log.cc 122 KB
Newer Older
1
/* Copyright (C) 2000-2003 MySQL AB
serg@serg.mylan's avatar
serg@serg.mylan committed
2

bk@work.mysql.com's avatar
bk@work.mysql.com committed
3 4 5 6
   This program is free software; you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
   the Free Software Foundation; either version 2 of the License, or
   (at your option) any later version.
serg@serg.mylan's avatar
serg@serg.mylan committed
7

bk@work.mysql.com's avatar
bk@work.mysql.com committed
8 9 10 11
   This program is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   GNU General Public License for more details.
serg@serg.mylan's avatar
serg@serg.mylan committed
12

bk@work.mysql.com's avatar
bk@work.mysql.com committed
13 14 15 16 17 18
   You should have received a copy of the GNU General Public License
   along with this program; if not, write to the Free Software
   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA */


/* logging of commands */
19
/* TODO: Abort logging when we get an error in reading or writing log files */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
20 21

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

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

rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
29 30 31 32
#ifdef __NT__
#include "message.h"
#endif

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

35 36 37 38 39 40 41 42 43 44 45
/* max size of the log message */
#define MAX_LOG_BUFFER_SIZE 1024
#define MAX_USER_HOST_SIZE 512
#define MAX_TIME_SIZE 32

/* we need this for log files intialization */
extern char *opt_logname, *opt_slow_logname;

LOGGER logger;

MYSQL_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 bool test_if_number(const char *str,
			   long *res, bool allow_wildcards);
50
static bool binlog_init();
51 52 53 54 55 56 57
static int binlog_close_connection(THD *thd);
static int binlog_savepoint_set(THD *thd, void *sv);
static int binlog_savepoint_rollback(THD *thd, void *sv);
static int binlog_commit(THD *thd, bool all);
static int binlog_rollback(THD *thd, bool all);
static int binlog_prepare(THD *thd, bool all);

58 59 60 61 62 63
/*
  This is a POD. Please keep it that way!

  Don't add constructors, destructors, or virtual functions.
*/
struct binlog_trx_data {
64 65
  bool empty() const
  {
66
#ifdef HAVE_ROW_BASED_REPLICATION
67
    return pending == NULL && my_b_tell(&trans_log) == 0;
68 69 70
#else
    return my_b_tell(&trans_log) == 0;
#endif
71
  }
72
  binlog_trx_data() {}
73
  IO_CACHE trans_log;                         // The transaction cache
74
#ifdef HAVE_ROW_BASED_REPLICATION
75
  Rows_log_event *pending;                // The pending binrows event
76
#endif
77 78
};

79 80 81 82
static const char binlog_hton_name[]= "binlog";
static const char binlog_hton_comment[]=
  "This is a meta storage engine to represent the binlog in a transaction";

83
handlerton binlog_hton = {
84
  MYSQL_HANDLERTON_INTERFACE_VERSION,
85
  binlog_hton_name,
86
  SHOW_OPTION_YES,
87
  binlog_hton_comment,
88
  DB_TYPE_BINLOG,               /* IGNORE  for now */
89
  binlog_init,
90 91 92 93 94 95 96 97 98 99 100
  0,
  sizeof(my_off_t),             /* savepoint size = binlog offset */
  binlog_close_connection,
  binlog_savepoint_set,
  binlog_savepoint_rollback,
  NULL,                         /* savepoint_release */
  binlog_commit,
  binlog_rollback,
  binlog_prepare,
  NULL,                         /* recover */
  NULL,                         /* commit_by_xid */
101 102 103
  NULL,                         /* rollback_by_xid */
  NULL,                         /* create_cursor_read_view */
  NULL,                         /* set_cursor_read_view */
104
  NULL,                         /* close_cursor_read_view */
105 106 107 108 109 110
  NULL,                         /* Create a new handler */
  NULL,                         /* Drop a database */
  NULL,                         /* Panic call */
  NULL,                         /* Start Consistent Snapshot */
  NULL,                         /* Flush logs */
  NULL,                         /* Show status */
111 112
  NULL,                         /* Partition flags */
  NULL,                         /* Alter table flags */
113
  NULL,                         /* Alter Tablespace */
114
  NULL,                         /* Fill FILES table */
115 116
  HTON_NOT_USER_SELECTABLE | HTON_HIDDEN,
  NULL,                         /* binlog_func */
117 118
  NULL,                         /* binlog_log_query */
  NULL				/* release_temporary_latches */
119 120
};

121

122 123 124 125 126 127 128

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

  SYNOPSIS
    open_log_table()

knielsen@mysql.com's avatar
knielsen@mysql.com committed
129 130
    log_type   type of the log table to open: QUERY_LOG_GENERAL
               or QUERY_LOG_SLOW
131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152

  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
*/

bool Log_to_csv_event_handler::open_log_table(uint log_type)
{
  THD *log_thd, *curr= current_thd;
  TABLE_LIST *table;
  bool error= FALSE;
  DBUG_ENTER("open_log_table");

  switch (log_type) {
153
  case QUERY_LOG_GENERAL:
154 155 156 157 158 159 160
    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;
161
  case QUERY_LOG_SLOW:
162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262
    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:
    DBUG_ASSERT(0);
  }

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

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

  log_thd->store_globals();

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

  if (simple_open_n_lock_tables(log_thd, table) ||
      table->table->file->extra(HA_EXTRA_MARK_AS_LOG_TABLE) ||
      table->table->file->ha_rnd_init(0))
    error= TRUE;
  else
    table->table->locked_by_logger= TRUE;

  /* restore thread settings */
  if (curr)
    curr->store_globals();
  else
  {
    my_pthread_setspecific_ptr(THR_THD,  0);
    my_pthread_setspecific_ptr(THR_MALLOC, 0);
  }

  DBUG_RETURN(error);
}


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

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


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

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


/*
  Reopen log table of a given type

  SYNOPSIS
    reopen_log_table()

knielsen@mysql.com's avatar
knielsen@mysql.com committed
263 264
    log_type   type of the log table to open: QUERY_LOG_GENERAL
               or QUERY_LOG_SLOW
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

  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
*/

bool Log_to_csv_event_handler::reopen_log_table(uint log_type)
{
  /* don't open the log table, if it wasn't enabled during startup */
  if (!logger.is_log_tables_initialized)
    return FALSE;
  return open_log_table(log_type);
}

void Log_to_csv_event_handler::cleanup()
{
291 292
  close_log_table(QUERY_LOG_GENERAL, FALSE);
  close_log_table(QUERY_LOG_SLOW, FALSE);
293 294 295 296 297 298 299 300 301
  logger.is_log_tables_initialized= FALSE;
}

/* log event handlers */

/*
  Log command to the general log table

  SYNOPSIS
302
    log_general()
303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326

    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,
327 328
              const char *sql_text, uint sql_text_len,
              CHARSET_INFO *client_cs)
329 330 331 332 333 334 335
{
  TABLE *table= general_log.table;

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

336 337 338 339 340
  /*
    NOTE: we do not call restore_record() here, as all fields are
    filled by the Logger (=> no need to load default ones).
  */

341 342 343
  /* log table entries are not replicated at the moment */
  tmp_disable_binlog(current_thd);

344
  /* Set current time. Required for CURRENT_TIMESTAMP to work */
345
  general_log_thd->start_time= event_time;
346 347 348 349 350

  /*
    We do not set a value for table->field[0], as it will use
    default value (which is CURRENT_TIMESTAMP).
  */
351

352
  table->field[1]->store(user_host, user_host_len, client_cs);
353
  table->field[1]->set_notnull();
354
  table->field[2]->store((longlong) thread_id);
355
  table->field[2]->set_notnull();
356
  table->field[3]->store((longlong) server_id);
357
  table->field[3]->set_notnull();
358
  table->field[4]->store(command_type, command_type_len, client_cs);
359
  table->field[4]->set_notnull();
360
  table->field[5]->store(sql_text, sql_text_len, client_cs);
361
  table->field[5]->set_notnull();
362 363 364 365 366 367 368 369 370 371 372 373
  table->file->ha_write_row(table->record[0]);

  reenable_binlog(current_thd);

  return FALSE;
}


/*
  Log a query to the slow log table

  SYNOPSIS
374
    log_slow()
375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406
    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;
407
  CHARSET_INFO *client_cs= thd->variables.character_set_client;
408

409
  DBUG_ENTER("log_slow");
410 411 412 413 414 415 416 417 418 419

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

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

  /*
     Set start time for CURRENT_TIMESTAMP to the start of the query.
420
     This will be default value for the field[0]
421 422
  */
  slow_log_thd->start_time= query_start_arg;
423
  restore_record(table, s->default_values);    // Get empty record
424

425 426 427 428
  /*
    We do not set a value for table->field[0], as it will use
    default value.
  */
429 430

  /* store the value */
431
  table->field[1]->store(user_host, user_host_len, client_cs);
432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451

  if (query_start_arg)
  {
    /* fill in query_time field */
    table->field[2]->store(query_time);
    /* lock_time */
    table->field[3]->store(lock_time);
    /* rows_sent */
    table->field[4]->store((longlong) thd->sent_row_count);
    /* rows_examined */
    table->field[5]->store((longlong) thd->examined_row_count);
  }
  else
  {
    table->field[2]->set_null();
    table->field[3]->set_null();
    table->field[4]->set_null();
    table->field[5]->set_null();
  }

452
  /* fill database field */
453
  if (thd->db)
454
  {
455
    table->field[6]->store(thd->db, thd->db_length, client_cs);
456 457
    table->field[6]->set_notnull();
  }
458 459

  if (thd->last_insert_id_used)
460
  {
461
    table->field[7]->store((longlong) thd->current_insert_id);
462 463
    table->field[7]->set_notnull();
  }
464 465 466

  /* set value if we do an insert on autoincrement column */
  if (thd->insert_id_used)
467
  {
468
    table->field[8]->store((longlong) thd->last_insert_id);
469 470
    table->field[8]->set_notnull();
  }
471 472

  table->field[9]->store((longlong) server_id);
473
  table->field[9]->set_notnull();
474 475

  /* sql_text */
476
  table->field[10]->store(sql_text,sql_text_len, client_cs);
477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531

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

  reenable_binlog(current_thd);

  DBUG_RETURN(0);
}

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

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

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


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

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


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

bool Log_to_file_event_handler::
  log_general(time_t event_time, const char *user_host,
              uint user_host_len, int thread_id,
              const char *command_type, uint command_type_len,
532 533
              const char *sql_text, uint sql_text_len,
              CHARSET_INFO *client_cs)
534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600
{
  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)
      mysql_slow_log.open_slow_log(opt_slow_logname);

    if (opt_log)
      mysql_log.open_query_log(opt_logname);

    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 */
  mysql_log.new_file(1);
  mysql_slow_log.new_file(1);
}

/*
  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
601
void LOGGER::cleanup_base()
602 603 604 605
{
  DBUG_ASSERT(inited == 1);
  (void) pthread_mutex_destroy(&LOCK_logger);
  if (table_log_handler)
petr@mysql.com's avatar
petr@mysql.com committed
606
  {
607
    table_log_handler->cleanup();
petr@mysql.com's avatar
petr@mysql.com committed
608 609
    delete table_log_handler;
  }
610 611 612 613 614
  if (file_log_handler)
    file_log_handler->cleanup();
}


petr@mysql.com's avatar
petr@mysql.com committed
615 616 617 618 619 620 621 622
void LOGGER::cleanup_end()
{
  DBUG_ASSERT(inited == 1);
  if (file_log_handler)
    delete file_log_handler;
}


623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646
void LOGGER::close_log_table(uint log_type, bool lock_in_use)
{
  table_log_handler->close_log_table(log_type, lock_in_use);
}


/*
  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 */
647
  init_error_log(LOG_FILE);
648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690

  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;
}


bool LOGGER::reopen_log_table(uint log_type)
{
  return table_log_handler->reopen_log_table(log_type);
}


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

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

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

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

691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720
  /* flush tables, in the case they are enabled */
  if (logger.is_log_tables_initialized)
  {
    /*
      This will lock and wait for all but the logger thread to release the
      tables. Then we could reopen log tables. Then release the name locks.

