log.cc 84.3 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
#include "mysql_priv.h"
26
#include "sql_repl.h"
bk@work.mysql.com's avatar
bk@work.mysql.com committed
27 28 29 30 31

#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
32 33 34 35
#ifdef __NT__
#include "message.h"
#endif

36
MYSQL_LOG mysql_log, mysql_slow_log, mysql_bin_log;
37
ulong sync_binlog_counter= 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
38 39 40

static bool test_if_number(const char *str,
			   long *res, bool allow_wildcards);
41 42 43 44 45 46 47 48
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 = {
serg@serg.mylan's avatar
serg@serg.mylan committed
49
  "binlog",
50 51 52 53 54 55 56 57 58 59 60
  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 */
61 62 63 64 65
  NULL,                         /* rollback_by_xid */
  NULL,                         /* create_cursor_read_view */
  NULL,                         /* set_cursor_read_view */
  NULL,    			/* close_cursor_read_view */
  HTON_NO_FLAGS
66 67 68 69 70 71 72 73 74 75
};

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

handlerton *binlog_init()
{
serg@serg.mylan's avatar
serg@serg.mylan committed
76
  return &binlog_hton;
77 78 79 80 81
}

static int binlog_close_connection(THD *thd)
{
  IO_CACHE *trans_log= (IO_CACHE*)thd->ha_data[binlog_hton.slot];
82
  DBUG_ASSERT(mysql_bin_log.is_open() && !my_b_tell(trans_log));
83 84 85 86 87
  close_cached_file(trans_log);
  my_free((gptr)trans_log, MYF(0));
  return 0;
}

88
static int binlog_end_trans(THD *thd, IO_CACHE *trans_log, Log_event *end_ev)
89
{
90 91 92 93 94
  int error=0;
  DBUG_ENTER("binlog_end_trans");
  if (end_ev)
    error= mysql_bin_log.write(thd, trans_log, end_ev);

95 96 97 98 99 100 101 102
  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;
103
  DBUG_RETURN(error);
104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121
}

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)
{
  IO_CACHE *trans_log= (IO_CACHE*)thd->ha_data[binlog_hton.slot];
  DBUG_ENTER("binlog_commit");
  DBUG_ASSERT(mysql_bin_log.is_open() &&
122
     (all || !(thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN))));
123 124 125 126 127 128

  if (!my_b_tell(trans_log))
  {
    // we're here because trans_log was flushed in MYSQL_LOG::log()
    DBUG_RETURN(0);
  }
129
  Query_log_event qev(thd, "COMMIT", 6, TRUE, FALSE);
130
  DBUG_RETURN(binlog_end_trans(thd, trans_log, &qev));
131 132 133 134 135 136 137 138
}

static int binlog_rollback(THD *thd, bool all)
{
  int error=0;
  IO_CACHE *trans_log= (IO_CACHE*)thd->ha_data[binlog_hton.slot];
  DBUG_ENTER("binlog_rollback");
  /*
serg@serg.mylan's avatar
serg@serg.mylan committed
139 140
    First assert is guaranteed - see trans_register_ha() call below.
    The second must be true. If it is not, we're registering
141 142
    unnecessary, doing extra work. The cause should be found and eliminated
  */
serg@serg.mylan's avatar
serg@serg.mylan committed
143 144
  DBUG_ASSERT(all || !(thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN)));
  DBUG_ASSERT(mysql_bin_log.is_open() && my_b_tell(trans_log));
145
  /*
146 147 148 149
    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...)
150 151
  */
  if (unlikely(thd->options & OPTION_STATUS_NO_TRANS_UPDATE))
152 153
  {
    Query_log_event qev(thd, "ROLLBACK", 8, TRUE, FALSE);
154
    error= binlog_end_trans(thd, trans_log, &qev);
155
  }
156 157
  else
    error= binlog_end_trans(thd, trans_log, 0);
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
  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");

239 240
  if ((file = my_open(log_file_name, O_RDONLY | O_BINARY | O_SHARE, 
                      MYF(MY_WME))) < 0)
241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266
  {
    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
267

rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
268 269
#ifdef __NT__
static int eventSource = 0;
serg@serg.mylan's avatar
serg@serg.mylan committed
270

271
void setup_windows_event_source()
monty@mysql.com's avatar
monty@mysql.com committed
272
{
273
  HKEY    hRegKey= NULL;
monty@mysql.com's avatar
monty@mysql.com committed
274 275 276
  DWORD   dwError= 0;
  TCHAR   szPath[MAX_PATH];
  DWORD dwTypes;
277

monty@mysql.com's avatar
monty@mysql.com committed
278 279 280
  if (eventSource)               // Ensure that we are only called once
    return;
  eventSource= 1;
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
281

monty@mysql.com's avatar
monty@mysql.com committed
282
  // Create the event source registry key
283
  dwError= RegCreateKey(HKEY_LOCAL_MACHINE,
monty@mysql.com's avatar
monty@mysql.com committed
284 285
                          "SYSTEM\\CurrentControlSet\\Services\\EventLog\\Application\\MySQL", 
                          &hRegKey);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
286

monty@mysql.com's avatar
monty@mysql.com committed
287 288
  /* Name of the PE module that contains the message resource */
  GetModuleFileName(NULL, szPath, MAX_PATH);
serg@serg.mylan's avatar
serg@serg.mylan committed
289

rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
290
  /* Register EventMessageFile */
291
  dwError = RegSetValueEx(hRegKey, "EventMessageFile", 0, REG_EXPAND_SZ,
monty@mysql.com's avatar
monty@mysql.com committed
292
                          (PBYTE) szPath, strlen(szPath)+1);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
293

monty@mysql.com's avatar
monty@mysql.com committed
294 295 296 297 298
  /* 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
299

monty@mysql.com's avatar
monty@mysql.com committed
300
  RegCloseKey(hRegKey);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
301 302
}

monty@mysql.com's avatar
monty@mysql.com committed
303
#endif /* __NT__ */
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
304 305


bk@work.mysql.com's avatar
bk@work.mysql.com committed
306 307 308 309 310 311 312 313
/****************************************************************************
** 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)
{
314 315 316 317
  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
318
  reg1 struct fileinfo *file_info;
319 320
  ulong                 max_found=0;

bk@work.mysql.com's avatar
bk@work.mysql.com committed
321 322
  DBUG_ENTER("find_uniq_filename");

323 324 325 326
  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
327
  *end='.';
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
328
  length= (uint) (end-start+1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346

  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++='.';
347
  sprintf(end,"%06ld",max_found+1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
348 349 350
  DBUG_RETURN(0);
}

351

352
MYSQL_LOG::MYSQL_LOG()
353
  :bytes_written(0), last_time(0), query_start(0), name(0),
354
   file_id(1), open_count(1), log_type(LOG_CLOSED), write_error(0), inited(0),
355
   need_start_event(1), prepared_xids(0), description_event_for_exec(0),
356
   description_event_for_queue(0)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
357 358
{
  /*
guilhem@mysql.com's avatar
guilhem@mysql.com committed
359 360
    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
361
    called only in main(). Doing initialization here would make it happen
362
    before main().
bk@work.mysql.com's avatar
bk@work.mysql.com committed
363
  */
364 365
  index_file_name[0] = 0;
  bzero((char*) &log_file,sizeof(log_file));
366
  bzero((char*) &index_file, sizeof(index_file));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
367 368
}

369
/* this is called only once */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
370

monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
371
void MYSQL_LOG::cleanup()
bk@work.mysql.com's avatar
bk@work.mysql.com committed
372
{
373
  DBUG_ENTER("cleanup");
bk@work.mysql.com's avatar
bk@work.mysql.com committed
374
  if (inited)
375
  {
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
376
    inited= 0;
377
    close(LOG_CLOSE_INDEX|LOG_CLOSE_STOP_EVENT);
378 379
    delete description_event_for_queue;
    delete description_event_for_exec;
380 381
    (void) pthread_mutex_destroy(&LOCK_log);
    (void) pthread_mutex_destroy(&LOCK_index);
382
    (void) pthread_cond_destroy(&update_cond);
383
  }
384
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
385 386
}

387

bk@work.mysql.com's avatar
bk@work.mysql.com committed
388
int MYSQL_LOG::generate_new_name(char *new_name, const char *log_name)
389
{
390 391
  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
392 393 394 395 396 397 398 399 400 401 402 403 404
  {
    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;
}

405

406
void MYSQL_LOG::init(enum_log_type log_type_arg,
407
		     enum cache_type io_cache_type_arg,
408 409
		     bool no_auto_events_arg,
                     ulong max_size_arg)
410
{
411
  DBUG_ENTER("MYSQL_LOG::init");
412
  log_type = log_type_arg;
413
  io_cache_type = io_cache_type_arg;
414
  no_auto_events = no_auto_events_arg;
415
  max_size=max_size_arg;
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
416
  DBUG_PRINT("info",("log_type: %d max_size: %lu", log_type, max_size));
417
  DBUG_VOID_RETURN;
418 419
}

