log.cc 81.2 KB
Newer Older
1
/* Copyright (C) 2000-2003 MySQL AB
serg@serg.mylan's avatar
serg@serg.mylan committed
2

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

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

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


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

21 22 23 24
#ifdef __EMX__
#include <io.h>
#endif

bk@work.mysql.com's avatar
bk@work.mysql.com committed
25 26
#include "mysql_priv.h"
#include "sql_acl.h"
27
#include "sql_repl.h"
bk@work.mysql.com's avatar
bk@work.mysql.com committed
28 29 30 31 32

#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
33 34 35 36
#ifdef __NT__
#include "message.h"
#endif

37
MYSQL_LOG mysql_log, mysql_slow_log, mysql_bin_log;
38
ulong sync_binlog_counter= 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
39 40 41

static bool test_if_number(const char *str,
			   long *res, bool allow_wildcards);
42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253
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);

static handlerton binlog_hton = {
  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 */
  NULL                          /* rollback_by_xid */
};

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

handlerton *binlog_init()
{
  return &binlog_hton;
}

static int binlog_close_connection(THD *thd)
{
  IO_CACHE *trans_log= (IO_CACHE*)thd->ha_data[binlog_hton.slot];
  close_cached_file(trans_log);
  my_free((gptr)trans_log, MYF(0));
  return 0;
}

static inline void binlog_cleanup_trans(IO_CACHE *trans_log)
{
  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;
}

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

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

  if (!my_b_tell(trans_log))
  {
    // we're here because trans_log was flushed in MYSQL_LOG::log()
    DBUG_RETURN(0);
  }

  /* Update the binary log as we have cached some queries */
  error= mysql_bin_log.write(thd, trans_log, 1);
  binlog_cleanup_trans(trans_log);
  DBUG_RETURN(error);
}

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");
  /*
    first two conditions here are guaranteed - see trans_register_ha()
    call below. The third one must be true. If it is not, we're registering
    unnecessary, doing extra work. The cause should be found and eliminated
  */
  DBUG_ASSERT(all && mysql_bin_log.is_open() && my_b_tell(trans_log));
  /*
     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...)
  */
  if (unlikely(thd->options & OPTION_STATUS_NO_TRANS_UPDATE))
    error= mysql_bin_log.write(thd, trans_log, 0);
  binlog_cleanup_trans(trans_log);
  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");
  DBUG_ASSERT(mysql_bin_log.is_open() && my_b_tell(trans_log));

  *(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");
  DBUG_ASSERT(mysql_bin_log.is_open() && my_b_tell(trans_log));

  /*
    Write ROLLBACK TO SAVEPOINT to the binlog cache if we have updated some
    non-transactional table. Otherwise, truncate the binlog cache starting
    from the SAVEPOINT command.
  */
  if (unlikely(thd->options & OPTION_STATUS_NO_TRANS_UPDATE))
  {
    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");

  if ((file = my_open(log_file_name, O_RDONLY | O_BINARY, MYF(MY_WME))) < 0)
  {
    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
254

rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
255 256
#ifdef __NT__
static int eventSource = 0;
serg@serg.mylan's avatar
serg@serg.mylan committed
257

258
void setup_windows_event_source()
monty@mysql.com's avatar
monty@mysql.com committed
259
{
260
  HKEY    hRegKey= NULL;
monty@mysql.com's avatar
monty@mysql.com committed
261 262 263
  DWORD   dwError= 0;
  TCHAR   szPath[MAX_PATH];
  DWORD dwTypes;
264

monty@mysql.com's avatar
monty@mysql.com committed
265 266 267
  if (eventSource)               // Ensure that we are only called once
    return;
  eventSource= 1;
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
268

monty@mysql.com's avatar
monty@mysql.com committed
269
  // Create the event source registry key
270
  dwError= RegCreateKey(HKEY_LOCAL_MACHINE,
monty@mysql.com's avatar
monty@mysql.com committed
271 272
                          "SYSTEM\\CurrentControlSet\\Services\\EventLog\\Application\\MySQL", 
                          &hRegKey);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
273

monty@mysql.com's avatar
monty@mysql.com committed
274 275
  /* Name of the PE module that contains the message resource */
  GetModuleFileName(NULL, szPath, MAX_PATH);
serg@serg.mylan's avatar
serg@serg.mylan committed
276

rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
277
  /* Register EventMessageFile */
278
  dwError = RegSetValueEx(hRegKey, "EventMessageFile", 0, REG_EXPAND_SZ,
monty@mysql.com's avatar
monty@mysql.com committed
279
                          (PBYTE) szPath, strlen(szPath)+1);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
280

monty@mysql.com's avatar
monty@mysql.com committed
281 282 283 284 285
  /* 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
286

monty@mysql.com's avatar
monty@mysql.com committed
287
  RegCloseKey(hRegKey);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
288 289
}

monty@mysql.com's avatar
monty@mysql.com committed
290
#endif /* __NT__ */
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
291 292


bk@work.mysql.com's avatar
bk@work.mysql.com committed
293 294 295 296 297 298 299 300
/****************************************************************************
** 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)
{
301 302 303 304
  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
305
  reg1 struct fileinfo *file_info;
306 307
  ulong                 max_found=0;

bk@work.mysql.com's avatar
bk@work.mysql.com committed
308 309
  DBUG_ENTER("find_uniq_filename");

310 311 312 313
  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
314
  *end='.';
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
315
  length= (uint) (end-start+1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333

  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++='.';
334
  sprintf(end,"%06ld",max_found+1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
335 336 337
  DBUG_RETURN(0);
}

338

339
MYSQL_LOG::MYSQL_LOG()
340
  :bytes_written(0), last_time(0), query_start(0), name(0),
341
   file_id(1), open_count(1), log_type(LOG_CLOSED), write_error(0), inited(0),
342
   need_start_event(1), prepared_xids(0), description_event_for_exec(0),
343
   description_event_for_queue(0)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
344 345
{
  /*
guilhem@mysql.com's avatar
guilhem@mysql.com committed
346 347
    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
348
    called only in main(). Doing initialization here would make it happen
349
    before main().
bk@work.mysql.com's avatar
bk@work.mysql.com committed
350
  */
351 352
  index_file_name[0] = 0;
  bzero((char*) &log_file,sizeof(log_file));
353
  bzero((char*) &index_file, sizeof(index_file));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
354 355
}

356

bk@work.mysql.com's avatar
bk@work.mysql.com committed
357
MYSQL_LOG::~MYSQL_LOG()
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
358 359 360 361
{
  cleanup();
}

362
/* this is called only once */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
363

monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
364
void MYSQL_LOG::cleanup()
bk@work.mysql.com's avatar
bk@work.mysql.com committed
365
{
366
  DBUG_ENTER("cleanup");
bk@work.mysql.com's avatar
bk@work.mysql.com committed
367
  if (inited)
368
  {
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
369
    inited= 0;
370
    close(LOG_CLOSE_INDEX|LOG_CLOSE_STOP_EVENT);
371 372
    delete description_event_for_queue;
    delete description_event_for_exec;
373 374
    (void) pthread_mutex_destroy(&LOCK_log);
    (void) pthread_mutex_destroy(&LOCK_index);
375
    (void) pthread_cond_destroy(&update_cond);
376
  }
377
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
378 379
}

380

bk@work.mysql.com's avatar
bk@work.mysql.com committed
381 382
int MYSQL_LOG::generate_new_name(char *new_name, const char *log_name)
{      
383 384
  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
385 386 387 388 389 390 391 392 393 394 395 396 397
  {
    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;
}

398

399
void MYSQL_LOG::init(enum_log_type log_type_arg,
400
		     enum cache_type io_cache_type_arg,
401 402
		     bool no_auto_events_arg,
                     ulong max_size_arg)
403
{
404
  DBUG_ENTER("MYSQL_LOG::init");
405
  log_type = log_type_arg;
406
  io_cache_type = io_cache_type_arg;
407
  no_auto_events = no_auto_events_arg;
408
  max_size=max_size_arg;
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
409
  DBUG_PRINT("info",("log_type: %d max_size: %lu", log_type, max_size));
410
  DBUG_VOID_RETURN;
411 412
}

