log.cc 86.4 KB
Newer Older
1
/* Copyright (C) 2000-2003 MySQL AB
serg@serg.mylan's avatar
serg@serg.mylan committed
2

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

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

bk@work.mysql.com's avatar
bk@work.mysql.com committed
12 13 14 15 16 17
   You should have received a copy of the GNU General Public License
   along with this program; if not, write to the Free Software
   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA */


/* logging of commands */
18
/* TODO: Abort logging when we get an error in reading or writing log files */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
19

20 21 22 23
#ifdef __EMX__
#include <io.h>
#endif

bk@work.mysql.com's avatar
bk@work.mysql.com committed
24
#include "mysql_priv.h"
25
#include "sql_repl.h"
bk@work.mysql.com's avatar
bk@work.mysql.com committed
26 27 28 29 30

#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
31 32 33 34
#ifdef __NT__
#include "message.h"
#endif

35
MYSQL_LOG mysql_log, mysql_slow_log, mysql_bin_log;
36
ulong sync_binlog_counter= 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
37

38 39
static Muted_query_log_event invisible_commit;

bk@work.mysql.com's avatar
bk@work.mysql.com committed
40 41
static bool test_if_number(const char *str,
			   long *res, bool allow_wildcards);
42
static bool binlog_init();
43 44 45 46 47 48 49
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);

50
handlerton binlog_hton = {
serg@serg.mylan's avatar
serg@serg.mylan committed
51
  "binlog",
52 53 54 55
  SHOW_OPTION_YES,
  "This is a meta storage engine to represent the binlog in a transaction",
  DB_TYPE_UNKNOWN,              /* IGNORE  for now */
  binlog_init,
56 57 58 59 60 61 62 63 64 65 66
  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 */
67 68 69
  NULL,                         /* rollback_by_xid */
  NULL,                         /* create_cursor_read_view */
  NULL,                         /* set_cursor_read_view */
70
  NULL,                         /* close_cursor_read_view */
71
  HTON_HIDDEN
72 73 74 75 76 77 78 79
};

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

80
bool binlog_init()
81
{
82
  return !opt_bin_log;
83 84 85 86 87
}

static int binlog_close_connection(THD *thd)
{
  IO_CACHE *trans_log= (IO_CACHE*)thd->ha_data[binlog_hton.slot];
88
  DBUG_ASSERT(!my_b_tell(trans_log));
89 90 91 92 93
  close_cached_file(trans_log);
  my_free((gptr)trans_log, MYF(0));
  return 0;
}

94
static int binlog_end_trans(THD *thd, IO_CACHE *trans_log, Log_event *end_ev)
95
{
96 97
  int error=0;
  DBUG_ENTER("binlog_end_trans");
98 99 100

  /* NULL denotes ROLLBACK with nothing to replicate */
  if (end_ev != NULL)
101 102
    error= mysql_bin_log.write(thd, trans_log, end_ev);

103 104 105 106 107 108 109 110
  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;
111
  DBUG_RETURN(error);
112 113 114 115 116 117 118 119
}

static int binlog_prepare(THD *thd, bool all)
{
  /*
    do nothing.
    just pretend we can do 2pc, so that MySQL won't
    switch to 1pc.
120
    real work will be done in MYSQL_LOG::log_xid()
121 122 123 124 125 126 127 128
  */
  return 0;
}

static int binlog_commit(THD *thd, bool all)
{
  IO_CACHE *trans_log= (IO_CACHE*)thd->ha_data[binlog_hton.slot];
  DBUG_ENTER("binlog_commit");
129
  DBUG_ASSERT(
130
     (all || !(thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN))));
131

132
  if (my_b_tell(trans_log) == 0)
133
  {
134
    // we're here because trans_log was flushed in MYSQL_LOG::log_xid()
135 136
    DBUG_RETURN(0);
  }
137 138 139 140 141 142 143 144
  if (all) 
  {
    Query_log_event qev(thd, STRING_WITH_LEN("COMMIT"), TRUE, FALSE);
    qev.error_code= 0; // see comment in MYSQL_LOG::write(THD, IO_CACHE)
    DBUG_RETURN(binlog_end_trans(thd, trans_log, &qev));
  }
  else
    DBUG_RETURN(binlog_end_trans(thd, trans_log, &invisible_commit));
145 146 147 148 149 150 151 152
}

static int binlog_rollback(THD *thd, bool all)
{
  int error=0;
  IO_CACHE *trans_log= (IO_CACHE*)thd->ha_data[binlog_hton.slot];
  DBUG_ENTER("binlog_rollback");
  /*
serg@serg.mylan's avatar
serg@serg.mylan committed
153 154
    First assert is guaranteed - see trans_register_ha() call below.
    The second must be true. If it is not, we're registering
155 156
    unnecessary, doing extra work. The cause should be found and eliminated
  */
serg@serg.mylan's avatar
serg@serg.mylan committed
157
  DBUG_ASSERT(all || !(thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN)));
158
  DBUG_ASSERT(my_b_tell(trans_log));
159
  /*
160 161 162 163
    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...)
164
  */
165
  if (unlikely(thd->no_trans_update.all))
166
  {
167
    Query_log_event qev(thd, STRING_WITH_LEN("ROLLBACK"), TRUE, FALSE);
168
    qev.error_code= 0; // see comment in MYSQL_LOG::write(THD, IO_CACHE)
169
    error= binlog_end_trans(thd, trans_log, &qev);
170
  }
171 172
  else
    error= binlog_end_trans(thd, trans_log, 0);
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
  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)
{
  IO_CACHE *trans_log= (IO_CACHE*)thd->ha_data[binlog_hton.slot];
  DBUG_ENTER("binlog_savepoint_set");
201
  DBUG_ASSERT(my_b_tell(trans_log));
202 203 204 205 206 207 208 209 210 211 212

  *(my_off_t *)sv= my_b_tell(trans_log);
  /* Write it to the binary log */
  Query_log_event qinfo(thd, thd->query, thd->query_length, TRUE, FALSE);
  DBUG_RETURN(mysql_bin_log.write(&qinfo));
}

static int binlog_savepoint_rollback(THD *thd, void *sv)
{
  IO_CACHE *trans_log= (IO_CACHE*)thd->ha_data[binlog_hton.slot];
  DBUG_ENTER("binlog_savepoint_rollback");
213
  DBUG_ASSERT(my_b_tell(trans_log));
214 215 216 217 218 219

  /*
    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.
  */
220
  if (unlikely(thd->no_trans_update.all))
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
  {
    Query_log_event qinfo(thd, thd->query, thd->query_length, TRUE, FALSE);
    DBUG_RETURN(mysql_bin_log.write(&qinfo));
  }
  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");

254 255
  if ((file = my_open(log_file_name, O_RDONLY | O_BINARY | O_SHARE, 
                      MYF(MY_WME))) < 0)
256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281
  {
    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
282

rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
283 284
#ifdef __NT__
static int eventSource = 0;
serg@serg.mylan's avatar
serg@serg.mylan committed
285

286
static void setup_windows_event_source()
monty@mysql.com's avatar
monty@mysql.com committed
287
{
288
  HKEY    hRegKey= NULL;
monty@mysql.com's avatar
monty@mysql.com committed
289 290 291
  DWORD   dwError= 0;
  TCHAR   szPath[MAX_PATH];
  DWORD dwTypes;
292

monty@mysql.com's avatar
monty@mysql.com committed
293 294 295
  if (eventSource)               // Ensure that we are only called once
    return;
  eventSource= 1;
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
296

monty@mysql.com's avatar
monty@mysql.com committed
297
  // Create the event source registry key
298
  dwError= RegCreateKey(HKEY_LOCAL_MACHINE,
monty@mysql.com's avatar
monty@mysql.com committed
299 300
                          "SYSTEM\\CurrentControlSet\\Services\\EventLog\\Application\\MySQL", 
                          &hRegKey);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
301

monty@mysql.com's avatar
monty@mysql.com committed
302 303
  /* Name of the PE module that contains the message resource */
  GetModuleFileName(NULL, szPath, MAX_PATH);
serg@serg.mylan's avatar
serg@serg.mylan committed
304

rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
305
  /* Register EventMessageFile */
306
  dwError = RegSetValueEx(hRegKey, "EventMessageFile", 0, REG_EXPAND_SZ,
307
                          (PBYTE) szPath, (DWORD) (strlen(szPath) + 1));
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
308

monty@mysql.com's avatar
monty@mysql.com committed
309 310 311 312 313
  /* 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
314

monty@mysql.com's avatar
monty@mysql.com committed
315
  RegCloseKey(hRegKey);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
316 317
}

monty@mysql.com's avatar
monty@mysql.com committed
318
#endif /* __NT__ */
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
319 320


bk@work.mysql.com's avatar
bk@work.mysql.com committed
321 322 323 324 325 326 327 328
/****************************************************************************
** 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)
{
329 330 331 332
  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
333
  reg1 struct fileinfo *file_info;
334 335
  ulong                 max_found=0;

bk@work.mysql.com's avatar
bk@work.mysql.com committed
336 337
  DBUG_ENTER("find_uniq_filename");

338 339 340 341
  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
342
  *end='.';
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
343
  length= (uint) (end-start+1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361

  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++='.';
362
  sprintf(end,"%06ld",max_found+1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
363 364 365
  DBUG_RETURN(0);
}

366

367
MYSQL_LOG::MYSQL_LOG()
368
  :bytes_written(0), last_time(0), query_start(0), name(0),
369
   prepared_xids(0), log_type(LOG_CLOSED), file_id(1), open_count(1),
370
   write_error(FALSE), inited(FALSE), need_start_event(TRUE),
371
   description_event_for_exec(0), description_event_for_queue(0)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
372 373
{
  /*
guilhem@mysql.com's avatar
guilhem@mysql.com committed
374 375
    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
376
    called only in main(). Doing initialization here would make it happen
377
    before main().
bk@work.mysql.com's avatar
bk@work.mysql.com committed
378
  */
379 380
  index_file_name[0] = 0;
  bzero((char*) &log_file,sizeof(log_file));
381
  bzero((char*) &index_file, sizeof(index_file));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
382 383
}