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

    /*
      Deny others from logging to general and slow log,
      while reopening tables.
    */
    logger.lock();

    table_log_handler->flush(table_log_handler->general_log_thd,
                             &close_slow_log, &close_general_log);

    /* end of log tables flush */
    logger.unlock();
  }
721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762
  return FALSE;
}


/*
  Log slow query with all enabled log event handlers

  SYNOPSIS
    slow_log_print()

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

  RETURN
    FALSE - OK
    TRUE - error occured
*/

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

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

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

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

763 764
    /* do not log slow queries from replication threads */
    if (thd->slave_thread)
765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863
      return 0;

    lock();

    /* 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 (thd->last_insert_id_used)
      last_insert_id= (longlong) thd->current_insert_id;

    /* set value if we do an insert on autoincrement column */
    if (thd->insert_id_used)
      insert_id= (longlong) thd->last_insert_id;

    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();
    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,
864 865
                   message_buff, message_buff_len,
                   thd->variables.character_set_client) || error;
866 867 868 869 870
    unlock();
  }
  return error;
}

871
void LOGGER::init_error_log(uint error_log_printer)
872
{
873 874
  if (error_log_printer & LOG_NONE)
  {
875
    error_log_handler_list[0]= 0;
876 877 878 879 880
    return;
  }

  switch (error_log_printer) {
  case LOG_FILE:
881 882 883 884
    error_log_handler_list[0]= file_log_handler;
    error_log_handler_list[1]= 0;
    break;
    /* these two are disabled for now */
885
  case LOG_TABLE:
886 887
    DBUG_ASSERT(0);
    break;
888
  case LOG_TABLE|LOG_FILE:
889 890 891 892 893
    DBUG_ASSERT(0);
    break;
  }
}

894
void LOGGER::init_slow_log(uint slow_log_printer)
895
{
896 897
  if (slow_log_printer & LOG_NONE)
  {
898
    slow_log_handler_list[0]= 0;
899 900 901 902 903
    return;
  }

  switch (slow_log_printer) {
  case LOG_FILE:
904 905 906
    slow_log_handler_list[0]= file_log_handler;
    slow_log_handler_list[1]= 0;
    break;
907
  case LOG_TABLE:
908 909 910
    slow_log_handler_list[0]= table_log_handler;
    slow_log_handler_list[1]= 0;
    break;
911
  case LOG_TABLE|LOG_FILE:
912 913 914 915 916 917 918
    slow_log_handler_list[0]= file_log_handler;
    slow_log_handler_list[1]= table_log_handler;
    slow_log_handler_list[2]= 0;
    break;
  }
}

919
void LOGGER::init_general_log(uint general_log_printer)
920
{
921 922
  if (general_log_printer & LOG_NONE)
  {
923
    general_log_handler_list[0]= 0;
924 925 926 927 928
    return;
  }

  switch (general_log_printer) {
  case LOG_FILE:
929 930 931
    general_log_handler_list[0]= file_log_handler;
    general_log_handler_list[1]= 0;
    break;
932
  case LOG_TABLE:
933 934 935
    general_log_handler_list[0]= table_log_handler;
    general_log_handler_list[1]= 0;
    break;
936
  case LOG_TABLE|LOG_FILE:
937 938 939 940 941 942 943 944 945 946 947 948
    general_log_handler_list[0]= file_log_handler;
    general_log_handler_list[1]= table_log_handler;
    general_log_handler_list[2]= 0;
    break;
  }
}


bool Log_to_csv_event_handler::flush(THD *thd, TABLE_LIST *close_slow_log,
                                     TABLE_LIST *close_general_log)
{
  VOID(pthread_mutex_lock(&LOCK_open));
949 950
  close_log_table(QUERY_LOG_GENERAL, TRUE);
  close_log_table(QUERY_LOG_SLOW, TRUE);
951 952 953 954 955
  close_general_log->next_local= close_slow_log;
  query_cache_invalidate3(thd, close_general_log, 0);
  unlock_table_name(thd, close_slow_log);
  unlock_table_name(thd, close_general_log);
  VOID(pthread_mutex_unlock(&LOCK_open));
knielsen@mysql.com's avatar
knielsen@mysql.com committed
956 957
  return reopen_log_table(QUERY_LOG_SLOW) ||
         reopen_log_table(QUERY_LOG_GENERAL);
958 959 960 961 962 963
}

/* the parameters are unused for the log tables */
bool Log_to_csv_event_handler::init()
{
  /* we always open log tables. even if the logging is disabled */
964
  return (open_log_table(QUERY_LOG_GENERAL) || open_log_table(QUERY_LOG_SLOW));
965 966
}

967 968 969
int LOGGER::set_handlers(uint error_log_printer,
                         uint slow_log_printer,
                         uint general_log_printer)
970 971
{
  /* error log table is not supported yet */
972
  DBUG_ASSERT(error_log_printer < LOG_TABLE);
973 974 975

  lock();

976
  if ((slow_log_printer & LOG_TABLE || general_log_printer & LOG_TABLE) &&
977 978
      !is_log_tables_initialized)
  {
979 980
    slow_log_printer= (slow_log_printer & ~LOG_TABLE) | LOG_FILE;
    general_log_printer= (general_log_printer & ~LOG_TABLE) | LOG_FILE;
981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001

    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()

knielsen@mysql.com's avatar
knielsen@mysql.com committed
1002 1003
    log_type       type of the log table to close: QUERY_LOG_GENERAL
                   or QUERY_LOG_SLOW
1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022
    lock_in_use    Set to TRUE if the caller owns LOCK_open. FALSE otherwise.

  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::
  close_log_table(uint log_type, bool lock_in_use)
{
  THD *log_thd, *curr= current_thd;
  TABLE_LIST *table;

  if (!logger.is_log_tables_initialized)
    return;                                     /* do nothing */

  switch (log_type) {
1023
  case QUERY_LOG_GENERAL:
1024 1025 1026
    log_thd= general_log_thd;
    table= &general_log;
    break;
1027
  case QUERY_LOG_SLOW:
1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060
    log_thd= slow_log_thd;
    table= &slow_log;
    break;
  default:
    DBUG_ASSERT(0);
  }

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

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

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


1061 1062 1063 1064 1065 1066
/*
  this function is mostly a placeholder.
  conceptually, binlog initialization (now mostly done in MYSQL_LOG::open)
  should be moved here.
*/

1067
bool binlog_init()
1068
{
1069
  return !opt_bin_log;
1070 1071 1072 1073
}

static int binlog_close_connection(THD *thd)
{
1074 1075 1076 1077
  binlog_trx_data *const trx_data=
    (binlog_trx_data*) thd->ha_data[binlog_hton.slot];
  IO_CACHE *trans_log= &trx_data->trans_log;
  DBUG_ASSERT(mysql_bin_log.is_open() && trx_data->empty());
1078
  close_cached_file(trans_log);
1079 1080
  thd->ha_data[binlog_hton.slot]= 0;
  my_free((gptr)trx_data, MYF(0));
1081 1082 1083
  return 0;
}

1084 1085
static int
binlog_end_trans(THD *thd, binlog_trx_data *trx_data, Log_event *end_ev)
1086
{
1087
  DBUG_ENTER("binlog_end_trans");
1088 1089 1090
  int error=0;
  IO_CACHE *trans_log= &trx_data->trans_log;

1091
  if (end_ev)
1092
  {
1093 1094 1095 1096 1097 1098
    /*
      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.
     */
1099
#ifdef HAVE_ROW_BASED_REPLICATION
1100
    thd->binlog_flush_pending_rows_event(TRUE);
1101
#endif
1102
    error= mysql_bin_log.write(thd, trans_log, end_ev);
1103
  }
1104
#ifdef HAVE_ROW_BASED_REPLICATION
1105 1106
  else
  {
1107
#ifdef HAVE_ROW_BASED_REPLICATION
1108
    thd->binlog_delete_pending_rows_event();
1109
#endif
1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122
  }

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

    We need to step the table map version after writing the
    transaction cache to disk.  In addition, we need to step the table
    map version on a rollback to ensure that a new table map event is
    generated instead of the one that was written to the thrown-away
    transaction cache.
  */
1123
  mysql_bin_log.update_table_map_version();
1124
#endif
1125

1126 1127 1128 1129 1130 1131 1132 1133
  statistic_increment(binlog_cache_use, &LOCK_status);
  if (trans_log->disk_writes != 0)
  {
    statistic_increment(binlog_cache_disk_use, &LOCK_status);
    trans_log->disk_writes= 0;
  }
  reinit_io_cache(trans_log, WRITE_CACHE, (my_off_t) 0, 0, 1); // cannot fail
  trans_log->end_of_file= max_binlog_cache_size;
1134
  DBUG_RETURN(error);
1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150
}

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

static int binlog_commit(THD *thd, bool all)
{
  DBUG_ENTER("binlog_commit");
1151 1152 1153
  binlog_trx_data *const trx_data=
    (binlog_trx_data*) thd->ha_data[binlog_hton.slot];
  IO_CACHE *trans_log= &trx_data->trans_log;
1154
  DBUG_ASSERT(mysql_bin_log.is_open() &&
1155
     (all || !(thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN))));
1156

1157
  if (trx_data->empty())
1158 1159 1160 1161
  {
    // we're here because trans_log was flushed in MYSQL_LOG::log()
    DBUG_RETURN(0);
  }
1162
  Query_log_event qev(thd, STRING_WITH_LEN("COMMIT"), TRUE, FALSE);
1163
  qev.error_code= 0; // see comment in MYSQL_LOG::write(THD, IO_CACHE)
1164
  DBUG_RETURN(binlog_end_trans(thd, trx_data, &qev));
1165 1166 1167 1168 1169
}

static int binlog_rollback(THD *thd, bool all)
{
  DBUG_ENTER("binlog_rollback");
1170 1171 1172 1173
  int error=0;
  binlog_trx_data *const trx_data=
    (binlog_trx_data*) thd->ha_data[binlog_hton.slot];
  IO_CACHE *trans_log= &trx_data->trans_log;
1174
  /*
serg@serg.mylan's avatar
serg@serg.mylan committed
1175 1176
    First assert is guaranteed - see trans_register_ha() call below.
    The second must be true. If it is not, we're registering
1177 1178
    unnecessary, doing extra work. The cause should be found and eliminated
  */
serg@serg.mylan's avatar
serg@serg.mylan committed
1179
  DBUG_ASSERT(all || !(thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN)));
1180
  DBUG_ASSERT(mysql_bin_log.is_open() && !trx_data->empty());
1181
  /*
1182 1183 1184 1185
    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...)
1186 1187
  */
  if (unlikely(thd->options & OPTION_STATUS_NO_TRANS_UPDATE))
1188
  {
1189
    Query_log_event qev(thd, STRING_WITH_LEN("ROLLBACK"), TRUE, FALSE);
1190
    qev.error_code= 0; // see comment in MYSQL_LOG::write(THD, IO_CACHE)
1191
    error= binlog_end_trans(thd, trx_data, &qev);
1192
  }
1193
  else
1194
    error= binlog_end_trans(thd, trx_data, 0);
1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221
  DBUG_RETURN(error);
}

/*
  NOTE: how do we handle this (unlikely but legal) case:
  [transaction] + [update to non-trans table] + [rollback to savepoint] ?
  The problem occurs when a savepoint is before the update to the
  non-transactional table. Then when there's a rollback to the savepoint, if we
  simply truncate the binlog cache, we lose the part of the binlog cache where
  the update is. If we want to not lose it, we need to write the SAVEPOINT
  command and the ROLLBACK TO SAVEPOINT command to the binlog cache. The latter
  is easy: it's just write at the end of the binlog cache, but the former
  should be *inserted* to the place where the user called SAVEPOINT. The
  solution is that when the user calls SAVEPOINT, we write it to the binlog
  cache (so no need to later insert it). As transactions are never intermixed
  in the binary log (i.e. they are serialized), we won't have conflicts with
  savepoint names when using mysqlbinlog or in the slave SQL thread.
  Then when ROLLBACK TO SAVEPOINT is called, if we updated some
  non-transactional table, we don't truncate the binlog cache but instead write
  ROLLBACK TO SAVEPOINT to it; otherwise we truncate the binlog cache (which
  will chop the SAVEPOINT command from the binlog cache, which is good as in
  that case there is no need to have it in the binlog).
*/

static int binlog_savepoint_set(THD *thd, void *sv)
{
  DBUG_ENTER("binlog_savepoint_set");
1222 1223
  binlog_trx_data *const trx_data=
    (binlog_trx_data*) thd->ha_data[binlog_hton.slot];
1224
  DBUG_ASSERT(mysql_bin_log.is_open() && my_b_tell(&trx_data->trans_log));
1225

1226
  *(my_off_t *)sv= my_b_tell(&trx_data->trans_log);
1227
  /* Write it to the binary log */
1228 1229 1230 1231 1232
  
  int const error=
    thd->binlog_query(THD::STMT_QUERY_TYPE,
                      thd->query, thd->query_length, TRUE, FALSE);
  DBUG_RETURN(error);
1233 1234 1235 1236 1237
}

static int binlog_savepoint_rollback(THD *thd, void *sv)
{
  DBUG_ENTER("binlog_savepoint_rollback");
1238 1239 1240
  binlog_trx_data *const trx_data=
    (binlog_trx_data*) thd->ha_data[binlog_hton.slot];
  IO_CACHE *trans_log= &trx_data->trans_log;
1241 1242 1243 1244 1245 1246 1247 1248 1249
  DBUG_ASSERT(mysql_bin_log.is_open() && my_b_tell(trans_log));

  /*
    Write ROLLBACK TO SAVEPOINT to the binlog cache if we have updated some
    non-transactional table. Otherwise, truncate the binlog cache starting
    from the SAVEPOINT command.
  */
  if (unlikely(thd->options & OPTION_STATUS_NO_TRANS_UPDATE))
  {
1250 1251 1252 1253
    int const error=
      thd->binlog_query(THD::STMT_QUERY_TYPE,
                        thd->query, thd->query_length, TRUE, FALSE);
    DBUG_RETURN(error);
1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283
  }
  reinit_io_cache(trans_log, WRITE_CACHE, *(my_off_t *)sv, 0, 0);
  DBUG_RETURN(0);
}

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

  if (my_b_read(log, (byte*) magic, sizeof(magic)))
  {
    *errmsg = "I/O error reading the header from the binary log";
    sql_print_error("%s, errno=%d, io cache code=%d", *errmsg, my_errno,
		    log->error);
    return 1;
  }
  if (memcmp(magic, BINLOG_MAGIC, sizeof(magic)))
  {
    *errmsg = "Binlog has bad magic number;  It's not a binary log file that can be used by this version of MySQL";
    return 1;
  }
  return 0;
}

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

1284 1285
  if ((file = my_open(log_file_name, O_RDONLY | O_BINARY | O_SHARE, 
                      MYF(MY_WME))) < 0)
1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311
  {
    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
1312

rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1313 1314
#ifdef __NT__
static int eventSource = 0;
serg@serg.mylan's avatar
serg@serg.mylan committed
1315