413

guilhem@mysql.com's avatar
guilhem@mysql.com committed
414 415
void MYSQL_LOG::init_pthread_objects()
{
416 417
  DBUG_ASSERT(inited == 0);
  inited= 1;
guilhem@mysql.com's avatar
guilhem@mysql.com committed
418 419 420
  (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);
421 422
}

bk@work.mysql.com's avatar
bk@work.mysql.com committed
423

424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440
/*
  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
*/

bool MYSQL_LOG::open(const char *log_name, enum_log_type log_type_arg,
		     const char *new_name, const char *index_file_name_arg,
		     enum cache_type io_cache_type_arg,
441
		     bool no_auto_events_arg,
442 443
                     ulong max_size_arg,
                     bool null_created_arg)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
444
{
445
  char buff[512];
446
  File file= -1, index_file_nr= -1;
447
  int open_flags = O_CREAT | O_BINARY;
448
  DBUG_ENTER("MYSQL_LOG::open");
449 450 451 452
  DBUG_PRINT("enter",("log_type: %d",(int) log_type));

  last_time=query_start=0;
  write_error=0;
453

454
  init(log_type_arg,io_cache_type_arg,no_auto_events_arg,max_size_arg);
455 456 457
  
  if (!(name=my_strdup(log_name,MYF(MY_WME))))
    goto err;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
458 459 460
  if (new_name)
    strmov(log_file_name,new_name);
  else if (generate_new_name(log_file_name, name))
461
    goto err;
462 463
  
  if (io_cache_type == SEQ_READ_APPEND)
464
    open_flags |= O_RDWR | O_APPEND;
465
  else
466
    open_flags |= O_WRONLY | (log_type == LOG_BIN ? 0 : O_APPEND);
467

bk@work.mysql.com's avatar
bk@work.mysql.com committed
468
  db[0]=0;
469
  open_count++;
470
  if ((file=my_open(log_file_name,open_flags,
471
		    MYF(MY_WME | ME_WAITTANG))) < 0 ||
472
      init_io_cache(&log_file, file, IO_SIZE, io_cache_type,
473 474
		    my_tell(file,MYF(MY_WME)), 0, MYF(MY_WME | MY_NABP)))
    goto err;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
475

476 477
  switch (log_type) {
  case LOG_NORMAL:
bk@work.mysql.com's avatar
bk@work.mysql.com committed
478
  {
479
    char *end;
480
    int len=my_snprintf(buff, sizeof(buff), "%s, Version: %s. "
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
481
#ifdef EMBEDDED_LIBRARY
482
		        "embedded library\n", my_progname, server_version
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
483
#elif __NT__
484 485
			"started with:\nTCP Port: %d, Named Pipe: %s\n",
			my_progname, server_version, mysqld_port, mysqld_unix_port
bk@work.mysql.com's avatar
bk@work.mysql.com committed
486
#else
487 488
			"started with:\nTcp port: %d  Unix socket: %s\n",
			my_progname,server_version,mysqld_port,mysqld_unix_port
bk@work.mysql.com's avatar
bk@work.mysql.com committed
489
#endif
serg@serg.mylan's avatar
serg@serg.mylan committed
490
                       );
491 492
    end=strnmov(buff+len,"Time                 Id Command    Argument\n",
                sizeof(buff)-len);
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
493
    if (my_b_write(&log_file, (byte*) buff,(uint) (end-buff)) ||
494 495
	flush_io_cache(&log_file))
      goto err;
496
    break;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
497
  }
498
  case LOG_NEW:
bk@work.mysql.com's avatar
bk@work.mysql.com committed
499
  {
500
    uint len;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
501 502
    time_t skr=time(NULL);
    struct tm tm_tmp;
503

bk@work.mysql.com's avatar
bk@work.mysql.com committed
504
    localtime_r(&skr,&tm_tmp);
505 506 507 508 509 510 511 512 513 514
    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) ||
515 516
	flush_io_cache(&log_file))
      goto err;
517
    break;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
518
  }
519
  case LOG_BIN:
bk@work.mysql.com's avatar
bk@work.mysql.com committed
520
  {
521 522 523 524
    bool write_file_name_to_index_file=0;

    myf opt= MY_UNPACK_FILENAME;
    if (!index_file_name_arg)
525
    {
526 527 528
      index_file_name_arg= name;	// Use same basename for index file
      opt= MY_UNPACK_FILENAME | MY_REPLACE_EXT;
    }
529

530 531 532 533 534 535 536 537
    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.
      */
538 539
      if (my_b_safe_write(&log_file, (byte*) BINLOG_MAGIC,
			  BIN_LOG_HEADER_SIZE))
540
        goto err;
541 542
      bytes_written+= BIN_LOG_HEADER_SIZE;
      write_file_name_to_index_file= 1;
543
    }
544

545 546 547 548 549 550 551 552 553 554 555 556
    if (!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.
      */
      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 ||
guilhem@mysql.com's avatar
guilhem@mysql.com committed
557
          my_sync(index_file_nr, MYF(MY_WME)) ||
558 559 560 561 562 563 564 565 566 567 568 569
	  init_io_cache(&index_file, index_file_nr,
			IO_SIZE, WRITE_CACHE,
			my_seek(index_file_nr,0L,MY_SEEK_END,MYF(0)),
			0, MYF(MY_WME)))
	goto err;
    }
    else
    {
      safe_mutex_assert_owner(&LOCK_index);
      reinit_io_cache(&index_file, WRITE_CACHE, my_b_filelength(&index_file),
		      0, 0);
    }
570
    if (need_start_event && !no_auto_events)
571
    {
572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595
      /*
        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);
      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
596
        Format_description_log_event::Format_description_log_event(char* buf,).
597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614
        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;
      
      if (description_event_for_queue->write(&log_file))
        goto err;
      bytes_written+= description_event_for_queue->data_written;
615
    }
guilhem@mysql.com's avatar
guilhem@mysql.com committed
616 617
    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
618
      goto err;
619 620 621

    if (write_file_name_to_index_file)
    {
guilhem@mysql.com's avatar
guilhem@mysql.com committed
622 623 624 625
      /*
        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.
      */
626 627 628
      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
629 630
	  flush_io_cache(&index_file) ||
          my_sync(index_file.file, MYF(MY_WME)))
631
	goto err;
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
632
    }
633
    break;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
634
  }
635
  case LOG_CLOSED:				// Impossible
636
  case LOG_TO_BE_OPENED:
637 638 639 640
    DBUG_ASSERT(1);
    break;
  }
  DBUG_RETURN(0);
641 642

err:
643 644 645 646
  sql_print_error("Could not use %s for logging (error %d). \
Turning logging off for the whole duration of the MySQL server process. \
To turn it on again: fix the cause, \
shutdown the MySQL server and restart it.", log_name, errno);
647 648
  if (file >= 0)
    my_close(file,MYF(0));
649 650
  if (index_file_nr >= 0)
    my_close(index_file_nr,MYF(0));
651
  end_io_cache(&log_file);
652 653
  end_io_cache(&index_file);
  safeFree(name);
654
  log_type= LOG_CLOSED;
655
  DBUG_RETURN(1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
656 657
}

658

bk@work.mysql.com's avatar
bk@work.mysql.com committed
659 660 661
int MYSQL_LOG::get_current_log(LOG_INFO* linfo)
{
  pthread_mutex_lock(&LOCK_log);
662 663
  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
664 665 666 667
  pthread_mutex_unlock(&LOCK_log);
  return 0;
}

668

669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710
/*
  Move all data up in a file in an filename index file

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

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

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

  RETURN VALUES
    0	ok
*/

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

  for (;; offset+= bytes_read)
  {
    (void) my_seek(file, offset, MY_SEEK_SET, MYF(0));
    if ((bytes_read= (int) my_read(file, io_buf, sizeof(io_buf), MYF(MY_WME)))
	< 0)
      goto err;
    if (!bytes_read)
      break;					// end of file
    (void) my_seek(file, offset-init_offset, MY_SEEK_SET, MYF(0));
    if (my_write(file, (byte*) io_buf, bytes_read, MYF(MY_WME | MY_NABP)))
      goto err;
  }
  /* The following will either truncate the file or fill the end with \n' */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
711 712
  if (my_chsize(file, offset - init_offset, '\n', MYF(MY_WME)) ||
      my_sync(file, MYF(MY_WME)))
713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732
    goto err;

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

err:
  DBUG_RETURN(1);
}


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

  SYNOPSIS
    find_log_pos()
    linfo		Store here the found log file name and position to
			the NEXT log file name in the index file.
    log_name		Filename to find in the index file.
			Is a null pointer if we want to read the first entry
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
733
    need_lock		Set this to 1 if the parent doesn't already have a
734 735 736
			lock on LOCK_index

  NOTE
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
737 738
    On systems without the truncate function the file will end with one or
    more empty lines.  These will be ignored when reading the file.
739 740 741 742 743 744 745 746 747

  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
748
{
749 750 751 752 753
  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
754

755 756 757 758
  /*
    Mutex needed because we need to make sure the file pointer does not move
    from under our feet
  */
759
  if (need_lock)
760
    pthread_mutex_lock(&LOCK_index);
761 762 763 764 765
  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);

766
  for (;;)
767
  {
768
    uint length;
769 770 771 772
    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
773
    {
774 775 776
      /* 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
777 778
    }

779 780
    // if the log entry matches, null string matching anything
    if (!log_name ||
sasha@mysql.sashanet.com's avatar
sasha@mysql.sashanet.com committed
781
	(log_name_len == length-1 && fname[log_name_len] == '\n' &&
782
	 !memcmp(fname, log_name, log_name_len)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
783
    {
784
      DBUG_PRINT("info",("Found log file entry"));
785
      fname[length-1]=0;			// remove last \n
786 787
      linfo->index_file_start_offset= offset;
      linfo->index_file_offset = my_b_tell(&index_file);
788
      break;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
789
    }
790
  }
791