384
/* this is called only once */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
385

monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
386
void MYSQL_LOG::cleanup()
bk@work.mysql.com's avatar
bk@work.mysql.com committed
387
{
388
  DBUG_ENTER("cleanup");
bk@work.mysql.com's avatar
bk@work.mysql.com committed
389
  if (inited)
390
  {
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
391
    inited= 0;
392
    close(LOG_CLOSE_INDEX|LOG_CLOSE_STOP_EVENT);
393 394
    delete description_event_for_queue;
    delete description_event_for_exec;
395 396
    (void) pthread_mutex_destroy(&LOCK_log);
    (void) pthread_mutex_destroy(&LOCK_index);
397
    (void) pthread_cond_destroy(&update_cond);
398
  }
399
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
400 401
}

402

bk@work.mysql.com's avatar
bk@work.mysql.com committed
403
int MYSQL_LOG::generate_new_name(char *new_name, const char *log_name)
404
{
405 406
  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
407 408 409 410 411 412 413 414 415 416 417 418 419
  {
    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;
}

420

421
void MYSQL_LOG::init(enum_log_type log_type_arg,
422
		     enum cache_type io_cache_type_arg,
423 424
		     bool no_auto_events_arg,
                     ulong max_size_arg)
425
{
426
  DBUG_ENTER("MYSQL_LOG::init");
427
  log_type = log_type_arg;
428
  io_cache_type = io_cache_type_arg;
429
  no_auto_events = no_auto_events_arg;
430
  max_size=max_size_arg;
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
431
  DBUG_PRINT("info",("log_type: %d max_size: %lu", log_type, max_size));
432
  DBUG_VOID_RETURN;
433 434
}

435

guilhem@mysql.com's avatar
guilhem@mysql.com committed
436 437
void MYSQL_LOG::init_pthread_objects()
{
438 439
  DBUG_ASSERT(inited == 0);
  inited= 1;
guilhem@mysql.com's avatar
guilhem@mysql.com committed
440 441 442
  (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);
443 444
}

445 446 447 448 449 450 451 452 453 454
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.
    */
455
    strmake(buff, pidfile_name,FN_REFLEN-5);
456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495
    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;
}

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
496
			0, MYF(MY_WME | MY_WAIT_IF_FULL)))
497 498 499 500 501 502 503 504
  {
    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
505

506 507 508 509 510 511 512 513 514 515 516 517 518 519
/*
  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
*/

520 521 522 523 524
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,
525 526
                     ulong max_size_arg,
                     bool null_created_arg)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
527
{
528 529
  char buff[FN_REFLEN];
  File file= -1;
530
  int open_flags = O_CREAT | O_BINARY;
531
  DBUG_ENTER("MYSQL_LOG::open");
532
  DBUG_PRINT("enter",("log_type: %d",(int) log_type_arg));
533 534 535

  last_time=query_start=0;
  write_error=0;
536

537
  init(log_type_arg,io_cache_type_arg,no_auto_events_arg,max_size_arg);
538

539
  if (!(name=my_strdup(log_name,MYF(MY_WME))))
540 541
  {
    name= (char *)log_name; // for the error message
542
    goto err;
543
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
544 545 546
  if (new_name)
    strmov(log_file_name,new_name);
  else if (generate_new_name(log_file_name, name))
547
    goto err;
548

549
  if (io_cache_type == SEQ_READ_APPEND)
550
    open_flags |= O_RDWR | O_APPEND;
551
  else
552
    open_flags |= O_WRONLY | (log_type == LOG_BIN ? 0 : O_APPEND);
553

bk@work.mysql.com's avatar
bk@work.mysql.com committed
554
  db[0]=0;
555
  open_count++;
556
  if ((file=my_open(log_file_name,open_flags,
557
		    MYF(MY_WME | ME_WAITTANG))) < 0 ||
558
      init_io_cache(&log_file, file, IO_SIZE, io_cache_type,
559 560 561
		    my_tell(file,MYF(MY_WME)), 0, 
                    MYF(MY_WME | MY_NABP |
                        ((log_type == LOG_BIN) ? MY_WAIT_IF_FULL : 0))))
562
    goto err;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
563

564 565
  switch (log_type) {
  case LOG_NORMAL:
bk@work.mysql.com's avatar
bk@work.mysql.com committed
566
  {
567
    char *end;
568
    int len=my_snprintf(buff, sizeof(buff), "%s, Version: %s (%s). "
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
569
#ifdef EMBEDDED_LIBRARY
570 571
		        "embedded library\n",
                        my_progname, server_version, MYSQL_COMPILATION_COMMENT
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
572
#elif __NT__
573
			"started with:\nTCP Port: %d, Named Pipe: %s\n",
574 575
			my_progname, server_version, MYSQL_COMPILATION_COMMENT,
                        mysqld_port, mysqld_unix_port
bk@work.mysql.com's avatar
bk@work.mysql.com committed
576
#else
577
			"started with:\nTcp port: %d  Unix socket: %s\n",
578 579
			my_progname, server_version, MYSQL_COMPILATION_COMMENT,
                        mysqld_port, mysqld_unix_port
bk@work.mysql.com's avatar
bk@work.mysql.com committed
580
#endif
serg@serg.mylan's avatar
serg@serg.mylan committed
581
                       );
582 583
    end=strnmov(buff+len,"Time                 Id Command    Argument\n",
                sizeof(buff)-len);
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
584
    if (my_b_write(&log_file, (byte*) buff,(uint) (end-buff)) ||
585 586
	flush_io_cache(&log_file))
      goto err;
587
    break;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
588
  }
589
  case LOG_NEW:
bk@work.mysql.com's avatar
bk@work.mysql.com committed
590
  {
591
    uint len;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
592 593
    time_t skr=time(NULL);
    struct tm tm_tmp;
594

bk@work.mysql.com's avatar
bk@work.mysql.com committed
595
    localtime_r(&skr,&tm_tmp);
596 597 598 599 600 601 602 603 604 605
    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) ||
606 607
	flush_io_cache(&log_file))
      goto err;
608
    break;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
609
  }
610
  case LOG_BIN:
bk@work.mysql.com's avatar
bk@work.mysql.com committed
611
  {
612 613 614 615 616 617 618 619 620 621
    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.
      */
622 623
      if (my_b_safe_write(&log_file, (byte*) BINLOG_MAGIC,
			  BIN_LOG_HEADER_SIZE))
624
        goto err;
625 626
      bytes_written+= BIN_LOG_HEADER_SIZE;
      write_file_name_to_index_file= 1;
627
    }
628

629
    DBUG_ASSERT(my_b_inited(&index_file) != 0);
630 631
    reinit_io_cache(&index_file, WRITE_CACHE,
                    my_b_filelength(&index_file), 0, 0);
632
    if (need_start_event && !no_auto_events)
633
    {
634 635 636 637 638
      /*
        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
639 640 641 642 643 644
      /*
        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;
645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663
      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
664
        Format_description_log_event::Format_description_log_event(char* buf,).
665 666 667 668 669 670 671 672 673 674 675 676 677 678
        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;
679

680 681 682
      if (description_event_for_queue->write(&log_file))
        goto err;
      bytes_written+= description_event_for_queue->data_written;
683
    }
guilhem@mysql.com's avatar
guilhem@mysql.com committed
684 685
    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
686
      goto err;
687 688 689

    if (write_file_name_to_index_file)
    {
guilhem@mysql.com's avatar
guilhem@mysql.com committed
690 691 692 693
      /*
        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.
      */
694 695 696
      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
697 698
	  flush_io_cache(&index_file) ||
          my_sync(index_file.file, MYF(MY_WME)))
699
	goto err;
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
700
    }
701
    break;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
702
  }
703
  case LOG_CLOSED:				// Impossible
704
  case LOG_TO_BE_OPENED:
705 706 707 708
    DBUG_ASSERT(1);
    break;
  }
  DBUG_RETURN(0);
709 710

err:
711 712 713
  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, \
714
shutdown the MySQL server and restart it.", name, errno);
715 716 717
  if (file >= 0)
    my_close(file,MYF(0));
  end_io_cache(&log_file);
718 719
  end_io_cache(&index_file);
  safeFree(name);
720
  log_type= LOG_CLOSED;
721
  DBUG_RETURN(1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
722 723
}

724

bk@work.mysql.com's avatar
bk@work.mysql.com committed
725 726 727
int MYSQL_LOG::get_current_log(LOG_INFO* linfo)
{
  pthread_mutex_lock(&LOCK_log);
728 729 730 731 732 733 734
  int ret = raw_get_current_log(linfo);
  pthread_mutex_unlock(&LOCK_log);
  return ret;
}

int MYSQL_LOG::raw_get_current_log(LOG_INFO* linfo)
{
735 736
  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
737 738 739
  return 0;
}

740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760
/*
  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
*/

761 762
#ifdef HAVE_REPLICATION

763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783
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
784 785
  if (my_chsize(file, offset - init_offset, '\n', MYF(MY_WME)) ||
      my_sync(file, MYF(MY_WME)))
786 787 788 789 790 791 792 793 794 795
    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);
}

796
#endif /* HAVE_REPLICATION */
797 798 799 800 801 802 803 804 805 806

/*
  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
807
    need_lock		Set this to 1 if the parent doesn't already have a
808 809 810
			lock on LOCK_index

  NOTE
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
811 812
    On systems without the truncate function the file will end with one or
    more empty lines.  These will be ignored when reading the file.
813 814 815 816 817 818 819 820 821

  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
822
{
823 824 825 826 827
  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
828

829
  /*
830 831
    Mutex needed because we need to make sure the file pointer does not
    move from under our feet
832
  */
833
  if (need_lock)
834
    pthread_mutex_lock(&LOCK_index);
835 836 837 838 839
  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);

840
  for (;;)
841
  {
842
    uint length;
843 844 845 846
    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
847
    {
848 849 850
      /* 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
851 852
    }

