log.cc 130 KB
Newer Older
1
/* Copyright (C) 2000-2003 MySQL AB
serg@serg.mylan's avatar
serg@serg.mylan committed
2

bk@work.mysql.com's avatar
bk@work.mysql.com committed
3 4 5 6
   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
   the Free Software Foundation; either version 2 of the License, or
   (at your option) any later version.
serg@serg.mylan's avatar
serg@serg.mylan committed
7

bk@work.mysql.com's avatar
bk@work.mysql.com committed
8 9 10 11
   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
12

bk@work.mysql.com's avatar
bk@work.mysql.com committed
13 14 15 16 17 18
   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 */


/* logging of commands */
19
/* 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
20 21

#include "mysql_priv.h"
22
#include "sql_repl.h"
lars@mysql.com's avatar
lars@mysql.com committed
23
#include "rpl_filter.h"
bk@work.mysql.com's avatar
bk@work.mysql.com committed
24 25 26 27 28

#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
29 30 31 32
#ifdef __NT__
#include "message.h"
#endif

acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
33 34
#include <mysql/plugin.h>

35 36 37 38 39 40 41
/* max size of the log message */
#define MAX_LOG_BUFFER_SIZE 1024
#define MAX_USER_HOST_SIZE 512
#define MAX_TIME_SIZE 32

LOGGER logger;

42
MYSQL_BIN_LOG mysql_bin_log;
43
ulong sync_binlog_counter= 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
44

45 46
static Muted_query_log_event invisible_commit;

bk@work.mysql.com's avatar
bk@work.mysql.com committed
47 48
static bool test_if_number(const char *str,
			   long *res, bool allow_wildcards);
49
static int binlog_init();
50 51 52 53 54 55 56
static int binlog_close_connection(THD *thd);
static int binlog_savepoint_set(THD *thd, void *sv);
static int binlog_savepoint_rollback(THD *thd, void *sv);
static int binlog_commit(THD *thd, bool all);
static int binlog_rollback(THD *thd, bool all);
static int binlog_prepare(THD *thd, bool all);

57 58 59 60 61 62 63 64
sql_print_message_func sql_print_message_handlers[3] =
{
  sql_print_information,
  sql_print_warning,
  sql_print_error
};


65 66 67 68 69 70 71
char *make_default_log_name(char *buff,const char* log_ext)
{
  strmake(buff, glob_hostname, FN_REFLEN-5);
  return fn_format(buff, buff, mysql_data_home, log_ext,
                   MYF(MY_UNPACK_FILENAME|MY_APPEND_EXT));
}

72 73 74 75 76 77
/*
  This is a POD. Please keep it that way!

  Don't add constructors, destructors, or virtual functions.
*/
struct binlog_trx_data {
78 79
  bool empty() const
  {
80
#ifdef HAVE_ROW_BASED_REPLICATION
81
    return pending == NULL && my_b_tell(&trans_log) == 0;
82 83 84
#else
    return my_b_tell(&trans_log) == 0;
#endif
85
  }
86
  binlog_trx_data() {}
87
  IO_CACHE trans_log;                         // The transaction cache
88
#ifdef HAVE_ROW_BASED_REPLICATION
89
  Rows_log_event *pending;                // The pending binrows event
90
#endif
91 92
};

93
handlerton *binlog_hton;
94 95 96 97 98 99 100

/*
  Open log table of a given type (general or slow log)

  SYNOPSIS
    open_log_table()

101 102
    log_table_type   type of the log table to open: QUERY_LOG_GENERAL
                     or QUERY_LOG_SLOW
103 104 105 106 107 108 109 110 111 112 113 114 115 116

  DESCRIPTION

    The function opens a log table and marks it as such. Log tables are open
    during the whole time, while server is running. Except for the moments
    when they have to be reopened: during FLUSH LOGS and TRUNCATE. This
    function is invoked directly only once during startup. All subsequent
    calls happen through reopen_log_table(), which performs additional check.

  RETURN
    FALSE - OK
    TRUE - error occured
*/

117
bool Log_to_csv_event_handler::open_log_table(uint log_table_type)
118 119 120 121 122 123
{
  THD *log_thd, *curr= current_thd;
  TABLE_LIST *table;
  bool error= FALSE;
  DBUG_ENTER("open_log_table");

124
  switch (log_table_type) {
125
  case QUERY_LOG_GENERAL:
126 127 128 129 130 131 132
    log_thd= general_log_thd;
    table= &general_log;
    /* clean up table before reuse/initial usage */
    bzero((char*) table, sizeof(TABLE_LIST));
    table->alias= table->table_name= (char*) "general_log";
    table->table_name_length= 11;
    break;
133
  case QUERY_LOG_SLOW:
134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181
    log_thd= slow_log_thd;
    table= &slow_log;
    bzero((char*) table, sizeof(TABLE_LIST));
    table->alias= table->table_name= (char*) "slow_log";
    table->table_name_length= 8;
    break;
  default:
    DBUG_ASSERT(0);
  }

  /*
    This way we check that appropriate log thd was created ok during
    initialization. We cannot check "is_log_tables_initialized" var, as
    the very initialization is not finished until this function is
    completed in the very first time.
  */
  if (!log_thd)
  {
    DBUG_PRINT("error",("Cannot initialize log tables"));
    DBUG_RETURN(TRUE);
  }

  /*
    Set THD's thread_stack. This is needed to perform stack overrun
    check, which is done by some routines (e.g. open_table()).
    In the case we are called by thread, which already has this parameter
    set, we use this value. Otherwise we do a wild guess. This won't help
    to correctly track the stack overrun in these exceptional cases (which
    could probably happen only during startup and shutdown) but at least
    lets us to pass asserts.
    The problem stems from the fact that logger THDs are not real threads.
  */
  if (curr)
    log_thd->thread_stack= curr->thread_stack;
  else
    log_thd->thread_stack= (char*) &log_thd;

  log_thd->store_globals();

  table->lock_type= TL_WRITE_CONCURRENT_INSERT;
  table->db= log_thd->db;
  table->db_length= log_thd->db_length;

  if (simple_open_n_lock_tables(log_thd, table) ||
      table->table->file->extra(HA_EXTRA_MARK_AS_LOG_TABLE) ||
      table->table->file->ha_rnd_init(0))
    error= TRUE;
  else
182 183
  {
    table->table->use_all_columns();
184
    table->table->locked_by_logger= TRUE;
185
  }
186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236
  /* restore thread settings */
  if (curr)
    curr->store_globals();
  else
  {
    my_pthread_setspecific_ptr(THR_THD,  0);
    my_pthread_setspecific_ptr(THR_MALLOC, 0);
  }

  DBUG_RETURN(error);
}


Log_to_csv_event_handler::Log_to_csv_event_handler()
{
  /* init artificial THD's */
  general_log_thd= new THD;
  /* logger thread always works with mysql database */
  general_log_thd->db= my_strdup("mysql", MYF(0));
  general_log_thd->db_length= 5;

  slow_log_thd= new THD;
  /* logger thread always works with mysql database */
  slow_log_thd->db= my_strdup("mysql", MYF(0));;
  slow_log_thd->db_length= 5;
}


Log_to_csv_event_handler::~Log_to_csv_event_handler()
{
  /* now cleanup the tables */
  if (general_log_thd)
  {
    delete general_log_thd;
    general_log_thd= NULL;
  }

  if (slow_log_thd)
  {
    delete slow_log_thd;
    slow_log_thd= NULL;
  }
}


/*
  Reopen log table of a given type

  SYNOPSIS
    reopen_log_table()

237 238
    log_table_type   type of the log table to open: QUERY_LOG_GENERAL
                     or QUERY_LOG_SLOW
239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254

  DESCRIPTION

    The function is a wrapper around open_log_table(). It is used during
    FLUSH LOGS and TRUNCATE of the log tables (i.e. when we need to close
    and reopen them). The difference is in the check of the
    logger.is_log_tables_initialized var, which can't be done in
    open_log_table(), as it makes no sense during startup.

    NOTE: this code assumes that we have logger mutex locked

  RETURN
    FALSE - ok
    TRUE - open_log_table() returned an error
*/

255
bool Log_to_csv_event_handler::reopen_log_table(uint log_table_type)
256 257 258 259
{
  /* don't open the log table, if it wasn't enabled during startup */
  if (!logger.is_log_tables_initialized)
    return FALSE;
260
  return open_log_table(log_table_type);
261 262 263 264
}

void Log_to_csv_event_handler::cleanup()
{
265 266 267 268
  if (opt_log)
    close_log_table(QUERY_LOG_GENERAL, FALSE);
  if (opt_slow_log)
    close_log_table(QUERY_LOG_SLOW, FALSE);
269 270 271 272 273 274 275 276 277
  logger.is_log_tables_initialized= FALSE;
}

/* log event handlers */

/*
  Log command to the general log table

  SYNOPSIS
278
    log_general()
279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302

    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 the general log table

  RETURN
    FALSE - OK
    TRUE - error occured
*/

bool Log_to_csv_event_handler::
  log_general(time_t event_time, const char *user_host,
              uint user_host_len, int thread_id,
              const char *command_type, uint command_type_len,
303 304
              const char *sql_text, uint sql_text_len,
              CHARSET_INFO *client_cs)
305 306 307 308 309 310 311
{
  TABLE *table= general_log.table;

  /* below should never happen */
  if (unlikely(!logger.is_log_tables_initialized))
    return FALSE;

312 313 314 315 316
  /*
    NOTE: we do not call restore_record() here, as all fields are
    filled by the Logger (=> no need to load default ones).
  */

317 318 319
  /* log table entries are not replicated at the moment */
  tmp_disable_binlog(current_thd);

320
  /* Set current time. Required for CURRENT_TIMESTAMP to work */
321
  general_log_thd->start_time= event_time;
322 323 324 325 326

  /*
    We do not set a value for table->field[0], as it will use
    default value (which is CURRENT_TIMESTAMP).
  */
327

328
  table->field[1]->store(user_host, user_host_len, client_cs);
329
  table->field[1]->set_notnull();
330
  table->field[2]->store((longlong) thread_id, TRUE);
331
  table->field[2]->set_notnull();
332
  table->field[3]->store((longlong) server_id, TRUE);
333
  table->field[3]->set_notnull();
334
  table->field[4]->store(command_type, command_type_len, client_cs);
335
  table->field[4]->set_notnull();
336
  table->field[5]->store(sql_text, sql_text_len, client_cs);
337
  table->field[5]->set_notnull();
338 339 340 341 342 343 344 345 346 347 348 349
  table->file->ha_write_row(table->record[0]);

  reenable_binlog(current_thd);

  return FALSE;
}


/*
  Log a query to the slow log table

  SYNOPSIS
350
    log_slow()
351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382
    thd               THD of the query
    current_time      current timestamp
    query_start_arg   command start timestamp
    user_host         the pointer to the string with user@host info
    user_host_len     length of the user_host string. this is computed once
                      and passed to all general log event handlers
    query_time        Amount of time the query took to execute (in seconds)
    lock_time         Amount of time the query was locked (in seconds)
    is_command        The flag, which determines, whether the sql_text is a
                      query or an administrator command (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,
           longlong query_time, longlong lock_time, bool is_command,
           const char *sql_text, uint sql_text_len)
{
  /* table variables */
  TABLE *table= slow_log.table;
383
  CHARSET_INFO *client_cs= thd->variables.character_set_client;
384

385
  DBUG_ENTER("log_slow");
386 387 388 389 390 391 392 393 394 395

  /* below should never happen */
  if (unlikely(!logger.is_log_tables_initialized))
    return FALSE;

  /* log table entries are not replicated at the moment */
  tmp_disable_binlog(current_thd);

  /*
     Set start time for CURRENT_TIMESTAMP to the start of the query.
396
     This will be default value for the field[0]
397 398
  */
  slow_log_thd->start_time= query_start_arg;
399
  restore_record(table, s->default_values);    // Get empty record
400

401 402 403 404
  /*
    We do not set a value for table->field[0], as it will use
    default value.
  */
405 406

  /* store the value */
407
  table->field[1]->store(user_host, user_host_len, client_cs);
408 409 410 411

  if (query_start_arg)
  {
    /* fill in query_time field */
412
    table->field[2]->store(query_time, TRUE);
413
    /* lock_time */
414
    table->field[3]->store(lock_time, TRUE);
415
    /* rows_sent */
416
    table->field[4]->store((longlong) thd->sent_row_count, TRUE);
417
    /* rows_examined */
418
    table->field[5]->store((longlong) thd->examined_row_count, TRUE);
419 420 421 422 423 424 425 426 427
  }
  else
  {
    table->field[2]->set_null();
    table->field[3]->set_null();
    table->field[4]->set_null();
    table->field[5]->set_null();
  }

428
  /* fill database field */
429
  if (thd->db)
430
  {
431
    table->field[6]->store(thd->db, thd->db_length, client_cs);
432 433
    table->field[6]->set_notnull();
  }
434

435
  if (thd->stmt_depends_on_first_successful_insert_id_in_prev_stmt)
436
  {
437 438
    table->field[7]->store((longlong)
                           thd->first_successful_insert_id_in_prev_stmt_for_binlog, TRUE);
439 440
    table->field[7]->set_notnull();
  }
441

442 443 444 445 446 447 448
  /*
    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)
449
  {
450 451
    table->field[8]->store((longlong)
                           thd->auto_inc_intervals_in_cur_stmt_for_binlog.minimum(), TRUE);
452 453
    table->field[8]->set_notnull();
  }
454

455
  table->field[9]->store((longlong) server_id, TRUE);
456
  table->field[9]->set_notnull();
457 458

  /* sql_text */
459
  table->field[10]->store(sql_text,sql_text_len, client_cs);
460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514

  /* write the row */
  table->file->ha_write_row(table->record[0]);

  reenable_binlog(current_thd);

  DBUG_RETURN(0);
}

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();
}


/* Wrapper around MYSQL_LOG::write() for slow log */

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,
           longlong query_time, longlong lock_time, bool is_command,
           const char *sql_text, uint sql_text_len)
{
  return mysql_slow_log.write(thd, current_time, query_start_arg,
                              user_host, user_host_len,
                              query_time, lock_time, is_command,
                              sql_text, sql_text_len);
}


/*
   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::
  log_general(time_t event_time, const char *user_host,
              uint user_host_len, int thread_id,
              const char *command_type, uint command_type_len,
515 516
              const char *sql_text, uint sql_text_len,
              CHARSET_INFO *client_cs)
517 518 519 520 521 522 523 524 525 526 527 528
{
  return mysql_log.write(event_time, user_host, user_host_len,
                         thread_id, command_type, command_type_len,
                         sql_text, sql_text_len);
}


bool Log_to_file_event_handler::init()
{
  if (!is_initialized)
  {
    if (opt_slow_log)
529
      mysql_slow_log.open_slow_log(sys_var_slow_log_path.value);
530 531

    if (opt_log)
532
      mysql_log.open_query_log(sys_var_general_log_path.value);
533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549

    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 */
550
  if (opt_log)
551
    mysql_log.reopen_file();
552
  if (opt_slow_log)
553
    mysql_slow_log.reopen_file();
554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585
}

/*
  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;
  Log_event_handler **current_handler= error_log_handler_list;

  /* currently we don't need locking here as there is no error_log table */
  while (*current_handler)
    error= (*current_handler++)->log_error(level, format, args) || error;

  return error;
}


petr@mysql.com's avatar
petr@mysql.com committed
586
void LOGGER::cleanup_base()
587 588 589 590
{
  DBUG_ASSERT(inited == 1);
  (void) pthread_mutex_destroy(&LOCK_logger);
  if (table_log_handler)
petr@mysql.com's avatar
petr@mysql.com committed
591
  {
592
    table_log_handler->cleanup();
petr@mysql.com's avatar
petr@mysql.com committed
593 594
    delete table_log_handler;
  }
595 596 597 598 599
  if (file_log_handler)
    file_log_handler->cleanup();
}


petr@mysql.com's avatar
petr@mysql.com committed
600 601 602 603 604 605 606 607
void LOGGER::cleanup_end()
{
  DBUG_ASSERT(inited == 1);
  if (file_log_handler)
    delete file_log_handler;
}


608
void LOGGER::close_log_table(uint log_table_type, bool lock_in_use)
609
{
610
  table_log_handler->close_log_table(log_table_type, lock_in_use);
611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631
}


/*
  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 */
632
  init_error_log(LOG_FILE);
633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649

  file_log_handler->init_pthread_objects();
  (void) pthread_mutex_init(&LOCK_logger, MY_MUTEX_INIT_SLOW);
}


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;
}


650
bool LOGGER::reopen_log_table(uint log_table_type)
651
{
652
  return table_log_handler->reopen_log_table(log_table_type);
653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672
}


bool LOGGER::flush_logs(THD *thd)
{
  TABLE_LIST close_slow_log, close_general_log;

  /* reopen log tables */
  bzero((char*) &close_slow_log, sizeof(TABLE_LIST));
  close_slow_log.alias= close_slow_log.table_name=(char*) "slow_log";
  close_slow_log.table_name_length= 8;
  close_slow_log.db= (char*) "mysql";
  close_slow_log.db_length= 5;

  bzero((char*) &close_general_log, sizeof(TABLE_LIST));
  close_general_log.alias= close_general_log.table_name=(char*) "general_log";
  close_general_log.table_name_length= 11;
  close_general_log.db= (char*) "mysql";
  close_general_log.db_length= 5;

673
  /* lock tables, in the case they are enabled */
674 675 676 677 678 679 680 681 682 683 684 685
  if (logger.is_log_tables_initialized)
  {
    /*
      This will lock and wait for all but the logger thread to release the
      tables. Then we could reopen log tables. Then release the name locks.

