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

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

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

bk@work.mysql.com's avatar
bk@work.mysql.com committed
12 13 14 15 16 17
   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 */
18
/* 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
19 20

#include "mysql_priv.h"
21
#include "sql_repl.h"
lars@mysql.com's avatar
lars@mysql.com committed
22
#include "rpl_filter.h"
23
#include "rpl_rli.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
/* max size of the log message */
#define MAX_LOG_BUFFER_SIZE 1024
#define MAX_USER_HOST_SIZE 512
#define MAX_TIME_SIZE 32
39
#define MY_OFF_T_UNDEF (~(my_off_t)0UL)
40

41
#define FLAGSTR(V,F) ((V)&(F)?#F" ":"")
42 43 44

LOGGER logger;

45
MYSQL_BIN_LOG mysql_bin_log;
46
ulong sync_binlog_counter= 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
47

48 49
static Muted_query_log_event invisible_commit;

bk@work.mysql.com's avatar
bk@work.mysql.com committed
50 51
static bool test_if_number(const char *str,
			   long *res, bool allow_wildcards);
52 53 54 55 56 57 58
static int binlog_init(void *p);
static int binlog_close_connection(handlerton *hton, THD *thd);
static int binlog_savepoint_set(handlerton *hton, THD *thd, void *sv);
static int binlog_savepoint_rollback(handlerton *hton, THD *thd, void *sv);
static int binlog_commit(handlerton *hton, THD *thd, bool all);
static int binlog_rollback(handlerton *hton, THD *thd, bool all);
static int binlog_prepare(handlerton *hton, THD *thd, bool all);
59

60 61 62 63 64 65 66 67
sql_print_message_func sql_print_message_handlers[3] =
{
  sql_print_information,
  sql_print_warning,
  sql_print_error
};


68 69
char *make_default_log_name(char *buff,const char* log_ext)
{
70
  strmake(buff, pidfile_name, FN_REFLEN-5);
71
  return fn_format(buff, buff, mysql_data_home, log_ext,
72
                   MYF(MY_UNPACK_FILENAME|MY_REPLACE_EXT));
73 74
}

75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109
/*
  Helper class to hold a mutex for the duration of the
  block.

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

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

private:
  pthread_mutex_t *m_mutex;

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

110
/*
111
  Helper class to store binary log transaction data.
112
*/
113 114 115
class binlog_trx_data {
public:
  binlog_trx_data()
116
    : m_pending(0), before_stmt_pos(MY_OFF_T_UNDEF)
117 118 119 120 121 122 123 124 125 126 127 128 129 130
  {
    trans_log.end_of_file= max_binlog_cache_size;
  }

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

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

131 132
  bool empty() const
  {
133
    return pending() == NULL && my_b_tell(&trans_log) == 0;
134
  }
135 136 137 138 139 140 141

  /*
    Truncate the transaction cache to a certain position. This
    includes deleting the pending event.
   */
  void truncate(my_off_t pos)
  {
142 143
    DBUG_PRINT("info", ("truncating to position %lu", (ulong) pos));
    DBUG_PRINT("info", ("before_stmt_pos=%lu", (ulong) pos));
144 145 146
    delete pending();
    set_pending(0);
    reinit_io_cache(&trans_log, WRITE_CACHE, pos, 0, 0);
147 148
    if (pos < before_stmt_pos)
      before_stmt_pos= MY_OFF_T_UNDEF;
149 150 151 152 153 154 155 156 157
  }

  /*
    Reset the entire contents of the transaction cache, emptying it
    completely.
   */
  void reset() {
    if (!empty())
      truncate(0);
158
    before_stmt_pos= MY_OFF_T_UNDEF;
159 160 161 162 163 164 165 166 167 168 169 170 171
    trans_log.end_of_file= max_binlog_cache_size;
  }

  Rows_log_event *pending() const
  {
    return m_pending;
  }

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

172
  IO_CACHE trans_log;                         // The transaction cache
173 174 175 176 177 178 179 180 181 182 183 184 185

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

public:
  /*
    Binlog position before the start of the current statement.
  */
  my_off_t before_stmt_pos;
186 187
};

188
handlerton *binlog_hton;
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
/* Check if a given table is opened log table */
int check_if_log_table(uint db_len, const char *db, uint table_name_len,
                       const char *table_name, uint check_if_opened)
{
  if (db_len == 5 &&
      !(lower_case_table_names ?
        my_strcasecmp(system_charset_info, db, "mysql") :
        strcmp(db, "mysql")))
  {
    if (table_name_len == 11 && !(lower_case_table_names ?
                                  my_strcasecmp(system_charset_info,
                                                table_name, "general_log") :
                                  strcmp(table_name, "general_log")) &&
        (!check_if_opened || logger.is_log_table_enabled(QUERY_LOG_GENERAL)))
      return QUERY_LOG_GENERAL;
    else
      if (table_name_len == 8 && !(lower_case_table_names ?
        my_strcasecmp(system_charset_info, table_name, "slow_log") :
        strcmp(table_name, "slow_log")) &&
          (!check_if_opened ||logger.is_log_table_enabled(QUERY_LOG_SLOW)))
        return QUERY_LOG_SLOW;
  }
  return 0;
}

216

217 218 219 220 221 222
/*
  Open log table of a given type (general or slow log)

  SYNOPSIS
    open_log_table()

223 224
    log_table_type   type of the log table to open: QUERY_LOG_GENERAL
                     or QUERY_LOG_SLOW
225 226 227 228 229 230 231 232 233 234 235 236 237 238

  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
*/

239
bool Log_to_csv_event_handler::open_log_table(uint log_table_type)
240 241 242 243 244 245
{
  THD *log_thd, *curr= current_thd;
  TABLE_LIST *table;
  bool error= FALSE;
  DBUG_ENTER("open_log_table");

246
  switch (log_table_type) {
247
  case QUERY_LOG_GENERAL:
248 249 250 251 252 253 254
    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;
255
  case QUERY_LOG_SLOW:
256 257 258 259 260 261 262
    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:
263
    assert(0);                                  // Impossible
264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298
  }

  /*
    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;

299 300
  lex_start(log_thd);
  log_thd->clear_error();
301 302 303 304 305
  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
306 307
  {
    table->table->use_all_columns();
308
    table->table->locked_by_logger= TRUE;
309
    table->table->no_replicate= TRUE;
310 311 312

    /* Honor next number columns if present */
    table->table->next_number_field= table->table->found_next_number_field;
313
  }
314 315 316 317 318 319 320 321 322
  /* restore thread settings */
  if (curr)
    curr->store_globals();
  else
  {
    my_pthread_setspecific_ptr(THR_THD,  0);
    my_pthread_setspecific_ptr(THR_MALLOC, 0);
  }

323 324 325 326 327 328
  /*
    After a log table was opened, we should clear privileged thread
    flag (which allows locking of a log table by a special thread, usually
    the one who closed log tables temporarily).
  */
  privileged_thread= 0;
329 330 331 332 333 334 335 336 337 338 339
  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;
340
  general_log.table= 0;
341 342 343 344 345

  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;
346
  slow_log.table= 0;
347 348
  /* no privileged thread exists at the moment */
  privileged_thread= 0;
349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374
}


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()

375 376
    log_table_type   type of the log table to open: QUERY_LOG_GENERAL
                     or QUERY_LOG_SLOW
377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392

  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
*/

393
bool Log_to_csv_event_handler::reopen_log_table(uint log_table_type)
394 395 396 397
{
  /* don't open the log table, if it wasn't enabled during startup */
  if (!logger.is_log_tables_initialized)
    return FALSE;
398
  return open_log_table(log_table_type);
399 400
}

401

402 403
void Log_to_csv_event_handler::cleanup()
{
404 405 406 407
  if (opt_log)
    close_log_table(QUERY_LOG_GENERAL, FALSE);
  if (opt_slow_log)
    close_log_table(QUERY_LOG_SLOW, FALSE);
408 409 410 411 412 413 414 415 416
  logger.is_log_tables_initialized= FALSE;
}

/* log event handlers */

/*
  Log command to the general log table

  SYNOPSIS
417
    log_general()
418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441

    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,
442 443
              const char *sql_text, uint sql_text_len,
              CHARSET_INFO *client_cs)
444 445
{
  TABLE *table= general_log.table;
malff/marcsql@weblab.(none)'s avatar
malff/marcsql@weblab.(none) committed
446
  uint field_index;
447

448 449 450 451 452 453 454 455 456 457 458 459
  /*
    "INSERT INTO general_log" can generate warning sometimes.
    Let's reset warnings from previous queries,
    otherwise warning list can grow too much,
    so thd->query gets spoiled as some point in time,
    and mysql_parse() receives a broken query.
    QQ: this problem needs to be studied in more details.
    Probably it's better to suppress warnings in logging INSERTs at all.
    Comment this line and run "cast.test" to see what's happening:
  */
  mysql_reset_errors(table->in_use, 1);

460 461 462 463
  /* below should never happen */
  if (unlikely(!logger.is_log_tables_initialized))
    return FALSE;

464 465 466 467 468 469
  /*
    NOTE: we do not call restore_record() here, as all fields are
    filled by the Logger (=> no need to load default ones).
  */

  /* Set current time. Required for CURRENT_TIMESTAMP to work */
470
  general_log_thd->start_time= event_time;
471 472 473 474 475

  /*
    We do not set a value for table->field[0], as it will use
    default value (which is CURRENT_TIMESTAMP).
  */
476

477 478 479 480 481 482 483 484 485 486 487 488 489 490
  /* check that all columns exist */
  if (!table->field[1] || !table->field[2] || !table->field[3] ||
      !table->field[4] || !table->field[5])
    goto err;

  /* do a write */
  if (table->field[1]->store(user_host, user_host_len, client_cs) ||
      table->field[2]->store((longlong) thread_id, TRUE) ||
      table->field[3]->store((longlong) server_id, TRUE) ||
      table->field[4]->store(command_type, command_type_len, client_cs) ||
      table->field[5]->store(sql_text, sql_text_len, client_cs))
    goto err;

  /* mark tables as not null */
491 492 493 494 495
  table->field[1]->set_notnull();
  table->field[2]->set_notnull();
  table->field[3]->set_notnull();
  table->field[4]->set_notnull();
  table->field[5]->set_notnull();
496

497 498 499 500 501 502
  /* Set any extra columns to their default values */
  for (field_index= 6 ; field_index < table->s->fields ; field_index++)
  {
    table->field[field_index]->set_default();
  }

503 504 505
  /* log table entries are not replicated at the moment */
  tmp_disable_binlog(current_thd);

506 507 508 509 510
  table->file->ha_write_row(table->record[0]);

  reenable_binlog(current_thd);

  return FALSE;
511 512
err:
  return TRUE;
513 514 515 516 517 518 519
}


/*
  Log a query to the slow log table

  SYNOPSIS
520
    log_slow()
521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552
    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;
553
  CHARSET_INFO *client_cs= thd->variables.character_set_client;
554

555
  DBUG_ENTER("log_slow");
556 557 558 559 560 561 562

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

  /*
     Set start time for CURRENT_TIMESTAMP to the start of the query.
563
     This will be default value for the field[0]
564 565
  */
  slow_log_thd->start_time= query_start_arg;
566
  restore_record(table, s->default_values);    // Get empty record
567

568 569 570 571
  /*
    We do not set a value for table->field[0], as it will use
    default value.
  */
572

573 574 575 576 577 578
  if (!table->field[1] || !table->field[2] || !table->field[3] ||
      !table->field[4] || !table->field[5] || !table->field[6] ||
      !table->field[7] || !table->field[8] || !table->field[9] ||
      !table->field[10])
    goto err;

579
  /* store the value */
580 581
  if (table->field[1]->store(user_host, user_host_len, client_cs))
    goto err;
582 583 584

  if (query_start_arg)
  {
585 586 587 588 589
    /*
      A TIME field can not hold the full longlong range; query_time or
      lock_time may be truncated without warning here, if greater than
      839 hours (~35 days)
    */
590
    MYSQL_TIME t;
591 592
    t.neg= 0;

593
    /* fill in query_time field */
594 595
    calc_time_from_sec(&t, (long) min(query_time, (longlong) TIME_MAX_VALUE_SECONDS), 0);
    if (table->field[2]->store_time(&t, MYSQL_TIMESTAMP_TIME))
596
      goto err;
597
    /* lock_time */
598 599
    calc_time_from_sec(&t, (long) min(lock_time, (longlong) TIME_MAX_VALUE_SECONDS), 0);
    if (table->field[3]->store_time(&t, MYSQL_TIMESTAMP_TIME))
600
      goto err;
601
    /* rows_sent */
602 603
    if (table->field[4]->store((longlong) thd->sent_row_count, TRUE))
      goto err;
604
    /* rows_examined */
605 606
    if (table->field[5]->store((longlong) thd->examined_row_count, TRUE))
      goto err;
607 608 609 610 611 612 613 614 615
  }
  else
  {
    table->field[2]->set_null();
    table->field[3]->set_null();
    table->field[4]->set_null();
    table->field[5]->set_null();
  }

616
  /* fill database field */
617
  if (thd->db)
618
  {
619 620
    if (table->field[6]->store(thd->db, thd->db_length, client_cs))
      goto err;
621 622
    table->field[6]->set_notnull();
  }
623

624
  if (thd->stmt_depends_on_first_successful_insert_id_in_prev_stmt)
625
  {
626 627 628 629 630
    if (table->
        field[7]->store((longlong)
                        thd->first_successful_insert_id_in_prev_stmt_for_binlog,
                        TRUE))
      goto err;
631 632
    table->field[7]->set_notnull();
  }
633

634 635 636 637 638 639 640
  /*
    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)
641
  {
642 643 644 645
    if (table->
        field[8]->store((longlong)
          thd->auto_inc_intervals_in_cur_stmt_for_binlog.minimum(), TRUE))
      goto err;
646 647
    table->field[8]->set_notnull();
  }
648

649 650
  if (table->field[9]->store((longlong) server_id, TRUE))
    goto err;
651
  table->field[9]->set_notnull();
652 653

  /* sql_text */
654 655 656 657 658
  if (table->field[10]->store(sql_text,sql_text_len, client_cs))
    goto err;

  /* log table entries are not replicated at the moment */
  tmp_disable_binlog(current_thd);
659 660 661 662 663 664 665

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

  reenable_binlog(current_thd);

  DBUG_RETURN(0);
666 667
err:
  DBUG_RETURN(1);
668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715
}

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,
716 717
              const char *sql_text, uint sql_text_len,
              CHARSET_INFO *client_cs)
718 719 720 721 722 723 724 725 726 727 728 729
{
  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)
730
      mysql_slow_log.open_slow_log(sys_var_slow_log_path.value);
731 732

    if (opt_log)
733
      mysql_log.open_query_log(sys_var_general_log_path.value);
734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750

    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 */
751
  if (opt_log)
752
    mysql_log.reopen_file();
753
  if (opt_slow_log)
754
    mysql_slow_log.reopen_file();
755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786
}