853 854
    // if the log entry matches, null string matching anything
    if (!log_name ||
sasha@mysql.sashanet.com's avatar
sasha@mysql.sashanet.com committed
855
	(log_name_len == length-1 && fname[log_name_len] == '\n' &&
856
	 !memcmp(fname, log_name, log_name_len)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
857
    {
858
      DBUG_PRINT("info",("Found log file entry"));
859
      fname[length-1]=0;			// remove last \n
860 861
      linfo->index_file_start_offset= offset;
      linfo->index_file_offset = my_b_tell(&index_file);
862
      break;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
863
    }
864
  }
865

866
  if (need_lock)
867
    pthread_mutex_unlock(&LOCK_index);
868
  DBUG_RETURN(error);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
869
}
870

871

872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893
/*
  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
*/

894
int MYSQL_LOG::find_next_log(LOG_INFO* linfo, bool need_lock)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
895
{
896
  int error= 0;
897
  uint length;
898 899
  char *fname= linfo->log_file_name;

900 901
  if (need_lock)
    pthread_mutex_lock(&LOCK_index);
902 903 904 905 906 907 908 909
  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)
910
  {
911
    error = !index_file.error ? LOG_INFO_EOF : LOG_INFO_IO;
912 913
    goto err;
  }
914
  fname[length-1]=0;				// kill \n
915
  linfo->index_file_offset = my_b_tell(&index_file);
916

bk@work.mysql.com's avatar
bk@work.mysql.com committed
917
err:
918 919
  if (need_lock)
    pthread_mutex_unlock(&LOCK_index);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
920 921 922
  return error;
}

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
923

924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942
/*
  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)
943 944
{
  LOG_INFO linfo;
945
  bool error=0;
946 947
  const char* save_name;
  enum_log_type save_log_type;
948
  DBUG_ENTER("reset_logs");
949

950 951 952 953
  /*
    We need to get both locks to be sure that no one is trying to
    write to the index log file.
  */
954
  pthread_mutex_lock(&LOCK_log);
955
  pthread_mutex_lock(&LOCK_index);
956

957 958 959 960 961 962 963
  /*
    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);
964 965 966 967 968

  /* Save variables so that we can reopen the log */
  save_name=name;
  name=0;					// Protect against free
  save_log_type=log_type;
969
  close(LOG_CLOSE_TO_BE_OPENED);
970 971 972 973

  /* First delete all old log files */

  if (find_log_pos(&linfo, NullS, 0))
974 975 976 977
  {
    error=1;
    goto err;
  }
978

979
  for (;;)
980
  {
981
    my_delete_allow_opened(linfo.log_file_name, MYF(MY_WME));
982
    if (find_next_log(&linfo, 0))
983 984
      break;
  }
985 986

  /* Start logging with a new file */
987
  close(LOG_CLOSE_INDEX);
988
  my_delete_allow_opened(index_file_name, MYF(MY_WME));	// Reset (open will update)
989
  if (!thd->slave_thread)
990
    need_start_event=1;
991 992 993
  if (!open_index_file(index_file_name, 0))
    open(save_name, save_log_type, 0,
         io_cache_type, no_auto_events, max_size, 0);
994
  my_free((gptr) save_name, MYF(0));
995

996
err:
997
  (void) pthread_mutex_unlock(&LOCK_thread_count);
998
  pthread_mutex_unlock(&LOCK_index);
999
  pthread_mutex_unlock(&LOCK_log);
1000
  DBUG_RETURN(error);
1001 1002
}

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
1003

1004
/*
1005 1006 1007
  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.
1008 1009 1010

  SYNOPSIS
    purge_first_log()
1011 1012 1013 1014 1015
    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).
1016

1017 1018
  NOTE
    - This is only called from the slave-execute thread when it has read
1019 1020 1021 1022 1023
      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).
1024 1025 1026

  IMPLEMENTATION
    - Protects index file with LOCK_index
1027 1028
    - Delete relevant relay log files
    - Copy all file names after these ones to the front of the index file
1029
    - If the OS has truncate, truncate the file, else fill it with \n'
1030
    - Read the next file name from the index file and store in rli->linfo
1031 1032

  RETURN VALUES
1033 1034 1035 1036
    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
1037 1038
*/

hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
1039
#ifdef HAVE_REPLICATION
1040

1041
int MYSQL_LOG::purge_first_log(struct st_relay_log_info* rli, bool included) 
1042
{
1043
  int error;
1044 1045
  DBUG_ENTER("purge_first_log");

1046 1047
  DBUG_ASSERT(is_open());
  DBUG_ASSERT(rli->slave_running == 1);
1048
  DBUG_ASSERT(!strcmp(rli->linfo.log_file_name,rli->event_relay_log_name));
1049

1050
  pthread_mutex_lock(&LOCK_index);
1051 1052 1053 1054 1055 1056
  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
1057 1058

  /*
1059 1060 1061
    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
1062
  */
1063 1064 1065 1066 1067
  pthread_cond_broadcast(&rli->log_space_cond);
  
  /*
    Read the next log file name from the index file and pass it back to
    the caller
1068 1069
    If included is true, we want the first relay log;
    otherwise we want the one after event_relay_log_name.
1070
  */
1071 1072 1073 1074
  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)))))
1075 1076
  {
    char buff[22];
1077 1078 1079 1080 1081
    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);
1082 1083
    goto err;
  }
1084

1085
  /*
1086
    Reset rli's coordinates to the current log.
1087
  */
1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101
  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);
1102
    rli->notify_group_relay_log_name_update();
1103
  }
1104 1105 1106

  /* 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
1107

1108
err:
1109
  pthread_mutex_unlock(&LOCK_index);
1110
  DBUG_RETURN(error);
1111 1112
}

1113 1114 1115 1116
/*
  Update log index_file
*/

1117
int MYSQL_LOG::update_log_index(LOG_INFO* log_info, bool need_update_threads)
1118 1119 1120 1121 1122
{
  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
1123 1124
  if (need_update_threads)
    adjust_linfo_offsets(log_info->index_file_start_offset);
1125 1126
  return 0;
}
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
1127

1128 1129 1130 1131 1132
/*
  Remove all logs before the given log from disk and from the index file.

  SYNOPSIS
    purge_logs()
1133 1134 1135 1136 1137 1138 1139
    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
1140 1141 1142 1143 1144 1145 1146 1147 1148 1149

  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
*/

1150 1151 1152 1153 1154
int MYSQL_LOG::purge_logs(const char *to_log, 
                          bool included,
                          bool need_mutex, 
                          bool need_update_threads, 
                          ulonglong *decrease_log_space)
1155 1156
{
  int error;
1157
  bool exit_loop= 0;
1158 1159
  LOG_INFO log_info;
  DBUG_ENTER("purge_logs");
1160
  DBUG_PRINT("info",("to_log= %s",to_log));
1161

1162 1163
  if (need_mutex)
    pthread_mutex_lock(&LOCK_index);
1164
  if ((error=find_log_pos(&log_info, to_log, 0 /*no mutex*/)))
1165 1166
    goto err;

1167
  /*
1168
    File name exists in index file; delete until we find this file
1169 1170 1171
    or a file that is used.
  */
  if ((error=find_log_pos(&log_info, NullS, 0 /*no mutex*/)))
1172
    goto err;
1173 1174
  while ((strcmp(to_log,log_info.log_file_name) || (exit_loop=included)) &&
         !log_in_use(log_info.log_file_name))
1175
  {
1176
    ulong file_size= 0;
1177 1178 1179
    if (decrease_log_space) //stat the file we want to delete
    {
      MY_STAT s;
1180 1181 1182 1183 1184 1185

      /* 
         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.
      */
1186
      if (my_stat(log_info.log_file_name,&s,MYF(0)))
1187
        file_size= s.st_size;
1188
      else
1189 1190
	sql_print_information("Failed to execute my_stat on file '%s'",
			      log_info.log_file_name);
1191 1192 1193 1194 1195 1196 1197
    }
    /*
      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)
1198
      *decrease_log_space-= file_size;
1199
    if (find_next_log(&log_info, 0) || exit_loop)
1200
      break;
1201
  }
1202

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
1203 1204 1205 1206
  /*
    If we get killed -9 here, the sysadmin would have to edit
    the log index file after restart - otherwise, this should be safe
  */
1207
  error= update_log_index(&log_info, need_update_threads);
1208

1209
err:
1210 1211
  if (need_mutex)
    pthread_mutex_unlock(&LOCK_index);
1212
  DBUG_RETURN(error);
1213 1214
}

1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232
/*
  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
*/

1233
int MYSQL_LOG::purge_logs_before_date(time_t purge_time)
1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248
{
  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*/)))
1249
    goto err;
1250 1251 1252 1253 1254

  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 */
1255 1256
    if (!my_stat(log_info.log_file_name, &stat_area, MYF(0)) ||
	stat_area.st_mtime >= purge_time)
1257
      break;
1258
    my_delete(log_info.log_file_name, MYF(0));
1259 1260
    if (find_next_log(&log_info, 0))
      break;
1261
  }
1262

1263 1264 1265 1266
  /*
    If we get killed -9 here, the sysadmin would have to edit
    the log index file after restart - otherwise, this should be safe
  */
1267
  error= update_log_index(&log_info, 1);
1268

1269 1270
err:
  pthread_mutex_unlock(&LOCK_index);
1271
  DBUG_RETURN(error);
1272
}
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
1273
#endif /* HAVE_REPLICATION */
1274

1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286

/*
  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
1287 1288
void MYSQL_LOG::make_log_name(char* buf, const char* log_ident)
{
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1289 1290 1291 1292 1293
  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
1294 1295
}

1296 1297 1298 1299 1300 1301

/*
  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
1302
{
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1303
  return !strcmp(log_file_name, log_file_name_arg);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1304 1305
}

1306 1307 1308 1309 1310 1311

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

  SYNOPSIS
    new_file()
1312
    need_lock		Set to 1 if caller has not locked LOCK_log
1313 1314 1315 1316 1317 1318