      NOTE: in fact, the first parameter used in lock_and_wait_for_table_name()
      and table_log_handler->flush() could be any non-NULL THD, as the
      underlying code makes certain assumptions about this.
      Here we use one of the logger handler THD's. Simply because it
      seems appropriate.
    */
686 687 688 689 690 691 692
    if (opt_slow_log)
      lock_and_wait_for_table_name(table_log_handler->general_log_thd,
				   &close_slow_log);
    if (opt_log)
      lock_and_wait_for_table_name(table_log_handler->general_log_thd,
				   &close_general_log);
  }
693

694 695 696 697 698 699 700 701
  /*
    Deny others from logging to general and slow log,
    while reopening tables.
  */
  logger.lock();

  /* reopen log files */
  file_log_handler->flush();
702

703 704
  /* flush tables, in the case they are enabled */
  if (logger.is_log_tables_initialized)
705 706
    table_log_handler->flush(table_log_handler->general_log_thd,
                             &close_slow_log, &close_general_log);
707 708
  /* end of log flush */
  logger.unlock();
709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749
  return FALSE;
}


/*
  Log slow query with all enabled log event handlers

  SYNOPSIS
    slow_log_print()

    thd               THD of the query being logged
    query             The query being logged
    query_length      The length of the query string
    query_start_arg   Query start timestamp

  RETURN
    FALSE - OK
    TRUE - error occured
*/

bool LOGGER::slow_log_print(THD *thd, const char *query, uint query_length,
                            time_t query_start_arg)
{
  bool error= FALSE;
  Log_event_handler **current_handler= slow_log_handler_list;
  bool is_command= FALSE;
  char user_host_buff[MAX_USER_HOST_SIZE];

  my_time_t current_time;
  Security_context *sctx= thd->security_ctx;
  uint message_buff_len= 0, user_host_len= 0;
  longlong query_time= 0, lock_time= 0;

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

  if (*slow_log_handler_list)
  {
    current_time= time(NULL);

750 751
    /* do not log slow queries from replication threads */
    if (thd->slave_thread)
752 753 754
      return 0;

    lock();
755 756 757 758 759
    if (!opt_slow_log)
    {
      unlock();
      return 0;
    }
760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826

    /* fill in user_host value: the format is "%s[%s] @ %s [%s]" */
    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;

    if (query_start_arg)
    {
      query_time= (longlong) (current_time - query_start_arg);
      lock_time= (longlong) (thd->time_after_lock - query_start_arg);
    }

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

    while (*current_handler)
      error= (*current_handler++)->log_slow(thd, current_time, query_start_arg,
                                            user_host_buff, user_host_len,
                                            query_time, lock_time, is_command,
                                            query, query_length) || error;

    unlock();
  }
  return error;
}

bool LOGGER::general_log_print(THD *thd, enum enum_server_command command,
                               const char *format, va_list args)
{
  bool error= FALSE;
  Log_event_handler **current_handler= general_log_handler_list;

  /*
    Print the message to the buffer 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)))
  {
    char message_buff[MAX_LOG_BUFFER_SIZE];
    char user_host_buff[MAX_USER_HOST_SIZE];
    Security_context *sctx= thd->security_ctx;
    ulong id;
    uint message_buff_len= 0, user_host_len= 0;

    if (thd)
    {                                           /* Normal thread */
      if ((thd->options & OPTION_LOG_OFF)
#ifndef NO_EMBEDDED_ACCESS_CHECKS
          && (sctx->master_access & SUPER_ACL)
#endif
         )
      {
        return 0;                         /* No logging */
      }
      id= thd->thread_id;
    }
    else
      id=0;                                     /* Log from connect handler */

    lock();
827 828 829 830 831
    if (!opt_log)
    {
      unlock();
      return 0;
    }
832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853
    time_t current_time= time(NULL);

    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;

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

    while (*current_handler)
      error+= (*current_handler++)->
        log_general(current_time, user_host_buff,
                   user_host_len, id,
                   command_name[(uint) command].str,
                   command_name[(uint) command].length,
854 855
                   message_buff, message_buff_len,
                   thd->variables.character_set_client) || error;
856 857 858 859 860
    unlock();
  }
  return error;
}

861
void LOGGER::init_error_log(uint error_log_printer)
862
{
863 864
  if (error_log_printer & LOG_NONE)
  {
865
    error_log_handler_list[0]= 0;
866 867 868 869 870
    return;
  }

  switch (error_log_printer) {
  case LOG_FILE:
871 872 873 874
    error_log_handler_list[0]= file_log_handler;
    error_log_handler_list[1]= 0;
    break;
    /* these two are disabled for now */
875
  case LOG_TABLE:
876 877
    DBUG_ASSERT(0);
    break;
878
  case LOG_TABLE|LOG_FILE:
879 880 881 882 883
    DBUG_ASSERT(0);
    break;
  }
}

884
void LOGGER::init_slow_log(uint slow_log_printer)
885
{
886 887
  if (slow_log_printer & LOG_NONE)
  {
888
    slow_log_handler_list[0]= 0;
889 890 891 892 893
    return;
  }

  switch (slow_log_printer) {
  case LOG_FILE:
894 895 896
    slow_log_handler_list[0]= file_log_handler;
    slow_log_handler_list[1]= 0;
    break;
897
  case LOG_TABLE:
898 899 900
    slow_log_handler_list[0]= table_log_handler;
    slow_log_handler_list[1]= 0;
    break;
901
  case LOG_TABLE|LOG_FILE:
902 903 904 905 906 907 908
    slow_log_handler_list[0]= file_log_handler;
    slow_log_handler_list[1]= table_log_handler;
    slow_log_handler_list[2]= 0;
    break;
  }
}

909
void LOGGER::init_general_log(uint general_log_printer)
910
{
911 912
  if (general_log_printer & LOG_NONE)
  {
913
    general_log_handler_list[0]= 0;
914 915 916 917 918
    return;
  }

  switch (general_log_printer) {
  case LOG_FILE:
919 920 921
    general_log_handler_list[0]= file_log_handler;
    general_log_handler_list[1]= 0;
    break;
922
  case LOG_TABLE:
923 924 925
    general_log_handler_list[0]= table_log_handler;
    general_log_handler_list[1]= 0;
    break;
926
  case LOG_TABLE|LOG_FILE:
927 928 929 930 931 932 933 934
    general_log_handler_list[0]= file_log_handler;
    general_log_handler_list[1]= table_log_handler;
    general_log_handler_list[2]= 0;
    break;
  }
}


935 936 937 938 939 940 941 942 943 944 945
bool LOGGER::activate_log_handler(THD* thd, uint log_type)
{
  bool res= 0;
  lock();
  switch (log_type) {
  case QUERY_LOG_SLOW:
    if (!opt_slow_log)
    {
      if ((res= reopen_log_table(log_type)))
        goto err;
      file_log_handler->get_mysql_slow_log()->
946
        open_slow_log(sys_var_slow_log_path.value);
947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016
      init_slow_log(log_output_options);
      opt_slow_log= TRUE;
    }
    break;
  case QUERY_LOG_GENERAL:
    if (!opt_log)
    {
      if ((res= reopen_log_table(log_type)))
        goto err;
      file_log_handler->get_mysql_log()->
        open_query_log(sys_var_general_log_path.value);
      init_general_log(log_output_options);
      opt_log= TRUE;
    }
    break;
  default:
    DBUG_ASSERT(0);
  }
err:
  unlock();
  return res;
}


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

  switch (log_type) {
  case QUERY_LOG_SLOW:
    table_list= &table_log_handler->slow_log;
    tmp_opt= &opt_slow_log;
    file_log= file_log_handler->get_mysql_slow_log();
    log_thd= table_log_handler->slow_log_thd;
    break;
  case QUERY_LOG_GENERAL:
    table_list= &table_log_handler->general_log;
    tmp_opt= &opt_log;
    file_log= file_log_handler->get_mysql_log();
    log_thd= table_log_handler->general_log_thd;
    break;
  default:
    DBUG_ASSERT(0);
  }

  if (!(*tmp_opt))
    return;

  if (is_log_tables_initialized)
    lock_and_wait_for_table_name(log_thd, table_list);
  lock();

  if (is_log_tables_initialized)
  {
    VOID(pthread_mutex_lock(&LOCK_open));
    close_log_table(log_type, TRUE);
    table_list->table= 0;
    query_cache_invalidate3(log_thd, table_list, 0);
    unlock_table_name(log_thd, table_list);
    VOID(pthread_mutex_unlock(&LOCK_open));
  }
  file_log->close(0);
  *tmp_opt= FALSE;
  unlock();
}


1017 1018 1019 1020
bool Log_to_csv_event_handler::flush(THD *thd, TABLE_LIST *close_slow_log,
                                     TABLE_LIST *close_general_log)
{
  VOID(pthread_mutex_lock(&LOCK_open));
1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032
  if (opt_log)
  {
    close_log_table(QUERY_LOG_GENERAL, TRUE);
    query_cache_invalidate3(thd, close_general_log, 0);
    unlock_table_name(thd, close_general_log);
  }
  if (opt_slow_log)
  {
    close_log_table(QUERY_LOG_SLOW, TRUE);
    query_cache_invalidate3(thd, close_slow_log, 0);
    unlock_table_name(thd, close_slow_log);
  }
1033
  VOID(pthread_mutex_unlock(&LOCK_open));
1034 1035 1036 1037 1038 1039 1040 1041
  /*
    we use | and not || here, to ensure that both reopen_log_table
    are called, even if the first one fails
  */
  if ((opt_slow_log && reopen_log_table(QUERY_LOG_SLOW)) |
      (opt_log && reopen_log_table(QUERY_LOG_GENERAL)))
    return 1;
  return 0;
1042 1043 1044 1045 1046
}

/* the parameters are unused for the log tables */
bool Log_to_csv_event_handler::init()
{
1047 1048 1049 1050 1051 1052 1053 1054
  /*
    we use | and not || here, to ensure that both open_log_table
    are called, even if the first one fails
  */
  if ((opt_log && open_log_table(QUERY_LOG_GENERAL)) |
      (opt_slow_log && open_log_table(QUERY_LOG_SLOW)))
    return 1;
  return 0;
1055 1056
}

1057 1058 1059
int LOGGER::set_handlers(uint error_log_printer,
                         uint slow_log_printer,
                         uint general_log_printer)
1060 1061
{
  /* error log table is not supported yet */
1062
  DBUG_ASSERT(error_log_printer < LOG_TABLE);
1063 1064 1065

  lock();

1066
  if ((slow_log_printer & LOG_TABLE || general_log_printer & LOG_TABLE) &&
1067 1068
      !is_log_tables_initialized)
  {
1069 1070
    slow_log_printer= (slow_log_printer & ~LOG_TABLE) | LOG_FILE;
    general_log_printer= (general_log_printer & ~LOG_TABLE) | LOG_FILE;
1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091

    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;
}


/*
  Close log table of a given type (general or slow log)

  SYNOPSIS
    close_log_table()

1092 1093 1094
    log_table_type   type of the log table to close: QUERY_LOG_GENERAL
                     or QUERY_LOG_SLOW
    lock_in_use      Set to TRUE if the caller owns LOCK_open. FALSE otherwise.
1095 1096 1097 1098 1099 1100 1101 1102 1103

  DESCRIPTION

    The function closes a log table. It is invoked (1) when we need to reopen
    log tables (e.g. FLUSH LOGS or TRUNCATE on the log table is being
    executed) or (2) during shutdown.
*/

void Log_to_csv_event_handler::
1104
  close_log_table(uint log_table_type, bool lock_in_use)
1105 1106 1107 1108
{
  THD *log_thd, *curr= current_thd;
  TABLE_LIST *table;

1109
  switch (log_table_type) {
1110
  case QUERY_LOG_GENERAL:
1111 1112
    if (!logger.is_general_log_table_enabled())
      return;                                     /* do nothing */
1113 1114 1115
    log_thd= general_log_thd;
    table= &general_log;
    break;
1116
  case QUERY_LOG_SLOW:
1117 1118
    if (!logger.is_slow_log_table_enabled())
      return;                                     /* do nothing */
1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151
    log_thd= slow_log_thd;
    table= &slow_log;
    break;
  default:
    DBUG_ASSERT(0);
  }

  /*
    Set thread stack start for the logger thread. See comment in
    open_log_table() for details.
  */
  if (curr)
    log_thd->thread_stack= curr->thread_stack;
  else
    log_thd->thread_stack= (char*) &log_thd;

  /* close the table */
  log_thd->store_globals();
  table->table->file->ha_rnd_end();
  /* discard logger mark before unlock*/
  table->table->locked_by_logger= FALSE;
  close_thread_tables(log_thd, lock_in_use);

  if (curr)
    curr->store_globals();
  else
  {
    my_pthread_setspecific_ptr(THR_THD,  0);
    my_pthread_setspecific_ptr(THR_MALLOC, 0);
  }
}


1152 1153
/*
  this function is mostly a placeholder.
1154
  conceptually, binlog initialization (now mostly done in MYSQL_BIN_LOG::open)
1155 1156 1157
  should be moved here.
*/

1158
int binlog_init(void *p)
1159
{
1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170
  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;
1171
  return 0;
1172 1173 1174 1175
}

static int binlog_close_connection(THD *thd)
{
1176
  binlog_trx_data *const trx_data=
1177
    (binlog_trx_data*) thd->ha_data[binlog_hton->slot];
1178 1179
  IO_CACHE *trans_log= &trx_data->trans_log;
  DBUG_ASSERT(mysql_bin_log.is_open() && trx_data->empty());
1180
  close_cached_file(trans_log);
1181
  thd->ha_data[binlog_hton->slot]= 0;
1182
  my_free((gptr)trx_data, MYF(0));
1183 1184 1185
  return 0;
}

1186 1187
static int
binlog_end_trans(THD *thd, binlog_trx_data *trx_data, Log_event *end_ev)
1188
{
1189
  DBUG_ENTER("binlog_end_trans");
1190 1191 1192
  int error=0;
  IO_CACHE *trans_log= &trx_data->trans_log;

1193 1194 1195

  /* NULL denotes ROLLBACK with nothing to replicate */
  if (end_ev != NULL)
1196
  {
1197 1198 1199 1200 1201 1202
    /*
      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.
     */
1203
#ifdef HAVE_ROW_BASED_REPLICATION
1204
    thd->binlog_flush_pending_rows_event(TRUE);
1205
#endif
1206
    error= mysql_bin_log.write(thd, trans_log, end_ev);
1207
  }
1208
#ifdef HAVE_ROW_BASED_REPLICATION
1209 1210
  else
  {
1211
#ifdef HAVE_ROW_BASED_REPLICATION
1212
    thd->binlog_delete_pending_rows_event();
1213
#endif
1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226
  }

  /*
    We need to step the table map version both after writing the
    entire transaction to the log file and after rolling back the
    transaction.