1316
void setup_windows_event_source()
monty@mysql.com's avatar
monty@mysql.com committed
1317
{
1318
  HKEY    hRegKey= NULL;
monty@mysql.com's avatar
monty@mysql.com committed
1319 1320 1321
  DWORD   dwError= 0;
  TCHAR   szPath[MAX_PATH];
  DWORD dwTypes;
1322

monty@mysql.com's avatar
monty@mysql.com committed
1323 1324 1325
  if (eventSource)               // Ensure that we are only called once
    return;
  eventSource= 1;
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1326

monty@mysql.com's avatar
monty@mysql.com committed
1327
  // Create the event source registry key
1328
  dwError= RegCreateKey(HKEY_LOCAL_MACHINE,
monty@mysql.com's avatar
monty@mysql.com committed
1329 1330
                          "SYSTEM\\CurrentControlSet\\Services\\EventLog\\Application\\MySQL", 
                          &hRegKey);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1331

monty@mysql.com's avatar
monty@mysql.com committed
1332 1333
  /* Name of the PE module that contains the message resource */
  GetModuleFileName(NULL, szPath, MAX_PATH);
serg@serg.mylan's avatar
serg@serg.mylan committed
1334

rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1335
  /* Register EventMessageFile */
1336
  dwError = RegSetValueEx(hRegKey, "EventMessageFile", 0, REG_EXPAND_SZ,
monty@mysql.com's avatar
monty@mysql.com committed
1337
                          (PBYTE) szPath, strlen(szPath)+1);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1338

monty@mysql.com's avatar
monty@mysql.com committed
1339 1340 1341 1342 1343
  /* 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
1344

monty@mysql.com's avatar
monty@mysql.com committed
1345
  RegCloseKey(hRegKey);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1346 1347
}

monty@mysql.com's avatar
monty@mysql.com committed
1348
#endif /* __NT__ */
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1349 1350


bk@work.mysql.com's avatar
bk@work.mysql.com committed
1351 1352 1353 1354 1355 1356 1357 1358
/****************************************************************************
** 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)
{
1359 1360 1361 1362
  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
1363
  reg1 struct fileinfo *file_info;
1364 1365
  ulong                 max_found=0;

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1366 1367
  DBUG_ENTER("find_uniq_filename");

1368 1369 1370 1371
  uint  length = dirname_part(buff,name);
  char *start  = name + length;
  char *end    = strend(start);

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1372
  *end='.';
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
1373
  length= (uint) (end-start+1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391

  if (!(dir_info = my_dir(buff,MYF(MY_DONT_SORT))))
  {						// This shouldn't happen
    strmov(end,".1");				// use name+1
    DBUG_RETURN(0);
  }
  file_info= dir_info->dir_entry;
  for (i=dir_info->number_off_files ; i-- ; file_info++)
  {
    if (bcmp(file_info->name,start,length) == 0 &&
	test_if_number(file_info->name+length, &number,0))
    {
      set_if_bigger(max_found,(ulong) number);
    }
  }
  my_dirend(dir_info);

  *end++='.';
1392
  sprintf(end,"%06ld",max_found+1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1393 1394 1395
  DBUG_RETURN(0);
}

1396

1397
MYSQL_LOG::MYSQL_LOG()
1398
  :bytes_written(0), last_time(0), query_start(0), name(0),
1399
   prepared_xids(0), log_type(LOG_CLOSED), file_id(1), open_count(1),
1400
   write_error(FALSE), inited(FALSE), need_start_event(TRUE),
1401
   m_table_map_version(0),
1402
   description_event_for_exec(0), description_event_for_queue(0)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1403 1404
{
  /*
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1405 1406
    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
1407
    called only in main(). Doing initialization here would make it happen
1408
    before main().
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1409
  */
1410 1411
  index_file_name[0] = 0;
  bzero((char*) &log_file,sizeof(log_file));
1412
  bzero((char*) &index_file, sizeof(index_file));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1413 1414
}

1415
/* this is called only once */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1416

monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
1417
void MYSQL_LOG::cleanup()
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1418
{
1419
  DBUG_ENTER("cleanup");
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1420
  if (inited)
1421
  {
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
1422
    inited= 0;
1423
    close(LOG_CLOSE_INDEX|LOG_CLOSE_STOP_EVENT);
1424 1425
    delete description_event_for_queue;
    delete description_event_for_exec;
1426 1427
    (void) pthread_mutex_destroy(&LOCK_log);
    (void) pthread_mutex_destroy(&LOCK_index);
1428
    (void) pthread_cond_destroy(&update_cond);
1429
  }
1430
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1431 1432
}

1433

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1434
int MYSQL_LOG::generate_new_name(char *new_name, const char *log_name)
1435
{
1436 1437
  fn_format(new_name,log_name,mysql_data_home,"",4);
  if (log_type != LOG_NORMAL)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450
  {
    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;
}

1451

1452
void MYSQL_LOG::init(enum_log_type log_type_arg,
1453
		     enum cache_type io_cache_type_arg,
1454 1455
		     bool no_auto_events_arg,
                     ulong max_size_arg)
1456
{
1457
  DBUG_ENTER("MYSQL_LOG::init");
1458
  log_type = log_type_arg;
1459
  io_cache_type = io_cache_type_arg;
1460
  no_auto_events = no_auto_events_arg;
1461
  max_size=max_size_arg;
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
1462
  DBUG_PRINT("info",("log_type: %d max_size: %lu", log_type, max_size));
1463
  DBUG_VOID_RETURN;
1464 1465
}

1466

guilhem@mysql.com's avatar
guilhem@mysql.com committed
1467 1468
void MYSQL_LOG::init_pthread_objects()
{
1469 1470
  DBUG_ASSERT(inited == 0);
  inited= 1;
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1471 1472 1473
  (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);
1474 1475
}

1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500
const char *MYSQL_LOG::generate_name(const char *log_name,
                                     const char *suffix,
                                     bool strip_ext, char *buff)
{
  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.
    */
    strmake(buff,glob_hostname,FN_REFLEN-5);
    strmov(fn_ext(buff),suffix);
    return (const char *)buff;
  }
  // get rid of extension if the log is binary to avoid problems
  if (strip_ext)
  {
    char *p = fn_ext(log_name);
    uint length=(uint) (p-log_name);
    strmake(buff,log_name,min(length,FN_REFLEN));
    return (const char*)buff;
  }
  return log_name;
}

1501

1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527
bool MYSQL_LOG::open_index_file(const char *index_file_name_arg,
                                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
1528
			0, MYF(MY_WME | MY_WAIT_IF_FULL)))
1529 1530 1531 1532 1533 1534 1535 1536
  {
    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
1537

1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551
/*
  Open a (new) log file.

  DESCRIPTION
  - If binary logs, also open the index file and register the new
    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
*/

1552 1553 1554 1555 1556
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,
                     bool no_auto_events_arg,
1557 1558
                     ulong max_size_arg,
                     bool null_created_arg)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1559
{
1560 1561
  char buff[FN_REFLEN];
  File file= -1;
1562
  int open_flags = O_CREAT | O_BINARY;
1563
  DBUG_ENTER("MYSQL_LOG::open");
1564
  DBUG_PRINT("enter",("log_type: %d",(int) log_type_arg));
1565 1566 1567

  last_time=query_start=0;
  write_error=0;
1568

1569
  init(log_type_arg,io_cache_type_arg,no_auto_events_arg,max_size_arg);
1570

1571
  if (!(name=my_strdup(log_name,MYF(MY_WME))))
1572 1573
  {
    name= (char *)log_name; // for the error message
1574
    goto err;
1575
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1576 1577 1578
  if (new_name)
    strmov(log_file_name,new_name);
  else if (generate_new_name(log_file_name, name))
1579
    goto err;
1580

1581
  if (io_cache_type == SEQ_READ_APPEND)
1582
    open_flags |= O_RDWR | O_APPEND;
1583
  else
1584
    open_flags |= O_WRONLY | (log_type == LOG_BIN ? 0 : O_APPEND);
1585

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1586
  db[0]=0;
1587
  open_count++;
1588
  if ((file=my_open(log_file_name,open_flags,
1589
		    MYF(MY_WME | ME_WAITTANG))) < 0 ||
1590
      init_io_cache(&log_file, file, IO_SIZE, io_cache_type,
1591 1592 1593
		    my_tell(file,MYF(MY_WME)), 0, 
                    MYF(MY_WME | MY_NABP |
                        ((log_type == LOG_BIN) ? MY_WAIT_IF_FULL : 0))))
1594
    goto err;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1595

1596 1597
  switch (log_type) {
  case LOG_NORMAL:
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1598
  {
1599
    char *end;
1600
    int len=my_snprintf(buff, sizeof(buff), "%s, Version: %s. "
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
1601
#ifdef EMBEDDED_LIBRARY
1602
		        "embedded library\n", my_progname, server_version
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
1603
#elif __NT__
1604 1605
			"started with:\nTCP Port: %d, Named Pipe: %s\n",
			my_progname, server_version, mysqld_port, mysqld_unix_port
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1606
#else
1607 1608
			"started with:\nTcp port: %d  Unix socket: %s\n",
			my_progname,server_version,mysqld_port,mysqld_unix_port
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1609
#endif
serg@serg.mylan's avatar
serg@serg.mylan committed
1610
                       );
1611 1612
    end=strnmov(buff+len,"Time                 Id Command    Argument\n",
                sizeof(buff)-len);
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
1613
    if (my_b_write(&log_file, (byte*) buff,(uint) (end-buff)) ||
1614 1615
	flush_io_cache(&log_file))
      goto err;
1616
    break;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1617
  }
1618
  case LOG_NEW:
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1619
  {
1620
    uint len;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1621 1622
    time_t skr=time(NULL);
    struct tm tm_tmp;
1623

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1624
    localtime_r(&skr,&tm_tmp);
1625 1626 1627 1628 1629 1630 1631 1632 1633 1634
    len= my_snprintf(buff,sizeof(buff),
		     "# %s, Version: %s at %02d%02d%02d %2d:%02d:%02d\n",
		     my_progname,server_version,
		     tm_tmp.tm_year % 100,
		     tm_tmp.tm_mon+1,
		     tm_tmp.tm_mday,
		     tm_tmp.tm_hour,
		     tm_tmp.tm_min,
		     tm_tmp.tm_sec);
    if (my_b_write(&log_file, (byte*) buff, len) ||
1635 1636
	flush_io_cache(&log_file))
      goto err;
1637
    break;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1638
  }
1639
  case LOG_BIN:
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1640
  {
1641 1642 1643 1644 1645 1646 1647 1648 1649 1650
    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.
      */
1651 1652
      if (my_b_safe_write(&log_file, (byte*) BINLOG_MAGIC,
			  BIN_LOG_HEADER_SIZE))
1653
        goto err;
1654 1655
      bytes_written+= BIN_LOG_HEADER_SIZE;
      write_file_name_to_index_file= 1;
1656
    }
1657

1658
    DBUG_ASSERT(my_b_inited(&index_file) != 0);
1659 1660
    reinit_io_cache(&index_file, WRITE_CACHE,
                    my_b_filelength(&index_file), 0, 0);
1661
    if (need_start_event && !no_auto_events)
1662
    {
1663 1664 1665 1666 1667
      /*
        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
1668 1669 1670 1671 1672 1673
      /*
        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;
1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692
      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
1693
        Format_description_log_event::Format_description_log_event(char* buf,).
1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707
        Why don't we want to write the description_event_for_queue if this
        event is for format<4 (3.23 or 4.x): this is because in that case, the
        description_event_for_queue describes the data received from the
        master, but not the data written to the relay log (*conversion*),
        which is in format 4 (slave's).
      */
      /*
        Set 'created' to 0, so that in next relay logs this event does not
        trigger cleaning actions on the slave in
        Format_description_log_event::exec_event().
      */
      description_event_for_queue->created= 0;
      /* Don't set log_pos in event header */
      description_event_for_queue->artificial_event=1;
1708

1709 1710 1711
      if (description_event_for_queue->write(&log_file))
        goto err;
      bytes_written+= description_event_for_queue->data_written;
1712
    }
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1713 1714
    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
1715
      goto err;
1716 1717 1718

    if (write_file_name_to_index_file)
    {
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1719 1720 1721 1722
      /*
        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.
      */
1723 1724 1725
      if (my_b_write(&index_file, (byte*) log_file_name,
		     strlen(log_file_name)) ||
	  my_b_write(&index_file, (byte*) "\n", 1) ||
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1726 1727
	  flush_io_cache(&index_file) ||
          my_sync(index_file.file, MYF(MY_WME)))
1728
	goto err;
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
1729
    }
1730
    break;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1731
  }
1732
  case LOG_CLOSED:				// Impossible
1733
  case LOG_TO_BE_OPENED:
1734 1735 1736 1737
    DBUG_ASSERT(1);
    break;
  }
  DBUG_RETURN(0);
1738 1739

err:
1740 1741 1742
  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, \
1743
shutdown the MySQL server and restart it.", name, errno);
1744 1745 1746
  if (file >= 0)
    my_close(file,MYF(0));
  end_io_cache(&log_file);
1747 1748
  end_io_cache(&index_file);
  safeFree(name);
1749
  log_type= LOG_CLOSED;
1750
  DBUG_RETURN(1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1751 1752
}

1753

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1754 1755 1756
int MYSQL_LOG::get_current_log(LOG_INFO* linfo)
{
  pthread_mutex_lock(&LOCK_log);
1757 1758
  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
1759 1760 1761 1762
  pthread_mutex_unlock(&LOCK_log);
  return 0;
}

1763

1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805
/*
  Move all data up in a file in an filename index file

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

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

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

  RETURN VALUES
    0	ok
*/

static bool copy_up_file_and_fill(IO_CACHE *index_file, my_off_t offset)
{
  int bytes_read;
  my_off_t init_offset= offset;
  File file= index_file->file;
  byte io_buf[IO_SIZE*2];
  DBUG_ENTER("copy_up_file_and_fill");

  for (;; offset+= bytes_read)
  {
    (void) my_seek(file, offset, MY_SEEK_SET, MYF(0));
    if ((bytes_read= (int) my_read(file, io_buf, sizeof(io_buf), MYF(MY_WME)))
	< 0)
      goto err;
    if (!bytes_read)
      break;					// end of file
    (void) my_seek(file, offset-init_offset, MY_SEEK_SET, MYF(0));
    if (my_write(file, (byte*) io_buf, bytes_read, MYF(MY_WME | MY_NABP)))
      goto err;
  }
  /* The following will either truncate the file or fill the end with \n' */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1806 1807
  if (my_chsize(file, offset - init_offset, '\n', MYF(MY_WME)) ||
      my_sync(file, MYF(MY_WME)))
1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827
    goto err;