  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
1319
{
1320 1321 1322
  char new_name[FN_REFLEN], *new_name_ptr, *old_name;
  enum_log_type save_log_type;

1323
  DBUG_ENTER("MYSQL_LOG::new_file");
1324
  if (!is_open())
1325 1326 1327 1328
  {
    DBUG_PRINT("info",("log is closed"));
    DBUG_VOID_RETURN;
  }
1329 1330 1331

  if (need_lock)
    pthread_mutex_lock(&LOCK_log);
1332 1333
  pthread_mutex_lock(&LOCK_index);

1334 1335 1336
  safe_mutex_assert_owner(&LOCK_log);
  safe_mutex_assert_owner(&LOCK_index);

1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354
  /*
    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);
  }

1355
  /* Reuse old name if not binlog and not update log */
1356
  new_name_ptr= name;
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
1357

1358
  /*
1359 1360 1361
    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.
1362
  */
1363 1364 1365
  if (generate_new_name(new_name, name))
    goto end;
  new_name_ptr=new_name;
1366

1367
  if (log_type == LOG_BIN)
1368
  {
1369
    if (!no_auto_events)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1370
    {
1371
      /*
1372 1373
        We log the whole file name for log file as the user may decide
        to change base names at some point.
1374
      */
monty@mysql.com's avatar
monty@mysql.com committed
1375
      THD *thd = current_thd; /* may be 0 if we are reacting to SIGHUP */
1376 1377
      Rotate_log_event r(thd,new_name+dirname_length(new_name),
                         0, LOG_EVENT_OFFSET, 0);
1378
      r.write(&log_file);
1379
      bytes_written += r.data_written;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1380
    }
1381 1382 1383 1384 1385
    /*
      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.
    */
1386
    signal_update();
1387 1388 1389 1390
  }
  old_name=name;
  save_log_type=log_type;
  name=0;				// Don't free name
1391
  close(LOG_CLOSE_TO_BE_OPENED);
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1392

1393
  /*
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1394
     Note that at this point, log_type != LOG_CLOSED (important for is_open()).
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1395
  */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1396

1397
  /*
1398
     new_file() is only used for rotation (in FLUSH LOGS or because size >
1399
     max_binlog_size or max_relay_log_size).
1400 1401 1402 1403
     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.
1404
  */
1405

1406 1407
  open(old_name, save_log_type, new_name_ptr,
       io_cache_type, no_auto_events, max_size, 1);
1408 1409
  my_free(old_name,MYF(0));

1410
end:
1411 1412
  if (need_lock)
    pthread_mutex_unlock(&LOCK_log);
1413 1414
  pthread_mutex_unlock(&LOCK_index);

1415
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1416 1417
}

1418

1419 1420 1421 1422
bool MYSQL_LOG::append(Log_event* ev)
{
  bool error = 0;
  pthread_mutex_lock(&LOCK_log);
1423 1424
  DBUG_ENTER("MYSQL_LOG::append");

1425
  DBUG_ASSERT(log_file.type == SEQ_READ_APPEND);
1426 1427 1428 1429
  /*
    Log_event::write() is smart enough to use my_b_write() or
    my_b_append() depending on the kind of cache we have.
  */
1430 1431 1432 1433 1434
  if (ev->write(&log_file))
  {
    error=1;
    goto err;
  }
1435
  bytes_written+= ev->data_written;
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
1436
  DBUG_PRINT("info",("max_size: %lu",max_size));
1437
  if ((uint) my_b_append_tell(&log_file) > max_size)
1438
    new_file(0);
1439

1440
err:
1441
  pthread_mutex_unlock(&LOCK_log);
1442
  signal_update();				// Safe as we don't call close
1443
  DBUG_RETURN(error);
1444 1445
}

1446

1447 1448
bool MYSQL_LOG::appendv(const char* buf, uint len,...)
{
1449
  bool error= 0;
1450
  DBUG_ENTER("MYSQL_LOG::appendv");
1451 1452
  va_list(args);
  va_start(args,len);
1453

1454
  DBUG_ASSERT(log_file.type == SEQ_READ_APPEND);
1455

1456
  safe_mutex_assert_owner(&LOCK_log);
1457 1458
  do
  {
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
1459
    if (my_b_append(&log_file,(byte*) buf,len))
1460
    {
1461 1462
      error= 1;
      goto err;
1463
    }
1464
    bytes_written += len;
1465
  } while ((buf=va_arg(args,const char*)) && (len=va_arg(args,uint)));
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
1466
  DBUG_PRINT("info",("max_size: %lu",max_size));
1467
  if ((uint) my_b_append_tell(&log_file) > max_size)
1468 1469 1470
    new_file(0);

err:
1471 1472
  if (!error)
    signal_update();
1473
  DBUG_RETURN(error);
1474
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1475

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
1476

1477 1478
/*
  Write to normal (not rotable) log
1479
  This is the format for the 'normal' log.
1480 1481
*/

1482
bool MYSQL_LOG::write(THD *thd,enum enum_server_command command,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1483 1484
		      const char *format,...)
{
1485
  if (is_open() && (what_to_log & (1L << (uint) command)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1486
  {
1487 1488
    uint length;
    int error= 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1489
    VOID(pthread_mutex_lock(&LOCK_log));
1490

nick@mysql.com's avatar
nick@mysql.com committed
1491 1492
    /* 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
1493 1494 1495
    {
      time_t skr;
      ulong id;
1496 1497 1498 1499
      va_list args;
      va_start(args,format);
      char buff[32];

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1500 1501
      if (thd)
      {						// Normal thread
hf@deer.(none)'s avatar
hf@deer.(none) committed
1502 1503
	if ((thd->options & OPTION_LOG_OFF)
#ifndef NO_EMBEDDED_ACCESS_CHECKS
1504
	    && (thd->security_ctx->master_access & SUPER_ACL)
hf@deer.(none)'s avatar
hf@deer.(none) committed
1505 1506
#endif
)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1507 1508
	{
	  VOID(pthread_mutex_unlock(&LOCK_log));
1509
	  return 0;				// No logging
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526
	}
	id=thd->thread_id;
	if (thd->user_time || !(skr=thd->query_start()))
	  skr=time(NULL);			// Connected
      }
      else
      {						// Log from connect handler
	skr=time(NULL);
	id=0;
      }
      if (skr != last_time)
      {
	last_time=skr;
	struct tm tm_tmp;
	struct tm *start;
	localtime_r(&skr,&tm_tmp);
	start=&tm_tmp;
1527 1528 1529 1530 1531 1532 1533 1534
	/* Note that my_b_write() assumes it knows the length for this */
	sprintf(buff,"%02d%02d%02d %2d:%02d:%02d\t",
		start->tm_year % 100,
		start->tm_mon+1,
		start->tm_mday,
		start->tm_hour,
		start->tm_min,
		start->tm_sec);
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
1535
	if (my_b_write(&log_file, (byte*) buff,16))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1536 1537
	  error=errno;
      }
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
1538
      else if (my_b_write(&log_file, (byte*) "\t\t",2) < 0)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1539
	error=errno;
1540 1541 1542 1543
      length=my_sprintf(buff,
			(buff, "%7ld %-11.11s", id,
			 command_name[(uint) command]));
      if (my_b_write(&log_file, (byte*) buff,length))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1544 1545 1546
	error=errno;
      if (format)
      {
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
1547
	if (my_b_write(&log_file, (byte*) " ",1) ||
1548
	    my_b_vprintf(&log_file,format,args) == (uint) -1)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1549 1550
	  error=errno;
      }
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
1551
      if (my_b_write(&log_file, (byte*) "\n",1) ||
1552
	  flush_io_cache(&log_file))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1553 1554 1555 1556 1557 1558
	error=errno;
      if (error && ! write_error)
      {
	write_error=1;
	sql_print_error(ER(ER_ERROR_ON_WRITE),name,error);
      }
1559
      va_end(args);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1560 1561
    }
    VOID(pthread_mutex_unlock(&LOCK_log));
1562
    return error != 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1563
  }
1564
  return 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1565 1566
}

1567
bool MYSQL_LOG::flush_and_sync()
1568
{
1569 1570 1571 1572 1573
  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)
1574 1575
  {
    sync_binlog_counter= 0;
1576
    err=my_sync(fd, MYF(MY_WME));
1577
  }
1578
  return err;
1579 1580
}

1581
void MYSQL_LOG::start_union_events(THD *thd, query_id_t query_id_param)
1582 1583 1584 1585 1586
{
  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;
1587
  thd->binlog_evt_union.first_query_id= query_id_param;
1588 1589 1590 1591 1592 1593 1594 1595
}

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

1596 1597 1598 1599 1600 1601
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);
}

1602 1603 1604 1605
/*
  Write an event to the binary log
*/

1606
bool MYSQL_LOG::write(Log_event *event_info)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1607
{
1608 1609 1610 1611
  THD *thd= event_info->thd;
  bool error= 1;
  DBUG_ENTER("MYSQL_LOG::write(Log_event *)");

1612 1613
  if (thd->binlog_evt_union.do_union)
  {
1614 1615 1616 1617
    /*
      In Stored function; Remember that function call caused an update.
      We will log the function call to the binary log on function exit
    */
1618 1619 1620 1621 1622
    thd->binlog_evt_union.unioned_events= TRUE;
    thd->binlog_evt_union.unioned_events_trans |= event_info->cache_stmt;
    DBUG_RETURN(0);
  }
  
1623 1624
  pthread_mutex_lock(&LOCK_log);

1625
  /*
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1626 1627 1628 1629
     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.
  */
1630
  if (likely(is_open()))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1631
  {
1632
    IO_CACHE *file= &log_file;
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
1633
#ifdef HAVE_REPLICATION
1634
    /*
1635 1636 1637
      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)"
1638
    */
1639
    const char *local_db= event_info->get_db();
1640
    if ((thd && !(thd->options & OPTION_BIN_LOG)) ||
1641
	(!db_ok(local_db, binlog_do_db, binlog_ignore_db)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1642
    {
1643
      VOID(pthread_mutex_unlock(&LOCK_log));
1644
      DBUG_PRINT("error",("!db_ok('%s')", local_db));
1645
      DBUG_RETURN(0);
1646
    }
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
1647
#endif /* HAVE_REPLICATION */
1648