/*
  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
787
void LOGGER::cleanup_base()
788 789 790 791
{
  DBUG_ASSERT(inited == 1);
  (void) pthread_mutex_destroy(&LOCK_logger);
  if (table_log_handler)
petr@mysql.com's avatar
petr@mysql.com committed
792
  {
793
    table_log_handler->cleanup();
petr@mysql.com's avatar
petr@mysql.com committed
794 795
    delete table_log_handler;
  }
796 797 798 799 800
  if (file_log_handler)
    file_log_handler->cleanup();
}


petr@mysql.com's avatar
petr@mysql.com committed
801 802 803 804 805 806 807 808
void LOGGER::cleanup_end()
{
  DBUG_ASSERT(inited == 1);
  if (file_log_handler)
    delete file_log_handler;
}


809
void LOGGER::close_log_table(uint log_table_type, bool lock_in_use)
810
{
811
  table_log_handler->close_log_table(log_table_type, lock_in_use);
812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832
}


/*
  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 */
833
  init_error_log(LOG_FILE);
834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850

  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;
}


851
bool LOGGER::reopen_log_table(uint log_table_type)
852
{
853
  return table_log_handler->reopen_log_table(log_table_type);
854 855
}

856
bool LOGGER::reopen_log_tables()
857
{
858 859 860 861 862 863 864 865 866
    /*
      we use | and not || here, to ensure that both reopen_log_table
      are called, even if the first one fails
    */
    if ((opt_slow_log && logger.reopen_log_table(QUERY_LOG_SLOW)) |
        (opt_log && logger.reopen_log_table(QUERY_LOG_GENERAL)))
      return TRUE;
    return FALSE;
}
867 868


869 870 871 872
void LOGGER::tmp_close_log_tables(THD *thd)
{
  table_log_handler->tmp_close_log_tables(thd);
}
873

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

878
  /*
879 880
    Now we lock logger, as nobody should be able to use logging routines while
    log tables are closed
881 882
  */
  logger.lock();
883 884
  if (logger.is_log_tables_initialized)
    table_log_handler->tmp_close_log_tables(thd); // the locking happens here
885 886 887

  /* reopen log files */
  file_log_handler->flush();
888

889
  /* reopen tables in the case they were enabled */
890
  if (logger.is_log_tables_initialized)
891 892 893 894
  {
    if (reopen_log_tables())
      rc= TRUE;
  }
895 896
  /* end of log flush */
  logger.unlock();
897
  return rc;
898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924
}


/*
  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];

925
  time_t current_time;
926
  Security_context *sctx= thd->security_ctx;
927
  uint user_host_len= 0;
928 929 930 931 932 933 934 935 936 937
  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);

938 939
    /* do not log slow queries from replication threads */
    if (thd->slave_thread)
940 941 942
      return 0;

    lock();
943 944 945 946 947
    if (!opt_slow_log)
    {
      unlock();
      return 0;
    }
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

    /* 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();
1015 1016 1017 1018 1019
    if (!opt_log)
    {
      unlock();
      return 0;
    }
1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041
    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,
1042 1043
                   message_buff, message_buff_len,
                   thd->variables.character_set_client) || error;
1044 1045 1046 1047 1048
    unlock();
  }
  return error;
}

1049
void LOGGER::init_error_log(uint error_log_printer)
1050
{
1051 1052
  if (error_log_printer & LOG_NONE)
  {
1053
    error_log_handler_list[0]= 0;
1054 1055 1056 1057 1058
    return;
  }

  switch (error_log_printer) {
  case LOG_FILE:
1059 1060 1061 1062
    error_log_handler_list[0]= file_log_handler;
    error_log_handler_list[1]= 0;
    break;
    /* these two are disabled for now */
1063
  case LOG_TABLE:
1064 1065
    DBUG_ASSERT(0);
    break;
1066
  case LOG_TABLE|LOG_FILE:
1067 1068 1069 1070 1071
    DBUG_ASSERT(0);
    break;
  }
}

1072
void LOGGER::init_slow_log(uint slow_log_printer)
1073
{
1074 1075
  if (slow_log_printer & LOG_NONE)
  {
1076
    slow_log_handler_list[0]= 0;
1077 1078 1079 1080 1081
    return;
  }

  switch (slow_log_printer) {
  case LOG_FILE:
1082 1083 1084
    slow_log_handler_list[0]= file_log_handler;
    slow_log_handler_list[1]= 0;
    break;
1085
  case LOG_TABLE:
1086 1087 1088
    slow_log_handler_list[0]= table_log_handler;
    slow_log_handler_list[1]= 0;
    break;
1089
  case LOG_TABLE|LOG_FILE:
1090 1091 1092 1093 1094 1095 1096
    slow_log_handler_list[0]= file_log_handler;
    slow_log_handler_list[1]= table_log_handler;
    slow_log_handler_list[2]= 0;
    break;
  }
}

1097
void LOGGER::init_general_log(uint general_log_printer)
1098
{
1099 1100
  if (general_log_printer & LOG_NONE)
  {
1101
    general_log_handler_list[0]= 0;
1102 1103 1104 1105 1106
    return;
  }

  switch (general_log_printer) {
  case LOG_FILE:
1107 1108 1109
    general_log_handler_list[0]= file_log_handler;
    general_log_handler_list[1]= 0;
    break;
1110
  case LOG_TABLE:
1111 1112 1113
    general_log_handler_list[0]= table_log_handler;
    general_log_handler_list[1]= 0;
    break;
1114
  case LOG_TABLE|LOG_FILE:
1115 1116 1117 1118 1119 1120 1121 1122
    general_log_handler_list[0]= file_log_handler;
    general_log_handler_list[1]= table_log_handler;
    general_log_handler_list[2]= 0;
    break;
  }
}


1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133
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()->
1134
        open_slow_log(sys_var_slow_log_path.value);
1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179
      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:
1180
    assert(0);                                  // Impossible
1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204
  }

  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();
}


1205 1206 1207 1208 1209 1210 1211
/*
  Close log tables temporarily. The thread which closed
  them this way can lock them in any mode it needs.
  NOTE: one should call logger.lock() before entering this
  function.
*/
void Log_to_csv_event_handler::tmp_close_log_tables(THD *thd)
1212
{
1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229
  TABLE_LIST close_slow_log, close_general_log;

  /* fill lists, we will need to perform operations on 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;

  privileged_thread= thd;

1230
  VOID(pthread_mutex_lock(&LOCK_open));
1231 1232 1233 1234 1235 1236 1237
  /*
    NOTE: in fact, the first parameter used in query_cache_invalidate3()
    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.
  */
1238 1239 1240
  if (opt_log)
  {
    close_log_table(QUERY_LOG_GENERAL, TRUE);
1241
    query_cache_invalidate3(general_log_thd, &close_general_log, 0);
1242 1243 1244 1245
  }
  if (opt_slow_log)
  {
    close_log_table(QUERY_LOG_SLOW, TRUE);
1246
    query_cache_invalidate3(general_log_thd, &close_slow_log, 0);
1247
  }
1248 1249 1250 1251 1252 1253
  VOID(pthread_mutex_unlock(&LOCK_open));
}

/* the parameters are unused for the log tables */
bool Log_to_csv_event_handler::init()
{
1254 1255 1256 1257 1258 1259 1260 1261
  /*
    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;
1262 1263
}

1264 1265 1266
int LOGGER::set_handlers(uint error_log_printer,
                         uint slow_log_printer,
                         uint general_log_printer)
1267 1268
{
  /* error log table is not supported yet */
1269
  DBUG_ASSERT(error_log_printer < LOG_TABLE);
1270 1271 1272

  lock();

1273
  if ((slow_log_printer & LOG_TABLE || general_log_printer & LOG_TABLE) &&
1274 1275
      !is_log_tables_initialized)
  {
1276 1277
    slow_log_printer= (slow_log_printer & ~LOG_TABLE) | LOG_FILE;
    general_log_printer= (general_log_printer & ~LOG_TABLE) | LOG_FILE;
1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298

    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()

1299 1300 1301
    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.
1302 1303 1304 1305 1306 1307 1308 1309 1310

  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::
1311
  close_log_table(uint log_table_type, bool lock_in_use)
1312 1313 1314 1315
{
  THD *log_thd, *curr= current_thd;
  TABLE_LIST *table;

1316 1317 1318
  if (!logger.is_log_table_enabled(log_table_type))
    return;                                     /* do nothing */

1319
  switch (log_table_type) {
1320
  case QUERY_LOG_GENERAL:
1321 1322 1323
    log_thd= general_log_thd;
    table= &general_log;
    break;
1324
  case QUERY_LOG_SLOW:
1325 1326 1327 1328
    log_thd= slow_log_thd;
    table= &slow_log;
    break;
  default:
1329
    assert(0);                                  // Impossible
1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343
  }

  /*
    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();
1344
  table->table->file->ha_release_auto_increment();
1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358
  /* 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);
  }
}


1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378
 /*
  Save position of binary log transaction cache.

  SYNPOSIS
    binlog_trans_log_savepos()

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

  DESCRIPTION

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

static void
binlog_trans_log_savepos(THD *thd, my_off_t *pos)
{
  DBUG_ENTER("binlog_trans_log_savepos");
  DBUG_ASSERT(pos != NULL);
mats@romeo.(none)'s avatar
mats@romeo.(none) committed
1379
  if (thd->ha_data[binlog_hton->slot] == NULL)
1380 1381
    thd->binlog_setup_trx_data();
  binlog_trx_data *const trx_data=
mats@romeo.(none)'s avatar
mats@romeo.(none) committed
1382
    (binlog_trx_data*) thd->ha_data[binlog_hton->slot];
1383 1384
  DBUG_ASSERT(mysql_bin_log.is_open());
  *pos= trx_data->position();
1385
  DBUG_PRINT("return", ("*pos: %lu", (ulong) *pos));
1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408
  DBUG_VOID_RETURN;
}


/*
  Truncate the binary log transaction cache.

  SYNPOSIS
    binlog_trans_log_truncate()

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

  DESCRIPTION

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

 */
static void
binlog_trans_log_truncate(THD *thd, my_off_t pos)
{
  DBUG_ENTER("binlog_trans_log_truncate");
1409
  DBUG_PRINT("enter", ("pos: %lu", (ulong) pos));
1410

mats@romeo.(none)'s avatar
mats@romeo.(none) committed
1411
  DBUG_ASSERT(thd->ha_data[binlog_hton->slot] != NULL);
1412 1413 1414 1415
  /* Only true if binlog_trans_log_savepos() wasn't called before */
  DBUG_ASSERT(pos != ~(my_off_t) 0);

  binlog_trx_data *const trx_data=
mats@romeo.(none)'s avatar
mats@romeo.(none) committed
1416
    (binlog_trx_data*) thd->ha_data[binlog_hton->slot];
1417 1418 1419 1420 1421
  trx_data->truncate(pos);
  DBUG_VOID_RETURN;
}


1422 1423
/*
  this function is mostly a placeholder.
1424
  conceptually, binlog initialization (now mostly done in MYSQL_BIN_LOG::open)
1425 1426 1427
  should be moved here.
*/

1428
int binlog_init(void *p)
1429
{
1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440
  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;
1441
  return 0;
1442 1443
}

1444
static int binlog_close_connection(handlerton *hton, THD *thd)
1445
{
1446
  binlog_trx_data *const trx_data=
1447
    (binlog_trx_data*) thd->ha_data[binlog_hton->slot];
1448
  DBUG_ASSERT(mysql_bin_log.is_open() && trx_data->empty());
1449
  thd->ha_data[binlog_hton->slot]= 0;
1450
  trx_data->~binlog_trx_data();
1451
  my_free((uchar*)trx_data, MYF(0));
1452 1453 1454
  return 0;
}

1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477
/*
  End a transaction.

  SYNOPSIS
    binlog_end_trans()

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

  DESCRIPTION

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

    If 'end_ev' is NULL, the transaction is a rollback of only
    transactional tables, so the transaction cache will be truncated
    to either just before the last opened statement transaction (if
    'all' is false), or reset completely (if 'all' is true).
 */
1478
static int
1479 1480
binlog_end_trans(THD *thd, binlog_trx_data *trx_data,
                 Log_event *end_ev, bool all)
1481
{
1482
  DBUG_ENTER("binlog_end_trans");
1483 1484
  int error=0;
  IO_CACHE *trans_log= &trx_data->trans_log;
1485 1486
  DBUG_PRINT("enter", ("transaction: %s  end_ev: 0x%lx",
                       all ? "all" : "stmt", (long) end_ev));
1487 1488 1489
  DBUG_PRINT("info", ("thd->options={ %s%s}",
                      FLAGSTR(thd->options, OPTION_NOT_AUTOCOMMIT),
                      FLAGSTR(thd->options, OPTION_BEGIN)));
1490

1491 1492 1493 1494 1495 1496
  /*
    NULL denotes ROLLBACK with nothing to replicate: i.e., rollback of
    only transactional tables.  If the transaction contain changes to
    any non-transactiona tables, we need write the transaction and log
    a ROLLBACK last.
  */
1497
  if (end_ev != NULL)
1498
  {
1499
    /*
1500 1501 1502 1503
      Doing a commit or a rollback including non-transactional tables,
      i.e., ending a transaction where we might write the transaction
      cache to the binary log.

1504 1505 1506 1507 1508
      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.
     */
1509
    thd->binlog_flush_pending_rows_event(TRUE);
1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530
    /*
      We write the transaction cache to the binary log if either we're
      committing the entire transaction, or if we are doing an
      autocommit outside a transaction.
     */
    if (all || !(thd->options & (OPTION_BEGIN | OPTION_NOT_AUTOCOMMIT)))
    {
      error= mysql_bin_log.write(thd, &trx_data->trans_log, end_ev);
      trx_data->reset();
      /*
        We need to step the table map version after writing the
        transaction cache to disk.
      */
      mysql_bin_log.update_table_map_version();
      statistic_increment(binlog_cache_use, &LOCK_status);
      if (trans_log->disk_writes != 0)
      {
        statistic_increment(binlog_cache_disk_use, &LOCK_status);
        trans_log->disk_writes= 0;
      }
    }
1531 1532 1533
  }
  else
  {
1534 1535 1536
    /*
      If rolling back an entire transaction or a single statement not
      inside a transaction, we reset the transaction cache.
1537

1538 1539 1540 1541 1542
      If rolling back a statement in a transaction, we truncate the
      transaction cache to remove the statement.
     */
    if (all || !(thd->options & (OPTION_BEGIN | OPTION_NOT_AUTOCOMMIT)))
      trx_data->reset();
1543 1544
    else                                        // ...statement
      trx_data->truncate(trx_data->before_stmt_pos);
1545 1546 1547 1548 1549 1550 1551

    /*
      We need to step the table map version on a rollback to ensure
      that a new table map event is generated instead of the one that
      was written to the thrown-away transaction cache.
    */
    mysql_bin_log.update_table_map_version();
1552
  }
1553

1554
  DBUG_RETURN(error);
1555 1556
}

1557
static int binlog_prepare(handlerton *hton, THD *thd, bool all)
1558 1559 1560 1561 1562
{
  /*
    do nothing.
    just pretend we can do 2pc, so that MySQL won't
    switch to 1pc.
1563
    real work will be done in MYSQL_BIN_LOG::log_xid()
1564 1565 1566 1567
  */
  return 0;
}

1568
static int binlog_commit(handlerton *hton, THD *thd, bool all)
1569 1570
{
  DBUG_ENTER("binlog_commit");
1571
  binlog_trx_data *const trx_data=
1572
    (binlog_trx_data*) thd->ha_data[binlog_hton->slot];
1573
  DBUG_ASSERT(mysql_bin_log.is_open());
1574