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

err:
  DBUG_RETURN(1);
}


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

  SYNOPSIS
    find_log_pos()
    linfo		Store here the found log file name and position to
			the NEXT log file name in the index file.
    log_name		Filename to find in the index file.
			Is a null pointer if we want to read the first entry
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
1828
    need_lock		Set this to 1 if the parent doesn't already have a
1829 1830 1831
			lock on LOCK_index

  NOTE
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
1832 1833
    On systems without the truncate function the file will end with one or
    more empty lines.  These will be ignored when reading the file.
1834 1835 1836 1837 1838 1839 1840 1841 1842

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

int MYSQL_LOG::find_log_pos(LOG_INFO *linfo, const char *log_name,
			    bool need_lock)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1843
{
1844 1845 1846 1847 1848
  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
1849

1850
  /*
1851 1852
    Mutex needed because we need to make sure the file pointer does not
    move from under our feet
1853
  */
1854
  if (need_lock)
1855
    pthread_mutex_lock(&LOCK_index);
1856 1857 1858 1859 1860
  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);

1861
  for (;;)
1862
  {
1863
    uint length;
1864 1865 1866 1867
    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
1868
    {
1869 1870 1871
      /* 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
1872 1873
    }

1874 1875
    // if the log entry matches, null string matching anything
    if (!log_name ||
sasha@mysql.sashanet.com's avatar
sasha@mysql.sashanet.com committed
1876
	(log_name_len == length-1 && fname[log_name_len] == '\n' &&
1877
	 !memcmp(fname, log_name, log_name_len)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1878
    {
1879
      DBUG_PRINT("info",("Found log file entry"));
1880
      fname[length-1]=0;			// remove last \n
1881 1882
      linfo->index_file_start_offset= offset;
      linfo->index_file_offset = my_b_tell(&index_file);
1883
      break;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1884
    }
1885
  }
1886

1887
  if (need_lock)
1888
    pthread_mutex_unlock(&LOCK_index);
1889
  DBUG_RETURN(error);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1890
}
1891

1892

1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914
/*
  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
*/

1915
int MYSQL_LOG::find_next_log(LOG_INFO* linfo, bool need_lock)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1916
{
1917
  int error= 0;
1918
  uint length;
1919 1920
  char *fname= linfo->log_file_name;

1921 1922
  if (need_lock)
    pthread_mutex_lock(&LOCK_index);
1923 1924 1925 1926 1927 1928 1929 1930
  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)
1931
  {
1932
    error = !index_file.error ? LOG_INFO_EOF : LOG_INFO_IO;
1933 1934
    goto err;
  }
1935
  fname[length-1]=0;				// kill \n
1936
  linfo->index_file_offset = my_b_tell(&index_file);
1937

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1938
err:
1939 1940
  if (need_lock)
    pthread_mutex_unlock(&LOCK_index);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1941 1942 1943
  return error;
}

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
1944

1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963
/*
  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
*/

bool MYSQL_LOG::reset_logs(THD* thd)
1964 1965
{
  LOG_INFO linfo;
1966
  bool error=0;
1967 1968
  const char* save_name;
  enum_log_type save_log_type;
1969
  DBUG_ENTER("reset_logs");
1970

tomas@poseidon.ndb.mysql.com's avatar
tomas@poseidon.ndb.mysql.com committed
1971
  ha_reset_logs(thd);
1972 1973 1974 1975
  /*
    We need to get both locks to be sure that no one is trying to
    write to the index log file.
  */
1976
  pthread_mutex_lock(&LOCK_log);
1977
  pthread_mutex_lock(&LOCK_index);
1978

1979 1980 1981 1982 1983 1984 1985
  /*
    The following mutex is needed to ensure that no threads call
    'delete thd' as we would then risk missing a 'rollback' from this
    thread. If the transaction involved MyISAM tables, it should go
    into binlog even on rollback.
  */
  (void) pthread_mutex_lock(&LOCK_thread_count);
1986 1987 1988 1989 1990

  /* Save variables so that we can reopen the log */
  save_name=name;
  name=0;					// Protect against free
  save_log_type=log_type;
1991
  close(LOG_CLOSE_TO_BE_OPENED);
1992 1993 1994 1995

  /* First delete all old log files */

  if (find_log_pos(&linfo, NullS, 0))
1996 1997 1998 1999
  {
    error=1;
    goto err;
  }
2000

2001
  for (;;)
2002 2003
  {
    my_delete(linfo.log_file_name, MYF(MY_WME));
2004
    if (find_next_log(&linfo, 0))
2005 2006
      break;
  }
2007 2008

  /* Start logging with a new file */
2009
  close(LOG_CLOSE_INDEX);
2010 2011
  my_delete(index_file_name, MYF(MY_WME));	// Reset (open will update)
  if (!thd->slave_thread)
2012
    need_start_event=1;
2013 2014 2015
  if (!open_index_file(index_file_name, 0))
    open(save_name, save_log_type, 0,
         io_cache_type, no_auto_events, max_size, 0);
2016
  my_free((gptr) save_name, MYF(0));
2017

2018
err:
2019
  (void) pthread_mutex_unlock(&LOCK_thread_count);
2020
  pthread_mutex_unlock(&LOCK_index);
2021
  pthread_mutex_unlock(&LOCK_log);
2022
  DBUG_RETURN(error);
2023 2024
}

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2025

2026
/*
2027 2028 2029
  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.
2030 2031 2032

  SYNOPSIS
    purge_first_log()
2033 2034 2035 2036 2037
    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).
2038

2039 2040
  NOTE
    - This is only called from the slave-execute thread when it has read
2041 2042 2043 2044 2045
      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).
2046 2047 2048

  IMPLEMENTATION
    - Protects index file with LOCK_index
2049 2050
    - Delete relevant relay log files
    - Copy all file names after these ones to the front of the index file
2051
    - If the OS has truncate, truncate the file, else fill it with \n'
2052
    - Read the next file name from the index file and store in rli->linfo
2053 2054

  RETURN VALUES
2055 2056 2057 2058
    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
2059 2060
*/

hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
2061
#ifdef HAVE_REPLICATION
2062

2063
int MYSQL_LOG::purge_first_log(struct st_relay_log_info* rli, bool included) 
2064
{
2065
  int error;
2066 2067
  DBUG_ENTER("purge_first_log");

2068 2069
  DBUG_ASSERT(is_open());
  DBUG_ASSERT(rli->slave_running == 1);
2070
  DBUG_ASSERT(!strcmp(rli->linfo.log_file_name,rli->event_relay_log_name));
2071

2072
  pthread_mutex_lock(&LOCK_index);
2073 2074 2075 2076 2077 2078
  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
2079 2080

  /*
2081 2082 2083
    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
2084
  */
2085 2086 2087 2088 2089
  pthread_cond_broadcast(&rli->log_space_cond);
  
  /*
    Read the next log file name from the index file and pass it back to
    the caller
2090 2091
    If included is true, we want the first relay log;
    otherwise we want the one after event_relay_log_name.
2092
  */
2093 2094 2095 2096
  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)))))
2097 2098
  {
    char buff[22];
2099 2100 2101 2102 2103
    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);
2104 2105
    goto err;
  }
2106

2107
  /*
2108
    Reset rli's coordinates to the current log.
2109
  */
2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123
  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);
2124
    rli->notify_group_relay_log_name_update();
2125
  }
2126 2127 2128

  /* 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
2129

2130
err:
2131
  pthread_mutex_unlock(&LOCK_index);
2132
  DBUG_RETURN(error);
2133 2134
}

2135 2136 2137 2138
/*
  Update log index_file
*/

2139
int MYSQL_LOG::update_log_index(LOG_INFO* log_info, bool need_update_threads)
2140 2141 2142 2143 2144
{
  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
2145 2146
  if (need_update_threads)
    adjust_linfo_offsets(log_info->index_file_start_offset);
2147 2148
  return 0;
}
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2149

2150 2151 2152 2153 2154
/*
  Remove all logs before the given log from disk and from the index file.

  SYNOPSIS
    purge_logs()
2155 2156 2157 2158 2159 2160 2161
    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
2162 2163 2164 2165 2166 2167 2168 2169 2170 2171

  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
*/

2172 2173 2174 2175 2176
int MYSQL_LOG::purge_logs(const char *to_log, 
                          bool included,
                          bool need_mutex, 
                          bool need_update_threads, 
                          ulonglong *decrease_log_space)
2177 2178
{
  int error;
2179
  bool exit_loop= 0;
2180 2181
  LOG_INFO log_info;
  DBUG_ENTER("purge_logs");
2182
  DBUG_PRINT("info",("to_log= %s",to_log));
2183

2184 2185
  if (need_mutex)
    pthread_mutex_lock(&LOCK_index);
2186
  if ((error=find_log_pos(&log_info, to_log, 0 /*no mutex*/)))
2187 2188
    goto err;

2189
  /*
2190
    File name exists in index file; delete until we find this file
2191 2192 2193
    or a file that is used.
  */
  if ((error=find_log_pos(&log_info, NullS, 0 /*no mutex*/)))
2194
    goto err;
2195 2196
  while ((strcmp(to_log,log_info.log_file_name) || (exit_loop=included)) &&
         !log_in_use(log_info.log_file_name))
2197
  {
2198
    ulong file_size= 0;
2199 2200 2201
    if (decrease_log_space) //stat the file we want to delete
    {
      MY_STAT s;
2202 2203 2204 2205 2206 2207

      /* 
         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.
      */
2208
      if (my_stat(log_info.log_file_name,&s,MYF(0)))
2209
        file_size= s.st_size;
2210
      else
2211 2212
	sql_print_information("Failed to execute my_stat on file '%s'",
			      log_info.log_file_name);
2213 2214 2215 2216 2217 2218 2219
    }
    /*
      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)
2220
      *decrease_log_space-= file_size;
tomas@poseidon.ndb.mysql.com's avatar
tomas@poseidon.ndb.mysql.com committed
2221 2222 2223

    ha_binlog_index_purge_file(current_thd, log_info.log_file_name);

2224
    if (find_next_log(&log_info, 0) || exit_loop)
2225
      break;
2226
  }
2227

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2228 2229 2230 2231
  /*
    If we get killed -9 here, the sysadmin would have to edit
    the log index file after restart - otherwise, this should be safe
  */
2232
  error= update_log_index(&log_info, need_update_threads);
2233

2234
err:
2235 2236
  if (need_mutex)
    pthread_mutex_unlock(&LOCK_index);
2237
  DBUG_RETURN(error);
2238 2239
}

2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257
/*
  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
*/

2258
int MYSQL_LOG::purge_logs_before_date(time_t purge_time)
2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273
{
  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*/)))
2274
    goto err;
2275 2276 2277 2278 2279

  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 */
2280 2281
    if (!my_stat(log_info.log_file_name, &stat_area, MYF(0)) ||
	stat_area.st_mtime >= purge_time)
2282
      break;
2283
    my_delete(log_info.log_file_name, MYF(0));
tomas@poseidon.ndb.mysql.com's avatar
tomas@poseidon.ndb.mysql.com committed
2284 2285 2286

    ha_binlog_index_purge_file(current_thd, log_info.log_file_name);

2287 2288
    if (find_next_log(&log_info, 0))
      break;
2289
  }
2290

2291 2292 2293 2294
  /*
    If we get killed -9 here, the sysadmin would have to edit
    the log index file after restart - otherwise, this should be safe
  */
2295
  error= update_log_index(&log_info, 1);
2296

2297 2298
err:
  pthread_mutex_unlock(&LOCK_index);
2299
  DBUG_RETURN(error);
2300 2301
}

2302

hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
2303
#endif /* HAVE_REPLICATION */
2304

2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316

/*
  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
*/

bk@work.mysql.com's avatar
bk@work.mysql.com committed
2317 2318
void MYSQL_LOG::make_log_name(char* buf, const char* log_ident)
{
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2319 2320 2321 2322 2323
  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
2324 2325
}

2326 2327 2328 2329 2330 2331

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

bool MYSQL_LOG::is_active(const char *log_file_name_arg)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2332
{
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2333
  return !strcmp(log_file_name, log_file_name_arg);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2334 2335
}

2336 2337 2338 2339 2340 2341

/*
  Start writing to a new log file or reopen the old file

  SYNOPSIS
    new_file()
2342
    need_lock		Set to 1 if caller has not locked LOCK_log
2343 2344 2345 2346 2347 2348

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

void MYSQL_LOG::new_file(bool need_lock)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2349
{
2350 2351 2352
  char new_name[FN_REFLEN], *new_name_ptr, *old_name;
  enum_log_type save_log_type;

2353
  DBUG_ENTER("MYSQL_LOG::new_file");
2354
  if (!is_open())
2355 2356 2357 2358
  {
    DBUG_PRINT("info",("log is closed"));
    DBUG_VOID_RETURN;
  }
2359 2360 2361

  if (need_lock)
    pthread_mutex_lock(&LOCK_log);
2362 2363
  pthread_mutex_lock(&LOCK_index);

2364 2365 2366
  safe_mutex_assert_owner(&LOCK_log);
  safe_mutex_assert_owner(&LOCK_index);

2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384
  /*
    if binlog is used as tc log, be sure all xids are "unlogged",
    so that on recover we only need to scan one - latest - binlog file
    for prepared xids. As this is expected to be a rare event,
    simple wait strategy is enough. We're locking LOCK_log to be sure no
    new Xid_log_event's are added to the log (and prepared_xids is not
    increased), and waiting on COND_prep_xids for late threads to
    catch up.
  */
  if (prepared_xids)
  {
    tc_log_page_waits++;
    pthread_mutex_lock(&LOCK_prep_xids);
    while (prepared_xids)
      pthread_cond_wait(&COND_prep_xids, &LOCK_prep_xids);
    pthread_mutex_unlock(&LOCK_prep_xids);
  }

2385
  /* Reuse old name if not binlog and not update log */
2386
  new_name_ptr= name;
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2387