1649 1650 1651 1652 1653 1654 1655 1656
#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)
1657
      - or if the event asks for it (cache_stmt == TRUE).
1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668
    */
    if (opt_using_transactions && thd)
    {
      IO_CACHE *trans_log= (IO_CACHE*)thd->ha_data[binlog_hton.slot];

      if (event_info->get_cache_stmt())
      {
        if (!trans_log)
        {
          thd->ha_data[binlog_hton.slot]= trans_log= (IO_CACHE *)
            my_malloc(sizeof(IO_CACHE), MYF(MY_ZEROFILL));
1669 1670
          if (!trans_log || open_cached_file(trans_log, mysql_tmpdir,
                                             LOG_PREFIX,
1671 1672 1673 1674 1675 1676 1677
                                             binlog_cache_size, MYF(MY_WME)))
          {
            my_free((gptr)trans_log, MYF(MY_ALLOW_ZERO_PTR));
            thd->ha_data[binlog_hton.slot]= trans_log= 0;
            goto err;
          }
          trans_log->end_of_file= max_binlog_cache_size;
1678
          trans_register_ha(thd,
1679 1680
                            test(thd->options & (OPTION_NOT_AUTOCOMMIT |
                                                 OPTION_BEGIN)),
1681
                            &binlog_hton);
1682 1683
        }
        else if (!my_b_tell(trans_log))
1684
          trans_register_ha(thd,
1685 1686
                            test(thd->options & (OPTION_NOT_AUTOCOMMIT |
                                                 OPTION_BEGIN)),
1687
                            &binlog_hton);
1688 1689 1690 1691 1692 1693 1694 1695
        file= trans_log;
      }
      else if (trans_log && my_b_tell(trans_log))
        file= trans_log;
    }
#endif
    DBUG_PRINT("info",("event type=%d",event_info->get_type_code()));

1696 1697 1698 1699
    /*
      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
1700 1701

    /*
1702 1703
      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
1704 1705
    */

1706
    if (thd)
1707
    {
1708
      if (thd->last_insert_id_used_bin_log)
1709 1710
      {
	Intvar_log_event e(thd,(uchar) LAST_INSERT_ID_EVENT,
1711
			   thd->current_insert_id);
1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726
	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;
      }
1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742
      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;
	}
      }
1743
    }
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
1744

1745 1746
    /*
       Write the SQL command
1747
     */
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
1748

monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
1749
    if (event_info->write(file))
1750
      goto err;
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
1751

1752
    if (file == &log_file) // we are writing to the real log (disk)
1753
    {
1754
      if (flush_and_sync())
monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
1755
	goto err;
1756 1757
      signal_update();
      rotate_and_purge(RP_LOCK_LOG_IS_ALREADY_LOCKED);
1758
    }
monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
1759
    error=0;
1760

1761 1762 1763 1764
err:
    if (error)
    {
      if (my_errno == EFBIG)
bell@sanja.is.com.ua's avatar
bell@sanja.is.com.ua committed
1765
	my_message(ER_TRANS_CACHE_FULL, ER(ER_TRANS_CACHE_FULL), MYF(0));
1766 1767 1768 1769 1770
      else
	my_error(ER_ERROR_ON_WRITE, MYF(0), name, errno);
      write_error=1;
    }
  }
1771 1772

  pthread_mutex_unlock(&LOCK_log);
1773
  DBUG_RETURN(error);
1774 1775
}

1776 1777
void MYSQL_LOG::rotate_and_purge(uint flags)
{
1778 1779
  if (!(flags & RP_LOCK_LOG_IS_ALREADY_LOCKED))
    pthread_mutex_lock(&LOCK_log);
1780 1781
  if ((flags & RP_FORCE_ROTATE) ||
      (my_b_tell(&log_file) >= (my_off_t) max_size))
1782
  {
1783
    new_file(0);
1784 1785 1786
#ifdef HAVE_REPLICATION
    if (expire_logs_days)
    {
1787
      time_t purge_time= time(0) - expire_logs_days*24*60*60;
1788 1789 1790 1791 1792
      if (purge_time >= 0)
        purge_logs_before_date(purge_time);
    }
#endif
  }
1793 1794
  if (!(flags & RP_LOCK_LOG_IS_ALREADY_LOCKED))
    pthread_mutex_unlock(&LOCK_log);
1795
}
1796

1797 1798 1799 1800 1801 1802 1803 1804 1805
uint MYSQL_LOG::next_file_id()
{
  uint res;
  pthread_mutex_lock(&LOCK_log);
  res = file_id++;
  pthread_mutex_unlock(&LOCK_log);
  return res;
}

1806

1807 1808
/*
  Write a cached log entry to the binary log
1809

guilhem@mysql.com's avatar
guilhem@mysql.com committed
1810 1811
  SYNOPSIS
    write()
1812
    thd
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1813 1814
    cache		The cache to copy to the binlog

1815 1816
  NOTE
    - We only come here if there is something in the cache.
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
1817
    - The thing in the cache is always a complete transaction
1818 1819 1820 1821
    - '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
1822
      with BEGIN/COMMIT or BEGIN/ROLLBACK in the binary log.
1823 1824 1825
      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.
1826 1827
*/

1828
bool MYSQL_LOG::write(THD *thd, IO_CACHE *cache, Log_event *commit_event)
1829
{
1830
  DBUG_ENTER("MYSQL_LOG::write(THD *, IO_CACHE *, Log_event *)");
1831
  VOID(pthread_mutex_lock(&LOCK_log));
1832

1833 1834 1835
  /* NULL would represent nothing to replicate after ROLLBACK */
  DBUG_ASSERT(commit_event != NULL);

1836
  if (likely(is_open()))                       // Should always be true
1837
  {
1838 1839
    uint length;

1840
    /*
1841 1842
      Log "BEGIN" at the beginning of the transaction.
      which may contain more than 1 SQL statement.
1843
    */
1844
    if (thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN))
1845
    {
1846
      Query_log_event qinfo(thd, STRING_WITH_LEN("BEGIN"), TRUE, FALSE);
1847 1848 1849 1850
      /*
        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
1851 1852 1853
        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.
1854 1855 1856
        This is safer than thd->clear_error() against kills at shutdown.
      */
      qinfo.error_code= 0;
1857 1858 1859
      /*
        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
1860 1861 1862
	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.
1863
      */
1864 1865 1866
      if (qinfo.write(&log_file))
	goto err;
    }
1867
    /* Read from the file used to cache the queries .*/
1868
    if (reinit_io_cache(cache, READ_CACHE, 0, 0, 0))
1869
      goto err;
1870
    length=my_b_bytes_in_cache(cache);
1871
    DBUG_EXECUTE_IF("half_binlogged_transaction", length-=100;);
1872
    do
1873
    {
1874
      /* Write data to the binary log file */
monty@bitch.mysql.fi's avatar
monty@bitch.mysql.fi committed
1875
      if (my_b_write(&log_file, cache->read_pos, length))
1876
	goto err;
monty@bitch.mysql.fi's avatar
monty@bitch.mysql.fi committed
1877
      cache->read_pos=cache->read_end;		// Mark buffer used up
1878
      DBUG_EXECUTE_IF("half_binlogged_transaction", goto DBUG_skip_commit;);
1879
    } while ((length=my_b_fill(cache)));
1880

1881 1882
    if (commit_event->write(&log_file))
      goto err;
1883
#ifndef DBUG_OFF
1884
DBUG_skip_commit:
1885
#endif
1886
    if (flush_and_sync())
1887
      goto err;
1888
    DBUG_EXECUTE_IF("half_binlogged_transaction", abort(););
1889 1890
    if (cache->error)				// Error on read
    {
1891
      sql_print_error(ER(ER_ERROR_ON_READ), cache->file_name, errno);
1892
      write_error=1;				// Don't give more errors
1893
      goto err;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1894
    }
1895
    signal_update();
1896 1897 1898 1899 1900 1901 1902 1903 1904
    /*
      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)
1905 1906 1907 1908 1909
    {
      pthread_mutex_lock(&LOCK_prep_xids);
      prepared_xids++;
      pthread_mutex_unlock(&LOCK_prep_xids);
    }
1910 1911
    else
      rotate_and_purge(RP_LOCK_LOG_IS_ALREADY_LOCKED);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1912
  }
1913
  VOID(pthread_mutex_unlock(&LOCK_log));
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
1914

1915
  DBUG_RETURN(0);
1916 1917

err:
1918 1919 1920 1921 1922
  if (!write_error)
  {
    write_error= 1;
    sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
  }
1923
  VOID(pthread_mutex_unlock(&LOCK_log));
1924
  DBUG_RETURN(1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1925 1926
}

1927

1928
/*
1929
  Write to the slow query log.
1930
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1931