1575
  if (trx_data->empty())
1576
  {
1577
    // we're here because trans_log was flushed in MYSQL_BIN_LOG::log_xid()
1578
    trx_data->reset();
1579 1580
    DBUG_RETURN(0);
  }
1581
  if (all)
1582 1583 1584
  {
    Query_log_event qev(thd, STRING_WITH_LEN("COMMIT"), TRUE, FALSE);
    qev.error_code= 0; // see comment in MYSQL_LOG::write(THD, IO_CACHE)
1585 1586
    int error= binlog_end_trans(thd, trx_data, &qev, all);
    DBUG_RETURN(error);
1587 1588
  }
  else
1589 1590 1591 1592
  {
    int error= binlog_end_trans(thd, trx_data, &invisible_commit, all);
    DBUG_RETURN(error);
  }
1593 1594
}

1595
static int binlog_rollback(handlerton *hton, THD *thd, bool all)
1596 1597
{
  DBUG_ENTER("binlog_rollback");
1598 1599
  int error=0;
  binlog_trx_data *const trx_data=
1600
    (binlog_trx_data*) thd->ha_data[binlog_hton->slot];
1601 1602 1603 1604 1605 1606 1607
  DBUG_ASSERT(mysql_bin_log.is_open());

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

1608
  /*
1609 1610 1611 1612
    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...)
1613
  */
1614
  if (unlikely(thd->no_trans_update.all || (thd->options & OPTION_KEEP_LOG)))
1615
  {
1616
    Query_log_event qev(thd, STRING_WITH_LEN("ROLLBACK"), TRUE, FALSE);
1617
    qev.error_code= 0; // see comment in MYSQL_LOG::write(THD, IO_CACHE)
1618
    error= binlog_end_trans(thd, trx_data, &qev, all);
1619
  }
1620
  else
1621
    error= binlog_end_trans(thd, trx_data, 0, all);
1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645
  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).
*/

1646
static int binlog_savepoint_set(handlerton *hton, THD *thd, void *sv)
1647 1648 1649
{
  DBUG_ENTER("binlog_savepoint_set");

1650
  binlog_trans_log_savepos(thd, (my_off_t*) sv);
1651
  /* Write it to the binary log */
1652 1653 1654 1655 1656
  
  int const error=
    thd->binlog_query(THD::STMT_QUERY_TYPE,
                      thd->query, thd->query_length, TRUE, FALSE);
  DBUG_RETURN(error);
1657 1658
}

1659
static int binlog_savepoint_rollback(handlerton *hton, THD *thd, void *sv)
1660 1661
{
  DBUG_ENTER("binlog_savepoint_rollback");
1662
  DBUG_ASSERT(mysql_bin_log.is_open());
1663 1664 1665 1666 1667 1668

  /*
    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.
  */
1669
  if (unlikely(thd->no_trans_update.all || (thd->options & OPTION_KEEP_LOG)))
1670
  {
1671
    int error=
1672 1673 1674
      thd->binlog_query(THD::STMT_QUERY_TYPE,
                        thd->query, thd->query_length, TRUE, FALSE);
    DBUG_RETURN(error);
1675
  }
1676
  binlog_trans_log_truncate(thd, *(my_off_t*)sv);
1677 1678 1679
  DBUG_RETURN(0);
}

1680

1681 1682 1683 1684 1685
int check_binlog_magic(IO_CACHE* log, const char** errmsg)
{
  char magic[4];
  DBUG_ASSERT(my_b_tell(log) == 0);

1686
  if (my_b_read(log, (uchar*) magic, sizeof(magic)))
1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700
  {
    *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;
}

1701

1702 1703 1704 1705 1706
File open_binlog(IO_CACHE *log, const char *log_file_name, const char **errmsg)
{
  File file;
  DBUG_ENTER("open_binlog");

1707 1708
  if ((file = my_open(log_file_name, O_RDONLY | O_BINARY | O_SHARE, 
                      MYF(MY_WME))) < 0)
1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734
  {
    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
1735

rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1736 1737
#ifdef __NT__
static int eventSource = 0;
serg@serg.mylan's avatar
serg@serg.mylan committed
1738

1739
static void setup_windows_event_source()
monty@mysql.com's avatar
monty@mysql.com committed
1740
{
1741
  HKEY    hRegKey= NULL;
monty@mysql.com's avatar
monty@mysql.com committed
1742 1743 1744
  DWORD   dwError= 0;
  TCHAR   szPath[MAX_PATH];
  DWORD dwTypes;
1745

monty@mysql.com's avatar
monty@mysql.com committed
1746 1747 1748
  if (eventSource)               // Ensure that we are only called once
    return;
  eventSource= 1;
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1749

monty@mysql.com's avatar
monty@mysql.com committed
1750
  // Create the event source registry key
1751
  dwError= RegCreateKey(HKEY_LOCAL_MACHINE,
monty@mysql.com's avatar
monty@mysql.com committed
1752 1753
                          "SYSTEM\\CurrentControlSet\\Services\\EventLog\\Application\\MySQL", 
                          &hRegKey);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1754

monty@mysql.com's avatar
monty@mysql.com committed
1755 1756
  /* Name of the PE module that contains the message resource */
  GetModuleFileName(NULL, szPath, MAX_PATH);
serg@serg.mylan's avatar
serg@serg.mylan committed
1757

rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1758
  /* Register EventMessageFile */
1759
  dwError = RegSetValueEx(hRegKey, "EventMessageFile", 0, REG_EXPAND_SZ,
1760
                          (PBYTE) szPath, (DWORD) (strlen(szPath) + 1));
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1761

monty@mysql.com's avatar
monty@mysql.com committed
1762 1763 1764 1765 1766
  /* 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
1767

monty@mysql.com's avatar
monty@mysql.com committed
1768
  RegCloseKey(hRegKey);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1769 1770
}

monty@mysql.com's avatar
monty@mysql.com committed
1771
#endif /* __NT__ */
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1772 1773


bk@work.mysql.com's avatar
bk@work.mysql.com committed
1774 1775 1776 1777 1778 1779 1780 1781
/****************************************************************************
** 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)
{
1782 1783 1784 1785
  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
1786
  reg1 struct fileinfo *file_info;
1787
  ulong                 max_found=0;
1788 1789
  size_t		buf_length, length;
  char			*start, *end;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1790 1791
  DBUG_ENTER("find_uniq_filename");

1792 1793 1794
  length= dirname_part(buff, name, &buf_length);
  start=  name + length;
  end=    strend(start);
1795

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1796
  *end='.';
1797
  length= (size_t) (end-start+1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1798 1799 1800 1801 1802 1803 1804 1805 1806

  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++)
  {
1807
    if (bcmp((uchar*) file_info->name, (uchar*) start, length) == 0 &&
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1808 1809 1810 1811 1812 1813 1814 1815
	test_if_number(file_info->name+length, &number,0))
    {
      set_if_bigger(max_found,(ulong) number);
    }
  }
  my_dirend(dir_info);

  *end++='.';
1816
  sprintf(end,"%06ld",max_found+1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1817 1818 1819
  DBUG_RETURN(0);
}

1820

1821
void MYSQL_LOG::init(enum_log_type log_type_arg,
1822
                     enum cache_type io_cache_type_arg)
1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 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 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891
{
  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;

1892
  if (log_type == LOG_NORMAL)
1893 1894
  {
    char *end;
1895
    int len=my_snprintf(buff, sizeof(buff), "%s, Version: %s (%s). "
1896
#ifdef EMBEDDED_LIBRARY
1897 1898
                        "embedded library\n",
                        my_progname, server_version, MYSQL_COMPILATION_COMMENT
1899 1900
#elif __NT__
			"started with:\nTCP Port: %d, Named Pipe: %s\n",
1901 1902
                        my_progname, server_version, MYSQL_COMPILATION_COMMENT,
                        mysqld_port, mysqld_unix_port
1903 1904
#else
			"started with:\nTcp port: %d  Unix socket: %s\n",
1905 1906
                        my_progname, server_version, MYSQL_COMPILATION_COMMENT,
                        mysqld_port, mysqld_unix_port
1907 1908 1909 1910
#endif
                       );
    end= strnmov(buff + len, "Time                 Id Command    Argument\n",
                 sizeof(buff) - len);
1911
    if (my_b_write(&log_file, (uchar*) buff, (uint) (end-buff)) ||
1912 1913 1914 1915
	flush_io_cache(&log_file))
      goto err;
  }

1916
  log_state= LOG_OPENED;
1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927
  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);
1928
  log_state= LOG_CLOSED;
1929 1930 1931
  DBUG_RETURN(1);
}

1932
MYSQL_LOG::MYSQL_LOG()
serg@janus.mylan's avatar
serg@janus.mylan committed
1933 1934
  : name(0), write_error(FALSE), inited(FALSE), log_type(LOG_UNKNOWN),
    log_state(LOG_CLOSED)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1935 1936
{
  /*
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1937 1938
    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
1939
    called only in main(). Doing initialization here would make it happen
1940
    before main().
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1941
  */
1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969
  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));
1970
  if (log_state == LOG_OPENED)
1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986
  {
    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);
    }
  }

1987
  log_state= (exiting & LOG_CLOSE_TO_BE_OPENED) ? LOG_TO_BE_OPENED : LOG_CLOSED;
1988 1989
  safeFree(name);
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1990 1991
}

1992
/* this is called only once */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1993

monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
1994
void MYSQL_LOG::cleanup()
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1995
{
1996
  DBUG_ENTER("cleanup");
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1997
  if (inited)
1998
  {
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
1999
    inited= 0;
2000
    (void) pthread_mutex_destroy(&LOCK_log);
2001
    close(0);
2002
  }
2003
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2004 2005
}

2006

bk@work.mysql.com's avatar
bk@work.mysql.com committed
2007
int MYSQL_LOG::generate_new_name(char *new_name, const char *log_name)
2008
{
2009 2010
  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
2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023
  {
    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;
}

2024

2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036
/*
  Reopen the log file

  SYNOPSIS
    reopen_file()

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


2037
void MYSQL_QUERY_LOG::reopen_file()
2038
{
2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054
  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);

  /*
2055
     Note that at this point, log_state != LOG_CLOSED (important for is_open()).
2056 2057
  */

2058
  open(save_name, log_type, 0, io_cache_type);
2059 2060 2061 2062
  my_free(save_name, MYF(0));

  pthread_mutex_unlock(&LOCK_log);

2063
  DBUG_VOID_RETURN;
2064 2065
}

2066

2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091
/*
  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
*/

2092 2093 2094 2095
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
2096
{
2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118
  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);

2119
        if (my_b_write(&log_file, (uchar*) &time_buff, time_buff_len))
2120 2121 2122
          goto err;
      }
      else
2123
        if (my_b_write(&log_file, (uchar*) "\t\t" ,2) < 0)
2124 2125
          goto err;

2126
      /* command_type, thread_id */
2127
      length= my_snprintf(buff, 32, "%5ld ", (long) thread_id);
2128

2129
    if (my_b_write(&log_file, (uchar*) buff, length))
2130 2131
      goto err;

2132
    if (my_b_write(&log_file, (uchar*) command_type, command_type_len))
2133 2134
      goto err;

2135
    if (my_b_write(&log_file, (uchar*) "\t", 1))
2136 2137 2138
      goto err;

    /* sql_text */
2139
    if (my_b_write(&log_file, (uchar*) sql_text, sql_text_len))
2140 2141
      goto err;

2142
    if (my_b_write(&log_file, (uchar*) "\n", 1) ||
2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155
        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;
2156 2157
}

2158

2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187
/*
  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
*/

2188 2189 2190 2191 2192
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
2193
{
2194
  bool error= 0;
2195
  DBUG_ENTER("MYSQL_QUERY_LOG::write");
2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221

  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))
    {
      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 */
2222
        if (my_b_write(&log_file, (uchar*) buff, buff_len))
2223 2224 2225 2226 2227 2228 2229
          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;
2230
      if (my_b_write(&log_file, (uchar*) "\n", 1))
2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246
        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);
    }
2247
    if (thd->stmt_depends_on_first_successful_insert_id_in_prev_stmt)
2248 2249
    {
      end=strmov(end, ",last_insert_id=");
2250 2251 2252
      end=longlong10_to_str((longlong)
                            thd->first_successful_insert_id_in_prev_stmt_for_binlog,
                            end, -10);
2253 2254
    }
    // Save value if we do an insert.
2255
    if (thd->auto_inc_intervals_in_cur_stmt_for_binlog.nb_elements() > 0)
2256 2257 2258 2259
    {
      if (!(specialflag & SPECIAL_SHORT_LOG_FORMAT))
      {
        end=strmov(end,",insert_id=");
2260 2261 2262
        end=longlong10_to_str((longlong)
                              thd->auto_inc_intervals_in_cur_stmt_for_binlog.minimum(),
                              end, -10);
2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277
      }
    }

    /*
      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';
2278 2279
      if (my_b_write(&log_file, (uchar*) "SET ", 4) ||
          my_b_write(&log_file, (uchar*) buff + 1, (uint) (end-buff)))
2280 2281 2282 2283 2284 2285
        tmp_errno= errno;
    }
    if (is_command)
    {
      end= strxmov(buff, "# administrator command: ", NullS);
      buff_len= (ulong) (end - buff);
2286
      my_b_write(&log_file, (uchar*) buff, buff_len);
2287
    }
2288 2289
    if (my_b_write(&log_file, (uchar*) sql_text, sql_text_len) ||
        my_b_write(&log_file, (uchar*) ";\n",2) ||
2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302
        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);
2303 2304
}

2305

2306
const char *MYSQL_LOG::generate_name(const char *log_name,
2307 2308
                                      const char *suffix,
                                      bool strip_ext, char *buff)
2309 2310 2311 2312 2313 2314 2315
{
  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.
    */
2316
    strmake(buff, pidfile_name, FN_REFLEN - 5);
2317
    strmov(fn_ext(buff), suffix);
2318 2319 2320 2321 2322
    return (const char *)buff;
  }
  // get rid of extension if the log is binary to avoid problems
  if (strip_ext)
  {
2323 2324 2325
    char *p= fn_ext(log_name);
    uint length= (uint) (p - log_name);
    strmake(buff, log_name, min(length, FN_REFLEN));
2326 2327 2328 2329 2330
    return (const char*)buff;
  }
  return log_name;
}

2331

2332

2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388
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,
2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413
                                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
2414
			0, MYF(MY_WME | MY_WAIT_IF_FULL)))
2415 2416 2417 2418 2419 2420 2421 2422
  {
    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
2423

2424
/*
2425
  Open a (new) binlog file.
2426 2427

  DESCRIPTION
2428
  - Open the log file and the index file. Register the new
2429 2430 2431 2432 2433 2434 2435 2436 2437
    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
*/

2438 2439 2440 2441 2442 2443 2444
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
2445
{
2446
  File file= -1;
2447
  DBUG_ENTER("MYSQL_BIN_LOG::open");
2448
  DBUG_PRINT("enter",("log_type: %d",(int) log_type_arg));
2449 2450

  write_error=0;
2451

2452 2453 2454
  /* 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 */
2455