420

guilhem@mysql.com's avatar
guilhem@mysql.com committed
421 422
void MYSQL_LOG::init_pthread_objects()
{
423 424
  DBUG_ASSERT(inited == 0);
  inited= 1;
guilhem@mysql.com's avatar
guilhem@mysql.com committed
425 426 427
  (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);
428 429
}

430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480
const char *MYSQL_LOG::generate_name(const char *log_name,
                                     const char *suffix,
                                     bool strip_ext, char *buff)
{
  if (!log_name || !log_name[0])
  {
    /*
      TODO: The following should be using fn_format();  We just need to
      first change fn_format() to cut the file name if it's too long.
    */
    strmake(buff,glob_hostname,FN_REFLEN-5);
    strmov(fn_ext(buff),suffix);
    return (const char *)buff;
  }
  // get rid of extension if the log is binary to avoid problems
  if (strip_ext)
  {
    char *p = fn_ext(log_name);
    uint length=(uint) (p-log_name);
    strmake(buff,log_name,min(length,FN_REFLEN));
    return (const char*)buff;
  }
  return log_name;
}

bool MYSQL_LOG::open_index_file(const char *index_file_name_arg,
                                const char *log_name)
{
  File index_file_nr= -1;
  DBUG_ASSERT(!my_b_inited(&index_file));

  /*
    First open of this class instance
    Create an index file that will hold all file names uses for logging.
    Add new entries to the end of it.
  */
  myf opt= MY_UNPACK_FILENAME;
  if (!index_file_name_arg)
  {
    index_file_name_arg= log_name;    // Use same basename for index file
    opt= MY_UNPACK_FILENAME | MY_REPLACE_EXT;
  }
  fn_format(index_file_name, index_file_name_arg, mysql_data_home,
            ".index", opt);
  if ((index_file_nr= my_open(index_file_name,
                              O_RDWR | O_CREAT | O_BINARY ,
                              MYF(MY_WME))) < 0 ||
       my_sync(index_file_nr, MYF(MY_WME)) ||
       init_io_cache(&index_file, index_file_nr,
                     IO_SIZE, WRITE_CACHE,
                     my_seek(index_file_nr,0L,MY_SEEK_END,MYF(0)),
serg@serg.mylan's avatar
serg@serg.mylan committed
481
			0, MYF(MY_WME | MY_WAIT_IF_FULL)))
482 483 484 485 486 487 488 489
  {
    if (index_file_nr >= 0)
      my_close(index_file_nr,MYF(0));
    return TRUE;
  }
  return FALSE;
}

bk@work.mysql.com's avatar
bk@work.mysql.com committed
490

491 492 493 494 495 496 497 498 499 500 501 502 503 504
/*
  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
*/

505 506 507 508 509
bool MYSQL_LOG::open(const char *log_name,
                     enum_log_type log_type_arg,
                     const char *new_name,
                     enum cache_type io_cache_type_arg,
                     bool no_auto_events_arg,
510 511
                     ulong max_size_arg,
                     bool null_created_arg)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
512
{
513 514
  char buff[FN_REFLEN];
  File file= -1;
515
  int open_flags = O_CREAT | O_BINARY;
516
  DBUG_ENTER("MYSQL_LOG::open");
517
  DBUG_PRINT("enter",("log_type: %d",(int) log_type_arg));
518 519 520

  last_time=query_start=0;
  write_error=0;
521

522
  init(log_type_arg,io_cache_type_arg,no_auto_events_arg,max_size_arg);
523

524
  if (!(name=my_strdup(log_name,MYF(MY_WME))))
525 526
  {
    name= (char *)log_name; // for the error message
527
    goto err;
528
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
529 530 531
  if (new_name)
    strmov(log_file_name,new_name);
  else if (generate_new_name(log_file_name, name))
532
    goto err;
533

534
  if (io_cache_type == SEQ_READ_APPEND)
535
    open_flags |= O_RDWR | O_APPEND;
536
  else
537
    open_flags |= O_WRONLY | (log_type == LOG_BIN ? 0 : O_APPEND);
538

bk@work.mysql.com's avatar
bk@work.mysql.com committed
539
  db[0]=0;
540
  open_count++;
541
  if ((file=my_open(log_file_name,open_flags,
542
		    MYF(MY_WME | ME_WAITTANG))) < 0 ||
543
      init_io_cache(&log_file, file, IO_SIZE, io_cache_type,
544 545 546
		    my_tell(file,MYF(MY_WME)), 0, 
                    MYF(MY_WME | MY_NABP |
                        ((log_type == LOG_BIN) ? MY_WAIT_IF_FULL : 0))))
547
    goto err;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
548

549 550
  switch (log_type) {
  case LOG_NORMAL:
bk@work.mysql.com's avatar
bk@work.mysql.com committed
551
  {
552
    char *end;
553
    int len=my_snprintf(buff, sizeof(buff), "%s, Version: %s. "
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
554
#ifdef EMBEDDED_LIBRARY
555
		        "embedded library\n", my_progname, server_version
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
556
#elif __NT__
557 558
			"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
559
#else
560 561
			"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
562
#endif
serg@serg.mylan's avatar
serg@serg.mylan committed
563
                       );
564 565
    end=strnmov(buff+len,"Time                 Id Command    Argument\n",
                sizeof(buff)-len);
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
566
    if (my_b_write(&log_file, (byte*) buff,(uint) (end-buff)) ||
567 568
	flush_io_cache(&log_file))
      goto err;
569
    break;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
570
  }
571
  case LOG_NEW:
bk@work.mysql.com's avatar
bk@work.mysql.com committed
572
  {
573
    uint len;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
574 575
    time_t skr=time(NULL);
    struct tm tm_tmp;
576

bk@work.mysql.com's avatar
bk@work.mysql.com committed
577
    localtime_r(&skr,&tm_tmp);
578 579 580 581 582 583 584 585 586 587
    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) ||
588 589
	flush_io_cache(&log_file))
      goto err;
590
    break;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
591
  }
592
  case LOG_BIN:
bk@work.mysql.com's avatar
bk@work.mysql.com committed
593
  {
594 595 596 597 598 599 600 601 602 603
    bool write_file_name_to_index_file=0;

    if (!my_b_filelength(&log_file))
    {
      /*
	The binary log file was empty (probably newly created)
	This is the normal case and happens when the user doesn't specify
	an extension for the binary log files.
	In this case we write a standard header to it.
      */
604 605
      if (my_b_safe_write(&log_file, (byte*) BINLOG_MAGIC,
			  BIN_LOG_HEADER_SIZE))
606
        goto err;
607 608
      bytes_written+= BIN_LOG_HEADER_SIZE;
      write_file_name_to_index_file= 1;
609
    }
610

611
    DBUG_ASSERT(my_b_inited(&index_file) != 0);
612 613
    reinit_io_cache(&index_file, WRITE_CACHE,
                    my_b_filelength(&index_file), 0, 0);
614
    if (need_start_event && !no_auto_events)
615
    {
616 617 618 619 620
      /*
        In 4.x we set need_start_event=0 here, but in 5.0 we want a Start event
        even if this is not the very first binlog.
      */
      Format_description_log_event s(BINLOG_VERSION);
serg@serg.mylan's avatar
serg@serg.mylan committed
621 622 623 624 625 626
      /*
        don't set LOG_EVENT_BINLOG_IN_USE_F for SEQ_READ_APPEND io_cache
        as we won't be able to reset it later
      */
      if (io_cache_type == WRITE_CACHE)
        s.flags|= LOG_EVENT_BINLOG_IN_USE_F;
627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645
      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
646
        Format_description_log_event::Format_description_log_event(char* buf,).
647 648 649 650 651 652 653 654 655 656 657 658 659 660
        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;
661

662 663 664
      if (description_event_for_queue->write(&log_file))
        goto err;
      bytes_written+= description_event_for_queue->data_written;
665
    }
guilhem@mysql.com's avatar
guilhem@mysql.com committed
666 667
    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
668
      goto err;
669 670 671

    if (write_file_name_to_index_file)
    {
guilhem@mysql.com's avatar
guilhem@mysql.com committed
672 673 674 675
      /*
        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.
      */
676 677 678
      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
679 680
	  flush_io_cache(&index_file) ||
          my_sync(index_file.file, MYF(MY_WME)))
681
	goto err;
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
682
    }
683
    break;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
684
  }
685
  case LOG_CLOSED:				// Impossible
686
  case LOG_TO_BE_OPENED:
687 688 689 690
    DBUG_ASSERT(1);
    break;
  }
  DBUG_RETURN(0);
691 692

err:
693 694 695
  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, \
696
shutdown the MySQL server and restart it.", name, errno);
697 698 699
  if (file >= 0)
    my_close(file,MYF(0));
  end_io_cache(&log_file);
700 701
  end_io_cache(&index_file);
  safeFree(name);
702
  log_type= LOG_CLOSED;
703
  DBUG_RETURN(1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
704 705
}