    We need to step the table map version after writing the
    transaction cache to disk.  In addition, we need to step the table
    map version on a rollback to ensure that a new table map event is
    generated instead of the one that was written to the thrown-away
    transaction cache.
  */
1227
  mysql_bin_log.update_table_map_version();
1228
#endif
1229

1230 1231 1232 1233 1234 1235 1236 1237
  statistic_increment(binlog_cache_use, &LOCK_status);
  if (trans_log->disk_writes != 0)
  {
    statistic_increment(binlog_cache_disk_use, &LOCK_status);
    trans_log->disk_writes= 0;
  }
  reinit_io_cache(trans_log, WRITE_CACHE, (my_off_t) 0, 0, 1); // cannot fail
  trans_log->end_of_file= max_binlog_cache_size;
1238
  DBUG_RETURN(error);
1239 1240 1241 1242 1243 1244 1245 1246
}

static int binlog_prepare(THD *thd, bool all)
{
  /*
    do nothing.
    just pretend we can do 2pc, so that MySQL won't
    switch to 1pc.
1247
    real work will be done in MYSQL_BIN_LOG::log()
1248 1249 1250 1251 1252 1253 1254
  */
  return 0;
}

static int binlog_commit(THD *thd, bool all)
{
  DBUG_ENTER("binlog_commit");
1255
  binlog_trx_data *const trx_data=
1256
    (binlog_trx_data*) thd->ha_data[binlog_hton->slot];
1257
  IO_CACHE *trans_log= &trx_data->trans_log;
1258
  DBUG_ASSERT(mysql_bin_log.is_open() &&
1259
     (all || !(thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN))));
1260

1261
  if (trx_data->empty())
1262
  {
1263
    // we're here because trans_log was flushed in MYSQL_BIN_LOG::log()
1264 1265
    DBUG_RETURN(0);
  }
1266 1267 1268 1269 1270 1271 1272 1273
  if (all) 
  {
    Query_log_event qev(thd, STRING_WITH_LEN("COMMIT"), TRUE, FALSE);
    qev.error_code= 0; // see comment in MYSQL_LOG::write(THD, IO_CACHE)
    DBUG_RETURN(binlog_end_trans(thd, trx_data, &qev));
  }
  else
    DBUG_RETURN(binlog_end_trans(thd, trx_data, &invisible_commit));
1274 1275 1276 1277 1278
}

static int binlog_rollback(THD *thd, bool all)
{
  DBUG_ENTER("binlog_rollback");
1279 1280
  int error=0;
  binlog_trx_data *const trx_data=
1281
    (binlog_trx_data*) thd->ha_data[binlog_hton->slot];
1282
  IO_CACHE *trans_log= &trx_data->trans_log;
1283
  /*
serg@serg.mylan's avatar
serg@serg.mylan committed
1284 1285
    First assert is guaranteed - see trans_register_ha() call below.
    The second must be true. If it is not, we're registering
1286 1287
    unnecessary, doing extra work. The cause should be found and eliminated
  */
serg@serg.mylan's avatar
serg@serg.mylan committed
1288
  DBUG_ASSERT(all || !(thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN)));
1289
  DBUG_ASSERT(mysql_bin_log.is_open() && !trx_data->empty());
1290
  /*
1291 1292 1293 1294
    Update the binary log with a BEGIN/ROLLBACK block if we have
    cached some queries and we updated some non-transactional
    table. Such cases should be rare (updating a
    non-transactional table inside a transaction...)
1295
  */
1296 1297
  if (unlikely(thd->options & (OPTION_STATUS_NO_TRANS_UPDATE |
                               OPTION_KEEP_LOG)))
1298
  {
1299
    Query_log_event qev(thd, STRING_WITH_LEN("ROLLBACK"), TRUE, FALSE);
1300
    qev.error_code= 0; // see comment in MYSQL_LOG::write(THD, IO_CACHE)
1301
    error= binlog_end_trans(thd, trx_data, &qev);
1302
  }
1303
  else
1304
    error= binlog_end_trans(thd, trx_data, 0);
1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331
  DBUG_RETURN(error);
}

/*
  NOTE: how do we handle this (unlikely but legal) case:
  [transaction] + [update to non-trans table] + [rollback to savepoint] ?
  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).
*/

static int binlog_savepoint_set(THD *thd, void *sv)
{
  DBUG_ENTER("binlog_savepoint_set");
1332
  binlog_trx_data *const trx_data=
1333
    (binlog_trx_data*) thd->ha_data[binlog_hton->slot];
1334
  DBUG_ASSERT(mysql_bin_log.is_open() && my_b_tell(&trx_data->trans_log));
1335

1336
  *(my_off_t *)sv= my_b_tell(&trx_data->trans_log);
1337
  /* Write it to the binary log */
1338 1339 1340 1341 1342
  
  int const error=
    thd->binlog_query(THD::STMT_QUERY_TYPE,
                      thd->query, thd->query_length, TRUE, FALSE);
  DBUG_RETURN(error);
1343 1344 1345 1346 1347
}

static int binlog_savepoint_rollback(THD *thd, void *sv)
{
  DBUG_ENTER("binlog_savepoint_rollback");
1348
  binlog_trx_data *const trx_data=
1349
    (binlog_trx_data*) thd->ha_data[binlog_hton->slot];
1350
  IO_CACHE *trans_log= &trx_data->trans_log;
1351 1352 1353 1354 1355 1356 1357
  DBUG_ASSERT(mysql_bin_log.is_open() && my_b_tell(trans_log));

  /*
    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.
  */
1358 1359
  if (unlikely(thd->options &
               (OPTION_STATUS_NO_TRANS_UPDATE | OPTION_KEEP_LOG)))
1360
  {
1361 1362 1363 1364
    int const error=
      thd->binlog_query(THD::STMT_QUERY_TYPE,
                        thd->query, thd->query_length, TRUE, FALSE);
    DBUG_RETURN(error);
1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394
  }
  reinit_io_cache(trans_log, WRITE_CACHE, *(my_off_t *)sv, 0, 0);
  DBUG_RETURN(0);
}

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

  if (my_b_read(log, (byte*) magic, sizeof(magic)))
  {
    *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;
}

File open_binlog(IO_CACHE *log, const char *log_file_name, const char **errmsg)
{
  File file;
  DBUG_ENTER("open_binlog");

1395 1396
  if ((file = my_open(log_file_name, O_RDONLY | O_BINARY | O_SHARE, 
                      MYF(MY_WME))) < 0)
1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422
  {
    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
1423

rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1424 1425
#ifdef __NT__
static int eventSource = 0;
serg@serg.mylan's avatar
serg@serg.mylan committed
1426

1427
void setup_windows_event_source()
monty@mysql.com's avatar
monty@mysql.com committed
1428
{
1429
  HKEY    hRegKey= NULL;
monty@mysql.com's avatar
monty@mysql.com committed
1430 1431 1432
  DWORD   dwError= 0;
  TCHAR   szPath[MAX_PATH];
  DWORD dwTypes;
1433

monty@mysql.com's avatar
monty@mysql.com committed
1434 1435 1436
  if (eventSource)               // Ensure that we are only called once
    return;
  eventSource= 1;
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1437

monty@mysql.com's avatar
monty@mysql.com committed
1438
  // Create the event source registry key
1439
  dwError= RegCreateKey(HKEY_LOCAL_MACHINE,
monty@mysql.com's avatar
monty@mysql.com committed
1440 1441
                          "SYSTEM\\CurrentControlSet\\Services\\EventLog\\Application\\MySQL", 
                          &hRegKey);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1442

monty@mysql.com's avatar
monty@mysql.com committed
1443 1444
  /* Name of the PE module that contains the message resource */
  GetModuleFileName(NULL, szPath, MAX_PATH);
serg@serg.mylan's avatar
serg@serg.mylan committed
1445

rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1446
  /* Register EventMessageFile */
1447
  dwError = RegSetValueEx(hRegKey, "EventMessageFile", 0, REG_EXPAND_SZ,
monty@mysql.com's avatar
monty@mysql.com committed
1448
                          (PBYTE) szPath, strlen(szPath)+1);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1449

monty@mysql.com's avatar
monty@mysql.com committed
1450 1451 1452 1453 1454
  /* 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
1455

monty@mysql.com's avatar
monty@mysql.com committed
1456
  RegCloseKey(hRegKey);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1457 1458
}

monty@mysql.com's avatar
monty@mysql.com committed
1459
#endif /* __NT__ */
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1460 1461


bk@work.mysql.com's avatar
bk@work.mysql.com committed
1462 1463 1464 1465 1466 1467 1468 1469
/****************************************************************************
** Find a uniq filename for 'filename.#'.
** Set # to a number as low as possible
** returns != 0 if not possible to get uniq filename
****************************************************************************/

static int find_uniq_filename(char *name)
{
1470 1471 1472 1473
  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
1474
  reg1 struct fileinfo *file_info;
1475 1476
  ulong                 max_found=0;

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1477 1478
  DBUG_ENTER("find_uniq_filename");

1479 1480 1481 1482
  uint  length = dirname_part(buff,name);
  char *start  = name + length;
  char *end    = strend(start);

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1483
  *end='.';
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
1484
  length= (uint) (end-start+1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502

  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++)
  {
    if (bcmp(file_info->name,start,length) == 0 &&
	test_if_number(file_info->name+length, &number,0))
    {
      set_if_bigger(max_found,(ulong) number);
    }
  }
  my_dirend(dir_info);

  *end++='.';
1503
  sprintf(end,"%06ld",max_found+1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1504 1505 1506
  DBUG_RETURN(0);
}

1507

1508
void MYSQL_LOG::init(enum_log_type log_type_arg,
1509
                     enum cache_type io_cache_type_arg)
1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578
{
  DBUG_ENTER("MYSQL_LOG::init");
  log_type= log_type_arg;
  io_cache_type= io_cache_type_arg;
  DBUG_PRINT("info",("log_type: %d", log_type));
  DBUG_VOID_RETURN;
}


/*
  Open a (new) log file.

  SYNOPSIS
    open()

    log_name            The name of the log to open
    log_type_arg        The type of the log. E.g. LOG_NORMAL
    new_name            The new name for the logfile. This is only needed
                        when the method is used to open the binlog file.
    io_cache_type_arg   The type of the IO_CACHE to use for this log file

  DESCRIPTION
    Open the logfile, init IO_CACHE and write startup messages
    (in case of general and slow query logs).

  RETURN VALUES
    0   ok
    1   error
*/

bool MYSQL_LOG::open(const char *log_name, enum_log_type log_type_arg,
                     const char *new_name, enum cache_type io_cache_type_arg)
{
  char buff[FN_REFLEN];
  File file= -1;
  int open_flags= O_CREAT | O_BINARY;
  DBUG_ENTER("MYSQL_LOG::open");
  DBUG_PRINT("enter", ("log_type: %d", (int) log_type_arg));

  write_error= 0;

  init(log_type_arg, io_cache_type_arg);

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

  if (new_name)
    strmov(log_file_name, new_name);
  else if (generate_new_name(log_file_name, name))
    goto err;

  if (io_cache_type == SEQ_READ_APPEND)
    open_flags |= O_RDWR | O_APPEND;
  else
    open_flags |= O_WRONLY | (log_type == LOG_BIN ? 0 : O_APPEND);

  db[0]= 0;

  if ((file= my_open(log_file_name, open_flags,
                     MYF(MY_WME | ME_WAITTANG))) < 0 ||
      init_io_cache(&log_file, file, IO_SIZE, io_cache_type,
                    my_tell(file, MYF(MY_WME)), 0,
                    MYF(MY_WME | MY_NABP |
                        ((log_type == LOG_BIN) ? MY_WAIT_IF_FULL : 0))))
    goto err;

1579
  if (log_type == LOG_NORMAL)
1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601
  {
    char *end;
    int len=my_snprintf(buff, sizeof(buff), "%s, Version: %s. "
#ifdef EMBEDDED_LIBRARY
		        "embedded library\n", my_progname, server_version
#elif __NT__
			"started with:\nTCP Port: %d, Named Pipe: %s\n",
			my_progname, server_version, mysqld_port,
                        mysqld_unix_port
#else
			"started with:\nTcp port: %d  Unix socket: %s\n",
			my_progname, server_version, mysqld_port,
                        mysqld_unix_port
#endif
                       );
    end= strnmov(buff + len, "Time                 Id Command    Argument\n",
                 sizeof(buff) - len);
    if (my_b_write(&log_file, (byte*) buff, (uint) (end-buff)) ||
	flush_io_cache(&log_file))
      goto err;
  }

1602
  log_state= LOG_OPENED;
1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613
  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);
1614
  log_state= LOG_CLOSED;
1615 1616 1617
  DBUG_RETURN(1);
}

1618
MYSQL_LOG::MYSQL_LOG()
1619 1620
  : name(0),  log_type(LOG_UNKNOWN), log_state(LOG_CLOSED), write_error(FALSE),
    inited(FALSE)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1621 1622
{
  /*
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1623 1624
    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
1625
    called only in main(). Doing initialization here would make it happen
1626
    before main().
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1627
  */
1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655
  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));
1656
  if (log_state == LOG_OPENED)
1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672
  {
    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);
    }
  }

1673
  log_state= (exiting & LOG_CLOSE_TO_BE_OPENED) ? LOG_TO_BE_OPENED : LOG_CLOSED;
1674 1675
  safeFree(name);
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1676 1677
}

1678
/* this is called only once */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1679

monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
1680
void MYSQL_LOG::cleanup()
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1681
{
1682
  DBUG_ENTER("cleanup");
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1683
  if (inited)
1684
  {
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
1685
    inited= 0;
1686
    (void) pthread_mutex_destroy(&LOCK_log);
1687
    close(0);
1688
  }
1689
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1690 1691
}

1692

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1693
int MYSQL_LOG::generate_new_name(char *new_name, const char *log_name)
1694
{
1695 1696
  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
1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709
  {
    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;
}

1710

1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722
/*
  Reopen the log file

  SYNOPSIS
    reopen_file()

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


1723
void MYSQL_QUERY_LOG::reopen_file()
1724
{
1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740
  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);

  /*
1741
     Note that at this point, log_state != LOG_CLOSED (important for is_open()).
1742 1743
  */

1744
  open(save_name, log_type, 0, io_cache_type);
1745 1746 1747 1748
  my_free(save_name, MYF(0));

  pthread_mutex_unlock(&LOCK_log);

1749
  DBUG_VOID_RETURN;
1750 1751
}

1752

1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777
/*
  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
*/

1778 1779 1780 1781
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
1782
{
1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841
  char buff[32];
  uint length= 0;
  char time_buff[MAX_TIME_SIZE];
  struct tm start;
  uint time_buff_len= 0;

  /* Test if someone closed between the is_open test and lock */
  if (is_open())
  {
    /* Note that my_b_write() assumes it knows the length for this */
      if (event_time != last_time)
      {
        last_time= event_time;

        localtime_r(&event_time, &start);

        time_buff_len= my_snprintf(time_buff, MAX_TIME_SIZE,
                                   "%02d%02d%02d %2d:%02d:%02d",
                                   start.tm_year % 100, start.tm_mon + 1,
                                   start.tm_mday, start.tm_hour,
                                   start.tm_min, start.tm_sec);

        if (my_b_write(&log_file, (byte*) &time_buff, time_buff_len))
          goto err;
      }
      else
        if (my_b_write(&log_file, (byte*) "\t\t" ,2) < 0)
          goto err;

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

    if (my_b_write(&log_file, (byte*) buff, length))
      goto err;

    if (my_b_write(&log_file, (byte*) command_type, command_type_len))
      goto err;

    if (my_b_write(&log_file, (byte*) "\t", 1))
      goto err;

    /* sql_text */
    if (my_b_write(&log_file, (byte*) sql_text, sql_text_len))
      goto err;

    if (my_b_write(&log_file, (byte*) "\n", 1) ||
        flush_io_cache(&log_file))
      goto err;
  }

  return FALSE;
err:

  if (!write_error)
  {
    write_error= 1;
    sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
  }
  return TRUE;
1842 1843
}

1844

1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873
/*
  Log a query to the traditional slow log file

  SYNOPSIS
    write()

    thd               THD of the query
    current_time      current timestamp
    query_start_arg   command start timestamp
    user_host         the pointer to the string with user@host info
    user_host_len     length of the user_host string. this is computed once
                      and passed to all general log event handlers
    query_time        Amount of time the query took to execute (in seconds)
    lock_time         Amount of time the query was locked (in seconds)
    is_command        The flag, which determines, whether the sql_text is a
                      query or an administrator command.
    sql_text          the very text of the query or administrator command
                      processed
    sql_text_len      the length of sql_text string

  DESCRIPTION

   Log a query to the slow log file.