2456
  init(no_auto_events_arg, max_size_arg);
2457

2458
  open_count++;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2459

2460
  DBUG_ASSERT(log_type == LOG_BIN);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2461 2462

  {
2463 2464 2465 2466 2467 2468 2469 2470 2471 2472
    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.
      */
2473
      if (my_b_safe_write(&log_file, (uchar*) BINLOG_MAGIC,
2474
			  BIN_LOG_HEADER_SIZE))
2475
        goto err;
2476 2477
      bytes_written+= BIN_LOG_HEADER_SIZE;
      write_file_name_to_index_file= 1;
2478
    }
2479

2480
    DBUG_ASSERT(my_b_inited(&index_file) != 0);
2481 2482
    reinit_io_cache(&index_file, WRITE_CACHE,
                    my_b_filelength(&index_file), 0, 0);
2483
    if (need_start_event && !no_auto_events)
2484
    {
2485 2486 2487 2488 2489
      /*
        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
2490 2491 2492 2493 2494 2495
      /*
        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;
2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514
      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
2515
        Format_description_log_event::Format_description_log_event(char* buf,).
2516 2517 2518 2519 2520 2521 2522 2523 2524
        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
2525
        Format_description_log_event::apply_event_impl().
2526 2527 2528 2529
      */
      description_event_for_queue->created= 0;
      /* Don't set log_pos in event header */
      description_event_for_queue->artificial_event=1;
2530

2531 2532 2533
      if (description_event_for_queue->write(&log_file))
        goto err;
      bytes_written+= description_event_for_queue->data_written;
2534
    }
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2535 2536
    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
2537
      goto err;
2538 2539 2540

    if (write_file_name_to_index_file)
    {
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2541 2542 2543 2544
      /*
        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.
      */
2545
      if (my_b_write(&index_file, (uchar*) log_file_name,
2546
		     strlen(log_file_name)) ||
2547
	  my_b_write(&index_file, (uchar*) "\n", 1) ||
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2548 2549
	  flush_io_cache(&index_file) ||
          my_sync(index_file.file, MYF(MY_WME)))
2550
	goto err;
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
2551
    }
2552
  }
2553 2554
  log_state= LOG_OPENED;

2555
  DBUG_RETURN(0);
2556 2557

err:
2558 2559 2560
  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, \
2561
shutdown the MySQL server and restart it.", name, errno);
2562 2563 2564
  if (file >= 0)
    my_close(file,MYF(0));
  end_io_cache(&log_file);
2565 2566
  end_io_cache(&index_file);
  safeFree(name);
2567
  log_state= LOG_CLOSED;
2568
  DBUG_RETURN(1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2569 2570
}

2571

2572
int MYSQL_BIN_LOG::get_current_log(LOG_INFO* linfo)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2573 2574
{
  pthread_mutex_lock(&LOCK_log);
2575 2576 2577 2578 2579
  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
2580
int MYSQL_BIN_LOG::raw_get_current_log(LOG_INFO* linfo)
2581
{
2582 2583
  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
2584 2585 2586
  return 0;
}

2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607
/*
  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
*/

2608 2609
#ifdef HAVE_REPLICATION

2610 2611 2612 2613 2614
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;
2615
  uchar io_buf[IO_SIZE*2];
2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626
  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));
2627
    if (my_write(file, io_buf, bytes_read, MYF(MY_WME | MY_NABP)))
2628 2629 2630
      goto err;
  }
  /* The following will either truncate the file or fill the end with \n' */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2631 2632
  if (my_chsize(file, offset - init_offset, '\n', MYF(MY_WME)) ||
      my_sync(file, MYF(MY_WME)))
2633 2634 2635 2636 2637 2638 2639 2640 2641 2642
    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);
}

2643
#endif /* HAVE_REPLICATION */
2644 2645 2646 2647 2648 2649 2650 2651 2652 2653

/*
  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
2654
    need_lock		Set this to 1 if the parent doesn't already have a
2655 2656 2657
			lock on LOCK_index

  NOTE
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
2658 2659
    On systems without the truncate function the file will end with one or
    more empty lines.  These will be ignored when reading the file.
2660 2661 2662 2663 2664 2665 2666

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

2667
int MYSQL_BIN_LOG::find_log_pos(LOG_INFO *linfo, const char *log_name,
2668
			    bool need_lock)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2669
{
2670 2671 2672 2673 2674
  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
2675

2676
  /*
2677 2678
    Mutex needed because we need to make sure the file pointer does not
    move from under our feet
2679
  */
2680
  if (need_lock)
2681
    pthread_mutex_lock(&LOCK_index);
2682 2683 2684 2685 2686
  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);

2687
  for (;;)
2688
  {
2689
    uint length;
2690 2691 2692 2693
    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
2694
    {
2695 2696 2697
      /* 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
2698 2699
    }

2700 2701
    // if the log entry matches, null string matching anything
    if (!log_name ||
sasha@mysql.sashanet.com's avatar
sasha@mysql.sashanet.com committed
2702
	(log_name_len == length-1 && fname[log_name_len] == '\n' &&
2703
	 !memcmp(fname, log_name, log_name_len)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2704
    {
2705
      DBUG_PRINT("info",("Found log file entry"));
2706
      fname[length-1]=0;			// remove last \n
2707 2708
      linfo->index_file_start_offset= offset;
      linfo->index_file_offset = my_b_tell(&index_file);
2709
      break;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2710
    }
2711
  }
2712

2713
  if (need_lock)
2714
    pthread_mutex_unlock(&LOCK_index);
2715
  DBUG_RETURN(error);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2716
}
2717

2718

2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740
/*
  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
*/

2741
int MYSQL_BIN_LOG::find_next_log(LOG_INFO* linfo, bool need_lock)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2742
{
2743
  int error= 0;
2744
  uint length;
2745 2746
  char *fname= linfo->log_file_name;

2747 2748
  if (need_lock)
    pthread_mutex_lock(&LOCK_index);
2749 2750 2751 2752 2753 2754 2755 2756
  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)
2757
  {
2758
    error = !index_file.error ? LOG_INFO_EOF : LOG_INFO_IO;
2759 2760
    goto err;
  }
2761
  fname[length-1]=0;				// kill \n
2762
  linfo->index_file_offset = my_b_tell(&index_file);
2763

bk@work.mysql.com's avatar
bk@work.mysql.com committed
2764
err:
2765 2766
  if (need_lock)
    pthread_mutex_unlock(&LOCK_index);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2767 2768 2769
  return error;
}

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2770

2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788
/*
  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
*/

2789
bool MYSQL_BIN_LOG::reset_logs(THD* thd)
2790 2791
{
  LOG_INFO linfo;
2792
  bool error=0;
2793
  const char* save_name;
2794
  DBUG_ENTER("reset_logs");
2795

tomas@poseidon.ndb.mysql.com's avatar
tomas@poseidon.ndb.mysql.com committed
2796
  ha_reset_logs(thd);
2797 2798 2799 2800
  /*
    We need to get both locks to be sure that no one is trying to
    write to the index log file.
  */
2801
  pthread_mutex_lock(&LOCK_log);
2802
  pthread_mutex_lock(&LOCK_index);
2803

2804 2805 2806 2807 2808 2809
  /*
    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.
  */
2810
  VOID(pthread_mutex_lock(&LOCK_thread_count));
2811 2812 2813 2814

  /* Save variables so that we can reopen the log */
  save_name=name;
  name=0;					// Protect against free
2815
  close(LOG_CLOSE_TO_BE_OPENED);
2816 2817 2818 2819

  /* First delete all old log files */

  if (find_log_pos(&linfo, NullS, 0))
2820 2821 2822 2823
  {
    error=1;
    goto err;
  }
2824

2825
  for (;;)
2826
  {
2827
    my_delete_allow_opened(linfo.log_file_name, MYF(MY_WME));
2828
    if (find_next_log(&linfo, 0))
2829 2830
      break;
  }
2831 2832

  /* Start logging with a new file */
2833
  close(LOG_CLOSE_INDEX);
2834
  my_delete_allow_opened(index_file_name, MYF(MY_WME));	// Reset (open will update)
2835
  if (!thd->slave_thread)
2836
    need_start_event=1;
2837
  if (!open_index_file(index_file_name, 0))
2838
    open(save_name, log_type, 0, io_cache_type, no_auto_events, max_size, 0);
2839
  my_free((uchar*) save_name, MYF(0));
2840

2841
err:
2842
  VOID(pthread_mutex_unlock(&LOCK_thread_count));
2843
  pthread_mutex_unlock(&LOCK_index);
2844
  pthread_mutex_unlock(&LOCK_log);
2845
  DBUG_RETURN(error);
2846 2847
}

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2848

2849
/*
2850 2851 2852
  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.
2853 2854 2855

  SYNOPSIS
    purge_first_log()
2856 2857 2858 2859 2860
    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).
2861

2862 2863
  NOTE
    - This is only called from the slave-execute thread when it has read
2864 2865 2866 2867 2868
      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).
2869 2870 2871

  IMPLEMENTATION
    - Protects index file with LOCK_index
2872 2873
    - Delete relevant relay log files
    - Copy all file names after these ones to the front of the index file
2874
    - If the OS has truncate, truncate the file, else fill it with \n'
2875
    - Read the next file name from the index file and store in rli->linfo
2876 2877

  RETURN VALUES
2878 2879 2880 2881
    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
2882 2883
*/

hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
2884
#ifdef HAVE_REPLICATION
2885

2886
int MYSQL_BIN_LOG::purge_first_log(struct st_relay_log_info* rli, bool included) 
2887
{
2888
  int error;
2889 2890
  DBUG_ENTER("purge_first_log");

2891 2892
  DBUG_ASSERT(is_open());
  DBUG_ASSERT(rli->slave_running == 1);
2893
  DBUG_ASSERT(!strcmp(rli->linfo.log_file_name,rli->event_relay_log_name));
2894

2895
  pthread_mutex_lock(&LOCK_index);
2896 2897 2898 2899 2900 2901
  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
2902 2903

  /*
2904 2905 2906
    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
2907
  */
2908 2909 2910 2911 2912
  pthread_cond_broadcast(&rli->log_space_cond);
  
  /*
    Read the next log file name from the index file and pass it back to
    the caller
2913 2914
    If included is true, we want the first relay log;
    otherwise we want the one after event_relay_log_name.
2915
  */
2916 2917 2918 2919
  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)))))
2920 2921
  {
    char buff[22];
2922 2923 2924 2925 2926
    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);
2927 2928
    goto err;
  }
2929

2930
  /*
2931
    Reset rli's coordinates to the current log.
2932
  */
2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946
  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);
2947
    rli->notify_group_relay_log_name_update();
2948
  }
2949 2950 2951

  /* 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
2952

2953
err:
2954
  pthread_mutex_unlock(&LOCK_index);
2955
  DBUG_RETURN(error);
2956 2957
}

2958 2959 2960 2961
/*
  Update log index_file
*/

2962
int MYSQL_BIN_LOG::update_log_index(LOG_INFO* log_info, bool need_update_threads)
2963 2964 2965 2966 2967
{
  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
2968 2969
  if (need_update_threads)
    adjust_linfo_offsets(log_info->index_file_start_offset);
2970 2971
  return 0;
}
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2972

2973 2974 2975 2976 2977
/*
  Remove all logs before the given log from disk and from the index file.

  SYNOPSIS
    purge_logs()
2978 2979 2980 2981 2982 2983 2984
    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
2985 2986 2987 2988 2989 2990 2991 2992 2993 2994

  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
*/

2995
int MYSQL_BIN_LOG::purge_logs(const char *to_log, 
2996 2997 2998 2999
                          bool included,
                          bool need_mutex, 
                          bool need_update_threads, 
                          ulonglong *decrease_log_space)
3000 3001
{
  int error;
3002
  int ret = 0;
3003
  bool exit_loop= 0;
3004 3005
  LOG_INFO log_info;
  DBUG_ENTER("purge_logs");
3006
  DBUG_PRINT("info",("to_log= %s",to_log));
3007

3008 3009
  if (need_mutex)
    pthread_mutex_lock(&LOCK_index);
3010
  if ((error=find_log_pos(&log_info, to_log, 0 /*no mutex*/)))
3011 3012
    goto err;

3013
  /*
3014
    File name exists in index file; delete until we find this file
3015 3016 3017
    or a file that is used.
  */
  if ((error=find_log_pos(&log_info, NullS, 0 /*no mutex*/)))
3018
    goto err;
3019 3020
  while ((strcmp(to_log,log_info.log_file_name) || (exit_loop=included)) &&
         !log_in_use(log_info.log_file_name))
3021
  {
3022
    ulong file_size= 0;
3023 3024 3025
    if (decrease_log_space) //stat the file we want to delete
    {
      MY_STAT s;
3026 3027 3028 3029 3030 3031

      /* 
         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.
      */
3032
      if (my_stat(log_info.log_file_name,&s,MYF(0)))
3033
        file_size= s.st_size;
3034
      else
3035 3036
	sql_print_information("Failed to execute my_stat on file '%s'",
			      log_info.log_file_name);
3037 3038 3039 3040 3041 3042 3043
    }
    /*
      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)
3044
      *decrease_log_space-= file_size;
tomas@poseidon.ndb.mysql.com's avatar
tomas@poseidon.ndb.mysql.com committed
3045 3046

    ha_binlog_index_purge_file(current_thd, log_info.log_file_name);
3047 3048 3049 3050 3051 3052 3053 3054
    if (current_thd->query_error) {
      DBUG_PRINT("info",("query error: %d", current_thd->query_error));
      if (my_errno == EMFILE) {
        DBUG_PRINT("info",("my_errno: %d, set ret = LOG_INFO_EMFILE", my_errno));
        ret = LOG_INFO_EMFILE;
        break;
      }
    }
tomas@poseidon.ndb.mysql.com's avatar
tomas@poseidon.ndb.mysql.com committed
3055

3056
    if (find_next_log(&log_info, 0) || exit_loop)
3057
      break;
3058
  }
3059

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
3060 3061 3062 3063
  /*
    If we get killed -9 here, the sysadmin would have to edit
    the log index file after restart - otherwise, this should be safe
  */
3064
  error= update_log_index(&log_info, need_update_threads);
3065 3066 3067
  if (error == 0) {
    error = ret;
  }
3068

3069
err:
3070 3071
  if (need_mutex)
    pthread_mutex_unlock(&LOCK_index);
3072
  DBUG_RETURN(error);
3073 3074
}

3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092
/*
  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
*/

3093
int MYSQL_BIN_LOG::purge_logs_before_date(time_t purge_time)
3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108
{
  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*/)))
3109
    goto err;
3110 3111 3112 3113 3114

  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 */
3115 3116
    if (!my_stat(log_info.log_file_name, &stat_area, MYF(0)) ||
	stat_area.st_mtime >= purge_time)
3117
      break;
3118
    my_delete(log_info.log_file_name, MYF(0));
tomas@poseidon.ndb.mysql.com's avatar
tomas@poseidon.ndb.mysql.com committed
3119 3120 3121

    ha_binlog_index_purge_file(current_thd, log_info.log_file_name);

3122 3123
    if (find_next_log(&log_info, 0))
      break;
3124
  }
3125

3126 3127 3128 3129
  /*
    If we get killed -9 here, the sysadmin would have to edit
    the log index file after restart - otherwise, this should be safe
  */
3130
  error= update_log_index(&log_info, 1);
3131