792
  if (need_lock)
793
    pthread_mutex_unlock(&LOCK_index);
794
  DBUG_RETURN(error);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
795
}
796

797

798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819
/*
  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
*/

820
int MYSQL_LOG::find_next_log(LOG_INFO* linfo, bool need_lock)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
821
{
822
  int error= 0;
823
  uint length;
824 825
  char *fname= linfo->log_file_name;

826 827
  if (need_lock)
    pthread_mutex_lock(&LOCK_index);
828 829 830 831 832 833 834 835
  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)
836
  {
837
    error = !index_file.error ? LOG_INFO_EOF : LOG_INFO_IO;
838 839
    goto err;
  }
840
  fname[length-1]=0;				// kill \n
841
  linfo->index_file_offset = my_b_tell(&index_file);
842

bk@work.mysql.com's avatar
bk@work.mysql.com committed
843
err:
844 845
  if (need_lock)
    pthread_mutex_unlock(&LOCK_index);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
846 847 848
  return error;
}

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
849

850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868
/*
  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)
869 870
{
  LOG_INFO linfo;
871
  bool error=0;
872 873
  const char* save_name;
  enum_log_type save_log_type;
874
  DBUG_ENTER("reset_logs");
875

876 877 878 879
  /*
    We need to get both locks to be sure that no one is trying to
    write to the index log file.
  */
880
  pthread_mutex_lock(&LOCK_log);
881 882 883 884 885 886
  pthread_mutex_lock(&LOCK_index);

  /* Save variables so that we can reopen the log */
  save_name=name;
  name=0;					// Protect against free
  save_log_type=log_type;
887
  close(LOG_CLOSE_TO_BE_OPENED);
888 889 890 891

  /* First delete all old log files */

  if (find_log_pos(&linfo, NullS, 0))
892 893 894 895
  {
    error=1;
    goto err;
  }
896

897
  for (;;)
898 899
  {
    my_delete(linfo.log_file_name, MYF(MY_WME));
900
    if (find_next_log(&linfo, 0))
901 902
      break;
  }
903 904

  /* Start logging with a new file */
905
  close(LOG_CLOSE_INDEX);
906 907
  my_delete(index_file_name, MYF(MY_WME));	// Reset (open will update)
  if (!thd->slave_thread)
908
    need_start_event=1;
909
  open(save_name, save_log_type, 0, index_file_name,
910
       io_cache_type, no_auto_events, max_size, 0);
911
  my_free((gptr) save_name, MYF(0));
912

913
err:  
914
  pthread_mutex_unlock(&LOCK_index);
915
  pthread_mutex_unlock(&LOCK_log);
916
  DBUG_RETURN(error);
917 918
}

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
919

920
/*
921 922 923
  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.
924 925 926

  SYNOPSIS
    purge_first_log()
927 928 929 930 931 932
    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).
    
933 934
  NOTE
    - This is only called from the slave-execute thread when it has read
935 936 937 938 939
      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).
940 941 942

  IMPLEMENTATION
    - Protects index file with LOCK_index
943 944
    - Delete relevant relay log files
    - Copy all file names after these ones to the front of the index file
945
    - If the OS has truncate, truncate the file, else fill it with \n'
946
    - Read the next file name from the index file and store in rli->linfo
947 948

  RETURN VALUES
949 950 951 952
    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
953 954
*/

hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
955
#ifdef HAVE_REPLICATION
956

957
int MYSQL_LOG::purge_first_log(struct st_relay_log_info* rli, bool included) 
958
{
959
  int error;
960 961
  DBUG_ENTER("purge_first_log");

962 963
  DBUG_ASSERT(is_open());
  DBUG_ASSERT(rli->slave_running == 1);
964
  DBUG_ASSERT(!strcmp(rli->linfo.log_file_name,rli->event_relay_log_name));
965

966
  pthread_mutex_lock(&LOCK_index);
967 968 969 970 971 972
  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
973 974

  /*
975 976 977
    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
978
  */
979 980 981 982 983
  pthread_cond_broadcast(&rli->log_space_cond);
  
  /*
    Read the next log file name from the index file and pass it back to
    the caller
984 985
    If included is true, we want the first relay log;
    otherwise we want the one after event_relay_log_name.
986
  */
987 988 989 990
  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)))))
991 992
  {
    char buff[22];
993 994 995 996 997
    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);
998 999
    goto err;
  }
1000

1001
  /*
1002
    Reset rli's coordinates to the current log.
1003
  */
1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017
  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);
1018
    rli->notify_group_relay_log_name_update();
1019
  }
1020 1021 1022

  /* 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
1023

1024
err:
1025
  pthread_mutex_unlock(&LOCK_index);
1026
  DBUG_RETURN(error);
1027 1028
}

1029 1030 1031 1032
/*
  Update log index_file
*/

1033
int MYSQL_LOG::update_log_index(LOG_INFO* log_info, bool need_update_threads)
1034 1035 1036 1037 1038
{
  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
1039 1040
  if (need_update_threads)
    adjust_linfo_offsets(log_info->index_file_start_offset);
1041 1042
  return 0;
}
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
1043

1044 1045 1046 1047 1048
/*
  Remove all logs before the given log from disk and from the index file.

  SYNOPSIS
    purge_logs()
1049 1050 1051 1052 1053 1054 1055
    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
1056 1057 1058 1059 1060 1061 1062 1063 1064 1065

  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
*/

1066 1067 1068 1069 1070
int MYSQL_LOG::purge_logs(const char *to_log, 
                          bool included,
                          bool need_mutex, 
                          bool need_update_threads, 
                          ulonglong *decrease_log_space)
1071 1072
{
  int error;
1073
  bool exit_loop= 0;
1074 1075
  LOG_INFO log_info;
  DBUG_ENTER("purge_logs");
1076
  DBUG_PRINT("info",("to_log= %s",to_log));
1077

1078 1079
  if (need_mutex)
    pthread_mutex_lock(&LOCK_index);
1080
  if ((error=find_log_pos(&log_info, to_log, 0 /*no mutex*/)))
1081 1082
    goto err;

1083
  /*
1084
    File name exists in index file; delete until we find this file
1085 1086 1087
    or a file that is used.
  */
  if ((error=find_log_pos(&log_info, NullS, 0 /*no mutex*/)))
1088
    goto err;
1089 1090
  while ((strcmp(to_log,log_info.log_file_name) || (exit_loop=included)) &&
         !log_in_use(log_info.log_file_name))
1091
  {
1092
    ulong file_size= 0;
1093 1094 1095
    if (decrease_log_space) //stat the file we want to delete
    {
      MY_STAT s;
1096 1097 1098 1099 1100 1101

      /* 
         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.
      */
1102
      if (my_stat(log_info.log_file_name,&s,MYF(0)))
1103
        file_size= s.st_size;
1104 1105 1106 1107 1108 1109 1110
    }
    /*
      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)
1111
      *decrease_log_space-= file_size;
1112
    if (find_next_log(&log_info, 0) || exit_loop)
1113
      break;
1114
  }
1115

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
1116 1117 1118 1119
  /*
    If we get killed -9 here, the sysadmin would have to edit
    the log index file after restart - otherwise, this should be safe
  */
1120
  error= update_log_index(&log_info, need_update_threads);
1121

1122
err:
1123 1124
  if (need_mutex)
    pthread_mutex_unlock(&LOCK_index);
1125
  DBUG_RETURN(error);
1126 1127
}

1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145
/*
  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
*/

1146
int MYSQL_LOG::purge_logs_before_date(time_t purge_time)
1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161
{
  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*/)))
1162
    goto err;
1163 1164 1165 1166 1167

  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 */
1168 1169
    if (!my_stat(log_info.log_file_name, &stat_area, MYF(0)) ||
	stat_area.st_mtime >= purge_time)
1170
      break;
1171
    my_delete(log_info.log_file_name, MYF(0));
1172 1173
    if (find_next_log(&log_info, 0))
      break;
1174
  }
1175

1176 1177 1178 1179
  /*
    If we get killed -9 here, the sysadmin would have to edit
    the log index file after restart - otherwise, this should be safe
  */
1180
  error= update_log_index(&log_info, 1);
1181

1182 1183
err:
  pthread_mutex_unlock(&LOCK_index);
1184
  DBUG_RETURN(error);
1185 1186
}

1187

hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
1188
#endif /* HAVE_REPLICATION */
1189

1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201

/*
  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
1202 1203
void MYSQL_LOG::make_log_name(char* buf, const char* log_ident)
{
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1204 1205 1206 1207 1208
  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
1209 1210
}

1211 1212 1213 1214 1215 1216

/*
  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
1217
{
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1218
  return !strcmp(log_file_name, log_file_name_arg);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1219 1220
}

1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234

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

  SYNOPSIS
    new_file()
    need_lock		Set to 1 (default) if caller has not locked
			LOCK_log and LOCK_index