2388
  /*
2389 2390 2391
    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.
2392
  */
2393 2394 2395
  if (generate_new_name(new_name, name))
    goto end;
  new_name_ptr=new_name;
2396

2397
  if (log_type == LOG_BIN)
2398
  {
2399
    if (!no_auto_events)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2400
    {
2401
      /*
2402 2403
        We log the whole file name for log file as the user may decide
        to change base names at some point.
2404
      */
monty@mysql.com's avatar
monty@mysql.com committed
2405
      THD *thd = current_thd; /* may be 0 if we are reacting to SIGHUP */
2406
      Rotate_log_event r(new_name+dirname_length(new_name),
2407
                         0, LOG_EVENT_OFFSET, 0);
2408
      r.write(&log_file);
2409
      bytes_written += r.data_written;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2410
    }
2411 2412 2413 2414 2415
    /*
      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.
    */
2416
    signal_update();
2417 2418 2419 2420
  }
  old_name=name;
  save_log_type=log_type;
  name=0;				// Don't free name
2421
  close(LOG_CLOSE_TO_BE_OPENED);
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2422

2423
  /*
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2424
     Note that at this point, log_type != LOG_CLOSED (important for is_open()).
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2425
  */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2426

2427
  /*
2428
     new_file() is only used for rotation (in FLUSH LOGS or because size >
2429
     max_binlog_size or max_relay_log_size).
2430 2431 2432 2433
     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.
2434
  */
2435

2436 2437
  open(old_name, save_log_type, new_name_ptr,
       io_cache_type, no_auto_events, max_size, 1);
2438 2439
  my_free(old_name,MYF(0));

2440
end:
2441 2442
  if (need_lock)
    pthread_mutex_unlock(&LOCK_log);
2443 2444
  pthread_mutex_unlock(&LOCK_index);

2445
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2446 2447
}

2448

2449 2450 2451 2452
bool MYSQL_LOG::append(Log_event* ev)
{
  bool error = 0;
  pthread_mutex_lock(&LOCK_log);
2453 2454
  DBUG_ENTER("MYSQL_LOG::append");

2455
  DBUG_ASSERT(log_file.type == SEQ_READ_APPEND);
2456 2457 2458 2459
  /*
    Log_event::write() is smart enough to use my_b_write() or
    my_b_append() depending on the kind of cache we have.
  */
2460 2461 2462 2463 2464
  if (ev->write(&log_file))
  {
    error=1;
    goto err;
  }
2465
  bytes_written+= ev->data_written;
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
2466
  DBUG_PRINT("info",("max_size: %lu",max_size));
2467
  if ((uint) my_b_append_tell(&log_file) > max_size)
2468
    new_file(0);
2469

2470
err:
2471
  pthread_mutex_unlock(&LOCK_log);
2472
  signal_update();				// Safe as we don't call close
2473
  DBUG_RETURN(error);
2474 2475
}

2476

2477 2478
bool MYSQL_LOG::appendv(const char* buf, uint len,...)
{
2479
  bool error= 0;
2480
  DBUG_ENTER("MYSQL_LOG::appendv");
2481 2482
  va_list(args);
  va_start(args,len);
2483

2484
  DBUG_ASSERT(log_file.type == SEQ_READ_APPEND);
2485

2486
  safe_mutex_assert_owner(&LOCK_log);
2487 2488
  do
  {
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2489
    if (my_b_append(&log_file,(byte*) buf,len))
2490
    {
2491 2492
      error= 1;
      goto err;
2493
    }
2494
    bytes_written += len;
2495
  } while ((buf=va_arg(args,const char*)) && (len=va_arg(args,uint)));
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
2496
  DBUG_PRINT("info",("max_size: %lu",max_size));
2497
  if ((uint) my_b_append_tell(&log_file) > max_size)
2498 2499 2500
    new_file(0);

err:
2501 2502
  if (!error)
    signal_update();
2503
  DBUG_RETURN(error);
2504
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2505

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2506

2507
/*
2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529
  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
2530 2531
*/

2532 2533 2534 2535
bool MYSQL_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)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2536
{
2537 2538 2539 2540 2541 2542 2543 2544
  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())
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2545
  {
2546 2547 2548 2549
    /* Note that my_b_write() assumes it knows the length for this */
      if (event_time != last_time)
      {
        last_time= event_time;
2550

2551 2552 2553 2554 2555 2556 2557 2558 2559 2560
        localtime_r(&event_time, &start);

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

        if (my_b_write(&log_file, (byte*) &time_buff, time_buff_len))
          goto err;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2561 2562
      }
      else
2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584
        if (my_b_write(&log_file, (byte*) "\t\t" ,2) < 0)
          goto err;

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

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

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

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

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

    if (my_b_write(&log_file, (byte*) "\n", 1) ||
        flush_io_cache(&log_file))
      goto err;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2585
  }
2586 2587 2588 2589 2590 2591 2592 2593 2594 2595

  return FALSE;
err:

  if (!write_error)
  {
    write_error= 1;
    sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
  }
  return TRUE;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2596 2597
}

2598

2599
bool MYSQL_LOG::flush_and_sync()
2600
{
2601 2602 2603 2604 2605
  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)
2606 2607
  {
    sync_binlog_counter= 0;
2608
    err=my_sync(fd, MYF(MY_WME));
2609
  }
2610
  return err;
2611 2612
}

2613 2614 2615 2616 2617 2618
void MYSQL_LOG::start_union_events(THD *thd)
{
  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;
2619
  thd->binlog_evt_union.first_query_id= thd->query_id;
2620 2621 2622 2623 2624 2625 2626 2627
}

void MYSQL_LOG::stop_union_events(THD *thd)
{
  DBUG_ASSERT(thd->binlog_evt_union.do_union);
  thd->binlog_evt_union.do_union= FALSE;
}

2628 2629 2630 2631 2632 2633
bool MYSQL_LOG::is_query_in_union(THD *thd, query_id_t query_id_param)
{
  return (thd->binlog_evt_union.do_union && 
          query_id_param >= thd->binlog_evt_union.first_query_id);
}

2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662

/*
  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=
    (binlog_trx_data*) ha_data[binlog_hton.slot];

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

  ha_data[binlog_hton.slot]= trx_data=
    (binlog_trx_data*) my_malloc(sizeof(binlog_trx_data), MYF(MY_ZEROFILL));
  if (!trx_data ||
      open_cached_file(&trx_data->trans_log, mysql_tmpdir,
                       LOG_PREFIX, binlog_cache_size, MYF(MY_WME)))
  {
    my_free((gptr)trx_data, MYF(MY_ALLOW_ZERO_PTR));
    ha_data[binlog_hton.slot]= 0;
    DBUG_RETURN(1);                      // Didn't manage to set it up
  }
  trx_data->trans_log.end_of_file= max_binlog_cache_size;
  DBUG_RETURN(0);
}

2663 2664 2665 2666 2667 2668 2669
/*
  Write a table map to the binary log.

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

2670
#ifdef HAVE_ROW_BASED_REPLICATION
2671 2672
int THD::binlog_write_table_map(TABLE *table, bool is_trans)
{
2673
  int error;
2674
  DBUG_ENTER("THD::binlog_write_table_map");
2675
  DBUG_PRINT("enter", ("table: %p  (%s: #%u)",
2676 2677 2678
                       table, table->s->table_name, table->s->table_map_id));

  /* Pre-conditions */
2679
  DBUG_ASSERT(current_stmt_binlog_row_based && mysql_bin_log.is_open());
2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691
  DBUG_ASSERT(table->s->table_map_id != ULONG_MAX);

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

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

  if (is_trans)
    trans_register_ha(this, options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN),
                      &binlog_hton);

2692
  if ((error= mysql_bin_log.write(&the_event)))
2693 2694
    DBUG_RETURN(error);

2695
  binlog_table_maps++;
2696 2697 2698 2699
  table->s->table_map_version= mysql_bin_log.table_map_version();
  DBUG_RETURN(0);
}

2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716
Rows_log_event*
THD::binlog_get_pending_rows_event() const
{
  binlog_trx_data *const trx_data=
    (binlog_trx_data*) ha_data[binlog_hton.slot];
  /*
    This is less than ideal, but here's the story: If there is no
    trx_data, prepare_pending_rows_event() has never been called
    (since the trx_data is set up there). In that case, we just return
    NULL.
   */
  return trx_data ? trx_data->pending : NULL;
}

void
THD::binlog_set_pending_rows_event(Rows_log_event* ev)
{
2717 2718 2719
  if (ha_data[binlog_hton.slot] == NULL)
    binlog_setup_trx_data();

2720 2721
  binlog_trx_data *const trx_data=
    (binlog_trx_data*) ha_data[binlog_hton.slot];
2722

2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735
  DBUG_ASSERT(trx_data);
  trx_data->pending= ev;
}


/*
  Moves the last bunch of rows from the pending Rows event to the binlog
  (either cached binlog if transaction, or disk binlog). Sets a new pending
  event.
*/
int MYSQL_LOG::flush_and_set_pending_rows_event(THD *thd, Rows_log_event* event)
{
  DBUG_ENTER("MYSQL_LOG::flush_and_set_pending_rows_event(event)");
2736
  DBUG_ASSERT(thd->current_stmt_binlog_row_based && mysql_bin_log.is_open());
2737 2738 2739 2740 2741 2742 2743 2744 2745
  DBUG_PRINT("enter", ("event=%p", event));

  int error= 0;

  binlog_trx_data *const trx_data=
    (binlog_trx_data*) thd->ha_data[binlog_hton.slot];

  DBUG_ASSERT(trx_data);

2746 2747
  DBUG_PRINT("info", ("trx_data->pending=%p", trx_data->pending));

2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808
  if (Rows_log_event* pending= trx_data->pending)
  {
    IO_CACHE *file= &log_file;

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

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

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

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

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

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

    delete pending;

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

    pthread_mutex_unlock(&LOCK_log);
  }

2809
  thd->binlog_set_pending_rows_event(event);
2810 2811 2812

  DBUG_RETURN(error);
}
2813
#endif /*HAVE_ROW_BASED_REPLICATION*/
2814

2815 2816 2817 2818
/*
  Write an event to the binary log
*/

2819
bool MYSQL_LOG::write(Log_event *event_info)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2820
{
2821 2822 2823 2824
  THD *thd= event_info->thd;
  bool error= 1;
  DBUG_ENTER("MYSQL_LOG::write(Log_event *)");

2825 2826
  if (thd->binlog_evt_union.do_union)
  {
2827 2828 2829 2830
    /*
      In Stored function; Remember that function call caused an update.
      We will log the function call to the binary log on function exit
    */
2831 2832 2833 2834
    thd->binlog_evt_union.unioned_events= TRUE;
    thd->binlog_evt_union.unioned_events_trans |= event_info->cache_stmt;
    DBUG_RETURN(0);
  }
2835 2836 2837 2838 2839 2840 2841

  /*
    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.

2842 2843 2844
    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.
2845
  */
2846
#ifdef HAVE_ROW_BASED_REPLICATION
2847 2848 2849
  bool const end_stmt=
    thd->prelocked_mode && thd->lex->requires_prelocking();
  thd->binlog_flush_pending_rows_event(end_stmt);
2850
#endif /*HAVE_ROW_BASED_REPLICATION*/
2851

2852 2853
  pthread_mutex_lock(&LOCK_log);

2854
  /*
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2855 2856 2857 2858
     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.
  */
2859
  if (likely(is_open()))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2860
  {
2861 2862
    const char *local_db= event_info->get_db();
    IO_CACHE *file= &log_file;
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
2863
#ifdef HAVE_REPLICATION
2864
    /*
2865 2866 2867 2868
       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)"
    */
2869
    if ((thd && !(thd->options & OPTION_BIN_LOG)) ||
lars@mysql.com's avatar
Merge  
lars@mysql.com committed
2870
	(!binlog_filter->db_ok(local_db)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2871
    {
2872
      VOID(pthread_mutex_unlock(&LOCK_log));
2873 2874 2875
      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)));
2876
      DBUG_RETURN(0);
2877
    }
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
2878
#endif /* HAVE_REPLICATION */
2879

2880 2881 2882 2883 2884 2885 2886 2887
#ifdef USING_TRANSACTIONS
    /*
      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)
2888
      - or if the event asks for it (cache_stmt == TRUE).
2889 2890 2891
    */
    if (opt_using_transactions && thd)
    {
2892
#ifdef HAVE_ROW_BASED_REPLICATION
2893 2894
      if (thd->binlog_setup_trx_data())
        goto err;
2895
#endif /*HAVE_ROW_BASED_REPLICATION*/
2896

2897 2898 2899
      binlog_trx_data *const trx_data=
        (binlog_trx_data*) thd->ha_data[binlog_hton.slot];
      IO_CACHE *trans_log= &trx_data->trans_log;
2900 2901
      bool trans_log_in_use= my_b_tell(trans_log) != 0;
      if (event_info->get_cache_stmt() && !trans_log_in_use)
2902
        trans_register_ha(thd,
2903 2904
                          (thd->options &
                           (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN)),
2905
                          &binlog_hton);
2906 2907 2908
      if (event_info->get_cache_stmt() || trans_log_in_use)
      {
        DBUG_PRINT("info", ("Using trans_log"));
2909
        file= trans_log;
2910
      }
2911
      /*
2912
        TODO as Mats suggested, for all the cases above where we write to
2913 2914
        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
2915
        LOCK_log.
2916
      */
2917 2918
    }
#endif
2919
    DBUG_PRINT("info",("event type: %d",event_info->get_type_code()));
2920

2921 2922 2923 2924
    /*
      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
2925 2926

    /*
2927 2928
      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
2929 2930
    */