3132 3133
err:
  pthread_mutex_unlock(&LOCK_index);
3134
  DBUG_RETURN(error);
3135
}
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
3136
#endif /* HAVE_REPLICATION */
3137

3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149

/*
  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
*/

3150
void MYSQL_BIN_LOG::make_log_name(char* buf, const char* log_ident)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3151
{
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3152 3153 3154 3155 3156
  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
3157 3158
}

3159 3160 3161 3162 3163

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

3164
bool MYSQL_BIN_LOG::is_active(const char *log_file_name_arg)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3165
{
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3166
  return !strcmp(log_file_name, log_file_name_arg);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3167 3168
}

3169

3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189
/*
  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);
}


3190 3191 3192 3193
/*
  Start writing to a new log file or reopen the old file

  SYNOPSIS
3194
    new_file_impl()
3195
    need_lock		Set to 1 if caller has not locked LOCK_log
3196 3197 3198 3199 3200

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

3201
void MYSQL_BIN_LOG::new_file_impl(bool need_lock)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3202
{
3203 3204
  char new_name[FN_REFLEN], *new_name_ptr, *old_name;

3205
  DBUG_ENTER("MYSQL_BIN_LOG::new_file_impl");
3206
  if (!is_open())
3207 3208 3209 3210
  {
    DBUG_PRINT("info",("log is closed"));
    DBUG_VOID_RETURN;
  }
3211 3212 3213

  if (need_lock)
    pthread_mutex_lock(&LOCK_log);
3214 3215
  pthread_mutex_lock(&LOCK_index);

3216 3217 3218
  safe_mutex_assert_owner(&LOCK_log);
  safe_mutex_assert_owner(&LOCK_index);

3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231
  /*
    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);
3232 3233
    while (prepared_xids) {
      DBUG_PRINT("info", ("prepared_xids=%lu", prepared_xids));
3234
      pthread_cond_wait(&COND_prep_xids, &LOCK_prep_xids);
3235
    }
3236 3237 3238
    pthread_mutex_unlock(&LOCK_prep_xids);
  }

3239
  /* Reuse old name if not binlog and not update log */
3240
  new_name_ptr= name;
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
3241

3242
  /*
3243 3244 3245
    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.
3246
  */
3247 3248 3249
  if (generate_new_name(new_name, name))
    goto end;
  new_name_ptr=new_name;
3250

3251
  if (log_type == LOG_BIN)
3252
  {
3253
    if (!no_auto_events)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3254
    {
3255
      /*
3256 3257
        We log the whole file name for log file as the user may decide
        to change base names at some point.
3258
      */
3259
      Rotate_log_event r(new_name+dirname_length(new_name),
3260
                         0, LOG_EVENT_OFFSET, 0);
3261
      r.write(&log_file);
3262
      bytes_written += r.data_written;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3263
    }
3264 3265 3266 3267 3268
    /*
      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.
    */
3269
    signal_update();
3270 3271 3272
  }
  old_name=name;
  name=0;				// Don't free name
3273
  close(LOG_CLOSE_TO_BE_OPENED);
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3274

3275
  /*
3276
     Note that at this point, log_state != LOG_CLOSED (important for is_open()).
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3277
  */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3278

3279
  /*
3280
     new_file() is only used for rotation (in FLUSH LOGS or because size >
3281
     max_binlog_size or max_relay_log_size).
3282 3283 3284 3285
     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.
3286
  */
3287

3288
  open(old_name, log_type, new_name_ptr,
3289
       io_cache_type, no_auto_events, max_size, 1);
3290 3291
  my_free(old_name,MYF(0));

3292
end:
3293 3294
  if (need_lock)
    pthread_mutex_unlock(&LOCK_log);
3295 3296
  pthread_mutex_unlock(&LOCK_index);

3297
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3298 3299
}

3300

3301
bool MYSQL_BIN_LOG::append(Log_event* ev)
3302 3303 3304
{
  bool error = 0;
  pthread_mutex_lock(&LOCK_log);
3305
  DBUG_ENTER("MYSQL_BIN_LOG::append");
3306

3307
  DBUG_ASSERT(log_file.type == SEQ_READ_APPEND);
3308 3309 3310 3311
  /*
    Log_event::write() is smart enough to use my_b_write() or
    my_b_append() depending on the kind of cache we have.
  */
3312 3313 3314 3315 3316
  if (ev->write(&log_file))
  {
    error=1;
    goto err;
  }
3317
  bytes_written+= ev->data_written;
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
3318
  DBUG_PRINT("info",("max_size: %lu",max_size));
3319
  if ((uint) my_b_append_tell(&log_file) > max_size)
3320
    new_file_without_locking();
3321

3322
err:
3323
  pthread_mutex_unlock(&LOCK_log);
3324
  signal_update();				// Safe as we don't call close
3325
  DBUG_RETURN(error);
3326 3327
}

3328

3329
bool MYSQL_BIN_LOG::appendv(const char* buf, uint len,...)
3330
{
3331
  bool error= 0;
3332
  DBUG_ENTER("MYSQL_BIN_LOG::appendv");
3333 3334
  va_list(args);
  va_start(args,len);
3335

3336
  DBUG_ASSERT(log_file.type == SEQ_READ_APPEND);
3337

3338
  safe_mutex_assert_owner(&LOCK_log);
3339 3340
  do
  {
3341
    if (my_b_append(&log_file,(uchar*) buf,len))
3342
    {
3343 3344
      error= 1;
      goto err;
3345
    }
3346
    bytes_written += len;
3347
  } while ((buf=va_arg(args,const char*)) && (len=va_arg(args,uint)));
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
3348
  DBUG_PRINT("info",("max_size: %lu",max_size));
3349
  if ((uint) my_b_append_tell(&log_file) > max_size)
3350
    new_file_without_locking();
3351 3352

err:
3353 3354
  if (!error)
    signal_update();
3355
  DBUG_RETURN(error);
3356
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3357

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
3358

3359
bool MYSQL_BIN_LOG::flush_and_sync()
3360
{
3361 3362 3363 3364 3365
  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)
3366 3367
  {
    sync_binlog_counter= 0;
3368
    err=my_sync(fd, MYF(MY_WME));
3369
  }
3370
  return err;
3371 3372
}

3373
void MYSQL_BIN_LOG::start_union_events(THD *thd, query_id_t query_id_param)
3374 3375 3376 3377 3378
{
  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;
3379
  thd->binlog_evt_union.first_query_id= query_id_param;
3380 3381
}

3382
void MYSQL_BIN_LOG::stop_union_events(THD *thd)
3383 3384 3385 3386 3387
{
  DBUG_ASSERT(thd->binlog_evt_union.do_union);
  thd->binlog_evt_union.do_union= FALSE;
}

3388
bool MYSQL_BIN_LOG::is_query_in_union(THD *thd, query_id_t query_id_param)
3389 3390 3391 3392 3393
{
  return (thd->binlog_evt_union.do_union && 
          query_id_param >= thd->binlog_evt_union.first_query_id);
}

3394 3395 3396 3397 3398 3399 3400 3401 3402 3403

/*
  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=
3404
    (binlog_trx_data*) ha_data[binlog_hton->slot];
3405 3406 3407 3408

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

3409
  ha_data[binlog_hton->slot]= trx_data=
3410 3411 3412 3413 3414
    (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)))
  {
3415
    my_free((uchar*)trx_data, MYF(MY_ALLOW_ZERO_PTR));
3416
    ha_data[binlog_hton->slot]= 0;
3417 3418
    DBUG_RETURN(1);                      // Didn't manage to set it up
  }
3419

mats@romeo.(none)'s avatar
mats@romeo.(none) committed
3420
  trx_data= new (ha_data[binlog_hton->slot]) binlog_trx_data;
3421

3422 3423 3424
  DBUG_RETURN(0);
}

3425
/*
3426 3427
  Function to start a statement and optionally a transaction for the
  binary log.
3428

3429 3430
  SYNOPSIS
    binlog_start_trans_and_stmt()
3431

3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455
  DESCRIPTION

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

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

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

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

void
THD::binlog_start_trans_and_stmt()
{
mats@romeo.(none)'s avatar
mats@romeo.(none) committed
3456
  binlog_trx_data *trx_data= (binlog_trx_data*) ha_data[binlog_hton->slot];
3457 3458 3459 3460 3461 3462
  DBUG_ENTER("binlog_start_trans_and_stmt");
  DBUG_PRINT("enter", ("trx_data: 0x%lx  trx_data->before_stmt_pos: %lu",
                       (long) trx_data,
                       (trx_data ? (ulong) trx_data->before_stmt_pos :
                        (ulong) 0)));

3463
  if (trx_data == NULL ||
3464
      trx_data->before_stmt_pos == MY_OFF_T_UNDEF)
3465
  {
3466
    this->binlog_set_stmt_begin();
3467
    if (options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN))
mats@romeo.(none)'s avatar
mats@romeo.(none) committed
3468 3469
      trans_register_ha(this, TRUE, binlog_hton);
    trans_register_ha(this, FALSE, binlog_hton);
3470 3471 3472 3473
  }
  DBUG_VOID_RETURN;
}

3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493
void THD::binlog_set_stmt_begin() {
  binlog_trx_data *trx_data=
    (binlog_trx_data*) ha_data[binlog_hton->slot];

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

int THD::binlog_flush_transaction_cache()
{
  DBUG_ENTER("binlog_flush_transaction_cache");
  binlog_trx_data *trx_data= (binlog_trx_data*) ha_data[binlog_hton->slot];
3494
  DBUG_PRINT("enter", ("trx_data=0x%lu", (ulong) trx_data));
3495
  if (trx_data)
3496 3497
    DBUG_PRINT("enter", ("trx_data->before_stmt_pos=%lu",
                         (ulong) trx_data->before_stmt_pos));
3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518

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

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

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

  DBUG_RETURN(0);
}


3519 3520
/*
  Write a table map to the binary log.
3521 3522
 */

3523 3524
int THD::binlog_write_table_map(TABLE *table, bool is_trans)
{
3525
  int error;
3526
  DBUG_ENTER("THD::binlog_write_table_map");
3527 3528
  DBUG_PRINT("enter", ("table: 0x%lx  (%s: #%lu)",
                       (long) table, table->s->table_name.str,
3529
                       table->s->table_map_id));
3530 3531

  /* Pre-conditions */
3532
  DBUG_ASSERT(current_stmt_binlog_row_based && mysql_bin_log.is_open());
3533 3534 3535 3536 3537 3538 3539 3540
  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);

3541 3542
  if (is_trans && binlog_table_maps == 0)
    binlog_start_trans_and_stmt();
3543

3544
  if ((error= mysql_bin_log.write(&the_event)))
3545 3546
    DBUG_RETURN(error);

3547
  binlog_table_maps++;
3548 3549 3550 3551
  table->s->table_map_version= mysql_bin_log.table_map_version();
  DBUG_RETURN(0);
}

3552 3553 3554 3555
Rows_log_event*
THD::binlog_get_pending_rows_event() const
{
  binlog_trx_data *const trx_data=
3556
    (binlog_trx_data*) ha_data[binlog_hton->slot];
3557 3558 3559 3560 3561 3562
  /*
    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.
   */
3563
  return trx_data ? trx_data->pending() : NULL;
3564 3565 3566 3567 3568
}

void
THD::binlog_set_pending_rows_event(Rows_log_event* ev)
{
3569
  if (ha_data[binlog_hton->slot] == NULL)
3570 3571
    binlog_setup_trx_data();

3572
  binlog_trx_data *const trx_data=
3573
    (binlog_trx_data*) ha_data[binlog_hton->slot];
3574

3575
  DBUG_ASSERT(trx_data);
3576
  trx_data->set_pending(ev);
3577 3578 3579 3580 3581 3582 3583 3584
}


/*
  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.
*/
3585 3586 3587
int
MYSQL_BIN_LOG::flush_and_set_pending_rows_event(THD *thd,
                                                Rows_log_event* event)
3588
{
3589
  DBUG_ENTER("MYSQL_BIN_LOG::flush_and_set_pending_rows_event(event)");
3590
  DBUG_ASSERT(mysql_bin_log.is_open());
3591
  DBUG_PRINT("enter", ("event: 0x%lx", (long) event));
3592 3593 3594 3595

  int error= 0;

  binlog_trx_data *const trx_data=
3596
    (binlog_trx_data*) thd->ha_data[binlog_hton->slot];
3597 3598 3599

  DBUG_ASSERT(trx_data);

3600
  DBUG_PRINT("info", ("trx_data->pending(): 0x%lx", (long) trx_data->pending()));
3601

3602
  if (Rows_log_event* pending= trx_data->pending())
3603 3604 3605 3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655 3656 3657 3658 3659 3660 3661 3662
  {
    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);
  }

3663
  thd->binlog_set_pending_rows_event(event);
3664 3665 3666 3667

  DBUG_RETURN(error);
}

3668 3669 3670 3671
/*
  Write an event to the binary log
*/

3672
bool MYSQL_BIN_LOG::write(Log_event *event_info)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3673
{
3674 3675
  THD *thd= event_info->thd;
  bool error= 1;
3676
  DBUG_ENTER("MYSQL_BIN_LOG::write(Log_event *)");
3677

3678 3679
  if (thd->binlog_evt_union.do_union)
  {
3680 3681 3682 3683
    /*
      In Stored function; Remember that function call caused an update.
      We will log the function call to the binary log on function exit
    */
3684 3685 3686 3687
    thd->binlog_evt_union.unioned_events= TRUE;
    thd->binlog_evt_union.unioned_events_trans |= event_info->cache_stmt;
    DBUG_RETURN(0);
  }
3688 3689 3690 3691 3692 3693 3694

  /*
    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.