706

bk@work.mysql.com's avatar
bk@work.mysql.com committed
707 708 709
int MYSQL_LOG::get_current_log(LOG_INFO* linfo)
{
  pthread_mutex_lock(&LOCK_log);
710 711
  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
712 713 714 715
  pthread_mutex_unlock(&LOCK_log);
  return 0;
}

716

717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758
/*
  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
759 760
  if (my_chsize(file, offset - init_offset, '\n', MYF(MY_WME)) ||
      my_sync(file, MYF(MY_WME)))
761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780
    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
781
    need_lock		Set this to 1 if the parent doesn't already have a
782 783 784
			lock on LOCK_index

  NOTE
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
785 786
    On systems without the truncate function the file will end with one or
    more empty lines.  These will be ignored when reading the file.
787 788 789 790 791 792 793 794 795

  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
796
{
797 798 799 800 801
  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
802

803
  /*
804 805
    Mutex needed because we need to make sure the file pointer does not
    move from under our feet
806
  */
807
  if (need_lock)
808
    pthread_mutex_lock(&LOCK_index);
809 810 811 812 813
  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);

814
  for (;;)
815
  {
816
    uint length;
817 818 819 820
    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
821
    {
822 823 824
      /* 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
825 826
    }

827 828
    // if the log entry matches, null string matching anything
    if (!log_name ||
sasha@mysql.sashanet.com's avatar
sasha@mysql.sashanet.com committed
829
	(log_name_len == length-1 && fname[log_name_len] == '\n' &&
830
	 !memcmp(fname, log_name, log_name_len)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
831
    {
832
      DBUG_PRINT("info",("Found log file entry"));
833
      fname[length-1]=0;			// remove last \n
834 835
      linfo->index_file_start_offset= offset;
      linfo->index_file_offset = my_b_tell(&index_file);
836
      break;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
837
    }
838
  }
839

840
  if (need_lock)
841
    pthread_mutex_unlock(&LOCK_index);
842
  DBUG_RETURN(error);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
843
}
844

845

846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867
/*
  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
*/

868
int MYSQL_LOG::find_next_log(LOG_INFO* linfo, bool need_lock)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
869
{
870
  int error= 0;
871
  uint length;
872 873
  char *fname= linfo->log_file_name;

874 875
  if (need_lock)
    pthread_mutex_lock(&LOCK_index);
876 877 878 879 880 881 882 883
  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)
884
  {
885
    error = !index_file.error ? LOG_INFO_EOF : LOG_INFO_IO;
886 887
    goto err;
  }
888
  fname[length-1]=0;				// kill \n
889
  linfo->index_file_offset = my_b_tell(&index_file);
890

bk@work.mysql.com's avatar
bk@work.mysql.com committed
891
err:
892 893
  if (need_lock)
    pthread_mutex_unlock(&LOCK_index);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
894 895 896
  return error;
}

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
897

898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916
/*
  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)
917 918
{
  LOG_INFO linfo;
919
  bool error=0;
920 921
  const char* save_name;
  enum_log_type save_log_type;
922
  DBUG_ENTER("reset_logs");
923

924 925 926 927
  /*
    We need to get both locks to be sure that no one is trying to
    write to the index log file.
  */
928
  pthread_mutex_lock(&LOCK_log);
929
  pthread_mutex_lock(&LOCK_index);
930 931 932 933 934 935 936
  /*
    The following mutex is needed to ensure that no threads call
    'delete thd' as we would then risk missing a 'rollback' from this
    thread. If the transaction involved MyISAM tables, it should go
    into binlog even on rollback.
  */
  (void) pthread_mutex_lock(&LOCK_thread_count);
937 938 939 940 941

  /* Save variables so that we can reopen the log */
  save_name=name;
  name=0;					// Protect against free
  save_log_type=log_type;
942
  close(LOG_CLOSE_TO_BE_OPENED);
943 944 945 946

  /* First delete all old log files */

  if (find_log_pos(&linfo, NullS, 0))
947 948 949 950
  {
    error=1;
    goto err;
  }
951

952
  for (;;)
953 954
  {
    my_delete(linfo.log_file_name, MYF(MY_WME));
955
    if (find_next_log(&linfo, 0))
956 957
      break;
  }
958 959

  /* Start logging with a new file */
960
  close(LOG_CLOSE_INDEX);
961 962
  my_delete(index_file_name, MYF(MY_WME));	// Reset (open will update)
  if (!thd->slave_thread)
963
    need_start_event=1;
964 965 966
  if (!open_index_file(index_file_name, 0))
    open(save_name, save_log_type, 0,
         io_cache_type, no_auto_events, max_size, 0);
967
  my_free((gptr) save_name, MYF(0));
968

969
err:
970
  (void) pthread_mutex_unlock(&LOCK_thread_count);
971
  pthread_mutex_unlock(&LOCK_index);
972
  pthread_mutex_unlock(&LOCK_log);
973
  DBUG_RETURN(error);
974 975
}

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
976

977
/*
978 979 980
  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.
981 982 983

  SYNOPSIS
    purge_first_log()
984 985 986 987 988
    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).
989

990 991
  NOTE
    - This is only called from the slave-execute thread when it has read
992 993 994 995 996
      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).
997 998 999

  IMPLEMENTATION
    - Protects index file with LOCK_index
1000 1001
    - Delete relevant relay log files
    - Copy all file names after these ones to the front of the index file
1002
    - If the OS has truncate, truncate the file, else fill it with \n'
1003
    - Read the next file name from the index file and store in rli->linfo
1004 1005

  RETURN VALUES
1006 1007 1008 1009
    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
1010 1011
*/

hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
1012
#ifdef HAVE_REPLICATION
1013

1014
int MYSQL_LOG::purge_first_log(struct st_relay_log_info* rli, bool included) 
1015
{
1016
  int error;
1017 1018
  DBUG_ENTER("purge_first_log");

1019 1020
  DBUG_ASSERT(is_open());
  DBUG_ASSERT(rli->slave_running == 1);
1021
  DBUG_ASSERT(!strcmp(rli->linfo.log_file_name,rli->event_relay_log_name));
1022

1023
  pthread_mutex_lock(&LOCK_index);
1024 1025 1026 1027 1028 1029
  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
1030 1031

  /*
1032 1033 1034
    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
1035
  */
1036 1037 1038 1039 1040
  pthread_cond_broadcast(&rli->log_space_cond);
  
  /*
    Read the next log file name from the index file and pass it back to
    the caller
1041 1042
    If included is true, we want the first relay log;
    otherwise we want the one after event_relay_log_name.
1043
  */
1044 1045 1046 1047
  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)))))
1048 1049
  {
    char buff[22];
1050 1051 1052 1053 1054
    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);
1055 1056
    goto err;
  }
1057

1058
  /*
1059
    Reset rli's coordinates to the current log.
1060
  */
1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074
  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);
1075
    rli->notify_group_relay_log_name_update();
1076
  }
1077 1078 1079

  /* 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
1080

1081
err:
1082
  pthread_mutex_unlock(&LOCK_index);
1083
  DBUG_RETURN(error);
1084 1085
}

1086 1087 1088 1089
/*
  Update log index_file
*/

1090
int MYSQL_LOG::update_log_index(LOG_INFO* log_info, bool need_update_threads)
1091 1092 1093 1094 1095
{
  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
1096 1097
  if (need_update_threads)
    adjust_linfo_offsets(log_info->index_file_start_offset);
1098 1099
  return 0;
}
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
1100

1101 1102 1103 1104 1105
/*
  Remove all logs before the given log from disk and from the index file.