  RETURN
    FALSE - OK
    TRUE - error occured
*/

1874 1875 1876 1877 1878
bool MYSQL_QUERY_LOG::write(THD *thd, time_t current_time,
                            time_t query_start_arg, const char *user_host,
                            uint user_host_len, longlong query_time,
                            longlong lock_time, bool is_command,
                            const char *sql_text, uint sql_text_len)
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1879
{
1880
  bool error= 0;
1881
  DBUG_ENTER("MYSQL_QUERY_LOG::write");
1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933

  if (!is_open())
    DBUG_RETURN(0);

  if (is_open())
  {						// Safety agains reopen
    int tmp_errno= 0;
    char buff[80], *end;
    uint buff_len;
    end= buff;

    if (!(specialflag & SPECIAL_SHORT_LOG_FORMAT))
    {
      Security_context *sctx= thd->security_ctx;
      if (current_time != last_time)
      {
        last_time= current_time;
        struct tm start;
        localtime_r(&current_time, &start);

        buff_len= my_snprintf(buff, sizeof buff,
                              "# Time: %02d%02d%02d %2d:%02d:%02d\n",
                              start.tm_year % 100, start.tm_mon + 1,
                              start.tm_mday, start.tm_hour,
                              start.tm_min, start.tm_sec);

        /* Note that my_b_write() assumes it knows the length for this */
        if (my_b_write(&log_file, (byte*) buff, buff_len))
          tmp_errno= errno;
      }
      if (my_b_printf(&log_file, "# User@Host: ", sizeof("# User@Host: ") - 1)
          != sizeof("# User@Host: ") - 1)
        tmp_errno= errno;
      if (my_b_printf(&log_file, user_host, user_host_len) != user_host_len)
        tmp_errno= errno;
      if (my_b_write(&log_file, (byte*) "\n", 1))
        tmp_errno= errno;
    }
    /* For slow query log */
    if (my_b_printf(&log_file,
                    "# Query_time: %lu  Lock_time: %lu"
                    " Rows_sent: %lu  Rows_examined: %lu\n",
                    (ulong) query_time, (ulong) lock_time,
                    (ulong) thd->sent_row_count,
                    (ulong) thd->examined_row_count) == (uint) -1)
      tmp_errno= errno;
    if (thd->db && strcmp(thd->db, db))
    {						// Database changed
      if (my_b_printf(&log_file,"use %s;\n",thd->db) == (uint) -1)
        tmp_errno= errno;
      strmov(db,thd->db);
    }
1934
    if (thd->stmt_depends_on_first_successful_insert_id_in_prev_stmt)
1935 1936
    {
      end=strmov(end, ",last_insert_id=");
1937 1938 1939
      end=longlong10_to_str((longlong)
                            thd->first_successful_insert_id_in_prev_stmt_for_binlog,
                            end, -10);
1940 1941
    }
    // Save value if we do an insert.
1942
    if (thd->auto_inc_intervals_in_cur_stmt_for_binlog.nb_elements() > 0)
1943 1944 1945 1946
    {
      if (!(specialflag & SPECIAL_SHORT_LOG_FORMAT))
      {
        end=strmov(end,",insert_id=");
1947 1948 1949
        end=longlong10_to_str((longlong)
                              thd->auto_inc_intervals_in_cur_stmt_for_binlog.minimum(),
                              end, -10);
1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989
      }
    }

    /*
      This info used to show up randomly, depending on whether the query
      checked the query start time or not. now we always write current
      timestamp to the slow log
    */
    end= strmov(end, ",timestamp=");
    end= int10_to_str((long) current_time, end, 10);

    if (end != buff)
    {
      *end++=';';
      *end='\n';
      if (my_b_write(&log_file, (byte*) "SET ", 4) ||
          my_b_write(&log_file, (byte*) buff + 1, (uint) (end-buff)))
        tmp_errno= errno;
    }
    if (is_command)
    {
      end= strxmov(buff, "# administrator command: ", NullS);
      buff_len= (ulong) (end - buff);
      my_b_write(&log_file, (byte*) buff, buff_len);
    }
    if (my_b_write(&log_file, (byte*) sql_text, sql_text_len) ||
        my_b_write(&log_file, (byte*) ";\n",2) ||
        flush_io_cache(&log_file))
      tmp_errno= errno;
    if (tmp_errno)
    {
      error= 1;
      if (! write_error)
      {
        write_error= 1;
        sql_print_error(ER(ER_ERROR_ON_WRITE), name, error);
      }
    }
  }
  DBUG_RETURN(error);
1990 1991
}

1992

1993
const char *MYSQL_LOG::generate_name(const char *log_name,
1994 1995
                                      const char *suffix,
                                      bool strip_ext, char *buff)
1996 1997 1998 1999 2000 2001 2002
{
  if (!log_name || !log_name[0])
  {
    /*
      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.
    */
2003 2004
    strmake(buff, glob_hostname, FN_REFLEN - 5);
    strmov(fn_ext(buff), suffix);
2005 2006 2007 2008 2009
    return (const char *)buff;
  }
  // get rid of extension if the log is binary to avoid problems
  if (strip_ext)
  {
2010 2011 2012
    char *p= fn_ext(log_name);
    uint length= (uint) (p - log_name);
    strmake(buff, log_name, min(length, FN_REFLEN));
2013 2014 2015 2016 2017
    return (const char*)buff;
  }
  return log_name;
}

2018

2019

2020 2021 2022 2023 2024 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 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075
MYSQL_BIN_LOG::MYSQL_BIN_LOG()
  :bytes_written(0), prepared_xids(0), file_id(1), open_count(1),
   need_start_event(TRUE), m_table_map_version(0),
   description_event_for_exec(0), description_event_for_queue(0)
{
  /*
    We don't want to initialize locks here as such initialization depends on
    safe_mutex (when using safe_mutex) which depends on MY_INIT(), which is
    called only in main(). Doing initialization here would make it happen
    before main().
  */
  index_file_name[0] = 0;
  bzero((char*) &index_file, sizeof(index_file));
}

/* this is called only once */

void MYSQL_BIN_LOG::cleanup()
{
  DBUG_ENTER("cleanup");
  if (inited)
  {
    inited= 0;
    close(LOG_CLOSE_INDEX|LOG_CLOSE_STOP_EVENT);
    delete description_event_for_queue;
    delete description_event_for_exec;
    (void) pthread_mutex_destroy(&LOCK_log);
    (void) pthread_mutex_destroy(&LOCK_index);
    (void) pthread_cond_destroy(&update_cond);
  }
  DBUG_VOID_RETURN;
}


/* Init binlog-specific vars */
void MYSQL_BIN_LOG::init(bool no_auto_events_arg, ulong max_size_arg)
{
  DBUG_ENTER("MYSQL_BIN_LOG::init");
  no_auto_events= no_auto_events_arg;
  max_size= max_size_arg;
  DBUG_PRINT("info",("max_size: %lu", max_size));
  DBUG_VOID_RETURN;
}


void MYSQL_BIN_LOG::init_pthread_objects()
{
  DBUG_ASSERT(inited == 0);
  inited= 1;
  (void) pthread_mutex_init(&LOCK_log, MY_MUTEX_INIT_SLOW);
  (void) pthread_mutex_init(&LOCK_index, MY_MUTEX_INIT_SLOW);
  (void) pthread_cond_init(&update_cond, 0);
}


bool MYSQL_BIN_LOG::open_index_file(const char *index_file_name_arg,
2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100
                                const char *log_name)
{
  File index_file_nr= -1;
  DBUG_ASSERT(!my_b_inited(&index_file));

  /*
    First open of this class instance
    Create an index file that will hold all file names uses for logging.
    Add new entries to the end of it.
  */
  myf opt= MY_UNPACK_FILENAME;
  if (!index_file_name_arg)
  {
    index_file_name_arg= log_name;    // Use same basename for index file
    opt= MY_UNPACK_FILENAME | MY_REPLACE_EXT;
  }
  fn_format(index_file_name, index_file_name_arg, mysql_data_home,
            ".index", opt);
  if ((index_file_nr= my_open(index_file_name,
                              O_RDWR | O_CREAT | O_BINARY ,
                              MYF(MY_WME))) < 0 ||
       my_sync(index_file_nr, MYF(MY_WME)) ||
       init_io_cache(&index_file, index_file_nr,
                     IO_SIZE, WRITE_CACHE,
                     my_seek(index_file_nr,0L,MY_SEEK_END,MYF(0)),
serg@serg.mylan's avatar
serg@serg.mylan committed
2101
			0, MYF(MY_WME | MY_WAIT_IF_FULL)))
2102 2103 2104 2105 2106 2107 2108 2109
  {
    if (index_file_nr >= 0)
      my_close(index_file_nr,MYF(0));
    return TRUE;
  }
  return FALSE;
}

bk@work.mysql.com's avatar
bk@work.mysql.com committed
2110

2111
/*
2112
  Open a (new) binlog file.
2113 2114

  DESCRIPTION
2115
  - Open the log file and the index file. Register the new
2116 2117 2118 2119 2120 2121 2122 2123 2124
    file name in it
  - When calling this when the file is in use, you must have a locks
    on LOCK_log and LOCK_index.

  RETURN VALUES
    0	ok
    1	error
*/

2125 2126 2127 2128 2129 2130 2131
bool MYSQL_BIN_LOG::open(const char *log_name,
                         enum_log_type log_type_arg,
                         const char *new_name,
                         enum cache_type io_cache_type_arg,
                         bool no_auto_events_arg,
                         ulong max_size_arg,
                         bool null_created_arg)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2132
{
2133 2134
  char buff[FN_REFLEN];
  File file= -1;
2135
  int open_flags = O_CREAT | O_BINARY;
2136
  DBUG_ENTER("MYSQL_BIN_LOG::open");
2137
  DBUG_PRINT("enter",("log_type: %d",(int) log_type_arg));
2138 2139

  write_error=0;
2140

2141 2142 2143
  /* open the main log file */
  if (MYSQL_LOG::open(log_name, log_type_arg, new_name, io_cache_type_arg))
    DBUG_RETURN(1);                            /* all warnings issued */
2144

2145
  init(no_auto_events_arg, max_size_arg);
2146

2147
  open_count++;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2148

2149
  DBUG_ASSERT(log_type == LOG_BIN);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2150 2151

  {
2152 2153 2154 2155 2156 2157 2158 2159 2160 2161
    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.
      */
2162 2163
      if (my_b_safe_write(&log_file, (byte*) BINLOG_MAGIC,
			  BIN_LOG_HEADER_SIZE))
2164
        goto err;
2165 2166
      bytes_written+= BIN_LOG_HEADER_SIZE;
      write_file_name_to_index_file= 1;
2167
    }
2168

2169
    DBUG_ASSERT(my_b_inited(&index_file) != 0);
2170 2171
    reinit_io_cache(&index_file, WRITE_CACHE,
                    my_b_filelength(&index_file), 0, 0);
2172
    if (need_start_event && !no_auto_events)
2173
    {
2174 2175 2176 2177 2178
      /*
        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
2179 2180 2181 2182 2183 2184
      /*
        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;
2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203
      if (!s.is_valid())
        goto err;
      if (null_created_arg)
        s.created= 0;
      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
2204
        Format_description_log_event::Format_description_log_event(char* buf,).
2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218
        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
        Format_description_log_event::exec_event().
      */
      description_event_for_queue->created= 0;
      /* Don't set log_pos in event header */
      description_event_for_queue->artificial_event=1;
2219

2220 2221 2222
      if (description_event_for_queue->write(&log_file))
        goto err;
      bytes_written+= description_event_for_queue->data_written;
2223
    }
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2224 2225
    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
2226
      goto err;
2227 2228 2229

    if (write_file_name_to_index_file)
    {
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2230 2231 2232 2233
      /*
        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.
      */
2234 2235 2236
      if (my_b_write(&index_file, (byte*) log_file_name,
		     strlen(log_file_name)) ||
	  my_b_write(&index_file, (byte*) "\n", 1) ||
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2237 2238
	  flush_io_cache(&index_file) ||
          my_sync(index_file.file, MYF(MY_WME)))
2239
	goto err;
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
2240
    }
2241
  }
2242 2243
  log_state= LOG_OPENED;

2244
  DBUG_RETURN(0);
2245 2246

err:
2247 2248 2249
  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, \
2250
shutdown the MySQL server and restart it.", name, errno);
2251 2252 2253
  if (file >= 0)
    my_close(file,MYF(0));
  end_io_cache(&log_file);
2254 2255
  end_io_cache(&index_file);
  safeFree(name);
2256
  log_state= LOG_CLOSED;
2257
  DBUG_RETURN(1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2258 2259
}

2260

2261
int MYSQL_BIN_LOG::get_current_log(LOG_INFO* linfo)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2262 2263
{
  pthread_mutex_lock(&LOCK_log);
2264 2265 2266 2267 2268
  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
2269
int MYSQL_BIN_LOG::raw_get_current_log(LOG_INFO* linfo)
2270
{
2271 2272
  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
2273 2274 2275
  return 0;
}

2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317
/*
  Move all data up in a file in an filename index file

  SYNOPSIS
    copy_up_file_and_fill()
    index_file			File to move
    offset			Move everything from here to beginning

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

  IMPLEMENTATION
    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.

  RETURN VALUES
    0	ok
*/

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;
  byte io_buf[IO_SIZE*2];
  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));
    if (my_write(file, (byte*) io_buf, bytes_read, MYF(MY_WME | MY_NABP)))
      goto err;
  }
  /* The following will either truncate the file or fill the end with \n' */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2318 2319
  if (my_chsize(file, offset - init_offset, '\n', MYF(MY_WME)) ||
      my_sync(file, MYF(MY_WME)))
2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339
    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);
}


/*
  Find the position in the log-index-file for the given log name

  SYNOPSIS
    find_log_pos()
    linfo		Store here the found log file name and position to
			the NEXT log file name in the index file.
    log_name		Filename to find in the index file.
			Is a null pointer if we want to read the first entry
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
2340
    need_lock		Set this to 1 if the parent doesn't already have a
2341 2342 2343
			lock on LOCK_index

  NOTE
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
2344 2345
    On systems without the truncate function the file will end with one or
    more empty lines.  These will be ignored when reading the file.
2346 2347 2348 2349 2350 2351 2352

  RETURN VALUES
    0			ok
    LOG_INFO_EOF	End of log-index-file found
    LOG_INFO_IO		Got IO error while reading file
*/

2353
int MYSQL_BIN_LOG::find_log_pos(LOG_INFO *linfo, const char *log_name,
2354
			    bool need_lock)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2355
{
2356 2357 2358 2359 2360
  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
2361

2362
  /*
2363 2364
    Mutex needed because we need to make sure the file pointer does not
    move from under our feet
2365
  */
2366
  if (need_lock)
2367
    pthread_mutex_lock(&LOCK_index);
2368 2369 2370 2371 2372
  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);

2373
  for (;;)
2374
  {
2375
    uint length;
2376 2377 2378 2379
    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
2380
    {
2381 2382 2383
      /* 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
2384 2385
    }

2386 2387
    // if the log entry matches, null string matching anything
    if (!log_name ||
sasha@mysql.sashanet.com's avatar
sasha@mysql.sashanet.com committed
2388
	(log_name_len == length-1 && fname[log_name_len] == '\n' &&
2389
	 !memcmp(fname, log_name, log_name_len)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2390
    {
2391
      DBUG_PRINT("info",("Found log file entry"));
2392
      fname[length-1]=0;			// remove last \n
2393 2394
      linfo->index_file_start_offset= offset;
      linfo->index_file_offset = my_b_tell(&index_file);
2395
      break;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2396
    }
2397
  }
2398

2399
  if (need_lock)
2400
    pthread_mutex_unlock(&LOCK_index);
2401
  DBUG_RETURN(error);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2402
}
2403

2404

2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426
/*
  Find the position in the log-index-file for the given log name

  SYNOPSIS
    find_next_log()
    linfo		Store here the next log file name and position to
			the file name after that.
    need_lock		Set this to 1 if the parent doesn't already have a
			lock on LOCK_index

  NOTE
    - Before calling this function, one has to call find_log_pos()
      to set up 'linfo'
    - Mutex needed because we need to make sure the file pointer does not move
      from under our feet

  RETURN VALUES
    0			ok
    LOG_INFO_EOF	End of log-index-file found
    LOG_INFO_IO		Got IO error while reading file
*/

2427
int MYSQL_BIN_LOG::find_next_log(LOG_INFO* linfo, bool need_lock)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2428
{
2429
  int error= 0;
2430
  uint length;
2431 2432
  char *fname= linfo->log_file_name;

2433 2434
  if (need_lock)
    pthread_mutex_lock(&LOCK_index);
2435 2436 2437 2438 2439 2440 2441 2442
  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)
2443
  {
2444
    error = !index_file.error ? LOG_INFO_EOF : LOG_INFO_IO;
2445 2446
    goto err;
  }
2447
  fname[length-1]=0;				// kill \n
2448
  linfo->index_file_offset = my_b_tell(&index_file);
2449

bk@work.mysql.com's avatar
bk@work.mysql.com committed
2450
err:
2451 2452
  if (need_lock)
    pthread_mutex_unlock(&LOCK_index);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2453 2454 2455
  return error;
}

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2456

2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474
/*
  Delete all logs refered to in the index file
  Start writing to a new log file.  The new index file will only contain
  this file.