2931 2932 2933 2934
    /*
      If row-based binlogging, Insert_id, Rand and other kind of "setting
      context" events are not needed.
    */
2935
    if (thd)
2936
    {
2937
      if (!thd->current_stmt_binlog_row_based)
2938
      {
2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973
        if (thd->last_insert_id_used)
        {
          Intvar_log_event e(thd,(uchar) LAST_INSERT_ID_EVENT,
                             thd->current_insert_id);
          if (e.write(file))
            goto err;
        }
        if (thd->insert_id_used)
        {
          Intvar_log_event e(thd,(uchar) INSERT_ID_EVENT,thd->last_insert_id);
          if (e.write(file))
            goto err;
        }
        if (thd->rand_used)
        {
          Rand_log_event e(thd,thd->rand_saved_seed1,thd->rand_saved_seed2);
          if (e.write(file))
            goto err;
        }
        if (thd->user_var_events.elements)
        {
          for (uint i= 0; i < thd->user_var_events.elements; i++)
          {
            BINLOG_USER_VAR_EVENT *user_var_event;
            get_dynamic(&thd->user_var_events,(gptr) &user_var_event, i);
            User_var_log_event e(thd, user_var_event->user_var_event->name.str,
                                 user_var_event->user_var_event->name.length,
                                 user_var_event->value,
                                 user_var_event->length,
                                 user_var_event->type,
                                 user_var_event->charset_number);
            if (e.write(file))
              goto err;
          }
        }
2974
      }
2975
    }
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
2976

2977 2978
    /*
       Write the SQL command
2979
     */
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
2980

monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
2981
    if (event_info->write(file))
2982
      goto err;
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
2983

2984
    if (file == &log_file) // we are writing to the real log (disk)
2985
    {
2986
      if (flush_and_sync())
monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
2987
	goto err;
2988 2989
      signal_update();
      rotate_and_purge(RP_LOCK_LOG_IS_ALREADY_LOCKED);
2990
    }
monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
2991
    error=0;
2992

2993 2994 2995 2996
err:
    if (error)
    {
      if (my_errno == EFBIG)
bell@sanja.is.com.ua's avatar
bell@sanja.is.com.ua committed
2997
	my_message(ER_TRANS_CACHE_FULL, ER(ER_TRANS_CACHE_FULL), MYF(0));
2998 2999 3000 3001 3002
      else
	my_error(ER_ERROR_ON_WRITE, MYF(0), name, errno);
      write_error=1;
    }
  }
3003

3004 3005 3006
  if (event_info->flags & LOG_EVENT_UPDATE_TABLE_MAP_VERSION_F)
    ++m_table_map_version;

3007
  pthread_mutex_unlock(&LOCK_log);
3008
  DBUG_RETURN(error);
3009 3010
}

3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038

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;
}

3039 3040
void MYSQL_LOG::rotate_and_purge(uint flags)
{
3041 3042
  if (!(flags & RP_LOCK_LOG_IS_ALREADY_LOCKED))
    pthread_mutex_lock(&LOCK_log);
3043 3044
  if ((flags & RP_FORCE_ROTATE) ||
      (my_b_tell(&log_file) >= (my_off_t) max_size))
3045
  {
3046
    new_file(0);
3047 3048 3049 3050 3051 3052 3053 3054 3055
#ifdef HAVE_REPLICATION
    if (expire_logs_days)
    {
      long purge_time= time(0) - expire_logs_days*24*60*60;
      if (purge_time >= 0)
        purge_logs_before_date(purge_time);
    }
#endif
  }
3056 3057
  if (!(flags & RP_LOCK_LOG_IS_ALREADY_LOCKED))
    pthread_mutex_unlock(&LOCK_log);
3058
}
3059

3060 3061 3062 3063 3064 3065 3066 3067 3068
uint MYSQL_LOG::next_file_id()
{
  uint res;
  pthread_mutex_lock(&LOCK_log);
  res = file_id++;
  pthread_mutex_unlock(&LOCK_log);
  return res;
}

3069

3070 3071
/*
  Write a cached log entry to the binary log
3072

guilhem@mysql.com's avatar
guilhem@mysql.com committed
3073 3074
  SYNOPSIS
    write()
3075
    thd
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3076 3077
    cache		The cache to copy to the binlog

3078 3079
  NOTE
    - We only come here if there is something in the cache.
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
3080
    - The thing in the cache is always a complete transaction
3081 3082 3083 3084
    - '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
3085
      with BEGIN/COMMIT or BEGIN/ROLLBACK in the binary log.
3086 3087 3088
      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.
3089 3090
*/

3091
bool MYSQL_LOG::write(THD *thd, IO_CACHE *cache, Log_event *commit_event)
3092
{
3093
  DBUG_ENTER("MYSQL_LOG::write(THD *, IO_CACHE *, Log_event *)");
3094
  VOID(pthread_mutex_lock(&LOCK_log));
3095

3096
  if (likely(is_open()))                       // Should always be true
3097
  {
3098 3099
    uint length;

3100
    /*
3101 3102
      Log "BEGIN" at the beginning of the transaction.
      which may contain more than 1 SQL statement.
3103
    */
3104
    if (thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN))
3105
    {
3106
      Query_log_event qinfo(thd, STRING_WITH_LEN("BEGIN"), TRUE, FALSE);
3107 3108 3109 3110
      /*
        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
3111 3112 3113
        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.
3114 3115 3116
        This is safer than thd->clear_error() against kills at shutdown.
      */
      qinfo.error_code= 0;
3117 3118 3119
      /*
        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
3120 3121 3122
	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.
3123
      */
3124 3125 3126
      if (qinfo.write(&log_file))
	goto err;
    }
3127
    /* Read from the file used to cache the queries .*/
3128
    if (reinit_io_cache(cache, READ_CACHE, 0, 0, 0))
3129
      goto err;
3130
    length=my_b_bytes_in_cache(cache);
3131
    DBUG_EXECUTE_IF("half_binlogged_transaction", length-=100;);
3132
    do
3133
    {
3134
      /* Write data to the binary log file */
monty@bitch.mysql.fi's avatar
monty@bitch.mysql.fi committed
3135
      if (my_b_write(&log_file, cache->read_pos, length))
3136
	goto err;
monty@bitch.mysql.fi's avatar
monty@bitch.mysql.fi committed
3137
      cache->read_pos=cache->read_end;		// Mark buffer used up
3138
      DBUG_EXECUTE_IF("half_binlogged_transaction", goto DBUG_skip_commit;);
3139
    } while ((length=my_b_fill(cache)));
3140

3141 3142
    if (commit_event->write(&log_file))
      goto err;
3143
#ifndef DBUG_OFF
3144
DBUG_skip_commit:
3145
#endif
3146
    if (flush_and_sync())
3147
      goto err;
3148
    DBUG_EXECUTE_IF("half_binlogged_transaction", abort(););
3149 3150
    if (cache->error)				// Error on read
    {
3151
      sql_print_error(ER(ER_ERROR_ON_READ), cache->file_name, errno);
3152
      write_error=1;				// Don't give more errors
3153
      goto err;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3154
    }
3155
    signal_update();
3156 3157 3158 3159 3160 3161 3162 3163 3164
    /*
      if commit_event is Xid_log_event, increase the number of
      prepared_xids (it's decreasd in ::unlog()). Binlog cannot be rotated
      if there're prepared xids in it - see the comment in new_file() for
      an explanation.
      If the commit_event is not Xid_log_event (then it's a Query_log_event)
      rotate binlog, if necessary.
    */
    if (commit_event->get_type_code() == XID_EVENT)
3165 3166 3167 3168 3169
    {
      pthread_mutex_lock(&LOCK_prep_xids);
      prepared_xids++;
      pthread_mutex_unlock(&LOCK_prep_xids);
    }
3170 3171
    else
      rotate_and_purge(RP_LOCK_LOG_IS_ALREADY_LOCKED);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3172
  }
3173
  VOID(pthread_mutex_unlock(&LOCK_log));
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
3174

3175
  DBUG_RETURN(0);
3176 3177

err:
3178 3179 3180 3181 3182
  if (!write_error)
  {
    write_error= 1;
    sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
  }
3183
  VOID(pthread_mutex_unlock(&LOCK_log));
3184
  DBUG_RETURN(1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3185 3186
}

3187

3188
/*
3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214
  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
3215
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3216

3217 3218 3219 3220
bool MYSQL_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)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3221
{
3222
  bool error= 0;
monty@mysql.com's avatar
monty@mysql.com committed
3223 3224
  DBUG_ENTER("MYSQL_LOG::write");

3225 3226 3227
  if (!is_open())
    DBUG_RETURN(0);

3228
  if (is_open())
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3229
  {						// Safety agains reopen
3230 3231 3232 3233 3234 3235
    int tmp_errno= 0;
    char buff[80], *end;
    uint buff_len;
    end= buff;

    if (!(specialflag & SPECIAL_SHORT_LOG_FORMAT))
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3236
    {
3237
      Security_context *sctx= thd->security_ctx;
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3238
      if (current_time != last_time)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3239
      {
3240 3241 3242 3243 3244 3245 3246 3247 3248 3249
        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);

guilhem@mysql.com's avatar
guilhem@mysql.com committed
3250
        /* Note that my_b_write() assumes it knows the length for this */
3251
        if (my_b_write(&log_file, (byte*) buff, buff_len))
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3252
          tmp_errno=errno;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3253
      }
3254
      if (my_b_printf(&log_file, "# User@Host: ", sizeof("# User@Host: ") - 1))
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3255
        tmp_errno=errno;
3256 3257 3258
      if (my_b_printf(&log_file, user_host, user_host_len))
        tmp_errno=errno;
      if (my_b_write(&log_file, (byte*) "\n", 1))
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3259 3260
        tmp_errno=errno;
    }
3261 3262 3263 3264 3265 3266 3267 3268
    /* 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;
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282
    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);
    }
    if (thd->last_insert_id_used)
    {
      end=strmov(end,",last_insert_id=");
      end=longlong10_to_str((longlong) thd->current_insert_id,end,-10);
    }
    // Save value if we do an insert.
    if (thd->insert_id_used)
    {
3283
      if (!(specialflag & SPECIAL_SHORT_LOG_FORMAT))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3284
      {
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3285 3286
        end=strmov(end,",insert_id=");
        end=longlong10_to_str((longlong) thd->last_insert_id,end,-10);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3287
      }
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3288
    }
3289 3290 3291 3292 3293 3294 3295 3296 3297

    /*
      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);

guilhem@mysql.com's avatar
guilhem@mysql.com committed
3298 3299 3300 3301 3302 3303 3304 3305
    if (end != buff)
    {
      *end++=';';
      *end='\n';
      if (my_b_write(&log_file, (byte*) "SET ",4) ||
          my_b_write(&log_file, (byte*) buff+1,(uint) (end-buff)))
        tmp_errno=errno;
    }
3306
    if (is_command)
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3307
    {
3308 3309 3310
      end= strxmov(buff, "# administrator command: ", NullS);
      buff_len= (ulong) (end - buff);
      my_b_write(&log_file, (byte*) buff, buff_len);
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3311
    }
3312
    if (my_b_write(&log_file, (byte*) sql_text, sql_text_len) ||
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3313 3314 3315 3316 3317 3318 3319
        my_b_write(&log_file, (byte*) ";\n",2) ||
        flush_io_cache(&log_file))
      tmp_errno=errno;
    if (tmp_errno)
    {
      error=1;
      if (! write_error)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3320
      {
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3321 3322
        write_error=1;
        sql_print_error(ER(ER_ERROR_ON_WRITE),name,error);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3323 3324 3325
      }
    }
  }
monty@mysql.com's avatar
monty@mysql.com committed
3326
  DBUG_RETURN(error);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3327 3328
}

3329

3330 3331 3332 3333 3334 3335
/*
  Wait until we get a signal that the binary log has been updated

  SYNOPSIS
    wait_for_update()
    thd			Thread variable
3336
    is_slave            If 0, the caller is the Binlog_dump thread from master;
3337 3338
                        if 1, the caller is the SQL thread from the slave. This
                        influences only thd->proc_info.
3339 3340 3341

  NOTES
    One must have a lock on LOCK_log before calling this function.
3342
    This lock will be released before return! That's required by
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3343
    THD::enter_cond() (see NOTES in sql_class.h).
3344 3345
*/

3346
void MYSQL_LOG::wait_for_update(THD* thd, bool is_slave)
3347
{
monty@mysql.com's avatar
monty@mysql.com committed
3348 3349
  const char *old_msg;
  DBUG_ENTER("wait_for_update");
3350

monty@mysql.com's avatar
monty@mysql.com committed
3351
  old_msg= thd->enter_cond(&update_cond, &LOCK_log,
3352
                           is_slave ?
monty@mysql.com's avatar
monty@mysql.com committed
3353
                           "Has read all relay log; waiting for the slave I/O "
3354
                           "thread to update it" :
monty@mysql.com's avatar
monty@mysql.com committed
3355
                           "Has sent all binlog to slave; waiting for binlog "
3356
                           "to be updated");
3357 3358
  pthread_cond_wait(&update_cond, &LOCK_log);
  thd->exit_cond(old_msg);
monty@mysql.com's avatar
monty@mysql.com committed
3359
  DBUG_VOID_RETURN;
3360
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3361

3362

3363 3364 3365 3366 3367
/*
  Close the log file

  SYNOPSIS
    close()
3368 3369 3370 3371 3372
    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
3373 3374 3375 3376

  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
3377 3378
*/

3379
void MYSQL_LOG::close(uint exiting)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3380
{					// One can't set log_type here!
3381 3382
  DBUG_ENTER("MYSQL_LOG::close");
  DBUG_PRINT("enter",("exiting: %d", (int) exiting));
3383
  if (log_type != LOG_CLOSED && log_type != LOG_TO_BE_OPENED)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3384
  {
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
3385
#ifdef HAVE_REPLICATION
3386 3387
    if (log_type == LOG_BIN && !no_auto_events &&
	(exiting & LOG_CLOSE_STOP_EVENT))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3388 3389
    {
      Stop_log_event s;
3390
      s.write(&log_file);
3391
      bytes_written+= s.data_written;
3392
      signal_update();
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3393
    }
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
3394
#endif /* HAVE_REPLICATION */
3395
    end_io_cache(&log_file);
3396 3397