  SYNOPSIS
    purge_logs()
1106 1107 1108 1109 1110 1111 1112
    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
1113 1114 1115 1116 1117 1118 1119 1120 1121 1122

  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
*/

1123 1124 1125 1126 1127
int MYSQL_LOG::purge_logs(const char *to_log, 
                          bool included,
                          bool need_mutex, 
                          bool need_update_threads, 
                          ulonglong *decrease_log_space)
1128 1129
{
  int error;
1130
  bool exit_loop= 0;
1131 1132
  LOG_INFO log_info;
  DBUG_ENTER("purge_logs");
1133
  DBUG_PRINT("info",("to_log= %s",to_log));
1134

1135 1136
  if (need_mutex)
    pthread_mutex_lock(&LOCK_index);
1137
  if ((error=find_log_pos(&log_info, to_log, 0 /*no mutex*/)))
1138 1139
    goto err;

1140
  /*
1141
    File name exists in index file; delete until we find this file
1142 1143 1144
    or a file that is used.
  */
  if ((error=find_log_pos(&log_info, NullS, 0 /*no mutex*/)))
1145
    goto err;
1146 1147
  while ((strcmp(to_log,log_info.log_file_name) || (exit_loop=included)) &&
         !log_in_use(log_info.log_file_name))
1148
  {
1149
    ulong file_size= 0;
1150 1151 1152
    if (decrease_log_space) //stat the file we want to delete
    {
      MY_STAT s;
1153 1154 1155 1156 1157 1158

      /* 
         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.
      */
1159
      if (my_stat(log_info.log_file_name,&s,MYF(0)))
1160
        file_size= s.st_size;
1161
      else
1162 1163
	sql_print_information("Failed to execute my_stat on file '%s'",
			      log_info.log_file_name);
1164 1165 1166 1167 1168 1169 1170
    }
    /*
      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)
1171
      *decrease_log_space-= file_size;
1172
    if (find_next_log(&log_info, 0) || exit_loop)
1173
      break;
1174
  }
1175

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
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, need_update_threads);
1181

1182
err:
1183 1184
  if (need_mutex)
    pthread_mutex_unlock(&LOCK_index);
1185
  DBUG_RETURN(error);
1186 1187
}

1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205
/*
  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
*/

1206
int MYSQL_LOG::purge_logs_before_date(time_t purge_time)
1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221
{
  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*/)))
1222
    goto err;
1223 1224 1225 1226 1227

  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 */
1228 1229
    if (!my_stat(log_info.log_file_name, &stat_area, MYF(0)) ||
	stat_area.st_mtime >= purge_time)
1230
      break;
1231
    my_delete(log_info.log_file_name, MYF(0));
1232 1233
    if (find_next_log(&log_info, 0))
      break;
1234
  }
1235

1236 1237 1238 1239
  /*
    If we get killed -9 here, the sysadmin would have to edit
    the log index file after restart - otherwise, this should be safe
  */
1240
  error= update_log_index(&log_info, 1);
1241

1242 1243
err:
  pthread_mutex_unlock(&LOCK_index);
1244
  DBUG_RETURN(error);
1245 1246
}

1247

hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
1248
#endif /* HAVE_REPLICATION */
1249

1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261

/*
  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
1262 1263
void MYSQL_LOG::make_log_name(char* buf, const char* log_ident)
{
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1264 1265 1266 1267 1268
  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
1269 1270
}

1271 1272 1273 1274 1275 1276

/*
  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
1277
{
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1278
  return !strcmp(log_file_name, log_file_name_arg);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1279 1280
}

1281 1282 1283 1284 1285 1286

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

  SYNOPSIS
    new_file()
1287
    need_lock		Set to 1 if caller has not locked LOCK_log
1288 1289 1290 1291 1292 1293

  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
1294
{
1295 1296 1297
  char new_name[FN_REFLEN], *new_name_ptr, *old_name;
  enum_log_type save_log_type;

1298
  DBUG_ENTER("MYSQL_LOG::new_file");
1299
  if (!is_open())
1300 1301 1302 1303
  {
    DBUG_PRINT("info",("log is closed"));
    DBUG_VOID_RETURN;
  }
1304 1305 1306

  if (need_lock)
    pthread_mutex_lock(&LOCK_log);
1307 1308
  pthread_mutex_lock(&LOCK_index);

1309 1310 1311
  safe_mutex_assert_owner(&LOCK_log);
  safe_mutex_assert_owner(&LOCK_index);

1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329
  /*
    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);
  }

1330
  /* Reuse old name if not binlog and not update log */
1331
  new_name_ptr= name;
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
1332

1333
  /*
1334 1335 1336
    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.
1337
  */
1338 1339 1340
  if (generate_new_name(new_name, name))
    goto end;
  new_name_ptr=new_name;
1341

1342
  if (log_type == LOG_BIN)
1343
  {
1344
    if (!no_auto_events)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1345
    {
1346
      /*
1347 1348
        We log the whole file name for log file as the user may decide
        to change base names at some point.
1349
      */
monty@mysql.com's avatar
monty@mysql.com committed
1350
      THD *thd = current_thd; /* may be 0 if we are reacting to SIGHUP */
1351 1352
      Rotate_log_event r(thd,new_name+dirname_length(new_name));
      r.write(&log_file);
1353
      bytes_written += r.data_written;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1354
    }
1355 1356 1357 1358 1359
    /*
      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.
    */
1360
    signal_update();
1361 1362 1363 1364
  }
  old_name=name;
  save_log_type=log_type;
  name=0;				// Don't free name
1365
  close(LOG_CLOSE_TO_BE_OPENED);
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1366

1367
  /*
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1368
     Note that at this point, log_type != LOG_CLOSED (important for is_open()).
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1369
  */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1370

1371
  /*
1372
     new_file() is only used for rotation (in FLUSH LOGS or because size >
1373
     max_binlog_size or max_relay_log_size).
1374 1375 1376 1377
     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.
1378
  */
1379

1380 1381
  open(old_name, save_log_type, new_name_ptr,
       io_cache_type, no_auto_events, max_size, 1);
1382 1383
  my_free(old_name,MYF(0));

1384
end:
1385 1386
  if (need_lock)
    pthread_mutex_unlock(&LOCK_log);
1387 1388
  pthread_mutex_unlock(&LOCK_index);

1389
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1390 1391
}

1392

1393 1394 1395 1396
bool MYSQL_LOG::append(Log_event* ev)
{
  bool error = 0;
  pthread_mutex_lock(&LOCK_log);
1397 1398
  DBUG_ENTER("MYSQL_LOG::append");

1399
  DBUG_ASSERT(log_file.type == SEQ_READ_APPEND);
1400 1401 1402 1403
  /*
    Log_event::write() is smart enough to use my_b_write() or
    my_b_append() depending on the kind of cache we have.
  */
1404 1405 1406 1407 1408
  if (ev->write(&log_file))
  {
    error=1;
    goto err;
  }
1409
  bytes_written+= ev->data_written;
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
1410
  DBUG_PRINT("info",("max_size: %lu",max_size));
1411
  if ((uint) my_b_append_tell(&log_file) > max_size)
1412
    new_file(0);
1413

1414
err:
1415
  pthread_mutex_unlock(&LOCK_log);
1416
  signal_update();				// Safe as we don't call close
1417
  DBUG_RETURN(error);
1418 1419
}

1420

1421 1422
bool MYSQL_LOG::appendv(const char* buf, uint len,...)
{
1423
  bool error= 0;
1424
  DBUG_ENTER("MYSQL_LOG::appendv");
1425 1426
  va_list(args);
  va_start(args,len);
1427

1428
  DBUG_ASSERT(log_file.type == SEQ_READ_APPEND);
1429

1430 1431 1432
  pthread_mutex_lock(&LOCK_log);
  do
  {
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
1433
    if (my_b_append(&log_file,(byte*) buf,len))
1434
    {
1435 1436
      error= 1;
      goto err;
1437
    }
1438
    bytes_written += len;
1439
  } while ((buf=va_arg(args,const char*)) && (len=va_arg(args,uint)));
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
1440
  DBUG_PRINT("info",("max_size: %lu",max_size));
1441
  if ((uint) my_b_append_tell(&log_file) > max_size)
1442 1443 1444 1445
    new_file(0);

err:
  pthread_mutex_unlock(&LOCK_log);
1446 1447
  if (!error)
    signal_update();
1448
  DBUG_RETURN(error);
1449
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1450

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
1451

1452 1453
/*
  Write to normal (not rotable) log
1454
  This is the format for the 'normal' log.
1455 1456
*/

1457
bool MYSQL_LOG::write(THD *thd,enum enum_server_command command,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1458 1459
		      const char *format,...)
{
1460
  if (is_open() && (what_to_log & (1L << (uint) command)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1461
  {
1462 1463
    uint length;
    int error= 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1464
    VOID(pthread_mutex_lock(&LOCK_log));
1465

nick@mysql.com's avatar
nick@mysql.com committed
1466 1467
    /* 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
1468 1469 1470
    {
      time_t skr;
      ulong id;
1471 1472 1473 1474
      va_list args;
      va_start(args,format);
      char buff[32];