3695 3696 3697
    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.
3698
  */
3699 3700 3701
  bool const end_stmt=
    thd->prelocked_mode && thd->lex->requires_prelocking();
  thd->binlog_flush_pending_rows_event(end_stmt);
3702

3703 3704
  pthread_mutex_lock(&LOCK_log);

3705
  /*
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3706 3707 3708 3709
     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.
  */
3710
  if (likely(is_open()))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3711
  {
3712
    IO_CACHE *file= &log_file;
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
3713
#ifdef HAVE_REPLICATION
3714
    /*
3715 3716 3717
      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)"
3718
    */
3719
    const char *local_db= event_info->get_db();
3720
    if ((thd && !(thd->options & OPTION_BIN_LOG)) ||
lars@mysql.com's avatar
Merge  
lars@mysql.com committed
3721
	(!binlog_filter->db_ok(local_db)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3722
    {
3723
      VOID(pthread_mutex_unlock(&LOCK_log));
3724 3725 3726
      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)));
3727
      DBUG_RETURN(0);
3728
    }
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
3729
#endif /* HAVE_REPLICATION */
3730

3731
#if defined(USING_TRANSACTIONS) 
3732 3733 3734 3735 3736 3737 3738
    /*
      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)
3739
      - or if the event asks for it (cache_stmt == TRUE).
3740 3741 3742
    */
    if (opt_using_transactions && thd)
    {
3743 3744
      if (thd->binlog_setup_trx_data())
        goto err;
3745

3746
      binlog_trx_data *const trx_data=
3747
        (binlog_trx_data*) thd->ha_data[binlog_hton->slot];
3748
      IO_CACHE *trans_log= &trx_data->trans_log;
3749 3750
      my_off_t trans_log_pos= my_b_tell(trans_log);
      if (event_info->get_cache_stmt() || trans_log_pos != 0)
3751
      {
3752
        DBUG_PRINT("info", ("Using trans_log: cache: %d, trans_log_pos: %lu",
3753
                            event_info->get_cache_stmt(),
3754
                            (ulong) trans_log_pos));
3755 3756
        if (trans_log_pos == 0)
          thd->binlog_start_trans_and_stmt();
3757
        file= trans_log;
3758
      }
3759
      /*
3760
        TODO as Mats suggested, for all the cases above where we write to
3761 3762
        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
3763
        LOCK_log.
3764
      */
3765
    }
3766
#endif /* USING_TRANSACTIONS */
3767
    DBUG_PRINT("info",("event type: %d",event_info->get_type_code()));
3768

3769 3770 3771 3772
    /*
      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
3773 3774

    /*
3775 3776
      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
3777 3778
    */

3779 3780 3781 3782
    /*
      If row-based binlogging, Insert_id, Rand and other kind of "setting
      context" events are not needed.
    */
3783
    if (thd)
3784
    {
3785
      if (!thd->current_stmt_binlog_row_based)
3786
      {
3787
        if (thd->stmt_depends_on_first_successful_insert_id_in_prev_stmt)
3788 3789
        {
          Intvar_log_event e(thd,(uchar) LAST_INSERT_ID_EVENT,
3790
                             thd->first_successful_insert_id_in_prev_stmt_for_binlog);
3791 3792 3793
          if (e.write(file))
            goto err;
        }
3794
        if (thd->auto_inc_intervals_in_cur_stmt_for_binlog.nb_elements() > 0)
3795
        {
3796 3797 3798
          DBUG_PRINT("info",("number of auto_inc intervals: %u",
                             thd->auto_inc_intervals_in_cur_stmt_for_binlog.
                             nb_elements()));
3799 3800
          /*
            If the auto_increment was second in a table's index (possible with
3801
            MyISAM or BDB) (table->next_number_keypart != 0), such event is
3802 3803
            in fact not necessary. We could avoid logging it.
          */
3804 3805 3806
          Intvar_log_event e(thd, (uchar) INSERT_ID_EVENT,
                             thd->auto_inc_intervals_in_cur_stmt_for_binlog.
                             minimum());
3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820
          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;
3821
            get_dynamic(&thd->user_var_events,(uchar*) &user_var_event, i);
3822 3823 3824 3825 3826 3827 3828 3829 3830 3831
            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;
          }
        }
3832
      }
3833
    }
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
3834

3835 3836
    /*
       Write the SQL command
3837
     */
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
3838

monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
3839
    if (event_info->write(file))
3840
      goto err;
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
3841

3842
    if (file == &log_file) // we are writing to the real log (disk)
3843
    {
3844
      if (flush_and_sync())
monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
3845
	goto err;
3846 3847
      signal_update();
      rotate_and_purge(RP_LOCK_LOG_IS_ALREADY_LOCKED);
3848
    }
monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
3849
    error=0;
3850

3851 3852 3853 3854
err:
    if (error)
    {
      if (my_errno == EFBIG)
bell@sanja.is.com.ua's avatar
bell@sanja.is.com.ua committed
3855
	my_message(ER_TRANS_CACHE_FULL, ER(ER_TRANS_CACHE_FULL), MYF(0));
3856 3857 3858 3859 3860
      else
	my_error(ER_ERROR_ON_WRITE, MYF(0), name, errno);
      write_error=1;
    }
  }
3861

3862 3863 3864
  if (event_info->flags & LOG_EVENT_UPDATE_TABLE_MAP_VERSION_F)
    ++m_table_map_version;

3865
  pthread_mutex_unlock(&LOCK_log);
3866
  DBUG_RETURN(error);
3867 3868
}

3869 3870 3871 3872 3873 3874 3875 3876 3877 3878 3879 3880 3881 3882 3883 3884 3885 3886 3887 3888 3889 3890 3891 3892 3893 3894 3895 3896

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;
}

3897
void MYSQL_BIN_LOG::rotate_and_purge(uint flags)
3898
{
3899 3900
  if (!(flags & RP_LOCK_LOG_IS_ALREADY_LOCKED))
    pthread_mutex_lock(&LOCK_log);
3901 3902
  if ((flags & RP_FORCE_ROTATE) ||
      (my_b_tell(&log_file) >= (my_off_t) max_size))
3903
  {
3904
    new_file_without_locking();
3905 3906 3907
#ifdef HAVE_REPLICATION
    if (expire_logs_days)
    {
3908
      time_t purge_time= time(0) - expire_logs_days*24*60*60;
3909 3910 3911 3912 3913
      if (purge_time >= 0)
        purge_logs_before_date(purge_time);
    }
#endif
  }
3914 3915
  if (!(flags & RP_LOCK_LOG_IS_ALREADY_LOCKED))
    pthread_mutex_unlock(&LOCK_log);
3916
}
3917

3918
uint MYSQL_BIN_LOG::next_file_id()
3919 3920 3921 3922 3923 3924 3925 3926
{
  uint res;
  pthread_mutex_lock(&LOCK_log);
  res = file_id++;
  pthread_mutex_unlock(&LOCK_log);
  return res;
}

3927

3928 3929 3930 3931 3932 3933 3934 3935 3936 3937 3938 3939 3940 3941 3942 3943 3944 3945 3946 3947
/*
  Write the contents of a cache to the binary log.

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

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

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

  if (reinit_io_cache(cache, READ_CACHE, 0, 0, 0))
    return ER_ERROR_ON_WRITE;
3948 3949 3950 3951
  uint bytes= my_b_bytes_in_cache(cache), group, carry, hdr_offs;
  long val;
  uchar header[LOG_EVENT_HEADER_LEN];

3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965
  /*
    The events in the buffer have incorrect end_log_pos data
    (relative to beginning of group rather than absolute),
    so we'll recalculate them in situ so the binlog is always
    correct, even in the middle of a group. This is possible
    because we now know the start position of the group (the
    offset of this cache in the log, if you will); all we need
    to do is to find all event-headers, and add the position of
    the group to the end_log_pos of each event.  This is pretty
    straight forward, except that we read the cache in segments,
    so an event-header might end up on the cache-border and get
    split.
  */

3966
  group= (uint)my_b_tell(&log_file);
3967 3968
  hdr_offs= carry= 0;

3969 3970
  do
  {
3971 3972 3973 3974 3975 3976

    /*
      if we only got a partial header in the last iteration,
      get the other half now and process a full header.
    */
    if (unlikely(carry > 0))
3977 3978 3979
    {
      DBUG_ASSERT(carry < LOG_EVENT_HEADER_LEN);

3980
      /* assemble both halves */
3981 3982
      memcpy(&header[carry], (char *)cache->read_pos, LOG_EVENT_HEADER_LEN - carry);

3983
      /* fix end_log_pos */
3984 3985 3986
      val= uint4korr(&header[LOG_POS_OFFSET]) + group;
      int4store(&header[LOG_POS_OFFSET], val);

3987
      /* write the first half of the split header */
3988 3989 3990
      if (my_b_write(&log_file, header, carry))
        return ER_ERROR_ON_WRITE;

3991 3992 3993 3994
      /*
        copy fixed second half of header to cache so the correct
        version will be written later.
      */
3995 3996
      memcpy((char *)cache->read_pos, &header[carry], LOG_EVENT_HEADER_LEN - carry);

3997
      /* next event header at ... */
3998
      hdr_offs = uint4korr(&header[EVENT_LEN_OFFSET]) - carry;
3999 4000 4001 4002

      carry= 0;
    }

4003 4004
    /* if there is anything to write, process it. */

4005
    if (likely(bytes > 0))
4006
    {
4007
      /*
4008 4009 4010 4011
        process all event-headers in this (partial) cache.
        if next header is beyond current read-buffer,
        we'll get it later (though not necessarily in the
        very next iteration, just "eventually").
4012
      */
4013

4014
      while (hdr_offs < bytes)
4015
      {
4016 4017 4018 4019
        /*
          partial header only? save what we can get, process once
          we get the rest.
        */
4020

4021 4022 4023 4024 4025 4026 4027 4028 4029
        if (hdr_offs + LOG_EVENT_HEADER_LEN > bytes)
        {
          carry= bytes - hdr_offs;
          memcpy(header, (char *)cache->read_pos + hdr_offs, carry);
          bytes= hdr_offs;
        }
        else
        {
          /* we've got a full event-header, and it came in one piece */
4030

4031
          uchar *log_pos= (uchar *)cache->read_pos + hdr_offs + LOG_POS_OFFSET;
4032

4033 4034 4035
          /* fix end_log_pos */
          val= uint4korr(log_pos) + group;
          int4store(log_pos, val);
4036

4037 4038 4039
          /* next event header at ... */
          log_pos= (uchar *)cache->read_pos + hdr_offs + EVENT_LEN_OFFSET;
          hdr_offs += uint4korr(log_pos);
4040

4041
        }
4042
      }
4043 4044 4045 4046 4047 4048 4049 4050 4051 4052

      /*
        Adjust hdr_offs. Note that this doesn't mean it will necessarily
        be valid in the next iteration; if the current event is very long,
        it may take a couple of read-iterations (and subsequent fixings
        of hdr_offs) for it to become valid again.
        if we had a split header, hdr_offs was already fixed above.
      */
      if (carry == 0)
        hdr_offs -= bytes;
4053 4054 4055
    }

    /* Write data to the binary log file */
4056 4057
    if (my_b_write(&log_file, cache->read_pos, bytes))
      return ER_ERROR_ON_WRITE;
4058 4059
    cache->read_pos=cache->read_end;		// Mark buffer used up
  } while ((bytes=my_b_fill(cache)));
4060 4061 4062 4063 4064 4065 4066

  if (sync_log)
    flush_and_sync();

  return 0;                                     // All OK
}

4067 4068
/*
  Write a cached log entry to the binary log
4069

guilhem@mysql.com's avatar
guilhem@mysql.com committed
4070 4071
  SYNOPSIS
    write()
4072
    thd
guilhem@mysql.com's avatar
guilhem@mysql.com committed
4073
    cache		The cache to copy to the binlog
4074 4075
    commit_event        The commit event to print after writing the
                        contents of the cache.
guilhem@mysql.com's avatar
guilhem@mysql.com committed
4076

4077 4078
  NOTE
    - We only come here if there is something in the cache.
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
4079
    - The thing in the cache is always a complete transaction
4080 4081 4082 4083
    - '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
4084
      with BEGIN/COMMIT or BEGIN/ROLLBACK in the binary log.
4085 4086 4087
      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.
4088 4089
*/

4090
bool MYSQL_BIN_LOG::write(THD *thd, IO_CACHE *cache, Log_event *commit_event)
4091
{
4092
  DBUG_ENTER("MYSQL_BIN_LOG::write(THD *, IO_CACHE *, Log_event *)");
4093
  VOID(pthread_mutex_lock(&LOCK_log));
4094

4095 4096 4097
  /* NULL would represent nothing to replicate after ROLLBACK */
  DBUG_ASSERT(commit_event != NULL);

4098
  if (likely(is_open()))                       // Should always be true
4099
  {
4100
    /*
4101 4102 4103 4104
      We only bother to write to the binary log if there is anything
      to write.
     */
    if (my_b_tell(cache) > 0)
4105
    {
4106
      /*
4107 4108
        Log "BEGIN" at the beginning of the transaction.
        which may contain more than 1 SQL statement.
4109
      */
4110 4111 4112 4113 4114 4115 4116 4117 4118 4119 4120 4121 4122 4123 4124 4125 4126 4127 4128 4129 4130 4131 4132
      if (thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN))
      {
        Query_log_event qinfo(thd, STRING_WITH_LEN("BEGIN"), TRUE, FALSE);
        /*
          Imagine this is rollback due to net timeout, after all statements of
          the transaction succeeded. Then we want a zero-error code in BEGIN.
          In other words, if there was a really serious error code it's already
          in the statement's events, there is no need to put it also in this
          internally generated event, and as this event is generated late it
          would lead to false alarms.
          This is safer than thd->clear_error() against kills at shutdown.
        */
        qinfo.error_code= 0;
        /*
          Now this Query_log_event has artificial log_pos 0. It must be adjusted
          to reflect the real position in the log. Not doing it would confuse the
          slave: it would prevent this one from knowing where he is in the
          master's binlog, which would result in wrong positions being shown to
          the user, MASTER_POS_WAIT undue waiting etc.
        */
        if (qinfo.write(&log_file))
          goto err;
      }
4133

4134 4135
      if ((write_error= write_cache(cache, false, false)))
        goto err;
4136

4137 4138 4139 4140 4141 4142 4143 4144 4145 4146 4147 4148
      if (commit_event && commit_event->write(&log_file))
        goto err;
      if (flush_and_sync())
        goto err;
      DBUG_EXECUTE_IF("half_binlogged_transaction", abort(););
      if (cache->error)				// Error on read
      {
        sql_print_error(ER(ER_ERROR_ON_READ), cache->file_name, errno);
        write_error=1;				// Don't give more errors
        goto err;
      }
      signal_update();
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4149
    }
4150

4151 4152 4153 4154 4155 4156 4157 4158
    /*
      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.
    */
4159
    if (commit_event && commit_event->get_type_code() == XID_EVENT)
4160 4161 4162 4163 4164
    {
      pthread_mutex_lock(&LOCK_prep_xids);
      prepared_xids++;
      pthread_mutex_unlock(&LOCK_prep_xids);
    }
4165 4166
    else
      rotate_and_purge(RP_LOCK_LOG_IS_ALREADY_LOCKED);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4167
  }
4168
  VOID(pthread_mutex_unlock(&LOCK_log));
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
4169

4170
  DBUG_RETURN(0);
4171 4172

err:
4173 4174 4175 4176 4177
  if (!write_error)
  {
    write_error= 1;
    sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
  }
4178
  VOID(pthread_mutex_unlock(&LOCK_log));
4179
  DBUG_RETURN(1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4180 4181
}

4182

4183 4184 4185 4186 4187 4188
/*
  Wait until we get a signal that the binary log has been updated

  SYNOPSIS
    wait_for_update()
    thd			Thread variable
4189
    is_slave            If 0, the caller is the Binlog_dump thread from master;
4190 4191
                        if 1, the caller is the SQL thread from the slave. This
                        influences only thd->proc_info.
4192 4193 4194

  NOTES
    One must have a lock on LOCK_log before calling this function.
4195
    This lock will be released before return! That's required by
guilhem@mysql.com's avatar
guilhem@mysql.com committed
4196
    THD::enter_cond() (see NOTES in sql_class.h).
4197 4198
*/

4199
void MYSQL_BIN_LOG::wait_for_update(THD* thd, bool is_slave)
4200
{
monty@mysql.com's avatar
monty@mysql.com committed
4201 4202
  const char *old_msg;
  DBUG_ENTER("wait_for_update");
4203

monty@mysql.com's avatar
monty@mysql.com committed
4204
  old_msg= thd->enter_cond(&update_cond, &LOCK_log,
4205
                           is_slave ?
monty@mysql.com's avatar
monty@mysql.com committed
4206
                           "Has read all relay log; waiting for the slave I/O "
4207
                           "thread to update it" :
monty@mysql.com's avatar
monty@mysql.com committed
4208
                           "Has sent all binlog to slave; waiting for binlog "
4209
                           "to be updated");
4210 4211
  pthread_cond_wait(&update_cond, &LOCK_log);
  thd->exit_cond(old_msg);
monty@mysql.com's avatar
monty@mysql.com committed
4212
  DBUG_VOID_RETURN;
4213
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4214