  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
1235
{
1236 1237 1238
  char new_name[FN_REFLEN], *new_name_ptr, *old_name;
  enum_log_type save_log_type;

1239
  DBUG_ENTER("MYSQL_LOG::new_file");
1240
  if (!is_open())
1241 1242 1243 1244
  {
    DBUG_PRINT("info",("log is closed"));
    DBUG_VOID_RETURN;
  }
1245 1246

  if (need_lock)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1247
  {
1248 1249 1250 1251 1252 1253
    pthread_mutex_lock(&LOCK_log);
    pthread_mutex_lock(&LOCK_index);
  }    
  safe_mutex_assert_owner(&LOCK_log);
  safe_mutex_assert_owner(&LOCK_index);

1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271
  /*
    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);
  }

1272
  /* Reuse old name if not binlog and not update log */
1273
  new_name_ptr= name;
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
1274

1275
  /*
1276 1277 1278
    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.
1279
  */
1280 1281 1282 1283 1284
  if (generate_new_name(new_name, name))
    goto end;
  new_name_ptr=new_name;
  
  if (log_type == LOG_BIN)
1285
  {
1286
    if (!no_auto_events)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1287
    {
1288
      /*
1289 1290
        We log the whole file name for log file as the user may decide
        to change base names at some point.
1291
      */
monty@mysql.com's avatar
monty@mysql.com committed
1292
      THD *thd = current_thd; /* may be 0 if we are reacting to SIGHUP */
1293 1294
      Rotate_log_event r(thd,new_name+dirname_length(new_name));
      r.write(&log_file);
1295
      bytes_written += r.data_written;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1296
    }
1297 1298 1299 1300 1301 1302
    /*
      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.
    */
    signal_update(); 
1303 1304 1305 1306
  }
  old_name=name;
  save_log_type=log_type;
  name=0;				// Don't free name
1307
  close(LOG_CLOSE_TO_BE_OPENED);
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1308

guilhem@mysql.com's avatar
guilhem@mysql.com committed
1309
  /* 
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1310
     Note that at this point, log_type != LOG_CLOSED (important for is_open()).
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1311
  */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1312

1313 1314 1315 1316 1317 1318 1319 1320 1321
  /* 
     new_file() is only used for rotation (in FLUSH LOGS or because size >
     max_binlog_size or max_relay_log_size). 
     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.
  */ 

1322
  open(old_name, save_log_type, new_name_ptr, index_file_name, io_cache_type,
1323
       no_auto_events, max_size, 1);
1324 1325
  my_free(old_name,MYF(0));

1326
end:
1327 1328 1329 1330
  if (need_lock)
  {
    pthread_mutex_unlock(&LOCK_index);
    pthread_mutex_unlock(&LOCK_log);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1331
  }
1332
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1333 1334
}

1335

1336 1337 1338 1339
bool MYSQL_LOG::append(Log_event* ev)
{
  bool error = 0;
  pthread_mutex_lock(&LOCK_log);
1340 1341
  DBUG_ENTER("MYSQL_LOG::append");

1342
  DBUG_ASSERT(log_file.type == SEQ_READ_APPEND);
1343 1344 1345 1346
  /*
    Log_event::write() is smart enough to use my_b_write() or
    my_b_append() depending on the kind of cache we have.
  */
1347 1348 1349 1350 1351
  if (ev->write(&log_file))
  {
    error=1;
    goto err;
  }
1352
  bytes_written+= ev->data_written;
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
1353
  DBUG_PRINT("info",("max_size: %lu",max_size));
1354
  if ((uint) my_b_append_tell(&log_file) > max_size)
1355
  {
1356 1357 1358
    pthread_mutex_lock(&LOCK_index);
    new_file(0);
    pthread_mutex_unlock(&LOCK_index);
1359
  }
1360

1361 1362
err:  
  pthread_mutex_unlock(&LOCK_log);
1363
  signal_update();				// Safe as we don't call close
1364
  DBUG_RETURN(error);
1365 1366
}

1367

1368 1369
bool MYSQL_LOG::appendv(const char* buf, uint len,...)
{
1370
  bool error= 0;
1371
  DBUG_ENTER("MYSQL_LOG::appendv");
1372 1373 1374
  va_list(args);
  va_start(args,len);
  
1375 1376
  DBUG_ASSERT(log_file.type == SEQ_READ_APPEND);
  
1377 1378 1379
  pthread_mutex_lock(&LOCK_log);
  do
  {
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
1380
    if (my_b_append(&log_file,(byte*) buf,len))
1381
    {
1382 1383
      error= 1;
      goto err;
1384
    }
1385
    bytes_written += len;
1386
  } while ((buf=va_arg(args,const char*)) && (len=va_arg(args,uint)));
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
1387
  DBUG_PRINT("info",("max_size: %lu",max_size));
1388
  if ((uint) my_b_append_tell(&log_file) > max_size)
1389
  {
1390 1391 1392
    pthread_mutex_lock(&LOCK_index);
    new_file(0);
    pthread_mutex_unlock(&LOCK_index);
1393
  }
1394 1395 1396

err:
  pthread_mutex_unlock(&LOCK_log);
1397 1398
  if (!error)
    signal_update();
1399
  DBUG_RETURN(error);
1400
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1401

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
1402

1403 1404
/*
  Write to normal (not rotable) log
1405
  This is the format for the 'normal' log.
1406 1407
*/

1408
bool MYSQL_LOG::write(THD *thd,enum enum_server_command command,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1409 1410
		      const char *format,...)
{
1411
  if (is_open() && (what_to_log & (1L << (uint) command)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1412
  {
1413 1414
    uint length;
    int error= 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1415
    VOID(pthread_mutex_lock(&LOCK_log));
1416

nick@mysql.com's avatar
nick@mysql.com committed
1417 1418
    /* 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
1419 1420 1421
    {
      time_t skr;
      ulong id;
1422 1423 1424 1425
      va_list args;
      va_start(args,format);
      char buff[32];

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1426 1427
      if (thd)
      {						// Normal thread
hf@deer.(none)'s avatar
hf@deer.(none) committed
1428 1429 1430 1431 1432
	if ((thd->options & OPTION_LOG_OFF)
#ifndef NO_EMBEDDED_ACCESS_CHECKS
	    && (thd->master_access & SUPER_ACL)
#endif
)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1433 1434
	{
	  VOID(pthread_mutex_unlock(&LOCK_log));
1435
	  return 0;				// No logging
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452
	}
	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;
1453 1454 1455 1456 1457 1458 1459 1460
	/* 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
1461
	if (my_b_write(&log_file, (byte*) buff,16))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1462 1463
	  error=errno;
      }
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
1464
      else if (my_b_write(&log_file, (byte*) "\t\t",2) < 0)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1465
	error=errno;
1466 1467 1468 1469
      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
1470 1471 1472
	error=errno;
      if (format)
      {
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
1473
	if (my_b_write(&log_file, (byte*) " ",1) ||
1474
	    my_b_vprintf(&log_file,format,args) == (uint) -1)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1475 1476
	  error=errno;
      }
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
1477
      if (my_b_write(&log_file, (byte*) "\n",1) ||
1478
	  flush_io_cache(&log_file))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1479 1480 1481 1482 1483 1484
	error=errno;
      if (error && ! write_error)
      {
	write_error=1;
	sql_print_error(ER(ER_ERROR_ON_WRITE),name,error);
      }
1485
      va_end(args);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1486 1487
    }
    VOID(pthread_mutex_unlock(&LOCK_log));
1488
    return error != 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1489
  }
1490
  return 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1491 1492 1493
}


1494 1495
inline bool sync_binlog(IO_CACHE *cache)
{
1496 1497 1498 1499 1500 1501
  if (sync_binlog_period == ++sync_binlog_counter && sync_binlog_period)
  {
    sync_binlog_counter= 0;
    return my_sync(cache->file, MYF(MY_WME));
  }
  return 0;
1502 1503
}

1504

1505 1506 1507 1508
/*
  Write an event to the binary log
*/

1509
bool MYSQL_LOG::write(Log_event* event_info)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1510
{
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
1511
  THD *thd=event_info->thd;
1512
  bool error=1;
1513
  bool should_rotate = 0;
1514
  DBUG_ENTER("MYSQL_LOG::write(event)");
1515
  
1516 1517
  pthread_mutex_lock(&LOCK_log);

guilhem@mysql.com's avatar
guilhem@mysql.com committed
1518 1519 1520 1521 1522
  /* 
     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.
  */
1523
  if (likely(is_open()))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1524
  {
1525 1526
    const char *local_db= event_info->get_db();
    IO_CACHE *file= &log_file;
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
1527
#ifdef HAVE_REPLICATION
1528 1529 1530 1531 1532
    /* 
       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)"
    */
1533
    if ((thd && !(thd->options & OPTION_BIN_LOG)) ||
1534
	(local_db && !db_ok(local_db, binlog_do_db, binlog_ignore_db)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1535
    {
1536
      VOID(pthread_mutex_unlock(&LOCK_log));
1537
      DBUG_PRINT("error",("!db_ok('%s')", local_db));
1538
      DBUG_RETURN(0);
1539
    }
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
1540
#endif /* HAVE_REPLICATION */
1541