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1475 1476
      if (thd)
      {						// Normal thread
hf@deer.(none)'s avatar
hf@deer.(none) committed
1477 1478 1479 1480 1481
	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
1482 1483
	{
	  VOID(pthread_mutex_unlock(&LOCK_log));
1484
	  return 0;				// No logging
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501
	}
	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;
1502 1503 1504 1505 1506 1507 1508 1509
	/* 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
1510
	if (my_b_write(&log_file, (byte*) buff,16))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1511 1512
	  error=errno;
      }
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
1513
      else if (my_b_write(&log_file, (byte*) "\t\t",2) < 0)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1514
	error=errno;
1515 1516 1517 1518
      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
1519 1520 1521
	error=errno;
      if (format)
      {
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
1522
	if (my_b_write(&log_file, (byte*) " ",1) ||
1523
	    my_b_vprintf(&log_file,format,args) == (uint) -1)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1524 1525
	  error=errno;
      }
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
1526
      if (my_b_write(&log_file, (byte*) "\n",1) ||
1527
	  flush_io_cache(&log_file))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1528 1529 1530 1531 1532 1533
	error=errno;
      if (error && ! write_error)
      {
	write_error=1;
	sql_print_error(ER(ER_ERROR_ON_WRITE),name,error);
      }
1534
      va_end(args);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1535 1536
    }
    VOID(pthread_mutex_unlock(&LOCK_log));
1537
    return error != 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1538
  }
1539
  return 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1540 1541
}

1542
bool MYSQL_LOG::flush_and_sync()
1543
{
1544 1545 1546 1547 1548
  int err=0, fd=log_file.file;
  safe_mutex_assert_owner(&LOCK_log);
  if (flush_io_cache(&log_file))
    return 1;
  if (++sync_binlog_counter >= sync_binlog_period && sync_binlog_period)
1549 1550
  {
    sync_binlog_counter= 0;
1551
    err=my_sync(fd, MYF(MY_WME));
1552
  }
1553
  return err;
1554 1555
}

1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569
void MYSQL_LOG::start_union_events(THD *thd)
{
  DBUG_ASSERT(!thd->binlog_evt_union.do_union);
  thd->binlog_evt_union.do_union= TRUE;
  thd->binlog_evt_union.unioned_events= FALSE;
  thd->binlog_evt_union.unioned_events_trans= FALSE;
}

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

1570 1571 1572 1573
/*
  Write an event to the binary log
*/

1574
bool MYSQL_LOG::write(Log_event *event_info)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1575
{
1576 1577 1578 1579
  THD *thd= event_info->thd;
  bool error= 1;
  DBUG_ENTER("MYSQL_LOG::write(Log_event *)");

1580 1581
  if (thd->binlog_evt_union.do_union)
  {
1582 1583 1584 1585
    /*
      In Stored function; Remember that function call caused an update.
      We will log the function call to the binary log on function exit
    */
1586 1587 1588 1589 1590
    thd->binlog_evt_union.unioned_events= TRUE;
    thd->binlog_evt_union.unioned_events_trans |= event_info->cache_stmt;
    DBUG_RETURN(0);
  }
  
1591 1592
  pthread_mutex_lock(&LOCK_log);

1593
  /*
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1594 1595 1596 1597
     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.
  */
1598
  if (likely(is_open()))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1599
  {
1600 1601
    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
1602
#ifdef HAVE_REPLICATION
1603
    /*
1604 1605 1606 1607
       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)"
    */
1608
    if ((thd && !(thd->options & OPTION_BIN_LOG)) ||
1609
	(!db_ok(local_db, binlog_do_db, binlog_ignore_db)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1610
    {
1611
      VOID(pthread_mutex_unlock(&LOCK_log));
1612
      DBUG_PRINT("error",("!db_ok('%s')", local_db));
1613
      DBUG_RETURN(0);
1614
    }
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
1615
#endif /* HAVE_REPLICATION */
1616

1617 1618 1619 1620 1621 1622 1623 1624
#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)
1625
      - or if the event asks for it (cache_stmt == TRUE).
1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636
    */
    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));
1637 1638
          if (!trans_log || open_cached_file(trans_log, mysql_tmpdir,
                                             LOG_PREFIX,
1639 1640 1641 1642 1643 1644 1645
                                             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;
1646
          trans_register_ha(thd,
1647 1648 1649
                            thd->options & (OPTION_NOT_AUTOCOMMIT |
                                            OPTION_BEGIN),
                            &binlog_hton);
1650 1651
        }
        else if (!my_b_tell(trans_log))
1652
          trans_register_ha(thd,
1653 1654 1655
                            thd->options & (OPTION_NOT_AUTOCOMMIT |
                                            OPTION_BEGIN),
                            &binlog_hton);
1656 1657 1658 1659 1660 1661 1662 1663
        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()));

1664 1665 1666 1667
    /*
      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
1668 1669

    /*
1670 1671
      1. Write first log events which describe the 'run environment'
      of the SQL command
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
1672 1673
    */

1674
    if (thd)
1675
    {
1676 1677 1678
      if (thd->last_insert_id_used)
      {
	Intvar_log_event e(thd,(uchar) LAST_INSERT_ID_EVENT,
1679
			   thd->current_insert_id);
1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694
	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;
      }
1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710
      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;
	}
      }
1711
    }
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
1712

1713 1714
    /*
       Write the SQL command
1715
     */
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
1716

monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
1717
    if (event_info->write(file))
1718
      goto err;
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
1719

1720
    if (file == &log_file) // we are writing to the real log (disk)
1721
    {
1722
      if (flush_and_sync())
monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
1723
	goto err;
1724 1725
      signal_update();
      rotate_and_purge(RP_LOCK_LOG_IS_ALREADY_LOCKED);
1726
    }
monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
1727
    error=0;
1728

1729 1730 1731 1732
err:
    if (error)
    {
      if (my_errno == EFBIG)
bell@sanja.is.com.ua's avatar
bell@sanja.is.com.ua committed
1733
	my_message(ER_TRANS_CACHE_FULL, ER(ER_TRANS_CACHE_FULL), MYF(0));
1734 1735 1736 1737 1738
      else
	my_error(ER_ERROR_ON_WRITE, MYF(0), name, errno);
      write_error=1;
    }
  }
1739 1740

  pthread_mutex_unlock(&LOCK_log);
1741
  DBUG_RETURN(error);
1742 1743
}

1744 1745
void MYSQL_LOG::rotate_and_purge(uint flags)
{
1746 1747
  if (!(flags & RP_LOCK_LOG_IS_ALREADY_LOCKED))
    pthread_mutex_lock(&LOCK_log);
1748 1749
  if ((flags & RP_FORCE_ROTATE) ||
      (my_b_tell(&log_file) >= (my_off_t) max_size))
1750
  {
1751
    new_file(0);
1752 1753 1754 1755 1756 1757 1758 1759 1760
#ifdef HAVE_REPLICATION
    if (expire_logs_days)
    {
      long purge_time= time(0) - expire_logs_days*24*60*60;
      if (purge_time >= 0)
        purge_logs_before_date(purge_time);
    }
#endif
  }
1761 1762
  if (!(flags & RP_LOCK_LOG_IS_ALREADY_LOCKED))
    pthread_mutex_unlock(&LOCK_log);
1763
}
1764

1765 1766 1767 1768 1769 1770 1771 1772 1773
uint MYSQL_LOG::next_file_id()
{
  uint res;
  pthread_mutex_lock(&LOCK_log);
  res = file_id++;
  pthread_mutex_unlock(&LOCK_log);
  return res;
}

1774

1775 1776
/*
  Write a cached log entry to the binary log
1777

guilhem@mysql.com's avatar
guilhem@mysql.com committed
1778 1779
  SYNOPSIS
    write()
1780
    thd
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1781 1782
    cache		The cache to copy to the binlog

1783 1784
  NOTE
    - We only come here if there is something in the cache.
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
1785
    - The thing in the cache is always a complete transaction
1786 1787 1788 1789
    - 'cache' needs to be reinitialized after this functions returns.