  SYNOPSIS
     reset_logs()
     thd		Thread

  NOTE
    If not called from slave thread, write start event to new log


  RETURN VALUES
    0	ok
    1   error
*/

2475
bool MYSQL_BIN_LOG::reset_logs(THD* thd)
2476 2477
{
  LOG_INFO linfo;
2478
  bool error=0;
2479
  const char* save_name;
2480
  DBUG_ENTER("reset_logs");
2481

tomas@poseidon.ndb.mysql.com's avatar
tomas@poseidon.ndb.mysql.com committed
2482
  ha_reset_logs(thd);
2483 2484 2485 2486
  /*
    We need to get both locks to be sure that no one is trying to
    write to the index log file.
  */
2487
  pthread_mutex_lock(&LOCK_log);
2488
  pthread_mutex_lock(&LOCK_index);
2489

2490 2491 2492 2493 2494 2495 2496
  /*
    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.
  */
  (void) pthread_mutex_lock(&LOCK_thread_count);
2497 2498 2499 2500

  /* Save variables so that we can reopen the log */
  save_name=name;
  name=0;					// Protect against free
2501
  close(LOG_CLOSE_TO_BE_OPENED);
2502 2503 2504 2505

  /* First delete all old log files */

  if (find_log_pos(&linfo, NullS, 0))
2506 2507 2508 2509
  {
    error=1;
    goto err;
  }
2510

2511
  for (;;)
2512
  {
2513
    my_delete_allow_opened(linfo.log_file_name, MYF(MY_WME));
2514
    if (find_next_log(&linfo, 0))
2515 2516
      break;
  }
2517 2518

  /* Start logging with a new file */
2519
  close(LOG_CLOSE_INDEX);
2520
  my_delete_allow_opened(index_file_name, MYF(MY_WME));	// Reset (open will update)
2521
  if (!thd->slave_thread)
2522
    need_start_event=1;
2523
  if (!open_index_file(index_file_name, 0))
2524
    open(save_name, log_type, 0, io_cache_type, no_auto_events, max_size, 0);
2525
  my_free((gptr) save_name, MYF(0));
2526

2527
err:
2528
  (void) pthread_mutex_unlock(&LOCK_thread_count);
2529
  pthread_mutex_unlock(&LOCK_index);
2530
  pthread_mutex_unlock(&LOCK_log);
2531
  DBUG_RETURN(error);
2532 2533
}

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2534

2535
/*
2536 2537 2538
  Delete relay log files prior to rli->group_relay_log_name
  (i.e. all logs which are not involved in a non-finished group
  (transaction)), remove them from the index file and start on next relay log.
2539 2540 2541

  SYNOPSIS
    purge_first_log()
2542 2543 2544 2545 2546
    rli		 Relay log information
    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).
2547

2548 2549
  NOTE
    - This is only called from the slave-execute thread when it has read
2550 2551 2552 2553 2554
      all commands from a relay log and want to switch to a new relay log.
    - When this happens, we can be in an active transaction as
      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).
2555 2556 2557

  IMPLEMENTATION
    - Protects index file with LOCK_index
2558 2559
    - Delete relevant relay log files
    - Copy all file names after these ones to the front of the index file
2560
    - If the OS has truncate, truncate the file, else fill it with \n'
2561
    - Read the next file name from the index file and store in rli->linfo
2562 2563

  RETURN VALUES
2564 2565 2566 2567
    0			ok
    LOG_INFO_EOF	End of log-index-file found
    LOG_INFO_SEEK	Could not allocate IO cache
    LOG_INFO_IO		Got IO error while reading file
2568 2569
*/

hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
2570
#ifdef HAVE_REPLICATION
2571

2572
int MYSQL_BIN_LOG::purge_first_log(struct st_relay_log_info* rli, bool included) 
2573
{
2574
  int error;
2575 2576
  DBUG_ENTER("purge_first_log");

2577 2578
  DBUG_ASSERT(is_open());
  DBUG_ASSERT(rli->slave_running == 1);
2579
  DBUG_ASSERT(!strcmp(rli->linfo.log_file_name,rli->event_relay_log_name));
2580

2581
  pthread_mutex_lock(&LOCK_index);
2582 2583 2584 2585 2586 2587
  pthread_mutex_lock(&rli->log_space_lock);
  rli->relay_log.purge_logs(rli->group_relay_log_name, included,
                            0, 0, &rli->log_space_total);
  // Tell the I/O thread to take the relay_log_space_limit into account
  rli->ignore_log_space_limit= 0;
  pthread_mutex_unlock(&rli->log_space_lock);
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2588 2589

  /*
2590 2591 2592
    Ok to broadcast after the critical region as there is no risk of
    the mutex being destroyed by this thread later - this helps save
    context switches
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2593
  */
2594 2595 2596 2597 2598
  pthread_cond_broadcast(&rli->log_space_cond);
  
  /*
    Read the next log file name from the index file and pass it back to
    the caller
2599 2600
    If included is true, we want the first relay log;
    otherwise we want the one after event_relay_log_name.
2601
  */
2602 2603 2604 2605
  if ((included && (error=find_log_pos(&rli->linfo, NullS, 0))) ||
      (!included &&
       ((error=find_log_pos(&rli->linfo, rli->event_relay_log_name, 0)) ||
        (error=find_next_log(&rli->linfo, 0)))))
2606 2607
  {
    char buff[22];
2608 2609 2610 2611 2612
    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);
2613 2614
    goto err;
  }
2615

2616
  /*
2617
    Reset rli's coordinates to the current log.
2618
  */
2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632
  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);
2633
    rli->notify_group_relay_log_name_update();
2634
  }
2635 2636 2637

  /* 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
2638

2639
err:
2640
  pthread_mutex_unlock(&LOCK_index);
2641
  DBUG_RETURN(error);
2642 2643
}

2644 2645 2646 2647
/*
  Update log index_file
*/

2648
int MYSQL_BIN_LOG::update_log_index(LOG_INFO* log_info, bool need_update_threads)
2649 2650 2651 2652 2653
{
  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
2654 2655
  if (need_update_threads)
    adjust_linfo_offsets(log_info->index_file_start_offset);
2656 2657
  return 0;
}
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2658

2659 2660 2661 2662 2663
/*
  Remove all logs before the given log from disk and from the index file.

  SYNOPSIS
    purge_logs()
2664 2665 2666 2667 2668 2669 2670
    to_log	        Delete all log file name before this file. 
    included            If true, to_log is deleted too.
    need_mutex
    need_update_threads If we want to update the log coordinates of
                        all threads. False for relay logs, true otherwise.
    freed_log_space     If not null, decrement this variable of
                        the amount of log space freed
2671 2672 2673 2674 2675 2676 2677 2678 2679 2680

  NOTES
    If any of the logs before the deleted one is in use,
    only purge logs up to this one.

  RETURN VALUES
    0				ok
    LOG_INFO_EOF		to_log not found
*/

2681
int MYSQL_BIN_LOG::purge_logs(const char *to_log, 
2682 2683 2684 2685
                          bool included,
                          bool need_mutex, 
                          bool need_update_threads, 
                          ulonglong *decrease_log_space)
2686 2687
{
  int error;
2688
  bool exit_loop= 0;
2689 2690
  LOG_INFO log_info;
  DBUG_ENTER("purge_logs");
2691
  DBUG_PRINT("info",("to_log= %s",to_log));
2692

2693 2694
  if (need_mutex)
    pthread_mutex_lock(&LOCK_index);
2695
  if ((error=find_log_pos(&log_info, to_log, 0 /*no mutex*/)))
2696 2697
    goto err;

2698
  /*
2699
    File name exists in index file; delete until we find this file
2700 2701 2702
    or a file that is used.
  */
  if ((error=find_log_pos(&log_info, NullS, 0 /*no mutex*/)))
2703
    goto err;
2704 2705
  while ((strcmp(to_log,log_info.log_file_name) || (exit_loop=included)) &&
         !log_in_use(log_info.log_file_name))
2706
  {
2707
    ulong file_size= 0;
2708 2709 2710
    if (decrease_log_space) //stat the file we want to delete
    {
      MY_STAT s;
2711 2712 2713 2714 2715 2716

      /* 
         If we could not stat, we can't know the amount
         of space that deletion will free. In most cases,
         deletion won't work either, so it's not a problem.
      */
2717
      if (my_stat(log_info.log_file_name,&s,MYF(0)))
2718
        file_size= s.st_size;
2719
      else
2720 2721
	sql_print_information("Failed to execute my_stat on file '%s'",
			      log_info.log_file_name);
2722 2723 2724 2725 2726 2727 2728
    }
    /*
      It's not fatal if we can't delete a log file ;
      if we could delete it, take its size into account
    */
    DBUG_PRINT("info",("purging %s",log_info.log_file_name));
    if (!my_delete(log_info.log_file_name, MYF(0)) && decrease_log_space)
2729
      *decrease_log_space-= file_size;
tomas@poseidon.ndb.mysql.com's avatar
tomas@poseidon.ndb.mysql.com committed
2730 2731 2732

    ha_binlog_index_purge_file(current_thd, log_info.log_file_name);

2733
    if (find_next_log(&log_info, 0) || exit_loop)
2734
      break;
2735
  }
2736

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2737 2738 2739 2740
  /*
    If we get killed -9 here, the sysadmin would have to edit
    the log index file after restart - otherwise, this should be safe
  */
2741
  error= update_log_index(&log_info, need_update_threads);
2742

2743
err:
2744 2745
  if (need_mutex)
    pthread_mutex_unlock(&LOCK_index);
2746
  DBUG_RETURN(error);
2747 2748
}

2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766
/*
  Remove all logs before the given file date from disk and from the
  index file.

  SYNOPSIS
    purge_logs_before_date()
    thd		Thread pointer
    before_date	Delete all log files before given date.

  NOTES
    If any of the logs before the deleted one is in use,
    only purge logs up to this one.

  RETURN VALUES
    0				ok
    LOG_INFO_PURGE_NO_ROTATE	Binary file that can't be rotated
*/

2767
int MYSQL_BIN_LOG::purge_logs_before_date(time_t purge_time)
2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782
{
  int error;
  LOG_INFO log_info;
  MY_STAT stat_area;

  DBUG_ENTER("purge_logs_before_date");

  pthread_mutex_lock(&LOCK_index);

  /*
    Delete until we find curren file
    or a file that is used or a file
    that is older than purge_time.
  */
  if ((error=find_log_pos(&log_info, NullS, 0 /*no mutex*/)))
2783
    goto err;
2784 2785 2786 2787 2788

  while (strcmp(log_file_name, log_info.log_file_name) &&
	 !log_in_use(log_info.log_file_name))
  {
    /* It's not fatal even if we can't delete a log file */
2789 2790
    if (!my_stat(log_info.log_file_name, &stat_area, MYF(0)) ||
	stat_area.st_mtime >= purge_time)
2791
      break;
2792
    my_delete(log_info.log_file_name, MYF(0));
tomas@poseidon.ndb.mysql.com's avatar
tomas@poseidon.ndb.mysql.com committed
2793 2794 2795

    ha_binlog_index_purge_file(current_thd, log_info.log_file_name);

2796 2797
    if (find_next_log(&log_info, 0))
      break;
2798
  }
2799

2800 2801 2802 2803
  /*
    If we get killed -9 here, the sysadmin would have to edit
    the log index file after restart - otherwise, this should be safe
  */
2804
  error= update_log_index(&log_info, 1);
2805

2806 2807
err:
  pthread_mutex_unlock(&LOCK_index);
2808
  DBUG_RETURN(error);
2809 2810
}

2811

hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
2812
#endif /* HAVE_REPLICATION */
2813

2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825

/*
  Create a new log file name

  SYNOPSIS
    make_log_name()
    buf			buf of at least FN_REFLEN where new name is stored

  NOTE
    If file name will be longer then FN_REFLEN it will be truncated
*/

2826
void MYSQL_BIN_LOG::make_log_name(char* buf, const char* log_ident)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2827
{
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2828 2829 2830 2831 2832
  uint dir_len = dirname_length(log_file_name); 
  if (dir_len > FN_REFLEN)
    dir_len=FN_REFLEN-1;
  strnmov(buf, log_file_name, dir_len);
  strmake(buf+dir_len, log_ident, FN_REFLEN - dir_len);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2833 2834
}

2835 2836 2837 2838 2839

/*
  Check if we are writing/reading to the given log file
*/

2840
bool MYSQL_BIN_LOG::is_active(const char *log_file_name_arg)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2841
{
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2842
  return !strcmp(log_file_name, log_file_name_arg);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2843 2844
}

2845

2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865
/*
  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);
}


2866 2867 2868 2869
/*
  Start writing to a new log file or reopen the old file

  SYNOPSIS
2870
    new_file_impl()
2871
    need_lock		Set to 1 if caller has not locked LOCK_log
2872 2873 2874 2875 2876

  NOTE
    The new file name is stored last in the index file
*/

2877
void MYSQL_BIN_LOG::new_file_impl(bool need_lock)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2878
{
2879 2880
  char new_name[FN_REFLEN], *new_name_ptr, *old_name;

2881
  DBUG_ENTER("MYSQL_BIN_LOG::new_file_impl");
2882
  if (!is_open())
2883 2884 2885 2886
  {
    DBUG_PRINT("info",("log is closed"));
    DBUG_VOID_RETURN;
  }
2887 2888 2889

  if (need_lock)
    pthread_mutex_lock(&LOCK_log);
2890 2891
  pthread_mutex_lock(&LOCK_index);

2892 2893 2894
  safe_mutex_assert_owner(&LOCK_log);
  safe_mutex_assert_owner(&LOCK_index);

2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912
  /*
    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);
    while (prepared_xids)
      pthread_cond_wait(&COND_prep_xids, &LOCK_prep_xids);
    pthread_mutex_unlock(&LOCK_prep_xids);
  }

2913
  /* Reuse old name if not binlog and not update log */
2914
  new_name_ptr= name;
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2915

2916
  /*
2917 2918 2919
    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.
2920
  */
2921 2922 2923
  if (generate_new_name(new_name, name))
    goto end;
  new_name_ptr=new_name;
2924

2925
  if (log_type == LOG_BIN)
2926
  {
2927
    if (!no_auto_events)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2928
    {
2929
      /*
2930 2931
        We log the whole file name for log file as the user may decide
        to change base names at some point.
2932
      */
monty@mysql.com's avatar
monty@mysql.com committed
2933
      THD *thd = current_thd; /* may be 0 if we are reacting to SIGHUP */
2934
      Rotate_log_event r(new_name+dirname_length(new_name),
2935
                         0, LOG_EVENT_OFFSET, 0);
2936
      r.write(&log_file);
2937
      bytes_written += r.data_written;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2938
    }
2939 2940 2941 2942 2943
    /*
      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.
    */
2944
    signal_update();
2945 2946 2947
  }
  old_name=name;
  name=0;				// Don't free name
2948
  close(LOG_CLOSE_TO_BE_OPENED);
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2949

2950
  /*
2951
     Note that at this point, log_state != LOG_CLOSED (important for is_open()).
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2952
  */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2953

2954
  /*
2955
     new_file() is only used for rotation (in FLUSH LOGS or because size >
2956
     max_binlog_size or max_relay_log_size).
2957 2958 2959 2960
     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.
2961
  */
2962

2963
  open(old_name, log_type, new_name_ptr,
2964
       io_cache_type, no_auto_events, max_size, 1);
2965 2966
  my_free(old_name,MYF(0));

2967
end:
2968 2969
  if (need_lock)
    pthread_mutex_unlock(&LOCK_log);
2970 2971
  pthread_mutex_unlock(&LOCK_index);

2972
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2973 2974
}

2975

2976
bool MYSQL_BIN_LOG::append(Log_event* ev)
2977 2978 2979
{
  bool error = 0;
  pthread_mutex_lock(&LOCK_log);
2980
  DBUG_ENTER("MYSQL_BIN_LOG::append");
2981

2982
  DBUG_ASSERT(log_file.type == SEQ_READ_APPEND);
2983 2984 2985 2986
  /*
    Log_event::write() is smart enough to use my_b_write() or
    my_b_append() depending on the kind of cache we have.
  */
2987 2988 2989 2990 2991
  if (ev->write(&log_file))
  {
    error=1;
    goto err;
  }
2992
  bytes_written+= ev->data_written;
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
2993
  DBUG_PRINT("info",("max_size: %lu",max_size));
2994
  if ((uint) my_b_append_tell(&log_file) > max_size)
2995
    new_file_without_locking();
2996

2997
err:
2998
  pthread_mutex_unlock(&LOCK_log);
2999
  signal_update();				// Safe as we don't call close
3000
  DBUG_RETURN(error);
3001 3002
}

3003

3004
bool MYSQL_BIN_LOG::appendv(const char* buf, uint len,...)
3005
{
3006
  bool error= 0;
3007
  DBUG_ENTER("MYSQL_BIN_LOG::appendv");
3008 3009
  va_list(args);
  va_start(args,len);
3010

3011
  DBUG_ASSERT(log_file.type == SEQ_READ_APPEND);
3012

3013
  safe_mutex_assert_owner(&LOCK_log);
3014 3015
  do
  {
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
3016
    if (my_b_append(&log_file,(byte*) buf,len))
3017
    {
3018 3019
      error= 1;
      goto err;
3020
    }
3021
    bytes_written += len;
3022
  } while ((buf=va_arg(args,const char*)) && (len=va_arg(args,uint)));
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
3023
  DBUG_PRINT("info",("max_size: %lu",max_size));
3024
  if ((uint) my_b_append_tell(&log_file) > max_size)
3025
    new_file_without_locking();
3026 3027

err:
3028 3029
  if (!error)
    signal_update();
3030
  DBUG_RETURN(error);
3031
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3032

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
3033

3034
bool MYSQL_BIN_LOG::flush_and_sync()
3035
{
3036 3037 3038 3039 3040
  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)
3041 3042
  {
    sync_binlog_counter= 0;
3043
    err=my_sync(fd, MYF(MY_WME));
3044
  }
3045
  return err;
3046 3047
}

3048
void MYSQL_BIN_LOG::start_union_events(THD *thd)
3049 3050 3051 3052 3053
{
  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;
3054
  thd->binlog_evt_union.first_query_id= thd->query_id;
3055 3056
}

3057
void MYSQL_BIN_LOG::stop_union_events(THD *thd)
3058 3059 3060 3061 3062
{
  DBUG_ASSERT(thd->binlog_evt_union.do_union);
  thd->binlog_evt_union.do_union= FALSE;
}

3063
bool MYSQL_BIN_LOG::is_query_in_union(THD *thd, query_id_t query_id_param)
3064 3065 3066 3067 3068
{
  return (thd->binlog_evt_union.do_union && 
          query_id_param >= thd->binlog_evt_union.first_query_id);
}

3069 3070 3071 3072 3073 3074 3075 3076 3077 3078

/*
  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=
3079
    (binlog_trx_data*) ha_data[binlog_hton->slot];
3080 3081 3082 3083

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

3084
  ha_data[binlog_hton->slot]= trx_data=
3085 3086 3087 3088 3089 3090
    (binlog_trx_data*) my_malloc(sizeof(binlog_trx_data), MYF(MY_ZEROFILL));
  if (!trx_data ||
      open_cached_file(&trx_data->trans_log, mysql_tmpdir,
                       LOG_PREFIX, binlog_cache_size, MYF(MY_WME)))
  {
    my_free((gptr)trx_data, MYF(MY_ALLOW_ZERO_PTR));
3091
    ha_data[binlog_hton->slot]= 0;
3092 3093 3094 3095 3096 3097
    DBUG_RETURN(1);                      // Didn't manage to set it up
  }
  trx_data->trans_log.end_of_file= max_binlog_cache_size;
  DBUG_RETURN(0);
}

3098 3099 3100 3101 3102 3103 3104
/*
  Write a table map to the binary log.