1932
bool MYSQL_LOG::write(THD *thd,const char *query, uint query_length,
1933
		      time_t query_start_arg)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1934
{
1935
  bool error=0;
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1936
  time_t current_time;
1937 1938
  if (!is_open())
    return 0;
monty@mysql.com's avatar
monty@mysql.com committed
1939 1940
  DBUG_ENTER("MYSQL_LOG::write");

guilhem@mysql.com's avatar
guilhem@mysql.com committed
1941
  VOID(pthread_mutex_lock(&LOCK_log));
1942
  if (is_open())
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1943 1944 1945 1946
  {						// Safety agains reopen
    int tmp_errno=0;
    char buff[80],*end;
    end=buff;
1947
    if (!(thd->options & OPTION_UPDATE_LOG))
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1948 1949
    {
      VOID(pthread_mutex_unlock(&LOCK_log));
monty@mysql.com's avatar
monty@mysql.com committed
1950
      DBUG_RETURN(0);
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1951
    }
1952
    if (!(specialflag & SPECIAL_SHORT_LOG_FORMAT) || query_start_arg)
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1953
    {
1954
      Security_context *sctx= thd->security_ctx;
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1955 1956
      current_time=time(NULL);
      if (current_time != last_time)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1957
      {
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972
        last_time=current_time;
        struct tm tm_tmp;
        struct tm *start;
        localtime_r(&current_time,&tm_tmp);
        start=&tm_tmp;
        /* Note that my_b_write() assumes it knows the length for this */
        sprintf(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);
        if (my_b_write(&log_file, (byte*) buff,24))
          tmp_errno=errno;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1973
      }
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1974
      if (my_b_printf(&log_file, "# User@Host: %s[%s] @ %s [%s]\n",
1975 1976 1977 1978 1979
                      sctx->priv_user ?
                      sctx->priv_user : "",
                      sctx->user ? sctx->user : "",
                      sctx->host ? sctx->host : "",
                      sctx->ip ? sctx->ip : "") ==
1980
          (uint) -1)
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999
        tmp_errno=errno;
    }
    if (query_start_arg)
    {
      /* For slow query log */
      if (my_b_printf(&log_file,
                      "# Query_time: %lu  Lock_time: %lu  Rows_sent: %lu  Rows_examined: %lu\n",
                      (ulong) (current_time - query_start_arg),
                      (ulong) (thd->time_after_lock - query_start_arg),
                      (ulong) thd->sent_row_count,
                      (ulong) thd->examined_row_count) == (uint) -1)
        tmp_errno=errno;
    }
    if (thd->db && strcmp(thd->db,db))
    {						// Database changed
      if (my_b_printf(&log_file,"use %s;\n",thd->db) == (uint) -1)
        tmp_errno=errno;
      strmov(db,thd->db);
    }
2000
    if (thd->last_insert_id_used_bin_log)
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2001 2002 2003 2004 2005 2006 2007
    {
      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)
    {
2008
      if (!(specialflag & SPECIAL_SHORT_LOG_FORMAT))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2009
      {
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2010 2011
        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
2012
      }
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2013 2014 2015 2016
    }
    if (thd->query_start_used)
    {
      if (query_start_arg != thd->query_start())
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2017
      {
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2018 2019 2020
        query_start_arg=thd->query_start();
        end=strmov(end,",timestamp=");
        end=int10_to_str((long) query_start_arg,end,10);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2021
      }
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045
    }
    if (end != buff)
    {
      *end++=';';
      *end='\n';
      if (my_b_write(&log_file, (byte*) "SET ",4) ||
          my_b_write(&log_file, (byte*) buff+1,(uint) (end-buff)))
        tmp_errno=errno;
    }
    if (!query)
    {
      end=strxmov(buff, "# administrator command: ",
                  command_name[thd->command], NullS);
      query_length=(ulong) (end-buff);
      query=buff;
    }
    if (my_b_write(&log_file, (byte*) query,query_length) ||
        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
2046
      {
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2047 2048
        write_error=1;
        sql_print_error(ER(ER_ERROR_ON_WRITE),name,error);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2049 2050 2051
      }
    }
  }
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2052
  VOID(pthread_mutex_unlock(&LOCK_log));
monty@mysql.com's avatar
monty@mysql.com committed
2053
  DBUG_RETURN(error);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2054 2055
}

2056

2057 2058 2059 2060 2061 2062
/*
  Wait until we get a signal that the binary log has been updated

  SYNOPSIS
    wait_for_update()
    thd			Thread variable
2063
    is_slave            If 0, the caller is the Binlog_dump thread from master;
2064 2065
                        if 1, the caller is the SQL thread from the slave. This
                        influences only thd->proc_info.
2066 2067 2068

  NOTES
    One must have a lock on LOCK_log before calling this function.
2069
    This lock will be released before return! That's required by
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2070
    THD::enter_cond() (see NOTES in sql_class.h).
2071 2072
*/

2073
void MYSQL_LOG::wait_for_update(THD* thd, bool is_slave)
2074
{
monty@mysql.com's avatar
monty@mysql.com committed
2075 2076
  const char *old_msg;
  DBUG_ENTER("wait_for_update");
2077

monty@mysql.com's avatar
monty@mysql.com committed
2078
  old_msg= thd->enter_cond(&update_cond, &LOCK_log,
2079
                           is_slave ?
monty@mysql.com's avatar
monty@mysql.com committed
2080
                           "Has read all relay log; waiting for the slave I/O "
2081
                           "thread to update it" :
monty@mysql.com's avatar
monty@mysql.com committed
2082
                           "Has sent all binlog to slave; waiting for binlog "
2083
                           "to be updated");
2084 2085
  pthread_cond_wait(&update_cond, &LOCK_log);
  thd->exit_cond(old_msg);
monty@mysql.com's avatar
monty@mysql.com committed
2086
  DBUG_VOID_RETURN;
2087
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2088

2089

2090 2091 2092 2093 2094
/*
  Close the log file

  SYNOPSIS
    close()
2095 2096 2097 2098 2099
    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
2100 2101 2102 2103

  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
2104 2105
*/

2106
void MYSQL_LOG::close(uint exiting)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2107
{					// One can't set log_type here!
2108 2109
  DBUG_ENTER("MYSQL_LOG::close");
  DBUG_PRINT("enter",("exiting: %d", (int) exiting));
2110
  if (log_type != LOG_CLOSED && log_type != LOG_TO_BE_OPENED)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2111
  {
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
2112
#ifdef HAVE_REPLICATION
2113 2114
    if (log_type == LOG_BIN && !no_auto_events &&
	(exiting & LOG_CLOSE_STOP_EVENT))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2115 2116
    {
      Stop_log_event s;
2117
      s.write(&log_file);
2118
      bytes_written+= s.data_written;
2119
      signal_update();
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2120
    }
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
2121
#endif /* HAVE_REPLICATION */
2122
    end_io_cache(&log_file);
2123 2124

    /* don't pwrite in a file opened with O_APPEND - it doesn't work */
2125
    if (log_file.type == WRITE_CACHE && log_type == LOG_BIN)
2126 2127
    {
      my_off_t offset= BIN_LOG_HEADER_SIZE + FLAGS_OFFSET;
serg@serg.mylan's avatar
serg@serg.mylan committed
2128
      byte flags=0; // clearing LOG_EVENT_BINLOG_IN_USE_F
2129
      my_pwrite(log_file.file, &flags, 1, offset, MYF(0));
2130 2131
    }

2132 2133 2134 2135 2136 2137
    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
2138 2139
    {
      write_error=1;
2140
      sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2141 2142
    }
  }
2143 2144 2145 2146 2147 2148

  /*
    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.
  */

2149
  if ((exiting & LOG_CLOSE_INDEX) && my_b_inited(&index_file))
2150
  {
2151 2152
    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
2153
    {
2154 2155
      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
2156
    }
2157
  }
2158
  log_type= (exiting & LOG_CLOSE_TO_BE_OPENED) ? LOG_TO_BE_OPENED : LOG_CLOSED;
2159
  safeFree(name);
2160
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2161 2162 2163
}


2164 2165 2166 2167 2168 2169 2170 2171 2172
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
2173 2174
  DBUG_ENTER("MYSQL_LOG::set_max_size");
  pthread_mutex_lock(&LOCK_log);
2175 2176
  if (is_open())
    max_size= max_size_arg;
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2177 2178
  pthread_mutex_unlock(&LOCK_log);
  DBUG_VOID_RETURN;
2179
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2180

2181

2182 2183
/*
  Check if a string is a valid number
2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197

  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
2198
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210

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++;
2211 2212
  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
2213 2214 2215 2216 2217 2218 2219
  {
    flag=1;
    str++;
  }
  if (*str == '.')
  {
    for (str++ ;
2220
	 my_isdigit(files_charset_info,*str) ||
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239
	   (allow_wildcards && (*str == wild_many || *str == wild_one)) ;
	 str++, flag=1) ;
  }
  if (*str != 0 || flag == 0)
    DBUG_RETURN(0);
  if (res)
    *res=atol(start);
  DBUG_RETURN(1);			/* Number ok */
} /* test_if_number */


void sql_perror(const char *message)
{
#ifdef HAVE_STRERROR
  sql_print_error("%s: %s",message, strerror(errno));
#else
  perror(message);
#endif
}
2240

monty@mysql.com's avatar
monty@mysql.com committed
2241

2242 2243 2244
bool flush_error_log()
{
  bool result=0;
2245
  if (opt_error_log)
2246 2247 2248 2249
  {
    char err_renamed[FN_REFLEN], *end;
    end= strmake(err_renamed,log_error_file,FN_REFLEN-4);
    strmov(end, "-old");
2250
    VOID(pthread_mutex_lock(&LOCK_error_log));
2251 2252 2253 2254 2255 2256
#ifdef __WIN__
    char err_temp[FN_REFLEN+4];
    /*
     On Windows is necessary a temporary file for to rename
     the current error file.
    */
2257
    strxmov(err_temp, err_renamed,"-tmp",NullS);
2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270
    (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)
2271
             my_fwrite(stderr, (byte*) buf, bytes, MYF(0));
2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284
        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
2285
    VOID(pthread_mutex_unlock(&LOCK_error_log));
2286 2287 2288
  }
   return result;
}
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
2289

monty@mysql.com's avatar
monty@mysql.com committed
2290 2291 2292 2293 2294 2295 2296
void MYSQL_LOG::signal_update()
{
  DBUG_ENTER("MYSQL_LOG::signal_update");
  pthread_cond_broadcast(&update_cond);
  DBUG_VOID_RETURN;
}

2297
#ifdef __NT__
2298
static void print_buffer_to_nt_eventlog(enum loglevel level, char *buff,
2299
                                        size_t length, size_t buffLen)
2300 2301
{
  HANDLE event;
2302
  char   *buffptr= buff;
serg@serg.mylan's avatar
serg@serg.mylan committed
2303
  DBUG_ENTER("print_buffer_to_nt_eventlog");
2304