4215

4216 4217 4218 4219 4220
/*
  Close the log file

  SYNOPSIS
    close()
4221 4222 4223 4224 4225
    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
4226 4227 4228 4229

  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
4230 4231
*/

4232
void MYSQL_BIN_LOG::close(uint exiting)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4233
{					// One can't set log_type here!
4234
  DBUG_ENTER("MYSQL_BIN_LOG::close");
4235
  DBUG_PRINT("enter",("exiting: %d", (int) exiting));
4236
  if (log_state == LOG_OPENED)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4237
  {
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
4238
#ifdef HAVE_REPLICATION
4239 4240
    if (log_type == LOG_BIN && !no_auto_events &&
	(exiting & LOG_CLOSE_STOP_EVENT))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4241 4242
    {
      Stop_log_event s;
4243
      s.write(&log_file);
4244
      bytes_written+= s.data_written;
4245
      signal_update();
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4246
    }
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
4247
#endif /* HAVE_REPLICATION */
4248 4249

    /* don't pwrite in a file opened with O_APPEND - it doesn't work */
4250
    if (log_file.type == WRITE_CACHE && log_type == LOG_BIN)
4251 4252
    {
      my_off_t offset= BIN_LOG_HEADER_SIZE + FLAGS_OFFSET;
4253 4254
      my_off_t org_position= my_tell(log_file.file, MYF(0));
      uchar flags= 0;            // clearing LOG_EVENT_BINLOG_IN_USE_F
4255
      my_pwrite(log_file.file, &flags, 1, offset, MYF(0));
4256 4257 4258 4259 4260 4261 4262
      /*
        Restore position so that anything we have in the IO_cache is written
        to the correct position.
        We need the seek here, as my_pwrite() is not guaranteed to keep the
        original position on system that doesn't support pwrite().
      */
      my_seek(log_file.file, org_position, MY_SEEK_SET, MYF(0));
4263 4264
    }

4265 4266
    /* 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
4267
  }
4268 4269 4270 4271 4272 4273

  /*
    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.
  */

4274
  if ((exiting & LOG_CLOSE_INDEX) && my_b_inited(&index_file))
4275
  {
4276 4277
    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
4278
    {
4279 4280
      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
4281
    }
4282
  }
4283
  log_state= (exiting & LOG_CLOSE_TO_BE_OPENED) ? LOG_TO_BE_OPENED : LOG_CLOSED;
4284
  safeFree(name);
4285
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4286 4287 4288
}


4289
void MYSQL_BIN_LOG::set_max_size(ulong max_size_arg)
4290 4291 4292 4293 4294 4295 4296 4297
{
  /*
    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.
  */
4298
  DBUG_ENTER("MYSQL_BIN_LOG::set_max_size");
guilhem@mysql.com's avatar
guilhem@mysql.com committed
4299
  pthread_mutex_lock(&LOCK_log);
4300 4301
  if (is_open())
    max_size= max_size_arg;
guilhem@mysql.com's avatar
guilhem@mysql.com committed
4302 4303
  pthread_mutex_unlock(&LOCK_log);
  DBUG_VOID_RETURN;
4304
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4305

4306

4307 4308
/*
  Check if a string is a valid number
4309 4310 4311 4312 4313 4314 4315 4316 4317 4318 4319 4320 4321 4322

  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
4323
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4324 4325 4326 4327 4328 4329 4330 4331 4332 4333 4334 4335

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++;
4336 4337
  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
4338 4339 4340 4341 4342 4343 4344
  {
    flag=1;
    str++;
  }
  if (*str == '.')
  {
    for (str++ ;
4345
	 my_isdigit(files_charset_info,*str) ||
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4346 4347 4348 4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359 4360 4361 4362 4363 4364
	   (allow_wildcards && (*str == wild_many || *str == wild_one)) ;
	 str++, flag=1) ;
  }
  if (*str != 0 || flag == 0)
    DBUG_RETURN(0);
  if (res)
    *res=atol(start);
  DBUG_RETURN(1);			/* Number ok */
} /* test_if_number */


void sql_perror(const char *message)
{
#ifdef HAVE_STRERROR
  sql_print_error("%s: %s",message, strerror(errno));
#else
  perror(message);
#endif
}
4365

monty@mysql.com's avatar
monty@mysql.com committed
4366

4367 4368 4369
bool flush_error_log()
{
  bool result=0;
4370
  if (opt_error_log)
4371 4372 4373 4374
  {
    char err_renamed[FN_REFLEN], *end;
    end= strmake(err_renamed,log_error_file,FN_REFLEN-4);
    strmov(end, "-old");
4375
    VOID(pthread_mutex_lock(&LOCK_error_log));
4376 4377 4378 4379 4380 4381
#ifdef __WIN__
    char err_temp[FN_REFLEN+4];
    /*
     On Windows is necessary a temporary file for to rename
     the current error file.
    */
4382
    strxmov(err_temp, err_renamed,"-tmp",NullS);
4383 4384 4385
    (void) my_delete(err_temp, MYF(0)); 
    if (freopen(err_temp,"a+",stdout))
    {
4386 4387 4388 4389
      int fd;
      size_t bytes;
      uchar buf[IO_SIZE];

4390 4391 4392 4393 4394
      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);
4395

4396 4397
      if ((fd = my_open(err_temp, O_RDONLY, MYF(0))) >= 0)
      {
4398 4399 4400
        while ((bytes= my_read(fd, buf, IO_SIZE, MYF(0))) &&
               bytes != MY_FILE_ERROR)
          my_fwrite(stderr, buf, bytes, MYF(0));
4401 4402 4403 4404 4405 4406 4407 4408 4409 4410 4411 4412 4413
        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
4414
    VOID(pthread_mutex_unlock(&LOCK_error_log));
4415 4416 4417
  }
   return result;
}
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
4418

4419
void MYSQL_BIN_LOG::signal_update()
monty@mysql.com's avatar
monty@mysql.com committed
4420
{
4421
  DBUG_ENTER("MYSQL_BIN_LOG::signal_update");
monty@mysql.com's avatar
monty@mysql.com committed
4422 4423 4424 4425
  pthread_cond_broadcast(&update_cond);
  DBUG_VOID_RETURN;
}

4426
#ifdef __NT__
4427
static void print_buffer_to_nt_eventlog(enum loglevel level, char *buff,
4428
                                        size_t length, size_t buffLen)
4429 4430
{
  HANDLE event;
4431
  char   *buffptr= buff;
serg@serg.mylan's avatar
serg@serg.mylan committed
4432
  DBUG_ENTER("print_buffer_to_nt_eventlog");
4433

4434 4435
  /* Add ending CR/LF's to string, overwrite last chars if necessary */
  strmov(buffptr+min(length, buffLen-5), "\r\n\r\n");
4436

monty@mysql.com's avatar
monty@mysql.com committed
4437 4438
  setup_windows_event_source();
  if ((event= RegisterEventSource(NULL,"MySQL")))
4439
  {
monty@mysql.com's avatar
monty@mysql.com committed
4440
    switch (level) {
4441
      case ERROR_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
4442
        ReportEvent(event, EVENTLOG_ERROR_TYPE, 0, MSG_DEFAULT, NULL, 1, 0,
4443
                    (LPCSTR*)&buffptr, NULL);
4444 4445
        break;
      case WARNING_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
4446
        ReportEvent(event, EVENTLOG_WARNING_TYPE, 0, MSG_DEFAULT, NULL, 1, 0,
4447
                    (LPCSTR*) &buffptr, NULL);
4448 4449
        break;
      case INFORMATION_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
4450
        ReportEvent(event, EVENTLOG_INFORMATION_TYPE, 0, MSG_DEFAULT, NULL, 1,
4451
                    0, (LPCSTR*) &buffptr, NULL);
4452 4453 4454 4455 4456 4457 4458
        break;
    }
    DeregisterEventSource(event);
  }

  DBUG_VOID_RETURN;
}
monty@mysql.com's avatar
monty@mysql.com committed
4459 4460
#endif /* __NT__ */

4461 4462

/*
monty@mysql.com's avatar
monty@mysql.com committed
4463 4464
  Prints a printf style message to the error log and, under NT, to the
  Windows event log.
4465 4466 4467

  SYNOPSIS
    vprint_msg_to_log()
monty@mysql.com's avatar
monty@mysql.com committed
4468 4469
    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
4470
    args                   va_list list of arguments for the message
4471 4472 4473 4474

  NOTE

  IMPLEMENTATION
monty@mysql.com's avatar
monty@mysql.com committed
4475 4476
    This function prints the message into a buffer and then sends that buffer
    to other functions to write that message to other logging sources.
4477 4478

  RETURN VALUES
4479 4480 4481
    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
4482
*/
monty@mysql.com's avatar
monty@mysql.com committed
4483

4484
#ifdef EMBEDDED_LIBRARY
rburnett@production.mysql.com's avatar
rburnett@production.mysql.com committed
4485
int vprint_msg_to_log(enum loglevel level __attribute__((unused)),
4486 4487
                       const char *format __attribute__((unused)),
                       va_list argsi __attribute__((unused)))
rburnett@production.mysql.com's avatar
rburnett@production.mysql.com committed
4488 4489 4490 4491
{
  DBUG_ENTER("vprint_msg_to_log");
  DBUG_RETURN(0);
}
4492
#else /*!EMBEDDED_LIBRARY*/
4493 4494 4495 4496 4497 4498 4499 4500 4501 4502 4503 4504 4505
static void print_buffer_to_file(enum loglevel level, const char *buffer)
{
  time_t skr;
  struct tm tm_tmp;
  struct tm *start;
  DBUG_ENTER("print_buffer_to_file");
  DBUG_PRINT("enter",("buffer: %s", buffer));

  VOID(pthread_mutex_lock(&LOCK_error_log));

  skr=time(NULL);
  localtime_r(&skr, &tm_tmp);
  start=&tm_tmp;
4506

4507 4508 4509 4510 4511 4512 4513 4514 4515 4516 4517 4518 4519 4520 4521 4522 4523 4524
  fprintf(stderr, "%02d%02d%02d %2d:%02d:%02d [%s] %s\n",
          start->tm_year % 100,
          start->tm_mon+1,
          start->tm_mday,
          start->tm_hour,
          start->tm_min,
          start->tm_sec,
          (level == ERROR_LEVEL ? "ERROR" : level == WARNING_LEVEL ?
           "Warning" : "Note"),
          buffer);

  fflush(stderr);

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


rburnett@production.mysql.com's avatar
rburnett@production.mysql.com committed
4525
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
4526 4527
{
  char   buff[1024];
4528
  size_t length;
4529
  DBUG_ENTER("vprint_msg_to_log");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4530

4531
  length= my_vsnprintf(buff, sizeof(buff), format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
4532
  print_buffer_to_file(level, buff);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4533 4534

#ifdef __NT__
monty@mysql.com's avatar
monty@mysql.com committed
4535
  print_buffer_to_nt_eventlog(level, buff, length, sizeof(buff));
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4536
#endif
4537

4538
  DBUG_RETURN(0);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4539
}
4540
#endif /*EMBEDDED_LIBRARY*/
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4541

4542

monty@mysql.com's avatar
monty@mysql.com committed
4543
void sql_print_error(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4544 4545
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
4546
  DBUG_ENTER("sql_print_error");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4547

serg@serg.mylan's avatar
serg@serg.mylan committed
4548
  va_start(args, format);
4549
  error_log_print(ERROR_LEVEL, format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
4550
  va_end(args);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4551 4552 4553 4554 4555

  DBUG_VOID_RETURN;
}


monty@mysql.com's avatar
monty@mysql.com committed
4556
void sql_print_warning(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4557 4558
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
4559
  DBUG_ENTER("sql_print_warning");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4560

serg@serg.mylan's avatar
serg@serg.mylan committed
4561
  va_start(args, format);
4562
  error_log_print(WARNING_LEVEL, format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
4563
  va_end(args);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4564 4565 4566 4567 4568

  DBUG_VOID_RETURN;
}


monty@mysql.com's avatar
monty@mysql.com committed
4569
void sql_print_information(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4570 4571
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
4572
  DBUG_ENTER("sql_print_information");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4573

serg@serg.mylan's avatar
serg@serg.mylan committed
4574
  va_start(args, format);
4575
  error_log_print(INFORMATION_LEVEL, format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
4576
  va_end(args);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4577

guilhem@mysql.com's avatar
guilhem@mysql.com committed
4578 4579
  DBUG_VOID_RETURN;
}
4580

4581

4582 4583 4584 4585 4586 4587 4588 4589
/********* 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
4590 4591
  states - active, syncing, pool.
  there could be only one page in active or syncing states,
4592
  but many in pool - pool is fifo queue.
4593
  usual lifecycle of a page is pool->active->syncing->pool
4594
  "active" page - is a page where new xid's are logged.
4595 4596
  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.
4597
  when the sync is done the page is moved to a pool and an active page
4598 4599 4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610
  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.
4611 4612 4613 4614
  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)

4615 4616
  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
4617 4618
  dirty - we don't sync removals to disk.
*/
4619 4620 4621 4622 4623

ulong tc_log_page_waits= 0;

#ifdef HAVE_MMAP

4624 4625
#define TC_LOG_HEADER_SIZE (sizeof(tc_log_magic)+1)

4626
static const char tc_log_magic[]={(char) 254, 0x23, 0x05, 0x74};
4627

4628
ulong opt_tc_log_size= TC_LOG_MIN_SIZE;
4629
ulong tc_log_max_pages_used=0, tc_log_page_size=0, tc_log_cur_pages_used=0;
4630 4631 4632 4633 4634 4635 4636

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

4637
  DBUG_ASSERT(total_ha_2pc > 1);
4638 4639
  DBUG_ASSERT(opt_name && opt_name[0]);

serg@serg.mylan's avatar
serg@serg.mylan committed
4640
  tc_log_page_size= my_getpagesize();
4641 4642 4643
  DBUG_ASSERT(TC_LOG_PAGE_SIZE % tc_log_page_size == 0);

  fn_format(logname,opt_name,mysql_data_home,"",MY_UNPACK_FILENAME);
4644
  if ((fd= my_open(logname, O_RDWR, MYF(0))) < 0)
4645
  {
4646 4647
    if (my_errno != ENOENT)
      goto err;
4648 4649
    if (using_heuristic_recover())
      return 1;
4650
    if ((fd= my_create(logname, CREATE_MODE, O_RDWR, MYF(MY_WME))) < 0)
4651 4652
      goto err;
    inited=1;
4653
    file_length= opt_tc_log_size;
4654 4655 4656 4657 4658
    if (my_chsize(fd, file_length, 0, MYF(MY_WME)))
      goto err;
  }
  else
  {
4659 4660
    inited= 1;
    crashed= TRUE;
serg@serg.mylan's avatar
serg@serg.mylan committed
4661
    sql_print_information("Recovering after a crash using %s", opt_name);
4662 4663 4664 4665 4666 4667
    if (tc_heuristic_recover)
    {
      sql_print_error("Cannot perform automatic crash recovery when "
                      "--tc-heuristic-recover is used");
      goto err;
    }
4668
    file_length= my_seek(fd, 0L, MY_SEEK_END, MYF(MY_WME+MY_FAE));
4669 4670 4671 4672
    if (file_length == MY_FILEPOS_ERROR || file_length % tc_log_page_size)
      goto err;
  }

