561
  /* found! store xid there and mark the page dirty */
  cookie= (ulong)((uchar *)p->ptr - data);      // can never be zero
5562 5563 5564 5565 5566
  *p->ptr++= xid;
  p->free--;
  p->state= DIRTY;

  /* to sync or not to sync - this is the question */
5567
  pthread_mutex_unlock(&LOCK_active);
5568 5569 5570
  pthread_mutex_lock(&LOCK_sync);
  pthread_mutex_unlock(&p->lock);

5571
  if (syncing)
5572 5573
  {                                          // somebody's syncing. let's wait
    p->waiters++;
5574
    /*
monty@mishka.local's avatar
monty@mishka.local committed
5575
      note - it must be while (), not do ... while () here
5576 5577 5578
      as p->state may be not DIRTY when we come here
    */
    while (p->state == DIRTY && syncing)
5579 5580 5581 5582 5583 5584 5585 5586 5587 5588 5589
      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
5590 5591 5592 5593
  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
5594 5595 5596 5597 5598 5599 5600 5601 5602 5603 5604 5605 5606
  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;

5607
  DBUG_ASSERT(syncing != active);
5608 5609 5610 5611 5612

  /*
    sit down and relax - this can take a while...
    note - no locks are held at this point
  */
5613
  err= my_msync(fd, syncing->start, 1, MS_SYNC);
5614

5615
  /* page is synced. let's move it to the pool */
5616
  pthread_mutex_lock(&LOCK_pool);
5617 5618 5619 5620 5621
  pool_last->next=syncing;
  pool_last=syncing;
  syncing->next=0;
  syncing->state= err ? ERROR : POOL;
  pthread_cond_broadcast(&syncing->cond);    // signal "sync done"
5622 5623 5624
  pthread_cond_signal(&COND_pool);           // in case somebody's waiting
  pthread_mutex_unlock(&LOCK_pool);

5625
  /* marking 'syncing' slot free */
5626
  pthread_mutex_lock(&LOCK_sync);
5627
  syncing=0;
5628 5629 5630 5631 5632
  pthread_cond_signal(&active->cond);        // wake up a new syncer
  pthread_mutex_unlock(&LOCK_sync);
  return err;
}

5633
/**
5634
  erase xid from the page, update page free space counters/pointers.
5635
  cookie points directly to the memory where xid was logged.
5636
*/
5637

5638 5639 5640 5641 5642 5643 5644 5645 5646 5647 5648 5649 5650
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);
5651
  if (p->free == p->size)               // the page is completely empty
5652 5653
    statistic_decrement(tc_log_cur_pages_used, &LOCK_status);
  if (p->waiters == 0)                 // the page is in pool and ready to rock
5654
    pthread_cond_signal(&COND_pool);   // ping ... for overflow()
5655 5656 5657 5658 5659
  pthread_mutex_unlock(&p->lock);
}

void TC_LOG_MMAP::close()
{
serg@serg.mylan's avatar
serg@serg.mylan committed
5660
  uint i;
5661 5662 5663 5664 5665 5666 5667 5668 5669
  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
5670
    for (i=0; i < npages; i++)
5671 5672 5673 5674 5675 5676 5677
    {
      if (pages[i].ptr == 0)
        break;
      pthread_mutex_destroy(&pages[i].lock);
      pthread_cond_destroy(&pages[i].cond);
    }
  case 3:
5678
    my_free((uchar*)pages, MYF(0));
5679
  case 2:
5680
    my_munmap((char*)data, (size_t)file_length);
5681 5682 5683 5684 5685 5686 5687 5688 5689 5690 5691 5692 5693 5694 5695 5696 5697 5698 5699 5700 5701 5702 5703 5704 5705
  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)
  {
5706
    sql_print_error("Recovery failed! You must enable "
5707 5708 5709
                    "exactly %d storage engines that support "
                    "two-phase commit protocol",
                    data[sizeof(tc_log_magic)]);
5710 5711 5712 5713
    goto err1;
  }