2305 2306
  /* Add ending CR/LF's to string, overwrite last chars if necessary */
  strmov(buffptr+min(length, buffLen-5), "\r\n\r\n");
2307

monty@mysql.com's avatar
monty@mysql.com committed
2308 2309
  setup_windows_event_source();
  if ((event= RegisterEventSource(NULL,"MySQL")))
2310
  {
monty@mysql.com's avatar
monty@mysql.com committed
2311
    switch (level) {
2312
      case ERROR_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
2313
        ReportEvent(event, EVENTLOG_ERROR_TYPE, 0, MSG_DEFAULT, NULL, 1, 0,
2314
                    (LPCSTR*)&buffptr, NULL);
2315 2316
        break;
      case WARNING_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
2317
        ReportEvent(event, EVENTLOG_WARNING_TYPE, 0, MSG_DEFAULT, NULL, 1, 0,
2318
                    (LPCSTR*) &buffptr, NULL);
2319 2320
        break;
      case INFORMATION_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
2321
        ReportEvent(event, EVENTLOG_INFORMATION_TYPE, 0, MSG_DEFAULT, NULL, 1,
2322
                    0, (LPCSTR*) &buffptr, NULL);
2323 2324 2325 2326 2327 2328 2329
        break;
    }
    DeregisterEventSource(event);
  }

  DBUG_VOID_RETURN;
}
monty@mysql.com's avatar
monty@mysql.com committed
2330 2331
#endif /* __NT__ */

2332 2333

/*
monty@mysql.com's avatar
monty@mysql.com committed
2334 2335
  Prints a printf style message to the error log and, under NT, to the
  Windows event log.
2336 2337 2338

  SYNOPSIS
    vprint_msg_to_log()
monty@mysql.com's avatar
monty@mysql.com committed
2339 2340
    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
2341
    args                   va_list list of arguments for the message
2342 2343 2344 2345

  NOTE

  IMPLEMENTATION
monty@mysql.com's avatar
monty@mysql.com committed
2346 2347
    This function prints the message into a buffer and then sends that buffer
    to other functions to write that message to other logging sources.
2348 2349 2350

  RETURN VALUES
    void
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2351
*/
monty@mysql.com's avatar
monty@mysql.com committed
2352

2353 2354 2355 2356 2357 2358
#ifdef EMBEDDED_LIBRARY
void vprint_msg_to_log(enum loglevel level __attribute__((unused)),
                       const char *format __attribute__((unused)),
                       va_list argsi __attribute__((unused)))
{}
#else /*!EMBEDDED_LIBRARY*/
2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389
static void print_buffer_to_file(enum loglevel level, const char *buffer)
{
  time_t skr;
  struct tm tm_tmp;
  struct tm *start;
  DBUG_ENTER("print_buffer_to_file");
  DBUG_PRINT("enter",("buffer: %s", buffer));

  VOID(pthread_mutex_lock(&LOCK_error_log));

  skr=time(NULL);
  localtime_r(&skr, &tm_tmp);
  start=&tm_tmp;
  fprintf(stderr, "%02d%02d%02d %2d:%02d:%02d [%s] %s\n",
          start->tm_year % 100,
          start->tm_mon+1,
          start->tm_mday,
          start->tm_hour,
          start->tm_min,
          start->tm_sec,
          (level == ERROR_LEVEL ? "ERROR" : level == WARNING_LEVEL ?
           "Warning" : "Note"),
          buffer);

  fflush(stderr);

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


2390
void vprint_msg_to_log(enum loglevel level, const char *format, va_list args)
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2391 2392
{
  char   buff[1024];
2393
  size_t length;
2394
  DBUG_ENTER("vprint_msg_to_log");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2395

2396
  length= my_vsnprintf(buff, sizeof(buff), format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
2397
  print_buffer_to_file(level, buff);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2398 2399

#ifdef __NT__
monty@mysql.com's avatar
monty@mysql.com committed
2400
  print_buffer_to_nt_eventlog(level, buff, length, sizeof(buff));
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2401
#endif
2402

rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2403 2404
  DBUG_VOID_RETURN;
}
2405
#endif /*EMBEDDED_LIBRARY*/
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2406

2407

monty@mysql.com's avatar
monty@mysql.com committed
2408
void sql_print_error(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2409 2410
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
2411
  DBUG_ENTER("sql_print_error");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2412

serg@serg.mylan's avatar
serg@serg.mylan committed
2413 2414 2415
  va_start(args, format);
  vprint_msg_to_log(ERROR_LEVEL, format, args);
  va_end(args);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2416 2417 2418 2419 2420

  DBUG_VOID_RETURN;
}


monty@mysql.com's avatar
monty@mysql.com committed
2421
void sql_print_warning(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2422 2423
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
2424
  DBUG_ENTER("sql_print_warning");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2425

serg@serg.mylan's avatar
serg@serg.mylan committed
2426 2427 2428
  va_start(args, format);
  vprint_msg_to_log(WARNING_LEVEL, format, args);
  va_end(args);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2429 2430 2431 2432 2433

  DBUG_VOID_RETURN;
}


monty@mysql.com's avatar
monty@mysql.com committed
2434
void sql_print_information(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2435 2436
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
2437
  DBUG_ENTER("sql_print_information");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2438

serg@serg.mylan's avatar
serg@serg.mylan committed
2439 2440 2441
  va_start(args, format);
  vprint_msg_to_log(INFORMATION_LEVEL, format, args);
  va_end(args);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2442

guilhem@mysql.com's avatar
guilhem@mysql.com committed
2443 2444
  DBUG_VOID_RETURN;
}
2445

2446

2447 2448 2449 2450 2451 2452 2453 2454
/********* 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
2455 2456
  states - active, syncing, pool.
  there could be only one page in active or syncing states,
2457
  but many in pool - pool is fifo queue.
2458
  usual lifecycle of a page is pool->active->syncing->pool
2459
  "active" page - is a page where new xid's are logged.
2460 2461
  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.
2462
  when the sync is done the page is moved to a pool and an active page
2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475
  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.
2476 2477 2478 2479
  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)

2480 2481
  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
2482 2483
  dirty - we don't sync removals to disk.
*/
2484 2485 2486 2487 2488

ulong tc_log_page_waits= 0;

#ifdef HAVE_MMAP

2489 2490
#define TC_LOG_HEADER_SIZE (sizeof(tc_log_magic)+1)

2491
static const char tc_log_magic[]={(char) 254, 0x23, 0x05, 0x74};
2492

2493
ulong opt_tc_log_size= TC_LOG_MIN_SIZE;
2494
ulong tc_log_max_pages_used=0, tc_log_page_size=0, tc_log_cur_pages_used=0;
2495 2496 2497 2498 2499 2500 2501

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

2502
  DBUG_ASSERT(total_ha_2pc > 1);
2503 2504
  DBUG_ASSERT(opt_name && opt_name[0]);

serg@serg.mylan's avatar
serg@serg.mylan committed
2505
  tc_log_page_size= my_getpagesize();
2506 2507 2508
  DBUG_ASSERT(TC_LOG_PAGE_SIZE % tc_log_page_size == 0);

  fn_format(logname,opt_name,mysql_data_home,"",MY_UNPACK_FILENAME);
2509
  if ((fd= my_open(logname, O_RDWR, MYF(0))) < 0)
2510
  {
2511 2512
    if (my_errno != ENOENT)
      goto err;
2513 2514
    if (using_heuristic_recover())
      return 1;
2515
    if ((fd= my_create(logname, CREATE_MODE, O_RDWR, MYF(MY_WME))) < 0)
2516 2517
      goto err;
    inited=1;
2518
    file_length= opt_tc_log_size;
2519 2520 2521 2522 2523
    if (my_chsize(fd, file_length, 0, MYF(MY_WME)))
      goto err;
  }
  else
  {
2524 2525
    inited= 1;
    crashed= TRUE;
serg@serg.mylan's avatar
serg@serg.mylan committed
2526
    sql_print_information("Recovering after a crash using %s", opt_name);
2527 2528 2529 2530 2531 2532
    if (tc_heuristic_recover)
    {
      sql_print_error("Cannot perform automatic crash recovery when "
                      "--tc-heuristic-recover is used");
      goto err;
    }
2533
    file_length= my_seek(fd, 0L, MY_SEEK_END, MYF(MY_WME+MY_FAE));
2534 2535 2536 2537
    if (file_length == MY_FILEPOS_ERROR || file_length % tc_log_page_size)
      goto err;
  }

2538
  data= (uchar *)my_mmap(0, (size_t)file_length, PROT_READ|PROT_WRITE,
2539 2540 2541 2542 2543 2544 2545 2546
                        MAP_NOSYNC|MAP_SHARED, fd, 0);
  if (data == MAP_FAILED)
  {
    my_errno=errno;
    goto err;
  }
  inited=2;

2547
  npages=(uint)file_length/tc_log_page_size;
2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564
  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);
2565
  pages[0].start=pages[0].end-pages[0].size;
2566 2567 2568
  pages[npages-1].next=0;
  inited=4;

2569
  if (crashed && recover())
2570
      goto err;
2571

2572
  memcpy(data, tc_log_magic, sizeof(tc_log_magic));
2573
  data[sizeof(tc_log_magic)]= (uchar)total_ha_2pc;
2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584
  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;

2585
  syncing= 0;
2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597
  active=pages;
  pool=pages+1;
  pool_last=pages+npages-1;

  return 0;

err:
  close();
  return 1;
}

/*
2598 2599
  there is no active page, let's got one from the pool

2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611
  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;

2612 2613 2614
  if (syncing)
    pthread_mutex_lock(&LOCK_pool);

2615 2616 2617
  do
  {
    best_p= p= &pool;
2618 2619
    if ((*p)->waiters == 0) // can the first page be used ?
      break;                // yes - take it.
2620

2621
    best_free=0;            // no - trying second strategy
2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633
    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;
2634
  if (active->free == active->size) // we've chosen an empty page
2635 2636 2637 2638 2639
  {
    tc_log_cur_pages_used++;
    set_if_bigger(tc_log_max_pages_used, tc_log_cur_pages_used);
  }