    /* don't pwrite in a file opened with O_APPEND - it doesn't work */
3398
    if (log_file.type == WRITE_CACHE && log_type == LOG_BIN)
3399 3400
    {
      my_off_t offset= BIN_LOG_HEADER_SIZE + FLAGS_OFFSET;
serg@serg.mylan's avatar
serg@serg.mylan committed
3401
      byte flags=0; // clearing LOG_EVENT_BINLOG_IN_USE_F
3402
      my_pwrite(log_file.file, &flags, 1, offset, MYF(0));
3403 3404
    }

3405 3406 3407 3408 3409 3410
    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)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3411 3412
    {
      write_error=1;
3413
      sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3414 3415
    }
  }
3416 3417 3418 3419 3420 3421

  /*
    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.
  */

3422
  if ((exiting & LOG_CLOSE_INDEX) && my_b_inited(&index_file))
3423
  {
3424 3425
    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
3426
    {
3427 3428
      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
3429
    }
3430
  }
3431
  log_type= (exiting & LOG_CLOSE_TO_BE_OPENED) ? LOG_TO_BE_OPENED : LOG_CLOSED;
3432
  safeFree(name);
3433
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3434 3435 3436
}


3437 3438 3439 3440 3441 3442 3443 3444 3445
void MYSQL_LOG::set_max_size(ulong max_size_arg)
{
  /*
    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.
  */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3446 3447
  DBUG_ENTER("MYSQL_LOG::set_max_size");
  pthread_mutex_lock(&LOCK_log);
3448 3449
  if (is_open())
    max_size= max_size_arg;
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3450 3451
  pthread_mutex_unlock(&LOCK_log);
  DBUG_VOID_RETURN;
3452
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3453

3454

3455 3456
/*
  Check if a string is a valid number
3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470

  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
3471
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483

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++;
3484 3485
  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
3486 3487 3488 3489 3490 3491 3492
  {
    flag=1;
    str++;
  }
  if (*str == '.')
  {
    for (str++ ;
3493
	 my_isdigit(files_charset_info,*str) ||
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504
	   (allow_wildcards && (*str == wild_many || *str == wild_one)) ;
	 str++, flag=1) ;
  }
  if (*str != 0 || flag == 0)
    DBUG_RETURN(0);
  if (res)
    *res=atol(start);
  DBUG_RETURN(1);			/* Number ok */
} /* test_if_number */


serg@serg.mylan's avatar
serg@serg.mylan committed
3505
void print_buffer_to_file(enum loglevel level, const char *buffer)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3506 3507 3508 3509
{
  time_t skr;
  struct tm tm_tmp;
  struct tm *start;
monty@mysql.com's avatar
monty@mysql.com committed
3510 3511
  DBUG_ENTER("print_buffer_to_file");
  DBUG_PRINT("enter",("buffer: %s", buffer));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3512 3513

  VOID(pthread_mutex_lock(&LOCK_error_log));
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
3514

bk@work.mysql.com's avatar
bk@work.mysql.com committed
3515
  skr=time(NULL);
3516
  localtime_r(&skr, &tm_tmp);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3517
  start=&tm_tmp;
3518

3519
  fprintf(stderr, "%02d%02d%02d %2d:%02d:%02d [%s] %s\n",
3520 3521 3522 3523 3524 3525
          start->tm_year % 100,
          start->tm_mon+1,
          start->tm_mday,
          start->tm_hour,
          start->tm_min,
          start->tm_sec,
monty@mysql.com's avatar
monty@mysql.com committed
3526
          (level == ERROR_LEVEL ? "ERROR" : level == WARNING_LEVEL ?
3527
           "Warning" : "Note"),
monty@mysql.com's avatar
monty@mysql.com committed
3528
          buffer);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
3529

bk@work.mysql.com's avatar
bk@work.mysql.com committed
3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544
  fflush(stderr);

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


void sql_perror(const char *message)
{
#ifdef HAVE_STRERROR
  sql_print_error("%s: %s",message, strerror(errno));
#else
  perror(message);
#endif
}
3545

monty@mysql.com's avatar
monty@mysql.com committed
3546

3547 3548 3549
bool flush_error_log()
{
  bool result=0;
3550
  if (opt_error_log)
3551 3552 3553 3554
  {
    char err_renamed[FN_REFLEN], *end;
    end= strmake(err_renamed,log_error_file,FN_REFLEN-4);
    strmov(end, "-old");
3555
    VOID(pthread_mutex_lock(&LOCK_error_log));
3556 3557 3558 3559 3560 3561
#ifdef __WIN__
    char err_temp[FN_REFLEN+4];
    /*
     On Windows is necessary a temporary file for to rename
     the current error file.
    */
3562
    strxmov(err_temp, err_renamed,"-tmp",NullS);
3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575
    (void) my_delete(err_temp, MYF(0)); 
    if (freopen(err_temp,"a+",stdout))
    {
      freopen(err_temp,"a+",stderr);
      (void) my_delete(err_renamed, MYF(0));
      my_rename(log_error_file,err_renamed,MYF(0));
      if (freopen(log_error_file,"a+",stdout))
        freopen(log_error_file,"a+",stderr);
      int fd, bytes;
      char buf[IO_SIZE];
      if ((fd = my_open(err_temp, O_RDONLY, MYF(0))) >= 0)
      {
        while ((bytes = (int) my_read(fd, (byte*) buf, IO_SIZE, MYF(0))) > 0)
3576
             my_fwrite(stderr, (byte*) buf, bytes, MYF(0));
3577 3578 3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589
        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
3590
    VOID(pthread_mutex_unlock(&LOCK_error_log));
3591 3592 3593
  }
   return result;
}
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
3594

monty@mysql.com's avatar
monty@mysql.com committed
3595 3596 3597 3598 3599 3600 3601
void MYSQL_LOG::signal_update()
{
  DBUG_ENTER("MYSQL_LOG::signal_update");
  pthread_cond_broadcast(&update_cond);
  DBUG_VOID_RETURN;
}

3602
#ifdef __NT__
monty@mysql.com's avatar
monty@mysql.com committed
3603 3604
void print_buffer_to_nt_eventlog(enum loglevel level, char *buff,
                                 uint length, int buffLen)
3605 3606 3607 3608
{
  HANDLE event;
  char   *buffptr;
  LPCSTR *buffmsgptr;
serg@serg.mylan's avatar
serg@serg.mylan committed
3609
  DBUG_ENTER("print_buffer_to_nt_eventlog");
3610

monty@mysql.com's avatar
monty@mysql.com committed
3611
  buffptr= buff;
3612
  if (length > (uint)(buffLen-5))
3613
  {
3614
    char *newBuff= new char[length + 5];
serg@serg.mylan's avatar
serg@serg.mylan committed
3615
    strcpy(newBuff, buff);
monty@mysql.com's avatar
monty@mysql.com committed
3616
    buffptr= newBuff;
3617
  }
monty@mysql.com's avatar
monty@mysql.com committed
3618 3619
  strmov(buffptr+length, "\r\n\r\n");
  buffmsgptr= (LPCSTR*) &buffptr;               // Keep windows happy
3620

monty@mysql.com's avatar
monty@mysql.com committed
3621 3622
  setup_windows_event_source();
  if ((event= RegisterEventSource(NULL,"MySQL")))
3623
  {
monty@mysql.com's avatar
monty@mysql.com committed
3624
    switch (level) {
3625
      case ERROR_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
3626 3627
        ReportEvent(event, EVENTLOG_ERROR_TYPE, 0, MSG_DEFAULT, NULL, 1, 0,
                    buffmsgptr, NULL);
3628 3629
        break;
      case WARNING_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
3630 3631
        ReportEvent(event, EVENTLOG_WARNING_TYPE, 0, MSG_DEFAULT, NULL, 1, 0,
                    buffmsgptr, NULL);
3632 3633
        break;
      case INFORMATION_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
3634 3635
        ReportEvent(event, EVENTLOG_INFORMATION_TYPE, 0, MSG_DEFAULT, NULL, 1,
                    0, buffmsgptr, NULL);
3636 3637 3638 3639 3640
        break;
    }
    DeregisterEventSource(event);
  }

monty@mysql.com's avatar
monty@mysql.com committed
3641
  /* if we created a string buffer, then delete it */
serg@serg.mylan's avatar
serg@serg.mylan committed
3642
  if (buffptr != buff)
3643 3644 3645 3646
    delete[] buffptr;

  DBUG_VOID_RETURN;
}
monty@mysql.com's avatar
monty@mysql.com committed
3647 3648
#endif /* __NT__ */

3649 3650

/*
monty@mysql.com's avatar
monty@mysql.com committed
3651 3652
  Prints a printf style message to the error log and, under NT, to the
  Windows event log.
3653 3654 3655

  SYNOPSIS
    vprint_msg_to_log()
monty@mysql.com's avatar
monty@mysql.com committed
3656 3657
    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
3658
    args                   va_list list of arguments for the message
3659 3660 3661 3662

  NOTE

  IMPLEMENTATION
monty@mysql.com's avatar
monty@mysql.com committed
3663 3664
    This function prints the message into a buffer and then sends that buffer
    to other functions to write that message to other logging sources.
3665 3666

  RETURN VALUES
3667 3668 3669
    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
3670
*/
monty@mysql.com's avatar
monty@mysql.com committed
3671

3672
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
3673 3674
{
  char   buff[1024];
monty@mysql.com's avatar
monty@mysql.com committed
3675
  uint length;
3676
  DBUG_ENTER("vprint_msg_to_log");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
3677

3678 3679
  /* "- 5" is because of print_buffer_to_nt_eventlog() */
  length= my_vsnprintf(buff, sizeof(buff) - 5, format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
3680
  print_buffer_to_file(level, buff);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
3681 3682

#ifdef __NT__
monty@mysql.com's avatar
monty@mysql.com committed
3683
  print_buffer_to_nt_eventlog(level, buff, length, sizeof(buff));
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
3684
#endif
3685

3686
  DBUG_RETURN(0);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
3687 3688
}

3689

monty@mysql.com's avatar
monty@mysql.com committed
3690
void sql_print_error(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
3691 3692
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
3693
  DBUG_ENTER("sql_print_error");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
3694

serg@serg.mylan's avatar
serg@serg.mylan committed
3695
  va_start(args, format);
3696
  error_log_print(ERROR_LEVEL, format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
3697
  va_end(args);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
3698 3699 3700 3701 3702

  DBUG_VOID_RETURN;
}


monty@mysql.com's avatar
monty@mysql.com committed
3703
void sql_print_warning(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
3704 3705
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
3706
  DBUG_ENTER("sql_print_warning");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
3707

serg@serg.mylan's avatar
serg@serg.mylan committed
3708
  va_start(args, format);
3709
  error_log_print(WARNING_LEVEL, format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
3710
  va_end(args);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
3711 3712 3713 3714 3715

  DBUG_VOID_RETURN;
}


monty@mysql.com's avatar
monty@mysql.com committed
3716
void sql_print_information(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
3717 3718
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
3719
  DBUG_ENTER("sql_print_information");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
3720

serg@serg.mylan's avatar
serg@serg.mylan committed
3721
  va_start(args, format);
3722
  error_log_print(INFORMATION_LEVEL, format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
3723
  va_end(args);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
3724

guilhem@mysql.com's avatar
guilhem@mysql.com committed
3725 3726
  DBUG_VOID_RETURN;
}
3727

3728

3729 3730 3731 3732 3733 3734 3735 3736
/********* 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
3737 3738
  states - active, syncing, pool.
  there could be only one page in active or syncing states,
3739
  but many in pool - pool is fifo queue.
3740
  usual lifecycle of a page is pool->active->syncing->pool
3741
  "active" page - is a page where new xid's are logged.
3742 3743
  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.
3744
  when the sync is done the page is moved to a pool and an active page
3745 3746 3747 3748 3749 3750 3751 3752 3753 3754 3755 3756 3757
  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.
3758 3759 3760 3761
  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)

3762 3763
  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
3764 3765
  dirty - we don't sync removals to disk.
*/
3766 3767 3768 3769 3770

ulong tc_log_page_waits= 0;

#ifdef HAVE_MMAP

3771 3772
#define TC_LOG_HEADER_SIZE (sizeof(tc_log_magic)+1)

3773
static const char tc_log_magic[]={(char) 254, 0x23, 0x05, 0x74};
3774

3775
ulong opt_tc_log_size= TC_LOG_MIN_SIZE;
3776
ulong tc_log_max_pages_used=0, tc_log_page_size=0, tc_log_cur_pages_used=0;
3777 3778 3779 3780 3781 3782 3783

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

3784
  DBUG_ASSERT(total_ha_2pc > 1);
3785 3786
  DBUG_ASSERT(opt_name && opt_name[0]);

serg@serg.mylan's avatar
serg@serg.mylan committed
3787
  tc_log_page_size= my_getpagesize();
3788 3789 3790
  DBUG_ASSERT(TC_LOG_PAGE_SIZE % tc_log_page_size == 0);

  fn_format(logname,opt_name,mysql_data_home,"",MY_UNPACK_FILENAME);
3791
  if ((fd= my_open(logname, O_RDWR, MYF(0))) < 0)
3792
  {
3793 3794
    if (my_errno != ENOENT)
      goto err;
3795 3796
    if (using_heuristic_recover())
      return 1;
3797
    if ((fd= my_create(logname, O_RDWR, 0, MYF(MY_WME))) < 0)
3798 3799
      goto err;
    inited=1;
3800
    file_length= opt_tc_log_size;
3801 3802 3803 3804 3805
    if (my_chsize(fd, file_length, 0, MYF(MY_WME)))
      goto err;
  }
  else
  {
3806 3807
    inited= 1;
    crashed= TRUE;
serg@serg.mylan's avatar
serg@serg.mylan committed
3808
    sql_print_information("Recovering after a crash using %s", opt_name);
3809 3810 3811 3812 3813 3814
    if (tc_heuristic_recover)
    {
      sql_print_error("Cannot perform automatic crash recovery when "
                      "--tc-heuristic-recover is used");
      goto err;
    }
3815
    file_length= my_seek(fd, 0L, MY_SEEK_END, MYF(MY_WME+MY_FAE));
3816 3817 3818 3819
    if (file_length == MY_FILEPOS_ERROR || file_length % tc_log_page_size)
      goto err;
  }