1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581
#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)
      - or if the event asks for it (cache_stmt == true).
    */
    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));
          if (!trans_log || open_cached_file(trans_log, mysql_tmpdir, LOG_PREFIX,
                                             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;
          trans_register_ha(thd, TRUE, &binlog_hton);
        }
        else if (!my_b_tell(trans_log))
          trans_register_ha(thd, TRUE, &binlog_hton);
        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()));

1582 1583 1584 1585
    /*
      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
1586 1587 1588 1589 1590 1591

    /*
    1. Write first log events which describe the 'run environment'
    of the SQL command
    */

1592
    if (thd)
1593
    {
1594
      /* NOTE: CHARSET AND TZ REPL WILL BE REWRITTEN SHORTLY */
1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612
      /*
        To make replication of charsets working in 4.1 we are writing values
        of charset related variables before every statement in the binlog,
        if values of those variables differ from global server-wide defaults.
        We are using SET ONE_SHOT command so that the charset vars get reset
        to default after the first non-SET statement.
        In the next 5.0 this won't be needed as we will use the new binlog
        format to store charset info.
      */
      if ((thd->variables.character_set_client->number !=
           global_system_variables.collation_server->number) ||
          (thd->variables.character_set_client->number !=
           thd->variables.collation_connection->number) ||
          (thd->variables.collation_server->number !=
           thd->variables.collation_connection->number))
      {
	char buf[200];
        int written= my_snprintf(buf, sizeof(buf)-1,
bar@mysql.com's avatar
bar@mysql.com committed
1613 1614 1615 1616 1617 1618
                    "SET ONE_SHOT CHARACTER_SET_CLIENT=%u,\
COLLATION_CONNECTION=%u,COLLATION_DATABASE=%u,COLLATION_SERVER=%u",
                             (uint) thd->variables.character_set_client->number,
                             (uint) thd->variables.collation_connection->number,
                             (uint) thd->variables.collation_database->number,
                             (uint) thd->variables.collation_server->number);
1619
	Query_log_event e(thd, buf, written, 0, FALSE);
1620 1621 1622
	if (e.write(file))
	  goto err;
      }
1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633
      /*
        We use the same ONE_SHOT trick for making replication of time zones 
        working in 4.1. Again in 5.0 we have better means for doing this.
      */
      if (thd->time_zone_used &&
          thd->variables.time_zone != global_system_variables.time_zone)
      {
        char buf[MAX_TIME_ZONE_NAME_LENGTH + 26];
        char *buf_end= strxmov(buf, "SET ONE_SHOT TIME_ZONE='", 
                               thd->variables.time_zone->get_name()->ptr(),
                               "'", NullS);
1634
        Query_log_event e(thd, buf, buf_end - buf, 0, FALSE);
1635 1636 1637
        if (e.write(file))
          goto err;
      }
1638

1639 1640 1641
      if (thd->last_insert_id_used)
      {
	Intvar_log_event e(thd,(uchar) LAST_INSERT_ID_EVENT,
1642
			   thd->current_insert_id);
1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657
	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;
      }
1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673
      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;
	}
      }
1674
    }
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
1675

monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
1676
    /* Write the SQL command */
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
1677

monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
1678
    if (event_info->write(file))
1679
      goto err;
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
1680

1681
    if (file == &log_file) // we are writing to the real log (disk)
1682
    {
1683
      if (flush_io_cache(file) || sync_binlog(file))
monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
1684
	goto err;
1685

1686
      /* check automatic rotation; */
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
1687
      DBUG_PRINT("info",("max_size: %lu",max_size));      
1688
      should_rotate= (my_b_tell(file) >= (my_off_t) max_size); 
1689
    }
monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
1690
    error=0;
1691

1692 1693 1694 1695
err:
    if (error)
    {
      if (my_errno == EFBIG)
bell@sanja.is.com.ua's avatar
bell@sanja.is.com.ua committed
1696
	my_message(ER_TRANS_CACHE_FULL, ER(ER_TRANS_CACHE_FULL), MYF(0));
1697 1698 1699 1700 1701
      else
	my_error(ER_ERROR_ON_WRITE, MYF(0), name, errno);
      write_error=1;
    }
    if (file == &log_file)
1702
      signal_update();
1703
    if (should_rotate)
1704
    {
1705
      pthread_mutex_lock(&LOCK_index);
1706 1707 1708
      new_file(0); // inside mutex
      pthread_mutex_unlock(&LOCK_index);
    }
1709
  }
1710 1711

  pthread_mutex_unlock(&LOCK_log);
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
1712

1713
#ifdef HAVE_REPLICATION
1714
  if (should_rotate && expire_logs_days)
1715 1716 1717
  {
    long purge_time= time(0) - expire_logs_days*24*60*60;
    if (purge_time >= 0)
1718
      error= purge_logs_before_date(purge_time);
1719
  }
1720
#endif
1721
  DBUG_RETURN(error);
1722 1723
}

1724

1725 1726 1727 1728 1729 1730 1731 1732 1733
uint MYSQL_LOG::next_file_id()
{
  uint res;
  pthread_mutex_lock(&LOCK_log);
  res = file_id++;
  pthread_mutex_unlock(&LOCK_log);
  return res;
}

1734

1735 1736
/*
  Write a cached log entry to the binary log
1737

guilhem@mysql.com's avatar
guilhem@mysql.com committed
1738 1739
  SYNOPSIS
    write()
1740
    thd
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1741
    cache		The cache to copy to the binlog
1742
    is_commit           If true, will write "COMMIT" in the end, if false will
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1743 1744
                        write "ROLLBACK".

1745 1746
  NOTE
    - We only come here if there is something in the cache.
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
1747
    - The thing in the cache is always a complete transaction
1748 1749
    - 'cache' needs to be reinitialized after this functions returns.

1750 1751 1752 1753
  TODO
      fix it to become atomic - either the complete cache is added to binlog
      or nothing (other storage engines rely on this, doing a ROLLBACK)

1754 1755
  IMPLEMENTATION
    - To support transaction over replication, we wrap the transaction
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1756 1757 1758 1759
      with BEGIN/COMMIT or BEGIN/ROLLBACK in the binary log.
      We want to write a BEGIN/ROLLBACK block when a non-transactional table was
      updated in a transaction which was rolled back. This is to ensure that the
      same updates are run on the slave.
1760 1761
*/

1762
bool MYSQL_LOG::write(THD *thd, IO_CACHE *cache, bool is_commit)
1763 1764
{
  VOID(pthread_mutex_lock(&LOCK_log));
1765
  DBUG_ENTER("MYSQL_LOG::write(cache");
1766
  
1767
  if (likely(is_open()))                       // Should always be true
1768
  {
1769 1770
    uint length;

1771
    /*
1772 1773 1774 1775 1776
      Add the "BEGIN" and "COMMIT" in the binlog around transactions
      which may contain more than 1 SQL statement. If we run with
      AUTOCOMMIT=1, then MySQL immediately writes each SQL statement to
      the binlog when the statement has been completed. No need to add
      "BEGIN" ... "COMMIT" around such statements. Otherwise, MySQL uses
1777 1778 1779
      trans_log (that is thd->ha_data[binlog_hton.slot]) to cache
      the SQL statements until the explicit commit, and at the commit writes
      the contents in trans_log to the binlog.
1780

1781
      We write the "BEGIN" mark first in the buffer (trans_log) where we
1782 1783 1784 1785
      store the SQL statements for a transaction. At the transaction commit
      we will add the "COMMIT mark and write the buffer to the binlog.
    */
    {
1786
      Query_log_event qinfo(thd, "BEGIN", 5, TRUE, FALSE);
1787 1788 1789 1790 1791 1792 1793 1794
      /*
        Imagine this is rollback due to net timeout, after all statements of
        the transaction succeeded. Then we want a zero-error code in BEGIN.
        In other words, if there was a really serious error code it's already
        in the statement's events.
        This is safer than thd->clear_error() against kills at shutdown.
      */
      qinfo.error_code= 0;
1795 1796 1797
      /*
        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
1798 1799 1800
	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.
1801
      */
1802 1803 1804
      if (qinfo.write(&log_file))
	goto err;
    }
1805
    /* Read from the file used to cache the queries .*/
1806
    if (reinit_io_cache(cache, READ_CACHE, 0, 0, 0))
1807
      goto err;
1808 1809
    length=my_b_bytes_in_cache(cache);
    do
1810
    {
1811
      /* Write data to the binary log file */
monty@bitch.mysql.fi's avatar
monty@bitch.mysql.fi committed
1812
      if (my_b_write(&log_file, cache->read_pos, length))
1813
	goto err;
monty@bitch.mysql.fi's avatar
monty@bitch.mysql.fi committed
1814
      cache->read_pos=cache->read_end;		// Mark buffer used up
1815
    } while ((length=my_b_fill(cache)));
1816 1817 1818 1819 1820 1821