  if (hash_init(&xids, &my_charset_bin, tc_log_page_size/3, 0,
5714
                sizeof(my_xid), 0, 0, MYF(0)))
5715 5716 5717 5718 5719
    goto err1;

  for ( ; p < end_p ; p++)
  {
    for (my_xid *x=p->start; x < p->end; x++)
5720
      if (*x && my_hash_insert(&xids, (uchar *)x))
5721 5722 5723 5724 5725 5726 5727
        goto err2; // OOM
  }

  if (ha_recover(&xids))
    goto err2;

  hash_free(&xids);
5728
  bzero(data, (size_t)file_length);
5729 5730 5731 5732 5733 5734 5735 5736 5737 5738 5739
  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
5740 5741 5742 5743 5744
#endif

TC_LOG *tc_log;
TC_LOG_DUMMY tc_log_dummy;
TC_LOG_MMAP  tc_log_mmap;
5745

5746 5747
/**
  Perform heuristic recovery, if --tc-heuristic-recover was used.
5748

5749
  @note
5750 5751
    no matter whether heuristic recovery was successful or not
    mysqld must exit. So, return value is the same in both cases.
5752 5753 5754 5755 5756

  @retval
    0	no heuristic recovery was requested
  @retval
    1   heuristic recovery was performed
5757 5758
*/

5759 5760 5761 5762 5763 5764 5765 5766 5767 5768 5769 5770 5771
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 ******/
5772
#define TC_LOG_BINLOG MYSQL_BIN_LOG
5773

5774 5775 5776
/**
  @todo
  keep in-memory list of prepared transactions
5777 5778 5779 5780 5781
  (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!
*/

5782
int TC_LOG_BINLOG::open(const char *opt_name)
5783
{
5784 5785
  LOG_INFO log_info;
  int      error= 1;
5786 5787

  DBUG_ASSERT(total_ha_2pc > 1);
5788
  DBUG_ASSERT(opt_name && opt_name[0]);
5789 5790 5791 5792

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

5793 5794 5795 5796 5797 5798 5799
  if (!my_b_inited(&index_file))
  {
    /* There was a failure to open the index file, can't open the binlog */
    cleanup();
    return 1;
  }

5800
  if (using_heuristic_recover())
5801 5802
  {
    /* generate a new binlog to mask a corrupted one */
5803
    open(opt_name, LOG_BIN, 0, WRITE_CACHE, 0, max_binlog_size, 0, TRUE);
5804
    cleanup();
5805
    return 1;
5806
  }
5807

5808
  if ((error= find_log_pos(&log_info, NullS, 1)))
5809
  {
5810 5811 5812 5813 5814
    if (error != LOG_INFO_EOF)
      sql_print_error("find_log_pos() failed (error: %d)", error);
    else
      error= 0;
    goto err;
5815
  }
5816 5817 5818 5819 5820

  {
    const char *errmsg;
    IO_CACHE    log;
    File        file;
5821 5822
    Log_event  *ev=0;
    Format_description_log_event fdle(BINLOG_VERSION);
5823
    char        log_name[FN_REFLEN];
5824 5825 5826 5827

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

5828
    do
5829
    {
5830 5831 5832 5833 5834 5835 5836
      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;
5837
    }
5838

5839
    if ((file= open_binlog(&log, log_name, &errmsg)) < 0)
5840 5841 5842 5843 5844
    {
      sql_print_error("%s", errmsg);
      goto err;
    }

5845 5846 5847
    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
5848 5849
    {
      sql_print_information("Recovering after a crash using %s", opt_name);
5850
      error= recover(&log, (Format_description_log_event *)ev);
serg@serg.mylan's avatar
serg@serg.mylan committed
5851
    }
5852 5853
    else
      error=0;
5854 5855 5856 5857 5858 5859 5860 5861 5862 5863