4673
  data= (uchar *)my_mmap(0, (size_t)file_length, PROT_READ|PROT_WRITE,
4674 4675 4676 4677 4678 4679 4680 4681
                        MAP_NOSYNC|MAP_SHARED, fd, 0);
  if (data == MAP_FAILED)
  {
    my_errno=errno;
    goto err;
  }
  inited=2;

4682
  npages=(uint)file_length/tc_log_page_size;
4683 4684 4685 4686 4687 4688 4689 4690 4691 4692 4693 4694 4695 4696 4697 4698 4699
  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);
4700
  pages[0].start=pages[0].end-pages[0].size;
4701 4702 4703
  pages[npages-1].next=0;
  inited=4;

4704
  if (crashed && recover())
4705
      goto err;
4706

4707
  memcpy(data, tc_log_magic, sizeof(tc_log_magic));
4708
  data[sizeof(tc_log_magic)]= (uchar)total_ha_2pc;
4709 4710 4711 4712 4713 4714 4715 4716 4717 4718 4719
  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;

4720
  syncing= 0;
4721 4722 4723 4724 4725 4726 4727 4728 4729 4730 4731 4732
  active=pages;
  pool=pages+1;
  pool_last=pages+npages-1;

  return 0;

err:
  close();
  return 1;
}

/*
4733 4734
  there is no active page, let's got one from the pool

4735 4736 4737 4738 4739 4740 4741 4742 4743 4744 4745 4746
  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;

4747 4748 4749
  if (syncing)
    pthread_mutex_lock(&LOCK_pool);

4750 4751 4752
  do
  {
    best_p= p= &pool;
4753 4754
    if ((*p)->waiters == 0) // can the first page be used ?
      break;                // yes - take it.
4755

4756
    best_free=0;            // no - trying second strategy
4757 4758 4759 4760 4761 4762 4763 4764 4765 4766 4767 4768
    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;
4769
  if (active->free == active->size) // we've chosen an empty page
4770 4771 4772 4773 4774
  {
    tc_log_cur_pages_used++;
    set_if_bigger(tc_log_max_pages_used, tc_log_cur_pages_used);
  }

4775
  if ((*best_p)->next)              // unlink the page from the pool
4776 4777 4778
    *best_p=(*best_p)->next;
  else
    pool_last=*best_p;
4779 4780 4781

  if (syncing)
    pthread_mutex_unlock(&LOCK_pool);
4782 4783 4784 4785 4786 4787 4788 4789 4790 4791 4792
}

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);
4793
  return 1; // always return 1
4794 4795 4796
}

/*
4797 4798 4799 4800 4801 4802 4803 4804 4805 4806 4807 4808 4809 4810 4811 4812 4813 4814
  Record that transaction XID is committed on the persistent storage

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

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

  IMPLEMENTATION
   If tc_log == MYSQL_LOG then tc_log writes transaction to binlog and
   records XID in a special Xid_log_event.
   If tc_log = TC_LOG_MMAP then xid is written in a special memory-mapped
   log.
4815 4816

  RETURN
4817 4818 4819 4820 4821
    0  Error
    #  "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.
4822 4823
*/

4824
int TC_LOG_MMAP::log_xid(THD *thd, my_xid xid)
4825 4826 4827 4828 4829 4830 4831
{
  int err;
  PAGE *p;
  ulong cookie;

  pthread_mutex_lock(&LOCK_active);

4832 4833 4834 4835 4836 4837 4838
  /*
    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.
  */
4839 4840 4841
  while (unlikely(active && active->free == 0))
    pthread_cond_wait(&COND_active, &LOCK_active);

4842
  /* no active page ? take one from the pool */
4843 4844 4845 4846 4847 4848
  if (active == 0)
    get_active_from_pool();

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

4849
  /* searching for an empty slot */
4850 4851 4852 4853 4854 4855
  while (*p->ptr)
  {
    p->ptr++;
    DBUG_ASSERT(p->ptr < p->end);               // because p->free > 0
  }

4856 4857
  /* found! store xid there and mark the page dirty */
  cookie= (ulong)((uchar *)p->ptr - data);      // can never be zero
4858 4859 4860 4861 4862
  *p->ptr++= xid;
  p->free--;
  p->state= DIRTY;

  /* to sync or not to sync - this is the question */
4863
  pthread_mutex_unlock(&LOCK_active);
4864 4865 4866
  pthread_mutex_lock(&LOCK_sync);
  pthread_mutex_unlock(&p->lock);

4867
  if (syncing)
4868 4869
  {                                          // somebody's syncing. let's wait
    p->waiters++;
4870
    /*
monty@mishka.local's avatar
monty@mishka.local committed
4871
      note - it must be while (), not do ... while () here
4872 4873 4874
      as p->state may be not DIRTY when we come here
    */
    while (p->state == DIRTY && syncing)
4875 4876 4877 4878 4879 4880 4881 4882 4883 4884 4885
      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
4886 4887 4888 4889
  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
4890 4891 4892 4893 4894 4895 4896 4897 4898 4899 4900 4901 4902
  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;

4903
  DBUG_ASSERT(syncing != active);
4904 4905 4906 4907 4908

  /*
    sit down and relax - this can take a while...
    note - no locks are held at this point
  */
4909
  err= my_msync(fd, syncing->start, 1, MS_SYNC);
4910

4911
  /* page is synced. let's move it to the pool */
4912
  pthread_mutex_lock(&LOCK_pool);
4913 4914 4915 4916 4917
  pool_last->next=syncing;
  pool_last=syncing;
  syncing->next=0;
  syncing->state= err ? ERROR : POOL;
  pthread_cond_broadcast(&syncing->cond);    // signal "sync done"
4918 4919 4920
  pthread_cond_signal(&COND_pool);           // in case somebody's waiting
  pthread_mutex_unlock(&LOCK_pool);

4921
  /* marking 'syncing' slot free */
4922
  pthread_mutex_lock(&LOCK_sync);
4923
  syncing=0;
4924 4925 4926 4927 4928
  pthread_cond_signal(&active->cond);        // wake up a new syncer
  pthread_mutex_unlock(&LOCK_sync);
  return err;
}

4929 4930 4931 4932
/*
  erase xid from the page, update page free space counters/pointers.
  cookie points directly to the memory where xid was logged
*/
4933

4934 4935 4936 4937 4938 4939 4940 4941 4942 4943 4944 4945 4946
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);
4947
  if (p->free == p->size)               // the page is completely empty
4948 4949
    statistic_decrement(tc_log_cur_pages_used, &LOCK_status);
  if (p->waiters == 0)                 // the page is in pool and ready to rock
4950
    pthread_cond_signal(&COND_pool);   // ping ... for overflow()
4951 4952 4953 4954 4955
  pthread_mutex_unlock(&p->lock);
}

void TC_LOG_MMAP::close()
{
serg@serg.mylan's avatar
serg@serg.mylan committed
4956
  uint i;
4957 4958 4959 4960 4961 4962 4963 4964 4965
  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
4966
    for (i=0; i < npages; i++)
4967 4968 4969 4970 4971 4972 4973
    {
      if (pages[i].ptr == 0)
        break;
      pthread_mutex_destroy(&pages[i].lock);
      pthread_cond_destroy(&pages[i].cond);
    }
  case 3:
4974
    my_free((uchar*)pages, MYF(0));
4975
  case 2:
4976
    my_munmap((char*)data, (size_t)file_length);
4977 4978 4979 4980 4981 4982 4983 4984 4985 4986 4987 4988 4989 4990 4991 4992 4993 4994 4995 4996 4997 4998 4999 5000 5001
  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)
  {
5002
    sql_print_error("Recovery failed! You must enable "
5003 5004 5005
                    "exactly %d storage engines that support "
                    "two-phase commit protocol",
                    data[sizeof(tc_log_magic)]);
5006 5007 5008 5009
    goto err1;
  }

  if (hash_init(&xids, &my_charset_bin, tc_log_page_size/3, 0,
5010
                sizeof(my_xid), 0, 0, MYF(0)))
5011 5012 5013 5014 5015
    goto err1;

  for ( ; p < end_p ; p++)
  {
    for (my_xid *x=p->start; x < p->end; x++)
5016
      if (*x && my_hash_insert(&xids, (uchar *)x))
5017 5018 5019 5020 5021 5022 5023
        goto err2; // OOM
  }

  if (ha_recover(&xids))
    goto err2;

  hash_free(&xids);
5024
  bzero(data, (size_t)file_length);
5025 5026 5027 5028 5029 5030 5031 5032 5033 5034 5035
  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
5036 5037 5038 5039 5040
#endif

TC_LOG *tc_log;
TC_LOG_DUMMY tc_log_dummy;
TC_LOG_MMAP  tc_log_mmap;
5041

5042 5043 5044 5045 5046 5047 5048 5049 5050 5051 5052 5053
/*
  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.
*/

5054 5055 5056 5057 5058 5059 5060 5061 5062 5063 5064 5065 5066
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 ******/
5067
#define TC_LOG_BINLOG MYSQL_BIN_LOG
5068 5069 5070 5071 5072 5073 5074 5075

/*
  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!
*/

5076
int TC_LOG_BINLOG::open(const char *opt_name)
5077
{
5078 5079
  LOG_INFO log_info;
  int      error= 1;
5080 5081

  DBUG_ASSERT(total_ha_2pc > 1);
5082
  DBUG_ASSERT(opt_name && opt_name[0]);
5083 5084 5085 5086

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

5087 5088 5089 5090 5091 5092 5093
  if (!my_b_inited(&index_file))
  {
    /* There was a failure to open the index file, can't open the binlog */
    cleanup();
    return 1;
  }

5094
  if (using_heuristic_recover())
5095 5096 5097 5098
  {
    /* generate a new binlog to mask a corrupted one */
    open(opt_name, LOG_BIN, 0, WRITE_CACHE, 0, max_binlog_size, 0);
    cleanup();
5099
    return 1;
5100
  }
5101

5102
  if ((error= find_log_pos(&log_info, NullS, 1)))
5103
  {
5104 5105 5106 5107 5108
    if (error != LOG_INFO_EOF)
      sql_print_error("find_log_pos() failed (error: %d)", error);
    else
      error= 0;
    goto err;
5109
  }
5110 5111 5112 5113 5114

  {
    const char *errmsg;
    IO_CACHE    log;
    File        file;
5115 5116
    Log_event  *ev=0;
    Format_description_log_event fdle(BINLOG_VERSION);
5117
    char        log_name[FN_REFLEN];
5118 5119 5120 5121

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

5122
    do
5123
    {
5124 5125 5126 5127 5128 5129 5130
      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;
5131
    }
5132

5133
    if ((file= open_binlog(&log, log_name, &errmsg)) < 0)
5134 5135 5136 5137 5138
    {
      sql_print_error("%s", errmsg);
      goto err;
    }

5139 5140 5141
    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
5142 5143
    {
      sql_print_information("Recovering after a crash using %s", opt_name);
5144
      error= recover(&log, (Format_description_log_event *)ev);
serg@serg.mylan's avatar
serg@serg.mylan committed
5145
    }
5146 5147
    else
      error=0;
5148 5149 5150 5151 5152 5153 5154 5155 5156 5157

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

    if (error)
      goto err;
  }

err:
5158
  return error;
5159 5160
}

5161 5162
/* this is called on shutdown, after ha_panic */
void TC_LOG_BINLOG::close()
5163 5164 5165 5166 5167 5168
{
  DBUG_ASSERT(prepared_xids==0);
  pthread_mutex_destroy(&LOCK_prep_xids);
  pthread_cond_destroy (&COND_prep_xids);
}

5169 5170 5171 5172 5173 5174 5175
/*
  TODO group commit

  RETURN
         0  - error
         1  - success
*/
5176
int TC_LOG_BINLOG::log_xid(THD *thd, my_xid xid)
5177
{
5178
  DBUG_ENTER("TC_LOG_BINLOG::log");
5179
  Xid_log_event xle(thd, xid);
5180
  binlog_trx_data *trx_data=
5181
    (binlog_trx_data*) thd->ha_data[binlog_hton->slot];
5182 5183 5184 5185 5186
  /*
    We always commit the entire transaction when writing an XID. Also
    note that the return value is inverted.
   */
  DBUG_RETURN(!binlog_end_trans(thd, trx_data, &xle, TRUE));
5187 5188
}

5189
void TC_LOG_BINLOG::unlog(ulong cookie, my_xid xid)
5190
{
5191
  pthread_mutex_lock(&LOCK_prep_xids);
5192
  DBUG_ASSERT(prepared_xids > 0);
5193 5194
  if (--prepared_xids == 0) {
    DBUG_PRINT("info", ("prepared_xids=%lu", prepared_xids));
5195
    pthread_cond_signal(&COND_prep_xids);
5196
  }
5197
  pthread_mutex_unlock(&LOCK_prep_xids);
5198
  rotate_and_purge(0);     // as ::write() did not rotate
5199 5200
}

5201
int TC_LOG_BINLOG::recover(IO_CACHE *log, Format_description_log_event *fdle)
5202 5203 5204 5205 5206 5207
{
  Log_event  *ev;
  HASH xids;
  MEM_ROOT mem_root;

  if (! fdle->is_valid() ||
5208
      hash_init(&xids, &my_charset_bin, TC_LOG_PAGE_SIZE/3, 0,
5209
                sizeof(my_xid), 0, 0, MYF(0)))
5210 5211
    goto err1;

5212
  init_alloc_root(&mem_root, TC_LOG_PAGE_SIZE, TC_LOG_PAGE_SIZE);
5213

5214 5215
  fdle->flags&= ~LOG_EVENT_BINLOG_IN_USE_F; // abort on the first error

5216 5217 5218 5219 5220
  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;
5221 5222
      uchar *x= (uchar *) memdup_root(&mem_root, (uchar*) &xev->xid,
                                      sizeof(xev->xid));
5223 5224 5225 5226 5227 5228 5229 5230 5231 5232 5233 5234 5235 5236 5237 5238 5239 5240 5241 5242 5243 5244 5245 5246 5247
      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;
}

5248
struct st_mysql_storage_engine binlog_storage_engine=
5249
{ MYSQL_HANDLERTON_INTERFACE_VERSION };
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
5250 5251 5252 5253

mysql_declare_plugin(binlog)
{
  MYSQL_STORAGE_ENGINE_PLUGIN,
5254 5255
  &binlog_storage_engine,
  "binlog",
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
5256
  "MySQL AB",
5257
  "This is a pseudo storage engine to represent the binlog in a transaction",
5258
  PLUGIN_LICENSE_GPL,
5259
  binlog_init, /* Plugin Init */
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
5260 5261
  NULL, /* Plugin Deinit */
  0x0100 /* 1.0 */,
5262 5263 5264
  NULL,                       /* status variables                */
  NULL,                       /* system variables                */
  NULL                        /* config options                  */
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
5265 5266
}
mysql_declare_plugin_end;