    /*
      We write the command "COMMIT" as the last SQL command in the
      binlog segment cached for this transaction
    */

1822
    {
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1823
      Query_log_event qinfo(thd, 
1824 1825
                            is_commit ? "COMMIT" : "ROLLBACK",
                            is_commit ?  6       :  8,
1826
                            TRUE, FALSE);
1827
      qinfo.error_code= 0;
1828 1829
      if (qinfo.write(&log_file) || flush_io_cache(&log_file) ||
          sync_binlog(&log_file))
1830
	goto err;
1831 1832 1833
    }
    if (cache->error)				// Error on read
    {
1834
      sql_print_error(ER(ER_ERROR_ON_READ), cache->file_name, errno);
1835
      write_error=1;				// Don't give more errors
1836
      goto err;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1837
    }
1838
    signal_update();
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
1839
    DBUG_PRINT("info",("max_size: %lu",max_size));
1840
    if (my_b_tell(&log_file) >= (my_off_t) max_size)
1841 1842 1843 1844 1845 1846
    {
      pthread_mutex_lock(&LOCK_index);
      new_file(0); // inside mutex
      pthread_mutex_unlock(&LOCK_index);
    }

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1847
  }
1848
  VOID(pthread_mutex_unlock(&LOCK_log));
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
1849

1850
  DBUG_RETURN(0);
1851 1852

err:
1853 1854 1855 1856 1857
  if (!write_error)
  {
    write_error= 1;
    sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
  }
1858
  VOID(pthread_mutex_unlock(&LOCK_log));
1859
  DBUG_RETURN(1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1860 1861
}

1862

1863
/*
1864
  Write to the slow query log.
1865
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1866

1867
bool MYSQL_LOG::write(THD *thd,const char *query, uint query_length,
1868
		      time_t query_start_arg)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1869
{
1870
  bool error=0;
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1871
  time_t current_time;
1872 1873
  if (!is_open())
    return 0;
monty@mysql.com's avatar
monty@mysql.com committed
1874 1875
  DBUG_ENTER("MYSQL_LOG::write");

guilhem@mysql.com's avatar
guilhem@mysql.com committed
1876
  VOID(pthread_mutex_lock(&LOCK_log));
1877
  if (is_open())
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1878 1879 1880 1881
  {						// Safety agains reopen
    int tmp_errno=0;
    char buff[80],*end;
    end=buff;
1882
    if (!(thd->options & OPTION_UPDATE_LOG))
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1883 1884
    {
      VOID(pthread_mutex_unlock(&LOCK_log));
monty@mysql.com's avatar
monty@mysql.com committed
1885
      DBUG_RETURN(0);
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1886
    }
1887
    if (!(specialflag & SPECIAL_SHORT_LOG_FORMAT) || query_start_arg)
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1888 1889 1890
    {
      current_time=time(NULL);
      if (current_time != last_time)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1891
      {
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906
        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
1907
      }
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1908
      if (my_b_printf(&log_file, "# User@Host: %s[%s] @ %s [%s]\n",
hf@deer.(none)'s avatar
hf@deer.(none) committed
1909 1910
                      thd->priv_user ? thd->priv_user : "",
                      thd->user ? thd->user : "",
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939
                      thd->host ? thd->host : "",
                      thd->ip ? thd->ip : "") == (uint) -1)
        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);
    }
    if (thd->last_insert_id_used)
    {
      end=strmov(end,",last_insert_id=");
      end=longlong10_to_str((longlong) thd->current_insert_id,end,-10);
    }
    // Save value if we do an insert.
    if (thd->insert_id_used)
    {
1940
      if (!(specialflag & SPECIAL_SHORT_LOG_FORMAT))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1941
      {
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1942 1943
        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
1944
      }
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1945 1946 1947 1948
    }
    if (thd->query_start_used)
    {
      if (query_start_arg != thd->query_start())
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1949
      {
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1950 1951 1952
        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
1953
      }
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977
    }
    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
1978
      {
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1979 1980
        write_error=1;
        sql_print_error(ER(ER_ERROR_ON_WRITE),name,error);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1981 1982 1983
      }
    }
  }
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1984
  VOID(pthread_mutex_unlock(&LOCK_log));
monty@mysql.com's avatar
monty@mysql.com committed
1985
  DBUG_RETURN(error);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1986 1987
}

1988

1989 1990 1991 1992 1993 1994
/*
  Wait until we get a signal that the binary log has been updated

  SYNOPSIS
    wait_for_update()
    thd			Thread variable
1995 1996 1997
    master_or_slave     If 0, the caller is the Binlog_dump thread from master;
                        if 1, the caller is the SQL thread from the slave. This
                        influences only thd->proc_info.
1998 1999 2000

  NOTES
    One must have a lock on LOCK_log before calling this function.
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2001 2002
    This lock will be freed before return! That's required by
    THD::enter_cond() (see NOTES in sql_class.h).
2003 2004
*/

monty@mysql.com's avatar
monty@mysql.com committed
2005
void MYSQL_LOG::wait_for_update(THD* thd, bool master_or_slave)
2006
{
monty@mysql.com's avatar
monty@mysql.com committed
2007 2008 2009 2010 2011 2012 2013 2014
  const char *old_msg;
  DBUG_ENTER("wait_for_update");
  old_msg= thd->enter_cond(&update_cond, &LOCK_log,
                           master_or_slave ?
                           "Has read all relay log; waiting for the slave I/O "
                           "thread to update it" : 
                           "Has sent all binlog to slave; waiting for binlog "
                           "to be updated"); 
2015 2016
  pthread_cond_wait(&update_cond, &LOCK_log);
  thd->exit_cond(old_msg);
monty@mysql.com's avatar
monty@mysql.com committed
2017
  DBUG_VOID_RETURN;
2018
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2019

2020

2021 2022 2023 2024 2025
/*
  Close the log file

  SYNOPSIS
    close()
2026 2027 2028 2029 2030
    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
2031 2032 2033 2034

  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
2035 2036
*/

2037
void MYSQL_LOG::close(uint exiting)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2038
{					// One can't set log_type here!
2039 2040
  DBUG_ENTER("MYSQL_LOG::close");
  DBUG_PRINT("enter",("exiting: %d", (int) exiting));
2041
  if (log_type != LOG_CLOSED && log_type != LOG_TO_BE_OPENED)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2042
  {
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
2043
#ifdef HAVE_REPLICATION
2044 2045
    if (log_type == LOG_BIN && !no_auto_events &&
	(exiting & LOG_CLOSE_STOP_EVENT))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2046 2047
    {
      Stop_log_event s;
2048
      s.write(&log_file);
2049
      bytes_written+= s.data_written;
2050
      signal_update();
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2051
    }
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
2052
#endif /* HAVE_REPLICATION */
2053
    end_io_cache(&log_file);
2054 2055 2056 2057 2058 2059 2060 2061 2062 2063

    /* don't pwrite in a file opened with O_APPEND - it doesn't work */
    if (log_file.type == WRITE_CACHE)
    {
      my_off_t offset= BIN_LOG_HEADER_SIZE + FLAGS_OFFSET;
      char flags=LOG_EVENT_BINLOG_CLOSED_F;
      my_pwrite(log_file.file, &flags, 1,
                BIN_LOG_HEADER_SIZE + FLAGS_OFFSET, MYF(0));
    }

2064
    if (my_close(log_file.file,MYF(0)) < 0 && ! write_error)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2065 2066
    {
      write_error=1;
2067
      sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2068 2069
    }
  }
2070 2071 2072 2073 2074 2075

  /*
    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.
  */

2076
  if ((exiting & LOG_CLOSE_INDEX) && my_b_inited(&index_file))
2077
  {
2078 2079
    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
2080
    {
2081 2082
      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
2083
    }
2084
  }
2085
  log_type= (exiting & LOG_CLOSE_TO_BE_OPENED) ? LOG_TO_BE_OPENED : LOG_CLOSED;
2086
  safeFree(name);
2087
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2088 2089 2090
}


2091 2092 2093 2094 2095 2096 2097 2098 2099
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
2100 2101
  DBUG_ENTER("MYSQL_LOG::set_max_size");
  pthread_mutex_lock(&LOCK_log);
2102 2103
  if (is_open())
    max_size= max_size_arg;
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2104 2105
  pthread_mutex_unlock(&LOCK_log);
  DBUG_VOID_RETURN;
2106
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2107

2108

2109 2110
/*
  Check if a string is a valid number
2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124

  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
2125
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137