  IMPLEMENTATION
    - To support transaction over replication, we wrap the transaction
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1790
      with BEGIN/COMMIT or BEGIN/ROLLBACK in the binary log.
1791 1792 1793
      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.
1794 1795
*/

1796
bool MYSQL_LOG::write(THD *thd, IO_CACHE *cache, Log_event *commit_event)
1797
{
1798
  DBUG_ENTER("MYSQL_LOG::write(THD *, IO_CACHE *, Log_event *)");
1799
  VOID(pthread_mutex_lock(&LOCK_log));
1800

1801
  if (likely(is_open()))                       // Should always be true
1802
  {
1803 1804
    uint length;

1805
    /*
1806 1807
      Log "BEGIN" at the beginning of the transaction.
      which may contain more than 1 SQL statement.
1808
    */
1809
    if (thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN))
1810
    {
1811
      Query_log_event qinfo(thd, "BEGIN", 5, TRUE, FALSE);
1812 1813 1814 1815 1816 1817 1818 1819
      /*
        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;
1820 1821 1822
      /*
        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
1823 1824 1825
	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.
1826
      */
1827 1828 1829
      if (qinfo.write(&log_file))
	goto err;
    }
1830
    /* Read from the file used to cache the queries .*/
1831
    if (reinit_io_cache(cache, READ_CACHE, 0, 0, 0))
1832
      goto err;
1833
    length=my_b_bytes_in_cache(cache);
1834
    DBUG_EXECUTE_IF("half_binlogged_transaction", length-=100;);
1835
    do
1836
    {
1837
      /* Write data to the binary log file */
monty@bitch.mysql.fi's avatar
monty@bitch.mysql.fi committed
1838
      if (my_b_write(&log_file, cache->read_pos, length))
1839
	goto err;
monty@bitch.mysql.fi's avatar
monty@bitch.mysql.fi committed
1840
      cache->read_pos=cache->read_end;		// Mark buffer used up
1841
      DBUG_EXECUTE_IF("half_binlogged_transaction", goto DBUG_skip_commit;);
1842
    } while ((length=my_b_fill(cache)));
1843

1844 1845 1846
    if (commit_event->write(&log_file))
      goto err;
DBUG_skip_commit:
1847
    if (flush_and_sync())
1848
      goto err;
1849
    DBUG_EXECUTE_IF("half_binlogged_transaction", abort(););
1850 1851
    if (cache->error)				// Error on read
    {
1852
      sql_print_error(ER(ER_ERROR_ON_READ), cache->file_name, errno);
1853
      write_error=1;				// Don't give more errors
1854
      goto err;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1855
    }
1856
    signal_update();
1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868
    /*
      if commit_event is Xid_log_event, increase the number of
      prepared_xids (it's decreasd in ::unlog()). Binlog cannot be rotated
      if there're prepared xids in it - see the comment in new_file() for
      an explanation.
      If the commit_event is not Xid_log_event (then it's a Query_log_event)
      rotate binlog, if necessary.
    */
    if (commit_event->get_type_code() == XID_EVENT)
      thread_safe_increment(prepared_xids, &LOCK_prep_xids);
    else
      rotate_and_purge(RP_LOCK_LOG_IS_ALREADY_LOCKED);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1869
  }
1870
  VOID(pthread_mutex_unlock(&LOCK_log));
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
1871

1872
  DBUG_RETURN(0);
1873 1874

err:
1875 1876 1877 1878 1879
  if (!write_error)
  {
    write_error= 1;
    sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
  }
1880
  VOID(pthread_mutex_unlock(&LOCK_log));
1881
  DBUG_RETURN(1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1882 1883
}

1884

1885
/*
1886
  Write to the slow query log.
1887
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1888

1889
bool MYSQL_LOG::write(THD *thd,const char *query, uint query_length,
1890
		      time_t query_start_arg)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1891
{
1892
  bool error=0;
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1893
  time_t current_time;
1894 1895
  if (!is_open())
    return 0;
monty@mysql.com's avatar
monty@mysql.com committed
1896 1897
  DBUG_ENTER("MYSQL_LOG::write");

guilhem@mysql.com's avatar
guilhem@mysql.com committed
1898
  VOID(pthread_mutex_lock(&LOCK_log));
1899
  if (is_open())
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1900 1901 1902 1903
  {						// Safety agains reopen
    int tmp_errno=0;
    char buff[80],*end;
    end=buff;
1904
    if (!(thd->options & OPTION_UPDATE_LOG))
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1905 1906
    {
      VOID(pthread_mutex_unlock(&LOCK_log));
monty@mysql.com's avatar
monty@mysql.com committed
1907
      DBUG_RETURN(0);
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1908
    }
1909
    if (!(specialflag & SPECIAL_SHORT_LOG_FORMAT) || query_start_arg)
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1910 1911 1912
    {
      current_time=time(NULL);
      if (current_time != last_time)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1913
      {
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928
        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
1929
      }
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1930
      if (my_b_printf(&log_file, "# User@Host: %s[%s] @ %s [%s]\n",
hf@deer.(none)'s avatar
hf@deer.(none) committed
1931 1932
                      thd->priv_user ? thd->priv_user : "",
                      thd->user ? thd->user : "",
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961
                      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)
    {
1962
      if (!(specialflag & SPECIAL_SHORT_LOG_FORMAT))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1963
      {
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1964 1965
        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
1966
      }
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1967 1968 1969 1970
    }
    if (thd->query_start_used)
    {
      if (query_start_arg != thd->query_start())
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1971
      {
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1972 1973 1974
        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
1975
      }
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999
    }
    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
2000
      {
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2001 2002
        write_error=1;
        sql_print_error(ER(ER_ERROR_ON_WRITE),name,error);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2003 2004 2005
      }
    }
  }
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2006
  VOID(pthread_mutex_unlock(&LOCK_log));
monty@mysql.com's avatar
monty@mysql.com committed
2007
  DBUG_RETURN(error);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2008 2009
}

2010

2011 2012 2013 2014 2015 2016
/*
  Wait until we get a signal that the binary log has been updated

  SYNOPSIS
    wait_for_update()
    thd			Thread variable
2017
    is_slave            If 0, the caller is the Binlog_dump thread from master;
2018 2019
                        if 1, the caller is the SQL thread from the slave. This
                        influences only thd->proc_info.
2020 2021 2022

  NOTES
    One must have a lock on LOCK_log before calling this function.
2023
    This lock will be released before return! That's required by
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2024
    THD::enter_cond() (see NOTES in sql_class.h).
2025 2026
*/

2027
void MYSQL_LOG::wait_for_update(THD* thd, bool is_slave)
2028
{
monty@mysql.com's avatar
monty@mysql.com committed
2029 2030 2031
  const char *old_msg;
  DBUG_ENTER("wait_for_update");
  old_msg= thd->enter_cond(&update_cond, &LOCK_log,
2032
                           is_slave ?
monty@mysql.com's avatar
monty@mysql.com committed
2033
                           "Has read all relay log; waiting for the slave I/O "
2034
                           "thread to update it" :
monty@mysql.com's avatar
monty@mysql.com committed
2035
                           "Has sent all binlog to slave; waiting for binlog "
2036
                           "to be updated");
2037 2038
  pthread_cond_wait(&update_cond, &LOCK_log);
  thd->exit_cond(old_msg);
monty@mysql.com's avatar
monty@mysql.com committed
2039
  DBUG_VOID_RETURN;
2040
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2041

2042

2043 2044 2045 2046 2047
/*
  Close the log file

  SYNOPSIS
    close()
2048 2049 2050 2051 2052
    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
2053 2054 2055 2056

  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
2057 2058
*/

2059
void MYSQL_LOG::close(uint exiting)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2060
{					// One can't set log_type here!
2061 2062
  DBUG_ENTER("MYSQL_LOG::close");
  DBUG_PRINT("enter",("exiting: %d", (int) exiting));
2063
  if (log_type != LOG_CLOSED && log_type != LOG_TO_BE_OPENED)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2064
  {
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
2065
#ifdef HAVE_REPLICATION
2066 2067
    if (log_type == LOG_BIN && !no_auto_events &&
	(exiting & LOG_CLOSE_STOP_EVENT))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2068 2069
    {
      Stop_log_event s;
2070
      s.write(&log_file);
2071
      bytes_written+= s.data_written;
2072
      signal_update();
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2073
    }
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
2074
#endif /* HAVE_REPLICATION */
2075
    end_io_cache(&log_file);
2076 2077

    /* don't pwrite in a file opened with O_APPEND - it doesn't work */
2078
    if (log_file.type == WRITE_CACHE && log_type == LOG_BIN)
2079 2080
    {
      my_off_t offset= BIN_LOG_HEADER_SIZE + FLAGS_OFFSET;
serg@serg.mylan's avatar
serg@serg.mylan committed
2081
      byte flags=0; // clearing LOG_EVENT_BINLOG_IN_USE_F
2082
      my_pwrite(log_file.file, &flags, 1, offset, MYF(0));
2083 2084
    }

2085 2086 2087 2088 2089 2090
    if (my_sync(log_file.file,MYF(MY_WME)) && ! write_error)
    {
      write_error=1;
      sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
    }
    if (my_close(log_file.file,MYF(MY_WME)) && ! write_error)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2091 2092
    {
      write_error=1;
2093
      sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2094 2095
    }
  }
2096 2097 2098 2099 2100 2101

  /*
    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.
  */

2102
  if ((exiting & LOG_CLOSE_INDEX) && my_b_inited(&index_file))
2103
  {
2104 2105
    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
2106
    {
2107 2108
      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
2109
    }
2110
  }
2111
  log_type= (exiting & LOG_CLOSE_TO_BE_OPENED) ? LOG_TO_BE_OPENED : LOG_CLOSED;
2112
  safeFree(name);
2113
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2114 2115 2116
}