3820
  data= (uchar *)my_mmap(0, (size_t)file_length, PROT_READ|PROT_WRITE,
3821 3822 3823 3824 3825 3826 3827 3828
                        MAP_NOSYNC|MAP_SHARED, fd, 0);
  if (data == MAP_FAILED)
  {
    my_errno=errno;
    goto err;
  }
  inited=2;

3829
  npages=(uint)file_length/tc_log_page_size;
3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845 3846
  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);
3847
  pages[0].start=pages[0].end-pages[0].size;
3848 3849 3850
  pages[npages-1].next=0;
  inited=4;

3851
  if (crashed && recover())
3852
      goto err;
3853

3854
  memcpy(data, tc_log_magic, sizeof(tc_log_magic));
3855
  data[sizeof(tc_log_magic)]= (uchar)total_ha_2pc;
3856 3857 3858 3859 3860 3861 3862 3863 3864 3865 3866
  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;

3867
  syncing= 0;
3868 3869 3870 3871 3872 3873 3874 3875 3876 3877 3878 3879
  active=pages;
  pool=pages+1;
  pool_last=pages+npages-1;

  return 0;

err:
  close();
  return 1;
}

/*
3880 3881
  there is no active page, let's got one from the pool

3882 3883 3884 3885 3886 3887 3888 3889 3890 3891 3892 3893
  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;

3894 3895 3896
  if (syncing)
    pthread_mutex_lock(&LOCK_pool);

3897 3898 3899
  do
  {
    best_p= p= &pool;
3900 3901
    if ((*p)->waiters == 0) // can the first page be used ?
      break;                // yes - take it.
3902

3903
    best_free=0;            // no - trying second strategy
3904 3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915
    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;
3916
  if (active->free == active->size) // we've chosen an empty page
3917 3918 3919 3920 3921
  {
    tc_log_cur_pages_used++;
    set_if_bigger(tc_log_max_pages_used, tc_log_cur_pages_used);
  }

3922
  if ((*best_p)->next)              // unlink the page from the pool
3923 3924 3925
    *best_p=(*best_p)->next;
  else
    pool_last=*best_p;
3926 3927 3928

  if (syncing)
    pthread_mutex_unlock(&LOCK_pool);
3929 3930 3931 3932 3933 3934 3935 3936 3937 3938 3939
}

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);
3940
  return 1; // always return 1
3941 3942 3943
}

/*
3944 3945 3946 3947 3948 3949 3950 3951 3952 3953 3954 3955
  all access to active page is serialized but it's not a problem, as
  we're assuming that fsync() will be a main bottleneck.
  That is, parallelizing writes to log pages we'll decrease number of
  threads waiting for a page, but then all these threads will be waiting
  for a fsync() anyway

  RETURN
         0  - error
  otherwise - "cookie", a number that will be passed as an argument
              to unlog() call. tc_log can define it any way it wants,
              and use for whatever purposes. TC_LOG_MMAP sets it
              to the position in memory where xid was logged to.
3956 3957 3958 3959 3960 3961 3962 3963 3964 3965
*/

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

  pthread_mutex_lock(&LOCK_active);

3966 3967 3968 3969 3970 3971 3972
  /*
    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.
  */
3973 3974 3975
  while (unlikely(active && active->free == 0))
    pthread_cond_wait(&COND_active, &LOCK_active);

3976
  /* no active page ? take one from the pool */
3977 3978 3979 3980 3981 3982
  if (active == 0)
    get_active_from_pool();

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

3983
  /* searching for an empty slot */
3984 3985 3986 3987 3988 3989
  while (*p->ptr)
  {
    p->ptr++;
    DBUG_ASSERT(p->ptr < p->end);               // because p->free > 0
  }

3990 3991
  /* found! store xid there and mark the page dirty */
  cookie= (ulong)((uchar *)p->ptr - data);      // can never be zero
3992 3993 3994 3995 3996
  *p->ptr++= xid;
  p->free--;
  p->state= DIRTY;

  /* to sync or not to sync - this is the question */
3997
  pthread_mutex_unlock(&LOCK_active);
3998 3999 4000
  pthread_mutex_lock(&LOCK_sync);
  pthread_mutex_unlock(&p->lock);

4001
  if (syncing)
4002 4003
  {                                          // somebody's syncing. let's wait
    p->waiters++;
4004
    /*
monty@mishka.local's avatar
monty@mishka.local committed
4005
      note - it must be while (), not do ... while () here
4006 4007 4008
      as p->state may be not DIRTY when we come here
    */
    while (p->state == DIRTY && syncing)
4009 4010 4011 4012 4013 4014 4015 4016 4017 4018 4019
      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
4020 4021 4022 4023
  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
4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036
  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;

4037
  DBUG_ASSERT(syncing != active);
4038 4039 4040 4041 4042

  /*
    sit down and relax - this can take a while...
    note - no locks are held at this point
  */
4043
  err= my_msync(fd, syncing->start, 1, MS_SYNC);
4044

4045
  /* page is synced. let's move it to the pool */
4046
  pthread_mutex_lock(&LOCK_pool);
4047 4048 4049 4050 4051
  pool_last->next=syncing;
  pool_last=syncing;
  syncing->next=0;
  syncing->state= err ? ERROR : POOL;
  pthread_cond_broadcast(&syncing->cond);    // signal "sync done"
4052 4053 4054
  pthread_cond_signal(&COND_pool);           // in case somebody's waiting
  pthread_mutex_unlock(&LOCK_pool);

4055
  /* marking 'syncing' slot free */
4056
  pthread_mutex_lock(&LOCK_sync);
4057
  syncing=0;
4058 4059 4060 4061 4062
  pthread_cond_signal(&active->cond);        // wake up a new syncer
  pthread_mutex_unlock(&LOCK_sync);
  return err;
}

4063 4064 4065 4066
/*
  erase xid from the page, update page free space counters/pointers.
  cookie points directly to the memory where xid was logged
*/
4067 4068 4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079
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);
4080
  if (p->free == p->size)               // the page is completely empty
4081 4082
    statistic_decrement(tc_log_cur_pages_used, &LOCK_status);
  if (p->waiters == 0)                 // the page is in pool and ready to rock
4083
    pthread_cond_signal(&COND_pool);   // ping ... for overflow()
4084 4085 4086 4087 4088
  pthread_mutex_unlock(&p->lock);
}

void TC_LOG_MMAP::close()
{
serg@serg.mylan's avatar
serg@serg.mylan committed
4089
  uint i;
4090 4091 4092 4093 4094 4095 4096 4097 4098
  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
4099
    for (i=0; i < npages; i++)
4100 4101 4102 4103 4104 4105 4106 4107 4108
    {
      if (pages[i].ptr == 0)
        break;
      pthread_mutex_destroy(&pages[i].lock);
      pthread_cond_destroy(&pages[i].cond);
    }
  case 3:
    my_free((gptr)pages, MYF(0));
  case 2:
kent@mysql.com's avatar
kent@mysql.com committed
4109
    my_munmap((byte*)data, (size_t)file_length);
4110 4111 4112 4113 4114 4115 4116 4117 4118 4119 4120 4121 4122 4123 4124 4125 4126 4127 4128 4129 4130 4131 4132 4133 4134
  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)
  {
4135
    sql_print_error("Recovery failed! You must enable "
4136 4137 4138
                    "exactly %d storage engines that support "
                    "two-phase commit protocol",
                    data[sizeof(tc_log_magic)]);
4139 4140 4141 4142 4143 4144 4145 4146 4147 4148 4149 4150 4151 4152 4153 4154 4155 4156
    goto err1;
  }

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

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

  if (ha_recover(&xids))
    goto err2;

  hash_free(&xids);
4157
  bzero(data, (size_t)file_length);
4158 4159 4160 4161 4162 4163 4164 4165 4166 4167 4168
  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
4169 4170 4171 4172 4173
#endif

TC_LOG *tc_log;
TC_LOG_DUMMY tc_log_dummy;
TC_LOG_MMAP  tc_log_mmap;
4174

4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185 4186
/*
  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.
*/

4187 4188 4189 4190 4191 4192 4193 4194 4195 4196 4197 4198 4199
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 ******/
4200
#define TC_LOG_BINLOG MYSQL_LOG
4201 4202 4203 4204 4205 4206 4207 4208

/*
  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!
*/

4209
int TC_LOG_BINLOG::open(const char *opt_name)
4210
{
4211 4212
  LOG_INFO log_info;
  int      error= 1;
4213 4214

  DBUG_ASSERT(total_ha_2pc > 1);
4215
  DBUG_ASSERT(opt_name && opt_name[0]);
4216 4217 4218 4219

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

4220 4221 4222 4223 4224 4225 4226
  if (!my_b_inited(&index_file))
  {
    /* There was a failure to open the index file, can't open the binlog */
    cleanup();
    return 1;
  }

4227
  if (using_heuristic_recover())
4228 4229 4230 4231
  {
    /* generate a new binlog to mask a corrupted one */
    open(opt_name, LOG_BIN, 0, WRITE_CACHE, 0, max_binlog_size, 0);
    cleanup();
4232
    return 1;
4233
  }
4234

4235
  if ((error= find_log_pos(&log_info, NullS, 1)))
4236
  {
4237 4238 4239 4240 4241
    if (error != LOG_INFO_EOF)
      sql_print_error("find_log_pos() failed (error: %d)", error);
    else
      error= 0;
    goto err;
4242
  }
4243 4244 4245 4246 4247

  {
    const char *errmsg;
    IO_CACHE    log;
    File        file;
4248 4249
    Log_event  *ev=0;
    Format_description_log_event fdle(BINLOG_VERSION);
4250
    char        log_name[FN_REFLEN];
4251 4252 4253 4254

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

4255
    do
4256
    {
4257 4258 4259 4260 4261 4262 4263
      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;
4264
    }
4265

4266
    if ((file= open_binlog(&log, log_name, &errmsg)) < 0)
4267 4268 4269 4270 4271
    {
      sql_print_error("%s", errmsg);
      goto err;
    }

4272 4273 4274
    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
4275 4276
    {
      sql_print_information("Recovering after a crash using %s", opt_name);
4277
      error= recover(&log, (Format_description_log_event *)ev);
serg@serg.mylan's avatar
serg@serg.mylan committed
4278
    }
4279 4280
    else
      error=0;
4281 4282 4283 4284 4285 4286 4287 4288 4289 4290

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

    if (error)
      goto err;
  }

err:
4291
  return error;
4292 4293
}

4294 4295
/* this is called on shutdown, after ha_panic */
void TC_LOG_BINLOG::close()
4296 4297 4298 4299 4300 4301
{
  DBUG_ASSERT(prepared_xids==0);
  pthread_mutex_destroy(&LOCK_prep_xids);
  pthread_cond_destroy (&COND_prep_xids);
}

4302 4303 4304 4305 4306 4307 4308 4309
/*
  TODO group commit

  RETURN
         0  - error
         1  - success
*/
int TC_LOG_BINLOG::log(THD *thd, my_xid xid)
4310
{
4311
  DBUG_ENTER("TC_LOG_BINLOG::log");
4312
  Xid_log_event xle(thd, xid);
4313 4314 4315
  binlog_trx_data *trx_data=
    (binlog_trx_data*) thd->ha_data[binlog_hton.slot];
  DBUG_RETURN(!binlog_end_trans(thd, trx_data, &xle));  // invert return value
4316 4317
}

4318
void TC_LOG_BINLOG::unlog(ulong cookie, my_xid xid)
4319
{
4320 4321
  pthread_mutex_lock(&LOCK_prep_xids);
  if (--prepared_xids == 0)
4322
    pthread_cond_signal(&COND_prep_xids);
4323
  pthread_mutex_unlock(&LOCK_prep_xids);
4324
  rotate_and_purge(0);     // as ::write() did not rotate
4325 4326
}

4327
int TC_LOG_BINLOG::recover(IO_CACHE *log, Format_description_log_event *fdle)
4328 4329 4330 4331 4332 4333
{
  Log_event  *ev;
  HASH xids;
  MEM_ROOT mem_root;

  if (! fdle->is_valid() ||
4334
      hash_init(&xids, &my_charset_bin, TC_LOG_PAGE_SIZE/3, 0,
4335 4336 4337
            sizeof(my_xid), 0, 0, MYF(0)))
    goto err1;

4338
  init_alloc_root(&mem_root, TC_LOG_PAGE_SIZE, TC_LOG_PAGE_SIZE);
4339

4340 4341
  fdle->flags&= ~LOG_EVENT_BINLOG_IN_USE_F; // abort on the first error

4342 4343 4344 4345 4346
  while ((ev= Log_event::read_log_event(log,0,fdle)) && ev->is_valid())
  {
    if (ev->get_type_code() == XID_EVENT)
    {
      Xid_log_event *xev=(Xid_log_event *)ev;
serg@serg.mylan's avatar
serg@serg.mylan committed
4347 4348
      byte *x=(byte *)memdup_root(&mem_root, (char *)& xev->xid,
                                  sizeof(xev->xid));
4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359 4360 4361 4362 4363 4364 4365 4366 4367 4368 4369 4370 4371 4372 4373
      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;
}

acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
4374 4375 4376 4377 4378

mysql_declare_plugin(binlog)
{
  MYSQL_STORAGE_ENGINE_PLUGIN,
  &binlog_hton,
4379
  binlog_hton_name,
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
4380
  "MySQL AB",
4381
  binlog_hton_comment,
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
4382 4383 4384 4385 4386
  NULL, /* Plugin Init */
  NULL, /* Plugin Deinit */
  0x0100 /* 1.0 */,
}
mysql_declare_plugin_end;