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++;
2138 2139
  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
2140 2141 2142 2143 2144 2145 2146
  {
    flag=1;
    str++;
  }
  if (*str == '.')
  {
    for (str++ ;
2147
	 my_isdigit(files_charset_info,*str) ||
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158
	   (allow_wildcards && (*str == wild_many || *str == wild_one)) ;
	 str++, flag=1) ;
  }
  if (*str != 0 || flag == 0)
    DBUG_RETURN(0);
  if (res)
    *res=atol(start);
  DBUG_RETURN(1);			/* Number ok */
} /* test_if_number */


serg@serg.mylan's avatar
serg@serg.mylan committed
2159
void print_buffer_to_file(enum loglevel level, const char *buffer)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2160 2161 2162 2163
{
  time_t skr;
  struct tm tm_tmp;
  struct tm *start;
monty@mysql.com's avatar
monty@mysql.com committed
2164 2165
  DBUG_ENTER("print_buffer_to_file");
  DBUG_PRINT("enter",("buffer: %s", buffer));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2166 2167

  VOID(pthread_mutex_lock(&LOCK_error_log));
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2168

bk@work.mysql.com's avatar
bk@work.mysql.com committed
2169
  skr=time(NULL);
2170
  localtime_r(&skr, &tm_tmp);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2171
  start=&tm_tmp;
2172
  fprintf(stderr, "%02d%02d%02d %2d:%02d:%02d [%s] %s\n",
2173 2174 2175 2176 2177 2178
          start->tm_year % 100,
          start->tm_mon+1,
          start->tm_mday,
          start->tm_hour,
          start->tm_min,
          start->tm_sec,
monty@mysql.com's avatar
monty@mysql.com committed
2179
          (level == ERROR_LEVEL ? "ERROR" : level == WARNING_LEVEL ?
2180
           "Warning" : "Note"),
monty@mysql.com's avatar
monty@mysql.com committed
2181
          buffer);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2182

bk@work.mysql.com's avatar
bk@work.mysql.com committed
2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197
  fflush(stderr);

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


void sql_perror(const char *message)
{
#ifdef HAVE_STRERROR
  sql_print_error("%s: %s",message, strerror(errno));
#else
  perror(message);
#endif
}
2198

monty@mysql.com's avatar
monty@mysql.com committed
2199

2200 2201 2202
bool flush_error_log()
{
  bool result=0;
2203
  if (opt_error_log)
2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244
  {
    char err_renamed[FN_REFLEN], *end;
    end= strmake(err_renamed,log_error_file,FN_REFLEN-4);
    strmov(end, "-old");
#ifdef __WIN__
    char err_temp[FN_REFLEN+4];
    /*
     On Windows is necessary a temporary file for to rename
     the current error file.
    */
    strmov(strmov(err_temp, err_renamed),"-tmp");
    (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)
             my_fwrite(stderr, (byte*) buf, (uint) strlen(buf),MYF(0));
        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
  }
   return result;
}
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
2245

monty@mysql.com's avatar
monty@mysql.com committed
2246 2247 2248 2249 2250 2251 2252 2253
void MYSQL_LOG::signal_update()
{
  DBUG_ENTER("MYSQL_LOG::signal_update");
  pthread_cond_broadcast(&update_cond);
  DBUG_VOID_RETURN;
}


2254
#ifdef __NT__
monty@mysql.com's avatar
monty@mysql.com committed
2255 2256
void print_buffer_to_nt_eventlog(enum loglevel level, char *buff,
                                 uint length, int buffLen)
2257 2258 2259 2260
{
  HANDLE event;
  char   *buffptr;
  LPCSTR *buffmsgptr;
serg@serg.mylan's avatar
serg@serg.mylan committed
2261
  DBUG_ENTER("print_buffer_to_nt_eventlog");
2262

monty@mysql.com's avatar
monty@mysql.com committed
2263 2264
  buffptr= buff;
  if (length > (uint)(buffLen-4))
2265
  {
monty@mysql.com's avatar
monty@mysql.com committed
2266
    char *newBuff= new char[length + 4];
serg@serg.mylan's avatar
serg@serg.mylan committed
2267
    strcpy(newBuff, buff);
monty@mysql.com's avatar
monty@mysql.com committed
2268
    buffptr= newBuff;
2269
  }
monty@mysql.com's avatar
monty@mysql.com committed
2270 2271
  strmov(buffptr+length, "\r\n\r\n");
  buffmsgptr= (LPCSTR*) &buffptr;               // Keep windows happy
2272

monty@mysql.com's avatar
monty@mysql.com committed
2273 2274
  setup_windows_event_source();
  if ((event= RegisterEventSource(NULL,"MySQL")))
2275
  {
monty@mysql.com's avatar
monty@mysql.com committed
2276
    switch (level) {
2277
      case ERROR_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
2278 2279
        ReportEvent(event, EVENTLOG_ERROR_TYPE, 0, MSG_DEFAULT, NULL, 1, 0,
                    buffmsgptr, NULL);
2280 2281
        break;
      case WARNING_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
2282 2283
        ReportEvent(event, EVENTLOG_WARNING_TYPE, 0, MSG_DEFAULT, NULL, 1, 0,
                    buffmsgptr, NULL);
2284 2285
        break;
      case INFORMATION_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
2286 2287
        ReportEvent(event, EVENTLOG_INFORMATION_TYPE, 0, MSG_DEFAULT, NULL, 1,
                    0, buffmsgptr, NULL);
2288 2289 2290 2291 2292
        break;
    }
    DeregisterEventSource(event);
  }

monty@mysql.com's avatar
monty@mysql.com committed
2293
  /* if we created a string buffer, then delete it */
serg@serg.mylan's avatar
serg@serg.mylan committed
2294
  if (buffptr != buff)
2295 2296 2297 2298
    delete[] buffptr;

  DBUG_VOID_RETURN;
}
monty@mysql.com's avatar
monty@mysql.com committed
2299 2300
#endif /* __NT__ */

2301 2302

/*
monty@mysql.com's avatar
monty@mysql.com committed
2303 2304
  Prints a printf style message to the error log and, under NT, to the
  Windows event log.
2305 2306 2307

  SYNOPSIS
    vprint_msg_to_log()
monty@mysql.com's avatar
monty@mysql.com committed
2308 2309 2310
    event_type             Type of event to write (Error, Warning, or Info)
    format                 Printf style format of message
    args                   va_list list of arguments for the message    
2311 2312 2313 2314

  NOTE

  IMPLEMENTATION
monty@mysql.com's avatar
monty@mysql.com committed
2315 2316
    This function prints the message into a buffer and then sends that buffer
    to other functions to write that message to other logging sources.
2317 2318 2319

  RETURN VALUES
    void
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2320
*/
monty@mysql.com's avatar
monty@mysql.com committed
2321

2322
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
2323 2324
{
  char   buff[1024];
monty@mysql.com's avatar
monty@mysql.com committed
2325
  uint length;
2326
  DBUG_ENTER("vprint_msg_to_log");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2327

monty@mysql.com's avatar
monty@mysql.com committed
2328
  length= my_vsnprintf(buff, sizeof(buff)-5, format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
2329
  print_buffer_to_file(level, buff);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2330 2331

#ifdef __NT__
monty@mysql.com's avatar
monty@mysql.com committed
2332
  print_buffer_to_nt_eventlog(level, buff, length, sizeof(buff));
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2333
#endif
2334

rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2335 2336 2337
  DBUG_VOID_RETURN;
}

2338

monty@mysql.com's avatar
monty@mysql.com committed
2339
void sql_print_error(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2340 2341
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
2342
  DBUG_ENTER("sql_print_error");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2343

serg@serg.mylan's avatar
serg@serg.mylan committed
2344 2345 2346
  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
2347 2348 2349 2350 2351

  DBUG_VOID_RETURN;
}


monty@mysql.com's avatar
monty@mysql.com committed
2352
void sql_print_warning(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2353 2354
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
2355
  DBUG_ENTER("sql_print_warning");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2356

serg@serg.mylan's avatar
serg@serg.mylan committed
2357 2358 2359
  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
2360 2361 2362 2363 2364

  DBUG_VOID_RETURN;
}


monty@mysql.com's avatar
monty@mysql.com committed
2365
void sql_print_information(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2366 2367
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
2368
  DBUG_ENTER("sql_print_information");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2369

serg@serg.mylan's avatar
serg@serg.mylan committed
2370 2371 2372
  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
2373

guilhem@mysql.com's avatar
guilhem@mysql.com committed
2374 2375
  DBUG_VOID_RETURN;
}
2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 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 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942

/********* 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
  states - active, in_sync, pool.
  there could be only one page in active or in_sync states,
  but many in pool - pool is fifo queue.
  usual lifecycle of a page is pool->active->in_sync->pool
  "active" page - is a page where new xid's are logged.
  the page stays active as long as in_sync slot is taken.
  "in_sync" page is being synced to disk. no new xid can be added to it.
  when the sync is done the page is moved to a pool and an active page
  becomes "in_sync".