2117 2118 2119 2120 2121 2122 2123 2124 2125
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
2126 2127
  DBUG_ENTER("MYSQL_LOG::set_max_size");
  pthread_mutex_lock(&LOCK_log);
2128 2129
  if (is_open())
    max_size= max_size_arg;
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2130 2131
  pthread_mutex_unlock(&LOCK_log);
  DBUG_VOID_RETURN;
2132
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2133

2134

2135 2136
/*
  Check if a string is a valid number
2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150

  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
2151
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163

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++;
2164 2165
  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
2166 2167 2168 2169 2170 2171 2172
  {
    flag=1;
    str++;
  }
  if (*str == '.')
  {
    for (str++ ;
2173
	 my_isdigit(files_charset_info,*str) ||
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184
	   (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
2185
void print_buffer_to_file(enum loglevel level, const char *buffer)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2186 2187 2188 2189
{
  time_t skr;
  struct tm tm_tmp;
  struct tm *start;
monty@mysql.com's avatar
monty@mysql.com committed
2190 2191
  DBUG_ENTER("print_buffer_to_file");
  DBUG_PRINT("enter",("buffer: %s", buffer));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2192 2193

  VOID(pthread_mutex_lock(&LOCK_error_log));
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2194

bk@work.mysql.com's avatar
bk@work.mysql.com committed
2195
  skr=time(NULL);
2196
  localtime_r(&skr, &tm_tmp);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2197
  start=&tm_tmp;
2198
  fprintf(stderr, "%02d%02d%02d %2d:%02d:%02d [%s] %s\n",
2199 2200 2201 2202 2203 2204
          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
2205
          (level == ERROR_LEVEL ? "ERROR" : level == WARNING_LEVEL ?
2206
           "Warning" : "Note"),
monty@mysql.com's avatar
monty@mysql.com committed
2207
          buffer);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2208

bk@work.mysql.com's avatar
bk@work.mysql.com committed
2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223
  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
}
2224

monty@mysql.com's avatar
monty@mysql.com committed
2225

2226 2227 2228
bool flush_error_log()
{
  bool result=0;
2229
  if (opt_error_log)
2230 2231 2232 2233
  {
    char err_renamed[FN_REFLEN], *end;
    end= strmake(err_renamed,log_error_file,FN_REFLEN-4);
    strmov(end, "-old");
2234
    VOID(pthread_mutex_lock(&LOCK_error_log));
2235 2236 2237 2238 2239 2240
#ifdef __WIN__
    char err_temp[FN_REFLEN+4];
    /*
     On Windows is necessary a temporary file for to rename
     the current error file.
    */
2241
    strxmov(err_temp, err_renamed,"-tmp",NullS);
2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254
    (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)
2255
             my_fwrite(stderr, (byte*) buf, bytes, MYF(0));
2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268
        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
2269
    VOID(pthread_mutex_unlock(&LOCK_error_log));
2270 2271 2272
  }
   return result;
}
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
2273

monty@mysql.com's avatar
monty@mysql.com committed
2274 2275 2276 2277 2278 2279 2280 2281
void MYSQL_LOG::signal_update()
{
  DBUG_ENTER("MYSQL_LOG::signal_update");
  pthread_cond_broadcast(&update_cond);
  DBUG_VOID_RETURN;
}


2282
#ifdef __NT__
monty@mysql.com's avatar
monty@mysql.com committed
2283 2284
void print_buffer_to_nt_eventlog(enum loglevel level, char *buff,
                                 uint length, int buffLen)
2285 2286 2287 2288
{
  HANDLE event;
  char   *buffptr;
  LPCSTR *buffmsgptr;
serg@serg.mylan's avatar
serg@serg.mylan committed
2289
  DBUG_ENTER("print_buffer_to_nt_eventlog");
2290

monty@mysql.com's avatar
monty@mysql.com committed
2291
  buffptr= buff;
2292
  if (length > (uint)(buffLen-5))
2293
  {
2294
    char *newBuff= new char[length + 5];
serg@serg.mylan's avatar
serg@serg.mylan committed
2295
    strcpy(newBuff, buff);
monty@mysql.com's avatar
monty@mysql.com committed
2296
    buffptr= newBuff;
2297
  }
monty@mysql.com's avatar
monty@mysql.com committed
2298 2299
  strmov(buffptr+length, "\r\n\r\n");
  buffmsgptr= (LPCSTR*) &buffptr;               // Keep windows happy
2300

monty@mysql.com's avatar
monty@mysql.com committed
2301 2302
  setup_windows_event_source();
  if ((event= RegisterEventSource(NULL,"MySQL")))
2303
  {
monty@mysql.com's avatar
monty@mysql.com committed
2304
    switch (level) {
2305
      case ERROR_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
2306 2307
        ReportEvent(event, EVENTLOG_ERROR_TYPE, 0, MSG_DEFAULT, NULL, 1, 0,
                    buffmsgptr, NULL);
2308 2309
        break;
      case WARNING_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
2310 2311
        ReportEvent(event, EVENTLOG_WARNING_TYPE, 0, MSG_DEFAULT, NULL, 1, 0,
                    buffmsgptr, NULL);
2312 2313
        break;
      case INFORMATION_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
2314 2315
        ReportEvent(event, EVENTLOG_INFORMATION_TYPE, 0, MSG_DEFAULT, NULL, 1,
                    0, buffmsgptr, NULL);
2316 2317 2318 2319 2320
        break;
    }
    DeregisterEventSource(event);
  }

monty@mysql.com's avatar
monty@mysql.com committed
2321
  /* if we created a string buffer, then delete it */
serg@serg.mylan's avatar
serg@serg.mylan committed
2322
  if (buffptr != buff)
2323 2324 2325 2326
    delete[] buffptr;

  DBUG_VOID_RETURN;
}
monty@mysql.com's avatar
monty@mysql.com committed
2327 2328
#endif /* __NT__ */

2329 2330

/*
monty@mysql.com's avatar
monty@mysql.com committed
2331 2332
  Prints a printf style message to the error log and, under NT, to the
  Windows event log.
2333 2334 2335

  SYNOPSIS
    vprint_msg_to_log()
monty@mysql.com's avatar
monty@mysql.com committed
2336 2337
    event_type             Type of event to write (Error, Warning, or Info)
    format                 Printf style format of message
serg@serg.mylan's avatar
serg@serg.mylan committed
2338
    args                   va_list list of arguments for the message
2339 2340 2341 2342

  NOTE

  IMPLEMENTATION
monty@mysql.com's avatar
monty@mysql.com committed
2343 2344
    This function prints the message into a buffer and then sends that buffer
    to other functions to write that message to other logging sources.
2345 2346 2347

  RETURN VALUES
    void
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2348
*/
monty@mysql.com's avatar
monty@mysql.com committed
2349

2350
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
2351 2352
{
  char   buff[1024];
monty@mysql.com's avatar
monty@mysql.com committed
2353
  uint length;
2354
  DBUG_ENTER("vprint_msg_to_log");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2355

monty@mysql.com's avatar
monty@mysql.com committed
2356
  length= my_vsnprintf(buff, sizeof(buff)-5, format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
2357
  print_buffer_to_file(level, buff);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2358 2359

#ifdef __NT__
monty@mysql.com's avatar
monty@mysql.com committed
2360
  print_buffer_to_nt_eventlog(level, buff, length, sizeof(buff));
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2361
#endif
2362

rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2363 2364 2365
  DBUG_VOID_RETURN;
}

2366

monty@mysql.com's avatar
monty@mysql.com committed
2367
void sql_print_error(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2368 2369
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
2370
  DBUG_ENTER("sql_print_error");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2371

serg@serg.mylan's avatar
serg@serg.mylan committed
2372 2373 2374
  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
2375 2376 2377 2378 2379

  DBUG_VOID_RETURN;
}


monty@mysql.com's avatar
monty@mysql.com committed
2380
void sql_print_warning(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2381 2382
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
2383
  DBUG_ENTER("sql_print_warning");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2384

serg@serg.mylan's avatar
serg@serg.mylan committed
2385 2386 2387
  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
2388 2389 2390 2391 2392

  DBUG_VOID_RETURN;
}


monty@mysql.com's avatar
monty@mysql.com committed
2393
void sql_print_information(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2394 2395
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
2396
  DBUG_ENTER("sql_print_information");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
2397

serg@serg.mylan's avatar
serg@serg.mylan committed
2398 2399 2400
  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
2401

guilhem@mysql.com's avatar
guilhem@mysql.com committed
2402 2403
  DBUG_VOID_RETURN;
}
2404

2405

2406 2407 2408 2409 2410 2411 2412 2413
/********* 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
2414 2415
  states - active, syncing, pool.
  there could be only one page in active or syncing states,
2416
  but many in pool - pool is fifo queue.
2417
  usual lifecycle of a page is pool->active->syncing->pool
2418
  "active" page - is a page where new xid's are logged.
2419 2420
  the page stays active as long as syncing slot is taken.
  "syncing" page is being synced to disk. no new xid can be added to it.
2421
  when the sync is done the page is moved to a pool and an active page
2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434
  becomes "syncing".