  This function is called from ha_external_lock() after the storage
  engine has registered for the transaction.
 */

3105
#ifdef HAVE_ROW_BASED_REPLICATION
3106 3107
int THD::binlog_write_table_map(TABLE *table, bool is_trans)
{
3108
  int error;
3109
  DBUG_ENTER("THD::binlog_write_table_map");
3110 3111 3112
  DBUG_PRINT("enter", ("table: %0xlx  (%s: #%u)",
                       (long) table, table->s->table_name,
                       table->s->table_map_id));
3113 3114

  /* Pre-conditions */
3115
  DBUG_ASSERT(current_stmt_binlog_row_based && mysql_bin_log.is_open());
3116 3117 3118 3119 3120 3121 3122 3123 3124
  DBUG_ASSERT(table->s->table_map_id != ULONG_MAX);

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

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

  if (is_trans)
3125 3126
    trans_register_ha(this,
                      (options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN)) != 0,
3127
                      binlog_hton);
3128

3129
  if ((error= mysql_bin_log.write(&the_event)))
3130 3131
    DBUG_RETURN(error);

3132
  binlog_table_maps++;
3133 3134 3135 3136
  table->s->table_map_version= mysql_bin_log.table_map_version();
  DBUG_RETURN(0);
}

3137 3138 3139 3140
Rows_log_event*
THD::binlog_get_pending_rows_event() const
{
  binlog_trx_data *const trx_data=
3141
    (binlog_trx_data*) ha_data[binlog_hton->slot];
3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153
  /*
    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.
   */
  return trx_data ? trx_data->pending : NULL;
}

void
THD::binlog_set_pending_rows_event(Rows_log_event* ev)
{
3154
  if (ha_data[binlog_hton->slot] == NULL)
3155 3156
    binlog_setup_trx_data();

3157
  binlog_trx_data *const trx_data=
3158
    (binlog_trx_data*) ha_data[binlog_hton->slot];
3159

3160 3161 3162 3163 3164 3165 3166 3167 3168 3169
  DBUG_ASSERT(trx_data);
  trx_data->pending= ev;
}


/*
  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.
*/
3170 3171
int MYSQL_BIN_LOG::
  flush_and_set_pending_rows_event(THD *thd, Rows_log_event* event)
3172
{
3173
  DBUG_ENTER("MYSQL_BIN_LOG::flush_and_set_pending_rows_event(event)");
3174
  DBUG_ASSERT(mysql_bin_log.is_open());
3175 3176 3177 3178 3179
  DBUG_PRINT("enter", ("event=%p", event));

  int error= 0;

  binlog_trx_data *const trx_data=
3180
    (binlog_trx_data*) thd->ha_data[binlog_hton->slot];
3181 3182 3183

  DBUG_ASSERT(trx_data);

3184 3185
  DBUG_PRINT("info", ("trx_data->pending=%p", trx_data->pending));

3186 3187 3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246
  if (Rows_log_event* pending= trx_data->pending)
  {
    IO_CACHE *file= &log_file;

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

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

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

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

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

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

    delete pending;

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

    pthread_mutex_unlock(&LOCK_log);
  }

3247
  thd->binlog_set_pending_rows_event(event);
3248 3249 3250

  DBUG_RETURN(error);
}
3251
#endif /*HAVE_ROW_BASED_REPLICATION*/
3252

3253 3254 3255 3256
/*
  Write an event to the binary log
*/

3257
bool MYSQL_BIN_LOG::write(Log_event *event_info)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3258
{
3259 3260
  THD *thd= event_info->thd;
  bool error= 1;
3261
  DBUG_ENTER("MYSQL_BIN_LOG::write(Log_event *)");
3262

3263 3264
  if (thd->binlog_evt_union.do_union)
  {
3265 3266 3267 3268
    /*
      In Stored function; Remember that function call caused an update.
      We will log the function call to the binary log on function exit
    */
3269 3270 3271 3272
    thd->binlog_evt_union.unioned_events= TRUE;
    thd->binlog_evt_union.unioned_events_trans |= event_info->cache_stmt;
    DBUG_RETURN(0);
  }
3273 3274 3275 3276 3277 3278 3279

  /*
    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.

3280 3281 3282
    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.
3283
  */
3284
#ifdef HAVE_ROW_BASED_REPLICATION
3285 3286 3287
  bool const end_stmt=
    thd->prelocked_mode && thd->lex->requires_prelocking();
  thd->binlog_flush_pending_rows_event(end_stmt);
3288
#endif /*HAVE_ROW_BASED_REPLICATION*/
3289

3290 3291
  pthread_mutex_lock(&LOCK_log);

3292
  /*
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3293 3294 3295 3296
     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.
  */
3297
  if (likely(is_open()))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3298
  {
3299 3300
    const char *local_db= event_info->get_db();
    IO_CACHE *file= &log_file;
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
3301
#ifdef HAVE_REPLICATION
3302
    /*
3303 3304 3305 3306
       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)"
    */
3307
    if ((thd && !(thd->options & OPTION_BIN_LOG)) ||
lars@mysql.com's avatar
Merge  
lars@mysql.com committed
3308
	(!binlog_filter->db_ok(local_db)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3309
    {
3310
      VOID(pthread_mutex_unlock(&LOCK_log));
3311 3312 3313
      DBUG_PRINT("info",("OPTION_BIN_LOG is %s, db_ok('%s') == %d",
                         (thd->options & OPTION_BIN_LOG) ? "set" : "clear",
                         local_db, binlog_filter->db_ok(local_db)));
3314
      DBUG_RETURN(0);
3315
    }
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
3316
#endif /* HAVE_REPLICATION */
3317

3318
#if defined(USING_TRANSACTIONS) && defined(HAVE_ROW_BASED_REPLICATION)
3319 3320 3321 3322 3323 3324 3325
    /*
      Should we write to the binlog cache or to the binlog on disk?
      Write to the binlog cache if:
      - it is already not empty (meaning we're in a transaction; note that the
     present event could be about a non-transactional table, but still we need
     to write to the binlog cache in that case to handle updates to mixed
     trans/non-trans table types the best possible in binlogging)
3326
      - or if the event asks for it (cache_stmt == TRUE).
3327 3328 3329
    */
    if (opt_using_transactions && thd)
    {
3330 3331
      if (thd->binlog_setup_trx_data())
        goto err;
3332

3333
      binlog_trx_data *const trx_data=
3334
        (binlog_trx_data*) thd->ha_data[binlog_hton->slot];
3335
      IO_CACHE *trans_log= &trx_data->trans_log;
3336 3337
      bool trans_log_in_use= my_b_tell(trans_log) != 0;
      if (event_info->get_cache_stmt() && !trans_log_in_use)
3338
        trans_register_ha(thd,
3339
                          (thd->options &
3340
                           (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN)) != 0,
3341
                          binlog_hton);
3342 3343 3344
      if (event_info->get_cache_stmt() || trans_log_in_use)
      {
        DBUG_PRINT("info", ("Using trans_log"));
3345
        file= trans_log;
3346
      }
3347
      /*
3348
        TODO as Mats suggested, for all the cases above where we write to
3349 3350
        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
3351
        LOCK_log.
3352
      */
3353
    }
3354
#endif /* USING_TRANSACTIONS && HAVE_ROW_BASED_REPLICATION */
3355
    DBUG_PRINT("info",("event type: %d",event_info->get_type_code()));
3356

3357 3358 3359 3360
    /*
      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
3361 3362

    /*
3363 3364
      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
3365 3366
    */

3367 3368 3369 3370
    /*
      If row-based binlogging, Insert_id, Rand and other kind of "setting
      context" events are not needed.
    */
3371
    if (thd)
3372
    {
3373
      if (!thd->current_stmt_binlog_row_based)
3374
      {
3375
        if (thd->stmt_depends_on_first_successful_insert_id_in_prev_stmt)
3376 3377
        {
          Intvar_log_event e(thd,(uchar) LAST_INSERT_ID_EVENT,
3378
                             thd->first_successful_insert_id_in_prev_stmt_for_binlog);
3379 3380 3381
          if (e.write(file))
            goto err;
        }
3382
        if (thd->auto_inc_intervals_in_cur_stmt_for_binlog.nb_elements() > 0)
3383
        {
3384 3385
          DBUG_PRINT("info",("number of auto_inc intervals: %lu",
                             thd->auto_inc_intervals_in_cur_stmt_for_binlog.nb_elements()));
3386 3387 3388 3389 3390
          /*
            If the auto_increment was second in a table's index (possible with
            MyISAM or BDB) (table->next_number_key_offset != 0), such event is
            in fact not necessary. We could avoid logging it.
          */
3391 3392
          Intvar_log_event e(thd,(uchar) INSERT_ID_EVENT,
                             thd->auto_inc_intervals_in_cur_stmt_for_binlog.minimum());
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
          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;
            get_dynamic(&thd->user_var_events,(gptr) &user_var_event, i);
            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;
          }
        }
3418
      }
3419
    }
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
3420

3421 3422
    /*
       Write the SQL command
3423
     */
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
3424

monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
3425
    if (event_info->write(file))
3426
      goto err;
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
3427

3428
    if (file == &log_file) // we are writing to the real log (disk)
3429
    {
3430
      if (flush_and_sync())
monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
3431
	goto err;
3432 3433
      signal_update();
      rotate_and_purge(RP_LOCK_LOG_IS_ALREADY_LOCKED);
3434
    }
monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
3435
    error=0;
3436

3437 3438 3439 3440
err:
    if (error)
    {
      if (my_errno == EFBIG)
bell@sanja.is.com.ua's avatar
bell@sanja.is.com.ua committed
3441
	my_message(ER_TRANS_CACHE_FULL, ER(ER_TRANS_CACHE_FULL), MYF(0));
3442 3443 3444 3445 3446
      else
	my_error(ER_ERROR_ON_WRITE, MYF(0), name, errno);
      write_error=1;
    }
  }
3447

3448 3449 3450
  if (event_info->flags & LOG_EVENT_UPDATE_TABLE_MAP_VERSION_F)
    ++m_table_map_version;

3451
  pthread_mutex_unlock(&LOCK_log);
3452
  DBUG_RETURN(error);
3453 3454
}

3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482

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,
                    time_t query_start_arg)
{
  return logger.slow_log_print(thd, query, query_length, query_start_arg);
}


bool general_log_print(THD *thd, enum enum_server_command command,
                       const char *format, ...)
{
  va_list args;
  uint error= 0;

  va_start(args, format);
  error= logger.general_log_print(thd, command, format, args);
  va_end(args);

  return error;
}

3483
void MYSQL_BIN_LOG::rotate_and_purge(uint flags)
3484
{
3485 3486
  if (!(flags & RP_LOCK_LOG_IS_ALREADY_LOCKED))
    pthread_mutex_lock(&LOCK_log);
3487 3488
  if ((flags & RP_FORCE_ROTATE) ||
      (my_b_tell(&log_file) >= (my_off_t) max_size))
3489
  {
3490
    new_file_without_locking();
3491 3492 3493 3494 3495 3496 3497 3498 3499
#ifdef HAVE_REPLICATION
    if (expire_logs_days)
    {
      long purge_time= time(0) - expire_logs_days*24*60*60;
      if (purge_time >= 0)
        purge_logs_before_date(purge_time);
    }
#endif
  }
3500 3501
  if (!(flags & RP_LOCK_LOG_IS_ALREADY_LOCKED))
    pthread_mutex_unlock(&LOCK_log);
3502
}
3503

3504
uint MYSQL_BIN_LOG::next_file_id()
3505 3506 3507 3508 3509 3510 3511 3512
{
  uint res;
  pthread_mutex_lock(&LOCK_log);
  res = file_id++;
  pthread_mutex_unlock(&LOCK_log);
  return res;
}

3513

3514 3515
/*
  Write a cached log entry to the binary log
3516

guilhem@mysql.com's avatar
guilhem@mysql.com committed
3517 3518
  SYNOPSIS
    write()
3519
    thd
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3520 3521
    cache		The cache to copy to the binlog

3522 3523
  NOTE
    - We only come here if there is something in the cache.
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
3524
    - The thing in the cache is always a complete transaction
3525 3526 3527 3528
    - 'cache' needs to be reinitialized after this functions returns.

  IMPLEMENTATION
    - To support transaction over replication, we wrap the transaction
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3529
      with BEGIN/COMMIT or BEGIN/ROLLBACK in the binary log.
3530 3531 3532
      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.
3533 3534
*/

3535
bool MYSQL_BIN_LOG::write(THD *thd, IO_CACHE *cache, Log_event *commit_event)
3536
{
3537
  DBUG_ENTER("MYSQL_BIN_LOG::write(THD *, IO_CACHE *, Log_event *)");
3538
  VOID(pthread_mutex_lock(&LOCK_log));
3539

3540 3541 3542
  /* NULL would represent nothing to replicate after ROLLBACK */
  DBUG_ASSERT(commit_event != NULL);

3543
  if (likely(is_open()))                       // Should always be true
3544
  {
3545 3546
    uint length;

3547
    /*
3548 3549
      Log "BEGIN" at the beginning of the transaction.
      which may contain more than 1 SQL statement.
3550
    */
3551
    if (thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN))
3552
    {
3553
      Query_log_event qinfo(thd, STRING_WITH_LEN("BEGIN"), TRUE, FALSE);
3554 3555 3556 3557
      /*
        Imagine this is rollback due to net timeout, after all statements of
        the transaction succeeded. Then we want a zero-error code in BEGIN.
        In other words, if there was a really serious error code it's already
3558 3559 3560
        in the statement's events, there is no need to put it also in this
        internally generated event, and as this event is generated late it
        would lead to false alarms.
3561 3562 3563
        This is safer than thd->clear_error() against kills at shutdown.
      */
      qinfo.error_code= 0;
3564 3565 3566
      /*
        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
3567 3568 3569
	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.
3570
      */
3571 3572 3573
      if (qinfo.write(&log_file))
	goto err;
    }
3574
    /* Read from the file used to cache the queries .*/
3575
    if (reinit_io_cache(cache, READ_CACHE, 0, 0, 0))
3576
      goto err;
3577
    length=my_b_bytes_in_cache(cache);
3578
    DBUG_EXECUTE_IF("half_binlogged_transaction", length-=100;);
3579
    do
3580
    {
3581
      /* Write data to the binary log file */
monty@bitch.mysql.fi's avatar
monty@bitch.mysql.fi committed
3582
      if (my_b_write(&log_file, cache->read_pos, length))
3583
	goto err;
monty@bitch.mysql.fi's avatar
monty@bitch.mysql.fi committed
3584
      cache->read_pos=cache->read_end;		// Mark buffer used up
3585
      DBUG_EXECUTE_IF("half_binlogged_transaction", goto DBUG_skip_commit;);
3586
    } while ((length=my_b_fill(cache)));
3587