  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 in_sync 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.
  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)

  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
  dirty - we don't sync removals to disk.
*/
#define TC_LOG_HEADER_SIZE (sizeof(tc_log_magic)+1)

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

uint opt_tc_log_size=TC_LOG_MIN_SIZE;
uint tc_log_max_pages_used=0, tc_log_page_size=0,
     tc_log_page_waits=0, tc_log_cur_pages_used=0;

TC_LOG *tc_log;
TC_LOG_MMAP tc_log_mmap;
TC_LOG_DUMMY tc_log_dummy;

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

  DBUG_ASSERT(total_ha_2pc);
  DBUG_ASSERT(opt_name && opt_name[0]);

#ifdef HAVE_GETPAGESIZE
  tc_log_page_size=getpagesize();
  DBUG_ASSERT(TC_LOG_PAGE_SIZE % tc_log_page_size == 0);
#else
  tc_log_page_size=TC_LOG_PAGE_SIZE;
#endif

  fn_format(logname,opt_name,mysql_data_home,"",MY_UNPACK_FILENAME);
  fd=my_open(logname, O_RDWR, MYF(0));  // TODO use O_CREAT and check len==0 ?
  if (fd == -1)
  {
    if (using_heuristic_recover())
      return 1;
    fd=my_open(logname, O_RDWR|O_CREAT, MYF(MY_WME));
    if (fd == -1)
      goto err;
    inited=1;
    file_length=opt_tc_log_size;
    if (my_chsize(fd, file_length, 0, MYF(MY_WME)))
      goto err;
  }
  else
  {
    inited=1;
    crashed=TRUE;
    if (tc_heuristic_recover)
    {
      sql_print_error("Cannot perform automatic crash recovery when "
                      "--tc-heuristic-recover is used");
      goto err;
    }
    file_length = my_seek(fd, 0L, MY_SEEK_END, MYF(MY_WME+MY_FAE));
    if (file_length == MY_FILEPOS_ERROR || file_length % tc_log_page_size)
      goto err;
  }

  data=(uchar *)my_mmap(0, file_length, PROT_READ|PROT_WRITE,
                        MAP_NOSYNC|MAP_SHARED, fd, 0);
  if (data == MAP_FAILED)
  {
    my_errno=errno;
    goto err;
  }
  inited=2;

  npages=file_length/tc_log_page_size;
  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].start=(my_xid *)(data+TC_LOG_HEADER_SIZE);
  pages[0].size=pages[0].free=
                (tc_log_page_size-TC_LOG_HEADER_SIZE)/sizeof(my_xid);
  pages[npages-1].next=0;
  inited=4;

  if (crashed)
  {
    sql_print_error("Recovering after a crash");
    if (recover())
      goto err;
  }
  memcpy(data, tc_log_magic, sizeof(tc_log_magic));
  data[sizeof(tc_log_magic)]= total_ha_2pc;
  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;

  in_sync= 0;
  active=pages;
  pool=pages+1;
  pool_last=pages+npages-1;

  return 0;

err:
  close();
  return 1;
}

/*
  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;

  do
  {
    best_p= p= &pool;
    if ((*p)->waiters == 0)
      break;

    best_free=0;
    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;
  if (active->free == active->size)
  {
    tc_log_cur_pages_used++;
    set_if_bigger(tc_log_max_pages_used, tc_log_cur_pages_used);
  }

  if ((*best_p)->next)
    *best_p=(*best_p)->next;
  else
    pool_last=*best_p;
}

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);
  return 1; // returns always 1
}

/*
  all access to active page is serialized but it's not a problem, as we're
  assuming that fsync() will be a bottleneck anyway.
  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 fsync()
*/

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

  pthread_mutex_lock(&LOCK_active);

  while (unlikely(active && active->free == 0))
    pthread_cond_wait(&COND_active, &LOCK_active);

  if (active == 0)
  {
    lock_queue(&LOCK_pool);
    get_active_from_pool();
    unlock_queue(&LOCK_pool);
  }

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

  while (*p->ptr)
  {
    p->ptr++;
    DBUG_ASSERT(p->ptr < p->end);               // because p->free > 0
  }

  cookie= (ulong)((uchar *)p->ptr - data);      // Can never be zero
  *p->ptr++= xid;
  p->free--;
  p->state= DIRTY;

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

  if (in_sync)
  {                                          // somebody's syncing. let's wait
    pthread_mutex_unlock(&LOCK_active);
    p->waiters++;
    do
    {
      pthread_cond_wait(&p->cond, &LOCK_sync);
    } while (p->state == DIRTY && in_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
  in_sync=p;                                 // place is vacant - take it
  DBUG_ASSERT(active == p);
  active=0;
  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;

  DBUG_ASSERT(in_sync != active);

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

  pthread_mutex_lock(&LOCK_pool);
  pool_last->next=in_sync;
  pool_last=in_sync;
  in_sync->next=0;
  in_sync->state= err ? ERROR : POOL;
  pthread_cond_broadcast(&in_sync->cond);    // signal "sync done"
  pthread_cond_signal(&COND_pool);           // in case somebody's waiting
  pthread_mutex_unlock(&LOCK_pool);

  pthread_mutex_lock(&LOCK_sync);
  in_sync=0;
  pthread_cond_signal(&active->cond);        // wake up a new syncer
  pthread_mutex_unlock(&LOCK_sync);
  return err;
}

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);
  if (p->free == p->size)
    statistic_decrement(tc_log_cur_pages_used, &LOCK_status);
  if (p->waiters == 0)                 // the page is in pool and ready to rock
    pthread_cond_signal(&COND_pool);   // ping ... in case somebody's waiting
  pthread_mutex_unlock(&p->lock);
}

void TC_LOG_MMAP::close()
{
  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:
    for (uint i=0; i < npages; i++)
    {
      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:
    my_munmap(data, file_length);
  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)
  {
    sql_print_error("For recovery to work all storage engines and binary log "
                    "must have exactly the same settings as before the crash!");
    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);
  bzero(data, file_length);
  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;
}

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 ******/

/*
  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!
*/

int MYSQL_LOG::open(const char *opt_name)
{
  LOG_INFO log_info, new_log_info;

  DBUG_ASSERT(total_ha_2pc > 1);

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

  if (using_heuristic_recover())
    return 1;

  /*
    read index file to get a last but one binlog filename
    note - there's no need to lock any mutex, mysqld is only starting
    up, no other threads are running yet.
    still, there's safe_mutex_assert_owner() in binlog code, so
    let's keep it happy.
  */

  if (find_log_pos(&new_log_info, NullS, 1))
    goto err;                           // er ? where's the current entry ?

  if (strcmp(log_file_name, new_log_info.log_file_name))
  {
    const char *errmsg;
    char        last_event_type=UNKNOWN_EVENT;
    IO_CACHE    log;
    File        file;
    int         error;
    Log_event  *ev;
    Format_description_log_event fdle(4);

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

    do
    {
      log_info.index_file_offset=new_log_info.index_file_offset;
      log_info.index_file_start_offset=new_log_info.index_file_offset;
      strcpy(log_info.log_file_name, new_log_info.log_file_name);
      if (find_next_log(&new_log_info, 1))
        goto err;                       // er ? where's the current entry ?
    } while (strcmp(log_file_name, new_log_info.log_file_name));

    if ((file= open_binlog(&log, log_info.log_file_name, &errmsg)) < 0)
    {
      sql_print_error("%s", errmsg);
      goto err;
    }

    if (((ev= Log_event::read_log_event(&log, 0, &fdle))) &&
         (ev->get_type_code() == FORMAT_DESCRIPTION_EVENT) &&
         (ev->flags & LOG_EVENT_BINLOG_CLOSED_F))
      error=0;
    else
      error= recover(&log, (Format_description_log_event *)ev);

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

    if (error)
      goto err;
  }

  return 0;

err:
  return 1;
}

void MYSQL_LOG::close()
{
  DBUG_ASSERT(prepared_xids==0);
  pthread_mutex_destroy(&LOCK_prep_xids);
  pthread_cond_destroy (&COND_prep_xids);
}

/* TODO group commit */
int MYSQL_LOG::log(THD *thd, my_xid xid)
{
  Xid_log_event xle(thd, xid);
  if (xle.write((IO_CACHE*)thd->ha_data[binlog_hton.slot]))
    return 0;
  thread_safe_increment(prepared_xids, &LOCK_prep_xids);
  return !binlog_commit(thd,1);                 // invert return value
}

void MYSQL_LOG::unlog(ulong cookie, my_xid xid)
{
  if (thread_safe_dec_and_test(prepared_xids, &LOCK_prep_xids))
    pthread_cond_signal(&COND_prep_xids);
}

int MYSQL_LOG::recover(IO_CACHE *log, Format_description_log_event *fdle)
{
  Log_event  *ev;
  HASH xids;
  MEM_ROOT mem_root;

  if (! fdle->is_valid() ||
      hash_init(&xids, &my_charset_bin, tc_log_page_size/3, 0,
            sizeof(my_xid), 0, 0, MYF(0)))
    goto err1;

  init_alloc_root(&mem_root, tc_log_page_size, tc_log_page_size);

  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;
      byte *x=memdup_root(&mem_root, (char *)& xev->xid, sizeof(xev->xid));
      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;
}