2640
  if ((*best_p)->next)              // unlink the page from the pool
2641 2642 2643
    *best_p=(*best_p)->next;
  else
    pool_last=*best_p;
2644 2645 2646

  if (syncing)
    pthread_mutex_unlock(&LOCK_pool);
2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657
}

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);
2658
  return 1; // always return 1
2659 2660 2661
}

/*
2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679
  Record that transaction XID is committed on the persistent storage

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

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

  IMPLEMENTATION
   If tc_log == MYSQL_LOG then tc_log writes transaction to binlog and
   records XID in a special Xid_log_event.
   If tc_log = TC_LOG_MMAP then xid is written in a special memory-mapped
   log.
2680 2681

  RETURN
2682 2683 2684 2685 2686
    0  Error
    #  "cookie", a number that will be passed as an argument
       to unlog() call. tc_log can define it any way it wants,
       and use for whatever purposes. TC_LOG_MMAP sets it
       to the position in memory where xid was logged to.
2687 2688
*/

2689
int TC_LOG_MMAP::log_xid(THD *thd, my_xid xid)
2690 2691 2692 2693 2694 2695 2696
{
  int err;
  PAGE *p;
  ulong cookie;

  pthread_mutex_lock(&LOCK_active);

2697 2698 2699 2700 2701 2702 2703
  /*
    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.
  */
2704 2705 2706
  while (unlikely(active && active->free == 0))
    pthread_cond_wait(&COND_active, &LOCK_active);

2707
  /* no active page ? take one from the pool */
2708 2709 2710 2711 2712 2713
  if (active == 0)
    get_active_from_pool();

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

2714
  /* searching for an empty slot */
2715 2716 2717 2718 2719 2720
  while (*p->ptr)
  {
    p->ptr++;
    DBUG_ASSERT(p->ptr < p->end);               // because p->free > 0
  }

2721 2722
  /* found! store xid there and mark the page dirty */
  cookie= (ulong)((uchar *)p->ptr - data);      // can never be zero
2723 2724 2725 2726 2727
  *p->ptr++= xid;
  p->free--;
  p->state= DIRTY;

  /* to sync or not to sync - this is the question */
2728
  pthread_mutex_unlock(&LOCK_active);
2729 2730 2731
  pthread_mutex_lock(&LOCK_sync);
  pthread_mutex_unlock(&p->lock);

2732
  if (syncing)
2733 2734
  {                                          // somebody's syncing. let's wait
    p->waiters++;
2735
    /*
monty@mishka.local's avatar
monty@mishka.local committed
2736
      note - it must be while (), not do ... while () here
2737 2738 2739
      as p->state may be not DIRTY when we come here
    */
    while (p->state == DIRTY && syncing)
2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750
      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
2751 2752 2753 2754
  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
2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767
  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;

2768
  DBUG_ASSERT(syncing != active);
2769 2770 2771 2772 2773

  /*
    sit down and relax - this can take a while...
    note - no locks are held at this point
  */
2774
  err= my_msync(fd, syncing->start, 1, MS_SYNC);
2775

2776
  /* page is synced. let's move it to the pool */
2777
  pthread_mutex_lock(&LOCK_pool);
2778 2779 2780 2781 2782
  pool_last->next=syncing;
  pool_last=syncing;
  syncing->next=0;
  syncing->state= err ? ERROR : POOL;
  pthread_cond_broadcast(&syncing->cond);    // signal "sync done"
2783 2784 2785
  pthread_cond_signal(&COND_pool);           // in case somebody's waiting
  pthread_mutex_unlock(&LOCK_pool);

2786
  /* marking 'syncing' slot free */
2787
  pthread_mutex_lock(&LOCK_sync);
2788
  syncing=0;
2789 2790 2791 2792 2793
  pthread_cond_signal(&active->cond);        // wake up a new syncer
  pthread_mutex_unlock(&LOCK_sync);
  return err;
}

2794 2795 2796 2797
/*
  erase xid from the page, update page free space counters/pointers.
  cookie points directly to the memory where xid was logged
*/
2798

2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811
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);
2812
  if (p->free == p->size)               // the page is completely empty
2813 2814
    statistic_decrement(tc_log_cur_pages_used, &LOCK_status);
  if (p->waiters == 0)                 // the page is in pool and ready to rock
2815
    pthread_cond_signal(&COND_pool);   // ping ... for overflow()
2816 2817 2818 2819 2820
  pthread_mutex_unlock(&p->lock);
}

void TC_LOG_MMAP::close()
{
serg@serg.mylan's avatar
serg@serg.mylan committed
2821
  uint i;
2822 2823 2824 2825 2826 2827 2828 2829 2830
  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
2831
    for (i=0; i < npages; i++)
2832 2833 2834 2835 2836 2837 2838 2839 2840
    {
      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
2841
    my_munmap((byte*)data, (size_t)file_length);
2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866
  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)
  {
2867
    sql_print_error("Recovery failed! You must enable "
2868 2869 2870
                    "exactly %d storage engines that support "
                    "two-phase commit protocol",
                    data[sizeof(tc_log_magic)]);
2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888
    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);
2889
  bzero(data, (size_t)file_length);
2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900
  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
2901 2902 2903 2904 2905
#endif

TC_LOG *tc_log;
TC_LOG_DUMMY tc_log_dummy;
TC_LOG_MMAP  tc_log_mmap;
2906

2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918
/*
  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.
*/

2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931
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 ******/
2932
#define TC_LOG_BINLOG MYSQL_LOG
2933 2934 2935 2936 2937 2938 2939 2940

/*
  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!
*/

2941
int TC_LOG_BINLOG::open(const char *opt_name)
2942
{
2943 2944
  LOG_INFO log_info;
  int      error= 1;
2945 2946

  DBUG_ASSERT(total_ha_2pc > 1);
2947
  DBUG_ASSERT(opt_name && opt_name[0]);
2948 2949 2950 2951

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

2952 2953 2954 2955 2956 2957 2958
  if (!my_b_inited(&index_file))
  {
    /* There was a failure to open the index file, can't open the binlog */
    cleanup();
    return 1;
  }

2959
  if (using_heuristic_recover())
2960 2961 2962 2963
  {
    /* generate a new binlog to mask a corrupted one */
    open(opt_name, LOG_BIN, 0, WRITE_CACHE, 0, max_binlog_size, 0);
    cleanup();
2964
    return 1;
2965
  }
2966

2967
  if ((error= find_log_pos(&log_info, NullS, 1)))
2968
  {
2969 2970 2971 2972 2973
    if (error != LOG_INFO_EOF)
      sql_print_error("find_log_pos() failed (error: %d)", error);
    else
      error= 0;
    goto err;
2974
  }
2975 2976 2977 2978 2979

  {
    const char *errmsg;
    IO_CACHE    log;
    File        file;
2980 2981
    Log_event  *ev=0;
    Format_description_log_event fdle(BINLOG_VERSION);
2982
    char        log_name[FN_REFLEN];
2983 2984 2985 2986

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

2987
    do
2988
    {
2989 2990 2991 2992 2993 2994 2995
      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;
2996
    }
2997

2998
    if ((file= open_binlog(&log, log_name, &errmsg)) < 0)
2999 3000 3001 3002 3003
    {
      sql_print_error("%s", errmsg);
      goto err;
    }

3004 3005 3006
    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
3007 3008
    {
      sql_print_information("Recovering after a crash using %s", opt_name);
3009
      error= recover(&log, (Format_description_log_event *)ev);
serg@serg.mylan's avatar
serg@serg.mylan committed
3010
    }
3011 3012
    else
      error=0;
3013 3014 3015 3016 3017 3018 3019 3020 3021 3022

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

    if (error)
      goto err;
  }

err:
3023
  return error;
3024 3025
}

3026 3027
/* this is called on shutdown, after ha_panic */
void TC_LOG_BINLOG::close()
3028 3029 3030 3031 3032 3033
{
  DBUG_ASSERT(prepared_xids==0);
  pthread_mutex_destroy(&LOCK_prep_xids);
  pthread_cond_destroy (&COND_prep_xids);
}

3034 3035 3036 3037 3038 3039 3040
/*
  TODO group commit

  RETURN
         0  - error
         1  - success
*/
3041
int TC_LOG_BINLOG::log_xid(THD *thd, my_xid xid)
3042 3043
{
  Xid_log_event xle(thd, xid);
3044
  IO_CACHE *trans_log= (IO_CACHE*)thd->ha_data[binlog_hton.slot];
3045
  return !binlog_end_trans(thd, trans_log, &xle);  // invert return value
3046 3047
}

3048
void TC_LOG_BINLOG::unlog(ulong cookie, my_xid xid)
3049
{
3050 3051
  pthread_mutex_lock(&LOCK_prep_xids);
  if (--prepared_xids == 0)
3052
    pthread_cond_signal(&COND_prep_xids);
3053
  pthread_mutex_unlock(&LOCK_prep_xids);
3054
  rotate_and_purge(0);     // as ::write() did not rotate
3055 3056
}

3057
int TC_LOG_BINLOG::recover(IO_CACHE *log, Format_description_log_event *fdle)
3058 3059 3060 3061 3062 3063
{
  Log_event  *ev;
  HASH xids;
  MEM_ROOT mem_root;

  if (! fdle->is_valid() ||
3064
      hash_init(&xids, &my_charset_bin, TC_LOG_PAGE_SIZE/3, 0,
3065 3066 3067
            sizeof(my_xid), 0, 0, MYF(0)))
    goto err1;

3068
  init_alloc_root(&mem_root, TC_LOG_PAGE_SIZE, TC_LOG_PAGE_SIZE);
3069

3070 3071
  fdle->flags&= ~LOG_EVENT_BINLOG_IN_USE_F; // abort on the first error

3072 3073 3074 3075 3076
  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
3077 3078
      byte *x=(byte *)memdup_root(&mem_root, (char *)& xev->xid,
                                  sizeof(xev->xid));
3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103
      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;
}