3588 3589
    if (commit_event->write(&log_file))
      goto err;
3590
#ifndef DBUG_OFF
3591
DBUG_skip_commit:
3592
#endif
3593
    if (flush_and_sync())
3594
      goto err;
3595
    DBUG_EXECUTE_IF("half_binlogged_transaction", abort(););
3596 3597
    if (cache->error)				// Error on read
    {
3598
      sql_print_error(ER(ER_ERROR_ON_READ), cache->file_name, errno);
3599
      write_error=1;				// Don't give more errors
3600
      goto err;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3601
    }
3602
    signal_update();
3603 3604 3605 3606 3607 3608 3609 3610 3611
    /*
      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.
    */
    if (commit_event->get_type_code() == XID_EVENT)
3612 3613 3614 3615 3616
    {
      pthread_mutex_lock(&LOCK_prep_xids);
      prepared_xids++;
      pthread_mutex_unlock(&LOCK_prep_xids);
    }
3617 3618
    else
      rotate_and_purge(RP_LOCK_LOG_IS_ALREADY_LOCKED);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3619
  }
3620
  VOID(pthread_mutex_unlock(&LOCK_log));
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
3621

3622
  DBUG_RETURN(0);
3623 3624

err:
3625 3626 3627 3628 3629
  if (!write_error)
  {
    write_error= 1;
    sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
  }
3630
  VOID(pthread_mutex_unlock(&LOCK_log));
3631
  DBUG_RETURN(1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3632 3633
}

3634

3635 3636 3637 3638 3639 3640
/*
  Wait until we get a signal that the binary log has been updated

  SYNOPSIS
    wait_for_update()
    thd			Thread variable
3641
    is_slave            If 0, the caller is the Binlog_dump thread from master;
3642 3643
                        if 1, the caller is the SQL thread from the slave. This
                        influences only thd->proc_info.
3644 3645 3646

  NOTES
    One must have a lock on LOCK_log before calling this function.
3647
    This lock will be released before return! That's required by
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3648
    THD::enter_cond() (see NOTES in sql_class.h).
3649 3650
*/

3651
void MYSQL_BIN_LOG::wait_for_update(THD* thd, bool is_slave)
3652
{
monty@mysql.com's avatar
monty@mysql.com committed
3653 3654
  const char *old_msg;
  DBUG_ENTER("wait_for_update");
3655

monty@mysql.com's avatar
monty@mysql.com committed
3656
  old_msg= thd->enter_cond(&update_cond, &LOCK_log,
3657
                           is_slave ?
monty@mysql.com's avatar
monty@mysql.com committed
3658
                           "Has read all relay log; waiting for the slave I/O "
3659
                           "thread to update it" :
monty@mysql.com's avatar
monty@mysql.com committed
3660
                           "Has sent all binlog to slave; waiting for binlog "
3661
                           "to be updated");
3662 3663
  pthread_cond_wait(&update_cond, &LOCK_log);
  thd->exit_cond(old_msg);
monty@mysql.com's avatar
monty@mysql.com committed
3664
  DBUG_VOID_RETURN;
3665
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3666

3667

3668 3669 3670 3671 3672
/*
  Close the log file

  SYNOPSIS
    close()
3673 3674 3675 3676 3677
    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
3678 3679 3680 3681

  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
3682 3683
*/

3684
void MYSQL_BIN_LOG::close(uint exiting)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3685
{					// One can't set log_type here!
3686
  DBUG_ENTER("MYSQL_BIN_LOG::close");
3687
  DBUG_PRINT("enter",("exiting: %d", (int) exiting));
3688
  if (log_state == LOG_OPENED)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3689
  {
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
3690
#ifdef HAVE_REPLICATION
3691 3692
    if (log_type == LOG_BIN && !no_auto_events &&
	(exiting & LOG_CLOSE_STOP_EVENT))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3693 3694
    {
      Stop_log_event s;
3695
      s.write(&log_file);
3696
      bytes_written+= s.data_written;
3697
      signal_update();
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3698
    }
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
3699
#endif /* HAVE_REPLICATION */
3700 3701

    /* don't pwrite in a file opened with O_APPEND - it doesn't work */
3702
    if (log_file.type == WRITE_CACHE && log_type == LOG_BIN)
3703 3704
    {
      my_off_t offset= BIN_LOG_HEADER_SIZE + FLAGS_OFFSET;
serg@serg.mylan's avatar
serg@serg.mylan committed
3705
      byte flags=0; // clearing LOG_EVENT_BINLOG_IN_USE_F
3706
      my_pwrite(log_file.file, &flags, 1, offset, MYF(0));
3707 3708
    }

3709 3710
    /* 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
3711
  }
3712 3713 3714 3715 3716 3717

  /*
    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.
  */

3718
  if ((exiting & LOG_CLOSE_INDEX) && my_b_inited(&index_file))
3719
  {
3720 3721
    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
3722
    {
3723 3724
      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
3725
    }
3726
  }
3727
  log_state= (exiting & LOG_CLOSE_TO_BE_OPENED) ? LOG_TO_BE_OPENED : LOG_CLOSED;
3728
  safeFree(name);
3729
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3730 3731 3732
}


3733
void MYSQL_BIN_LOG::set_max_size(ulong max_size_arg)
3734 3735 3736 3737 3738 3739 3740 3741
{
  /*
    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.
  */
3742
  DBUG_ENTER("MYSQL_BIN_LOG::set_max_size");
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3743
  pthread_mutex_lock(&LOCK_log);
3744 3745
  if (is_open())
    max_size= max_size_arg;
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3746 3747
  pthread_mutex_unlock(&LOCK_log);
  DBUG_VOID_RETURN;
3748
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3749

3750

3751 3752
/*
  Check if a string is a valid number
3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766

  SYNOPSIS
    test_if_number()
    str			String to test
    res			Store value here
    allow_wildcards	Set to 1 if we should ignore '%' and '_'

  NOTE
    For the moment the allow_wildcards argument is not used
    Should be move to some other file.

  RETURN VALUES
    1	String is a number
    0	Error
3767
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779

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++;
3780 3781
  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
3782 3783 3784 3785 3786 3787 3788
  {
    flag=1;
    str++;
  }
  if (*str == '.')
  {
    for (str++ ;
3789
	 my_isdigit(files_charset_info,*str) ||
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800
	   (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 */


serg@serg.mylan's avatar
serg@serg.mylan committed
3801
void print_buffer_to_file(enum loglevel level, const char *buffer)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3802 3803 3804 3805
{
  time_t skr;
  struct tm tm_tmp;
  struct tm *start;
monty@mysql.com's avatar
monty@mysql.com committed
3806 3807
  DBUG_ENTER("print_buffer_to_file");
  DBUG_PRINT("enter",("buffer: %s", buffer));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3808 3809

  VOID(pthread_mutex_lock(&LOCK_error_log));
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
3810

bk@work.mysql.com's avatar
bk@work.mysql.com committed
3811
  skr=time(NULL);
3812
  localtime_r(&skr, &tm_tmp);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3813
  start=&tm_tmp;
3814

3815
  fprintf(stderr, "%02d%02d%02d %2d:%02d:%02d [%s] %s\n",
3816 3817 3818 3819 3820 3821
          start->tm_year % 100,
          start->tm_mon+1,
          start->tm_mday,
          start->tm_hour,
          start->tm_min,
          start->tm_sec,
monty@mysql.com's avatar
monty@mysql.com committed
3822
          (level == ERROR_LEVEL ? "ERROR" : level == WARNING_LEVEL ?
3823
           "Warning" : "Note"),
monty@mysql.com's avatar
monty@mysql.com committed
3824
          buffer);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
3825

bk@work.mysql.com's avatar
bk@work.mysql.com committed
3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840
  fflush(stderr);

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


void sql_perror(const char *message)
{
#ifdef HAVE_STRERROR
  sql_print_error("%s: %s",message, strerror(errno));
#else
  perror(message);
#endif
}
3841

monty@mysql.com's avatar
monty@mysql.com committed
3842

3843 3844 3845
bool flush_error_log()
{
  bool result=0;
3846
  if (opt_error_log)
3847 3848 3849 3850
  {
    char err_renamed[FN_REFLEN], *end;
    end= strmake(err_renamed,log_error_file,FN_REFLEN-4);
    strmov(end, "-old");
3851
    VOID(pthread_mutex_lock(&LOCK_error_log));
3852 3853 3854 3855 3856 3857
#ifdef __WIN__
    char err_temp[FN_REFLEN+4];
    /*
     On Windows is necessary a temporary file for to rename
     the current error file.
    */
3858
    strxmov(err_temp, err_renamed,"-tmp",NullS);
3859 3860 3861 3862 3863 3864 3865 3866 3867 3868 3869 3870 3871
    (void) my_delete(err_temp, MYF(0)); 
    if (freopen(err_temp,"a+",stdout))
    {
      freopen(err_temp,"a+",stderr);
      (void) my_delete(err_renamed, MYF(0));
      my_rename(log_error_file,err_renamed,MYF(0));
      if (freopen(log_error_file,"a+",stdout))
        freopen(log_error_file,"a+",stderr);
      int fd, bytes;
      char buf[IO_SIZE];
      if ((fd = my_open(err_temp, O_RDONLY, MYF(0))) >= 0)
      {
        while ((bytes = (int) my_read(fd, (byte*) buf, IO_SIZE, MYF(0))) > 0)
3872
             my_fwrite(stderr, (byte*) buf, bytes, MYF(0));
3873 3874 3875 3876 3877 3878 3879 3880 3881 3882 3883 3884 3885
        my_close(fd, MYF(0));
      }
      (void) my_delete(err_temp, MYF(0)); 
    }
    else
     result= 1;
#else
   my_rename(log_error_file,err_renamed,MYF(0));
   if (freopen(log_error_file,"a+",stdout))
     freopen(log_error_file,"a+",stderr);
   else
     result= 1;
#endif
3886
    VOID(pthread_mutex_unlock(&LOCK_error_log));
3887 3888 3889
  }
   return result;
}
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
3890

3891
void MYSQL_BIN_LOG::signal_update()
monty@mysql.com's avatar
monty@mysql.com committed
3892
{
3893
  DBUG_ENTER("MYSQL_BIN_LOG::signal_update");
monty@mysql.com's avatar
monty@mysql.com committed
3894 3895 3896 3897
  pthread_cond_broadcast(&update_cond);
  DBUG_VOID_RETURN;
}

3898
#ifdef __NT__
monty@mysql.com's avatar
monty@mysql.com committed
3899 3900
void print_buffer_to_nt_eventlog(enum loglevel level, char *buff,
                                 uint length, int buffLen)
3901 3902 3903 3904
{
  HANDLE event;
  char   *buffptr;
  LPCSTR *buffmsgptr;
serg@serg.mylan's avatar
serg@serg.mylan committed
3905
  DBUG_ENTER("print_buffer_to_nt_eventlog");
3906

monty@mysql.com's avatar
monty@mysql.com committed
3907
  buffptr= buff;
3908
  if (length > (uint)(buffLen-5))
3909
  {
3910
    char *newBuff= new char[length + 5];
serg@serg.mylan's avatar
serg@serg.mylan committed
3911
    strcpy(newBuff, buff);
monty@mysql.com's avatar
monty@mysql.com committed
3912
    buffptr= newBuff;
3913
  }
monty@mysql.com's avatar
monty@mysql.com committed
3914 3915
  strmov(buffptr+length, "\r\n\r\n");
  buffmsgptr= (LPCSTR*) &buffptr;               // Keep windows happy
3916

monty@mysql.com's avatar
monty@mysql.com committed
3917 3918
  setup_windows_event_source();
  if ((event= RegisterEventSource(NULL,"MySQL")))
3919
  {
monty@mysql.com's avatar
monty@mysql.com committed
3920
    switch (level) {
3921
      case ERROR_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
3922 3923
        ReportEvent(event, EVENTLOG_ERROR_TYPE, 0, MSG_DEFAULT, NULL, 1, 0,
                    buffmsgptr, NULL);
3924 3925
        break;
      case WARNING_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
3926 3927
        ReportEvent(event, EVENTLOG_WARNING_TYPE, 0, MSG_DEFAULT, NULL, 1, 0,
                    buffmsgptr, NULL);
3928 3929
        break;
      case INFORMATION_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
3930 3931
        ReportEvent(event, EVENTLOG_INFORMATION_TYPE, 0, MSG_DEFAULT, NULL, 1,
                    0, buffmsgptr, NULL);
3932 3933 3934 3935 3936
        break;
    }
    DeregisterEventSource(event);
  }

monty@mysql.com's avatar
monty@mysql.com committed
3937
  /* if we created a string buffer, then delete it */
serg@serg.mylan's avatar
serg@serg.mylan committed
3938
  if (buffptr != buff)
3939 3940 3941 3942
    delete[] buffptr;

  DBUG_VOID_RETURN;
}
monty@mysql.com's avatar
monty@mysql.com committed
3943 3944
#endif /* __NT__ */

3945 3946

/*
monty@mysql.com's avatar
monty@mysql.com committed
3947 3948
  Prints a printf style message to the error log and, under NT, to the
  Windows event log.
3949 3950 3951

  SYNOPSIS
    vprint_msg_to_log()
monty@mysql.com's avatar
monty@mysql.com committed
3952 3953
    event_type             Type of event to write (Error, Warning, or Info)
    format                 Printf style format of message
serg@serg.mylan's avatar
serg@serg.mylan committed
3954
    args                   va_list list of arguments for the message
3955 3956 3957 3958

  NOTE

  IMPLEMENTATION
monty@mysql.com's avatar
monty@mysql.com committed
3959 3960
    This function prints the message into a buffer and then sends that buffer
    to other functions to write that message to other logging sources.
3961 3962

  RETURN VALUES
3963 3964 3965
    The function always returns 0. The return value is present in the
    signature to be compatible with other logging routines, which could
    return an error (e.g. logging to the log tables)
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
3966
*/
monty@mysql.com's avatar
monty@mysql.com committed
3967

3968
#ifdef EMBEDDED_LIBRARY
rburnett@production.mysql.com's avatar
rburnett@production.mysql.com committed
3969
int vprint_msg_to_log(enum loglevel level __attribute__((unused)),
3970 3971
                       const char *format __attribute__((unused)),
                       va_list argsi __attribute__((unused)))
rburnett@production.mysql.com's avatar
rburnett@production.mysql.com committed
3972 3973 3974 3975
{
  DBUG_ENTER("vprint_msg_to_log");
  DBUG_RETURN(0);
}
3976
#else /*!EMBEDDED_LIBRARY*/
rburnett@production.mysql.com's avatar
rburnett@production.mysql.com committed
3977
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
3978 3979
{
  char   buff[1024];
monty@mysql.com's avatar
monty@mysql.com committed
3980
  uint length;
3981
  DBUG_ENTER("vprint_msg_to_log");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
3982

3983 3984
  /* "- 5" is because of print_buffer_to_nt_eventlog() */
  length= my_vsnprintf(buff, sizeof(buff) - 5, format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
3985
  print_buffer_to_file(level, buff);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
3986 3987

#ifdef __NT__
monty@mysql.com's avatar
monty@mysql.com committed
3988
  print_buffer_to_nt_eventlog(level, buff, length, sizeof(buff));
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
3989
#endif
3990

3991
  DBUG_RETURN(0);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
3992
}
3993
#endif /*EMBEDDED_LIBRARY*/
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
3994

3995

monty@mysql.com's avatar
monty@mysql.com committed
3996
void sql_print_error(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
3997 3998
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
3999
  DBUG_ENTER("sql_print_error");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4000

serg@serg.mylan's avatar
serg@serg.mylan committed
4001
  va_start(args, format);
4002
  error_log_print(ERROR_LEVEL, format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
4003
  va_end(args);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4004 4005 4006 4007 4008

  DBUG_VOID_RETURN;
}


monty@mysql.com's avatar
monty@mysql.com committed
4009
void sql_print_warning(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4010 4011
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
4012
  DBUG_ENTER("sql_print_warning");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4013

serg@serg.mylan's avatar
serg@serg.mylan committed
4014
  va_start(args, format);
4015
  error_log_print(WARNING_LEVEL, format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
4016
  va_end(args);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4017 4018 4019 4020 4021

  DBUG_VOID_RETURN;
}


monty@mysql.com's avatar
monty@mysql.com committed
4022
void sql_print_information(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4023 4024
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
4025
  DBUG_ENTER("sql_print_information");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4026

serg@serg.mylan's avatar
serg@serg.mylan committed
4027
  va_start(args, format);
4028
  error_log_print(INFORMATION_LEVEL, format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
4029
  va_end(args);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4030

guilhem@mysql.com's avatar
guilhem@mysql.com committed
4031 4032
  DBUG_VOID_RETURN;
}
4033

4034

4035 4036 4037 4038 4039 4040 4041 4042
/********* 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
4043 4044
  states - active, syncing, pool.
  there could be only one page in active or syncing states,
4045
  but many in pool - pool is fifo queue.
4046
  usual lifecycle of a page is pool->active->syncing->pool
4047
  "active" page - is a page where new xid's are logged.
4048 4049
  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.
4050
  when the sync is done the page is moved to a pool and an active page
4051 4052 4053 4054 4055 4056 4057 4058 4059 4060 4061 4062 4063
  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.
4064 4065 4066 4067
  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)

4068 4069
  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
4070 4071
  dirty - we don't sync removals to disk.
*/
4072 4073 4074 4075 4076

ulong tc_log_page_waits= 0;