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

    if (error)
      goto err;
  }

err:
5864
  return error;
5865 5866
}

5867
/** This is called on shutdown, after ha_panic. */
5868
void TC_LOG_BINLOG::close()
5869 5870 5871 5872 5873 5874
{
  DBUG_ASSERT(prepared_xids==0);
  pthread_mutex_destroy(&LOCK_prep_xids);
  pthread_cond_destroy (&COND_prep_xids);
}

5875 5876 5877
/**
  @todo
  group commit
5878

5879 5880 5881 5882
  @retval
    0    error
  @retval
    1    success
5883
*/
5884
int TC_LOG_BINLOG::log_xid(THD *thd, my_xid xid)
5885
{
5886
  DBUG_ENTER("TC_LOG_BINLOG::log");
5887
  Xid_log_event xle(thd, xid);
5888
  binlog_trx_data *trx_data=
5889
    (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
5890 5891 5892 5893 5894
  /*
    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));
5895 5896
}

5897
void TC_LOG_BINLOG::unlog(ulong cookie, my_xid xid)
5898
{
5899
  pthread_mutex_lock(&LOCK_prep_xids);
5900
  DBUG_ASSERT(prepared_xids > 0);
5901 5902
  if (--prepared_xids == 0) {
    DBUG_PRINT("info", ("prepared_xids=%lu", prepared_xids));
5903
    pthread_cond_signal(&COND_prep_xids);
5904
  }
5905
  pthread_mutex_unlock(&LOCK_prep_xids);
5906
  rotate_and_purge(0);     // as ::write() did not rotate
5907 5908
}

5909
int TC_LOG_BINLOG::recover(IO_CACHE *log, Format_description_log_event *fdle)
5910 5911 5912 5913 5914 5915
{
  Log_event  *ev;
  HASH xids;
  MEM_ROOT mem_root;

  if (! fdle->is_valid() ||
5916
      hash_init(&xids, &my_charset_bin, TC_LOG_PAGE_SIZE/3, 0,
5917
                sizeof(my_xid), 0, 0, MYF(0)))
5918 5919
    goto err1;

5920
  init_alloc_root(&mem_root, TC_LOG_PAGE_SIZE, TC_LOG_PAGE_SIZE);
5921

5922 5923
  fdle->flags&= ~LOG_EVENT_BINLOG_IN_USE_F; // abort on the first error

5924 5925 5926 5927 5928
  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;
5929 5930
      uchar *x= (uchar *) memdup_root(&mem_root, (uchar*) &xev->xid,
                                      sizeof(xev->xid));
5931
      if (!x || my_hash_insert(&xids, x))
5932 5933 5934 5935 5936 5937 5938 5939 5940 5941 5942 5943 5944 5945 5946 5947 5948 5949 5950 5951 5952 5953 5954
        goto err2;
    }
    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;
}

5955 5956 5957 5958 5959 5960 5961 5962 5963 5964 5965 5966 5967 5968 5969 5970 5971 5972 5973 5974 5975 5976 5977

#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 */


5978
struct st_mysql_storage_engine binlog_storage_engine=
5979
{ MYSQL_HANDLERTON_INTERFACE_VERSION };
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
5980 5981 5982 5983

mysql_declare_plugin(binlog)
{
  MYSQL_STORAGE_ENGINE_PLUGIN,
5984 5985
  &binlog_storage_engine,
  "binlog",
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
5986
  "MySQL AB",
5987
  "This is a pseudo storage engine to represent the binlog in a transaction",
5988
  PLUGIN_LICENSE_GPL,
5989
  binlog_init, /* Plugin Init */
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
5990 5991
  NULL, /* Plugin Deinit */
  0x0100 /* 1.0 */,
5992 5993 5994
  NULL,                       /* status variables                */
  NULL,                       /* system variables                */
  NULL                        /* config options                  */
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
5995 5996
}
mysql_declare_plugin_end;