#ifdef HAVE_MMAP

4077 4078
#define TC_LOG_HEADER_SIZE (sizeof(tc_log_magic)+1)

4079
static const char tc_log_magic[]={(char) 254, 0x23, 0x05, 0x74};
4080

4081
ulong opt_tc_log_size= TC_LOG_MIN_SIZE;
4082
ulong tc_log_max_pages_used=0, tc_log_page_size=0, tc_log_cur_pages_used=0;
4083 4084 4085 4086 4087 4088 4089

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

4090
  DBUG_ASSERT(total_ha_2pc > 1);
4091 4092
  DBUG_ASSERT(opt_name && opt_name[0]);

serg@serg.mylan's avatar
serg@serg.mylan committed
4093
  tc_log_page_size= my_getpagesize();
4094 4095 4096
  DBUG_ASSERT(TC_LOG_PAGE_SIZE % tc_log_page_size == 0);

  fn_format(logname,opt_name,mysql_data_home,"",MY_UNPACK_FILENAME);
4097
  if ((fd= my_open(logname, O_RDWR, MYF(0))) < 0)
4098
  {
4099 4100
    if (my_errno != ENOENT)
      goto err;
4101 4102
    if (using_heuristic_recover())
      return 1;
4103
    if ((fd= my_create(logname, O_RDWR, 0, MYF(MY_WME))) < 0)
4104 4105
      goto err;
    inited=1;
4106
    file_length= opt_tc_log_size;
4107 4108 4109 4110 4111
    if (my_chsize(fd, file_length, 0, MYF(MY_WME)))
      goto err;
  }
  else
  {
4112 4113
    inited= 1;
    crashed= TRUE;
serg@serg.mylan's avatar
serg@serg.mylan committed
4114
    sql_print_information("Recovering after a crash using %s", opt_name);
4115 4116 4117 4118 4119 4120
    if (tc_heuristic_recover)
    {
      sql_print_error("Cannot perform automatic crash recovery when "
                      "--tc-heuristic-recover is used");
      goto err;
    }
4121
    file_length= my_seek(fd, 0L, MY_SEEK_END, MYF(MY_WME+MY_FAE));
4122 4123 4124 4125
    if (file_length == MY_FILEPOS_ERROR || file_length % tc_log_page_size)
      goto err;
  }

4126
  data= (uchar *)my_mmap(0, (size_t)file_length, PROT_READ|PROT_WRITE,
4127 4128 4129 4130 4131 4132 4133 4134
                        MAP_NOSYNC|MAP_SHARED, fd, 0);
  if (data == MAP_FAILED)
  {
    my_errno=errno;
    goto err;
  }
  inited=2;

4135
  npages=(uint)file_length/tc_log_page_size;
4136 4137 4138 4139 4140 4141 4142 4143 4144 4145 4146 4147 4148 4149 4150 4151 4152
  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);
4153
  pages[0].start=pages[0].end-pages[0].size;
4154 4155 4156
  pages[npages-1].next=0;
  inited=4;

4157
  if (crashed && recover())
4158
      goto err;
4159

4160
  memcpy(data, tc_log_magic, sizeof(tc_log_magic));
4161
  data[sizeof(tc_log_magic)]= (uchar)total_ha_2pc;
4162 4163 4164 4165 4166 4167 4168 4169 4170 4171 4172
  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;

4173
  syncing= 0;
4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185
  active=pages;
  pool=pages+1;
  pool_last=pages+npages-1;

  return 0;

err:
  close();
  return 1;
}

/*
4186 4187
  there is no active page, let's got one from the pool

4188 4189 4190 4191 4192 4193 4194 4195 4196 4197 4198 4199
  two strategies here:
  1. take the first from the pool
  2. if there're waiters - take the one with the most free space

  TODO page merging. try to allocate adjacent page first,
  so that they can be flushed both in one sync
*/
void TC_LOG_MMAP::get_active_from_pool()
{
  PAGE **p, **best_p=0;
  int best_free;

4200 4201 4202
  if (syncing)
    pthread_mutex_lock(&LOCK_pool);

4203 4204 4205
  do
  {
    best_p= p= &pool;
4206 4207
    if ((*p)->waiters == 0) // can the first page be used ?
      break;                // yes - take it.
4208

4209
    best_free=0;            // no - trying second strategy
4210 4211 4212 4213 4214 4215 4216 4217 4218 4219 4220 4221
    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;
4222
  if (active->free == active->size) // we've chosen an empty page
4223 4224 4225 4226 4227
  {
    tc_log_cur_pages_used++;
    set_if_bigger(tc_log_max_pages_used, tc_log_cur_pages_used);
  }

4228
  if ((*best_p)->next)              // unlink the page from the pool
4229 4230 4231
    *best_p=(*best_p)->next;
  else
    pool_last=*best_p;
4232 4233 4234

  if (syncing)
    pthread_mutex_unlock(&LOCK_pool);
4235 4236 4237 4238 4239 4240 4241 4242 4243 4244 4245
}

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);
4246
  return 1; // always return 1
4247 4248 4249
}

/*
4250 4251 4252 4253 4254 4255 4256 4257 4258 4259 4260 4261
  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

  RETURN
         0  - error
  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.
4262 4263 4264 4265 4266 4267 4268 4269 4270 4271
*/

int TC_LOG_MMAP::log(THD *thd, my_xid xid)
{
  int err;
  PAGE *p;
  ulong cookie;

  pthread_mutex_lock(&LOCK_active);

4272 4273 4274 4275 4276 4277 4278
  /*
    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.
  */
4279 4280 4281
  while (unlikely(active && active->free == 0))
    pthread_cond_wait(&COND_active, &LOCK_active);

4282
  /* no active page ? take one from the pool */
4283 4284 4285 4286 4287 4288
  if (active == 0)
    get_active_from_pool();

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

4289
  /* searching for an empty slot */
4290 4291 4292 4293 4294 4295
  while (*p->ptr)
  {
    p->ptr++;
    DBUG_ASSERT(p->ptr < p->end);               // because p->free > 0
  }

4296 4297
  /* found! store xid there and mark the page dirty */
  cookie= (ulong)((uchar *)p->ptr - data);      // can never be zero
4298 4299 4300 4301 4302
  *p->ptr++= xid;
  p->free--;
  p->state= DIRTY;

  /* to sync or not to sync - this is the question */
4303
  pthread_mutex_unlock(&LOCK_active);
4304 4305 4306
  pthread_mutex_lock(&LOCK_sync);
  pthread_mutex_unlock(&p->lock);

4307
  if (syncing)
4308 4309
  {                                          // somebody's syncing. let's wait
    p->waiters++;
4310
    /*
monty@mishka.local's avatar
monty@mishka.local committed
4311
      note - it must be while (), not do ... while () here
4312 4313 4314
      as p->state may be not DIRTY when we come here
    */
    while (p->state == DIRTY && syncing)
4315 4316 4317 4318 4319 4320 4321 4322 4323 4324 4325
      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
4326 4327 4328 4329
  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
4330 4331 4332 4333 4334 4335 4336 4337 4338 4339 4340 4341 4342
  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;

4343
  DBUG_ASSERT(syncing != active);
4344 4345 4346 4347 4348

  /*
    sit down and relax - this can take a while...
    note - no locks are held at this point
  */
4349
  err= my_msync(fd, syncing->start, 1, MS_SYNC);
4350

4351
  /* page is synced. let's move it to the pool */
4352
  pthread_mutex_lock(&LOCK_pool);
4353 4354 4355 4356 4357
  pool_last->next=syncing;
  pool_last=syncing;
  syncing->next=0;
  syncing->state= err ? ERROR : POOL;
  pthread_cond_broadcast(&syncing->cond);    // signal "sync done"
4358 4359 4360
  pthread_cond_signal(&COND_pool);           // in case somebody's waiting
  pthread_mutex_unlock(&LOCK_pool);

4361
  /* marking 'syncing' slot free */
4362
  pthread_mutex_lock(&LOCK_sync);
4363
  syncing=0;
4364 4365 4366 4367 4368
  pthread_cond_signal(&active->cond);        // wake up a new syncer
  pthread_mutex_unlock(&LOCK_sync);
  return err;
}

4369 4370 4371 4372
/*
  erase xid from the page, update page free space counters/pointers.
  cookie points directly to the memory where xid was logged
*/
4373 4374 4375 4376 4377 4378 4379 4380 4381 4382 4383 4384 4385
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);
4386
  if (p->free == p->size)               // the page is completely empty
4387 4388
    statistic_decrement(tc_log_cur_pages_used, &LOCK_status);
  if (p->waiters == 0)                 // the page is in pool and ready to rock
4389
    pthread_cond_signal(&COND_pool);   // ping ... for overflow()
4390 4391 4392 4393 4394
  pthread_mutex_unlock(&p->lock);
}

void TC_LOG_MMAP::close()
{
serg@serg.mylan's avatar
serg@serg.mylan committed
4395
  uint i;
4396 4397 4398 4399 4400 4401 4402 4403 4404
  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
4405
    for (i=0; i < npages; i++)
4406 4407 4408 4409 4410 4411 4412 4413 4414
    {
      if (pages[i].ptr == 0)
        break;
      pthread_mutex_destroy(&pages[i].lock);
      pthread_cond_destroy(&pages[i].cond);
    }
  case 3:
    my_free((gptr)pages, MYF(0));
  case 2:
kent@mysql.com's avatar
kent@mysql.com committed
4415
    my_munmap((byte*)data, (size_t)file_length);
4416 4417 4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430 4431 4432 4433 4434 4435 4436 4437 4438 4439 4440
  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)
  {
4441
    sql_print_error("Recovery failed! You must enable "
4442 4443 4444
                    "exactly %d storage engines that support "
                    "two-phase commit protocol",
                    data[sizeof(tc_log_magic)]);
4445 4446 4447 4448 4449 4450 4451 4452 4453 4454 4455 4456 4457 4458 4459 4460 4461 4462
    goto err1;
  }

  if (hash_init(&xids, &my_charset_bin, tc_log_page_size/3, 0,
            sizeof(my_xid), 0, 0, MYF(0)))
    goto err1;

  for ( ; p < end_p ; p++)
  {
    for (my_xid *x=p->start; x < p->end; x++)
      if (*x && my_hash_insert(&xids, (byte *)x))
        goto err2; // OOM
  }

  if (ha_recover(&xids))
    goto err2;

  hash_free(&xids);
4463
  bzero(data, (size_t)file_length);
4464 4465 4466 4467 4468 4469 4470 4471 4472 4473 4474
  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
4475 4476 4477 4478 4479
#endif

TC_LOG *tc_log;
TC_LOG_DUMMY tc_log_dummy;
TC_LOG_MMAP  tc_log_mmap;
4480

4481 4482 4483 4484 4485 4486 4487 4488 4489 4490 4491 4492
/*
  Perform heuristic recovery, if --tc-heuristic-recover was used

  RETURN VALUE
    0	no heuristic recovery was requested
    1   heuristic recovery was performed

  NOTE
    no matter whether heuristic recovery was successful or not
    mysqld must exit. So, return value is the same in both cases.
*/

4493 4494 4495 4496 4497 4498 4499 4500 4501 4502 4503 4504 4505
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 ******/
4506
#define TC_LOG_BINLOG MYSQL_BIN_LOG
4507 4508 4509 4510 4511 4512 4513 4514

/*
  TODO keep in-memory list of prepared transactions
  (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!
*/

4515
int TC_LOG_BINLOG::open(const char *opt_name)
4516
{
4517 4518
  LOG_INFO log_info;
  int      error= 1;
4519 4520

  DBUG_ASSERT(total_ha_2pc > 1);
4521
  DBUG_ASSERT(opt_name && opt_name[0]);
4522 4523 4524 4525

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

4526 4527 4528 4529 4530 4531 4532
  if (!my_b_inited(&index_file))
  {
    /* There was a failure to open the index file, can't open the binlog */
    cleanup();
    return 1;
  }

4533
  if (using_heuristic_recover())
4534 4535 4536 4537
  {
    /* generate a new binlog to mask a corrupted one */
    open(opt_name, LOG_BIN, 0, WRITE_CACHE, 0, max_binlog_size, 0);
    cleanup();
4538
    return 1;
4539
  }
4540

4541
  if ((error= find_log_pos(&log_info, NullS, 1)))
4542
  {
4543 4544 4545 4546 4547
    if (error != LOG_INFO_EOF)
      sql_print_error("find_log_pos() failed (error: %d)", error);
    else
      error= 0;
    goto err;
4548
  }
4549 4550 4551 4552 4553

  {
    const char *errmsg;
    IO_CACHE    log;
    File        file;
4554 4555
    Log_event  *ev=0;
    Format_description_log_event fdle(BINLOG_VERSION);
4556
    char        log_name[FN_REFLEN];
4557 4558 4559 4560

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

4561
    do
4562
    {
4563 4564 4565 4566 4567 4568 4569
      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;
4570
    }
4571

4572
    if ((file= open_binlog(&log, log_name, &errmsg)) < 0)
4573 4574 4575 4576 4577
    {
      sql_print_error("%s", errmsg);
      goto err;
    }

4578 4579 4580
    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
4581 4582
    {
      sql_print_information("Recovering after a crash using %s", opt_name);
4583
      error= recover(&log, (Format_description_log_event *)ev);
serg@serg.mylan's avatar
serg@serg.mylan committed
4584
    }
4585 4586
    else
      error=0;
4587 4588 4589 4590 4591 4592 4593 4594 4595 4596

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

    if (error)
      goto err;
  }

err:
4597
  return error;
4598 4599
}

4600 4601
/* this is called on shutdown, after ha_panic */
void TC_LOG_BINLOG::close()
4602 4603 4604 4605 4606 4607
{
  DBUG_ASSERT(prepared_xids==0);
  pthread_mutex_destroy(&LOCK_prep_xids);
  pthread_cond_destroy (&COND_prep_xids);
}

4608 4609 4610 4611 4612 4613 4614 4615
/*
  TODO group commit

  RETURN
         0  - error
         1  - success
*/
int TC_LOG_BINLOG::log(THD *thd, my_xid xid)
4616
{
4617
  DBUG_ENTER("TC_LOG_BINLOG::log");
4618
  Xid_log_event xle(thd, xid);
4619
  binlog_trx_data *trx_data=
4620
    (binlog_trx_data*) thd->ha_data[binlog_hton->slot];
4621
  DBUG_RETURN(!binlog_end_trans(thd, trx_data, &xle));  // invert return value
4622 4623
}

4624
void TC_LOG_BINLOG::unlog(ulong cookie, my_xid xid)
4625
{
4626 4627
  pthread_mutex_lock(&LOCK_prep_xids);
  if (--prepared_xids == 0)
4628
    pthread_cond_signal(&COND_prep_xids);
4629
  pthread_mutex_unlock(&LOCK_prep_xids);
4630
  rotate_and_purge(0);     // as ::write() did not rotate
4631 4632
}

4633
int TC_LOG_BINLOG::recover(IO_CACHE *log, Format_description_log_event *fdle)
4634 4635 4636 4637 4638 4639
{
  Log_event  *ev;
  HASH xids;
  MEM_ROOT mem_root;

  if (! fdle->is_valid() ||
4640
      hash_init(&xids, &my_charset_bin, TC_LOG_PAGE_SIZE/3, 0,
4641 4642 4643
            sizeof(my_xid), 0, 0, MYF(0)))
    goto err1;

4644
  init_alloc_root(&mem_root, TC_LOG_PAGE_SIZE, TC_LOG_PAGE_SIZE);
4645

4646 4647
  fdle->flags&= ~LOG_EVENT_BINLOG_IN_USE_F; // abort on the first error

4648 4649 4650 4651 4652
  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;
serg@serg.mylan's avatar
serg@serg.mylan committed
4653 4654
      byte *x=(byte *)memdup_root(&mem_root, (char *)& xev->xid,
                                  sizeof(xev->xid));
4655 4656 4657 4658 4659 4660 4661 4662 4663 4664 4665 4666 4667 4668 4669 4670 4671 4672 4673 4674 4675 4676 4677 4678 4679
      if (! x)
        goto err2;
      my_hash_insert(&xids, x);
    }
    delete ev;
  }

  if (ha_recover(&xids))
    goto err2;

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

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

4680
struct st_mysql_storage_engine binlog_storage_engine=
4681
{ MYSQL_HANDLERTON_INTERFACE_VERSION, binlog_hton };
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
4682 4683 4684 4685

mysql_declare_plugin(binlog)
{
  MYSQL_STORAGE_ENGINE_PLUGIN,
4686 4687
  &binlog_storage_engine,
  "binlog",
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
4688
  "MySQL AB",
4689 4690
  "This is a pseudo storage engine to represent the binlog in a transaction",
  binlog_init, /* Plugin Init */
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
4691 4692
  NULL, /* Plugin Deinit */
  0x0100 /* 1.0 */,
4693 4694 4695
  NULL,                       /* status variables                */
  NULL,                       /* system variables                */
  NULL                        /* config options                  */
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
4696 4697
}
mysql_declare_plugin_end;