  the result of such an architecture is a natural "commit grouping" -
  If commits are coming faster than the system can sync, they do not
  stall. Instead, all commit that came since the last sync are
  logged to the same page, and they all are synced with the next -
  one - sync. Thus, thought individual commits are delayed, throughput
  is not decreasing.

  when a xid is added to an active page, the thread of this xid waits
  for a page's condition until the page is synced. when syncing slot
  becomes vacant one of these waiters is awaken to take care of syncing.
  it syncs the page and signals all waiters that the page is synced.
2435 2436 2437 2438
  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)

2439 2440
  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
2441 2442
  dirty - we don't sync removals to disk.
*/
2443 2444 2445 2446 2447

ulong tc_log_page_waits= 0;

#ifdef HAVE_MMAP

2448 2449
#define TC_LOG_HEADER_SIZE (sizeof(tc_log_magic)+1)

2450
static const char tc_log_magic[]={(char) 254, 0x23, 0x05, 0x74};
2451

2452
ulong opt_tc_log_size= TC_LOG_MIN_SIZE;
2453
ulong tc_log_max_pages_used=0, tc_log_page_size=0, tc_log_cur_pages_used=0;
2454 2455 2456 2457 2458 2459 2460

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

2461
  DBUG_ASSERT(total_ha_2pc > 1);
2462 2463
  DBUG_ASSERT(opt_name && opt_name[0]);

serg@serg.mylan's avatar
serg@serg.mylan committed
2464
  tc_log_page_size= my_getpagesize();
2465 2466 2467
  DBUG_ASSERT(TC_LOG_PAGE_SIZE % tc_log_page_size == 0);

  fn_format(logname,opt_name,mysql_data_home,"",MY_UNPACK_FILENAME);
2468
  if ((fd= my_open(logname, O_RDWR, MYF(0))) < 0)
2469
  {
2470 2471
    if (my_errno != ENOENT)
      goto err;
2472 2473
    if (using_heuristic_recover())
      return 1;
2474
    if ((fd= my_create(logname, O_RDWR, 0, MYF(MY_WME))) < 0)
2475 2476
      goto err;
    inited=1;
2477
    file_length= opt_tc_log_size;
2478 2479 2480 2481 2482
    if (my_chsize(fd, file_length, 0, MYF(MY_WME)))
      goto err;
  }
  else
  {
2483 2484
    inited= 1;
    crashed= TRUE;
serg@serg.mylan's avatar
serg@serg.mylan committed
2485
    sql_print_information("Recovering after a crash using %s", opt_name);
2486 2487 2488 2489 2490 2491
    if (tc_heuristic_recover)
    {
      sql_print_error("Cannot perform automatic crash recovery when "
                      "--tc-heuristic-recover is used");
      goto err;
    }
2492
    file_length= my_seek(fd, 0L, MY_SEEK_END, MYF(MY_WME+MY_FAE));
2493 2494 2495 2496
    if (file_length == MY_FILEPOS_ERROR || file_length % tc_log_page_size)
      goto err;
  }

2497
  data= (uchar *)my_mmap(0, (size_t)file_length, PROT_READ|PROT_WRITE,
2498 2499 2500 2501 2502 2503 2504 2505
                        MAP_NOSYNC|MAP_SHARED, fd, 0);
  if (data == MAP_FAILED)
  {
    my_errno=errno;
    goto err;
  }
  inited=2;

2506
  npages=(uint)file_length/tc_log_page_size;
2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523
  DBUG_ASSERT(npages >= 3);             // to guarantee non-empty pool
  if (!(pages=(PAGE *)my_malloc(npages*sizeof(PAGE), MYF(MY_WME|MY_ZEROFILL))))
    goto err;
  inited=3;
  for (pg=pages, i=0; i < npages; i++, pg++)
  {
    pg->next=pg+1;
    pg->waiters=0;
    pg->state=POOL;
    pthread_mutex_init(&pg->lock, MY_MUTEX_INIT_FAST);
    pthread_cond_init (&pg->cond, 0);
    pg->start=(my_xid *)(data + i*tc_log_page_size);
    pg->end=(my_xid *)(pg->start + tc_log_page_size);
    pg->size=pg->free=tc_log_page_size/sizeof(my_xid);
  }
  pages[0].size=pages[0].free=
                (tc_log_page_size-TC_LOG_HEADER_SIZE)/sizeof(my_xid);
2524
  pages[0].start=pages[0].end-pages[0].size;
2525 2526 2527
  pages[npages-1].next=0;
  inited=4;

2528
  if (crashed && recover())
2529
      goto err;
2530

2531
  memcpy(data, tc_log_magic, sizeof(tc_log_magic));
2532
  data[sizeof(tc_log_magic)]= (uchar)total_ha_2pc;
2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543
  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;

2544
  syncing= 0;
2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556
  active=pages;
  pool=pages+1;
  pool_last=pages+npages-1;

  return 0;

err:
  close();
  return 1;
}

/*
2557 2558
  there is no active page, let's got one from the pool

2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570
  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;

2571 2572 2573
  if (syncing)
    pthread_mutex_lock(&LOCK_pool);

2574 2575 2576
  do
  {
    best_p= p= &pool;
2577 2578
    if ((*p)->waiters == 0) // can the first page be used ?
      break;                // yes - take it.
2579

2580
    best_free=0;            // no - trying second strategy
2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592
    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;
2593
  if (active->free == active->size) // we've chosen an empty page
2594 2595 2596 2597 2598
  {
    tc_log_cur_pages_used++;
    set_if_bigger(tc_log_max_pages_used, tc_log_cur_pages_used);
  }

2599
  if ((*best_p)->next)              // unlink the page from the pool
2600 2601 2602
    *best_p=(*best_p)->next;
  else
    pool_last=*best_p;
2603 2604 2605

  if (syncing)
    pthread_mutex_unlock(&LOCK_pool);
2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616
}

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);
2617
  return 1; // always return 1
2618 2619 2620
}

/*
2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632
  all access to active page is serialized but it's not a problem, as
  we're assuming that fsync() will be a main bottleneck.
  That is, parallelizing writes to log pages we'll decrease number of
  threads waiting for a page, but then all these threads will be waiting
  for a fsync() anyway

  RETURN
         0  - error
  otherwise - "cookie", a number that will be passed as an argument
              to unlog() call. tc_log can define it any way it wants,
              and use for whatever purposes. TC_LOG_MMAP sets it
              to the position in memory where xid was logged to.
2633 2634 2635 2636 2637 2638 2639 2640 2641 2642
*/

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

  pthread_mutex_lock(&LOCK_active);

2643 2644 2645 2646 2647 2648 2649
  /*
    if active page is full - just wait...
    frankly speaking, active->free here accessed outside of mutex
    protection, but it's safe, because it only means we may miss an
    unlog() for the active page, and we're not waiting for it here -
    unlog() does not signal COND_active.
  */
2650 2651 2652
  while (unlikely(active && active->free == 0))
    pthread_cond_wait(&COND_active, &LOCK_active);

2653
  /* no active page ? take one from the pool */
2654 2655 2656 2657 2658 2659
  if (active == 0)
    get_active_from_pool();

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

2660
  /* searching for an empty slot */
2661 2662 2663 2664 2665 2666
  while (*p->ptr)
  {
    p->ptr++;
    DBUG_ASSERT(p->ptr < p->end);               // because p->free > 0
  }

2667 2668
  /* found! store xid there and mark the page dirty */
  cookie= (ulong)((uchar *)p->ptr - data);      // can never be zero
2669 2670 2671 2672 2673
  *p->ptr++= xid;
  p->free--;
  p->state= DIRTY;

  /* to sync or not to sync - this is the question */
2674
  pthread_mutex_unlock(&LOCK_active);
2675 2676 2677
  pthread_mutex_lock(&LOCK_sync);
  pthread_mutex_unlock(&p->lock);

2678
  if (syncing)
2679 2680
  {                                          // somebody's syncing. let's wait
    p->waiters++;
2681
    /*
monty@mishka.local's avatar
monty@mishka.local committed
2682
      note - it must be while (), not do ... while () here
2683 2684 2685
      as p->state may be not DIRTY when we come here
    */
    while (p->state == DIRTY && syncing)
2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696
      pthread_cond_wait(&p->cond, &LOCK_sync);
    p->waiters--;
    err= p->state == ERROR;
    if (p->state != DIRTY)                   // page was synced
    {
      if (p->waiters == 0)
        pthread_cond_signal(&COND_pool);     // in case somebody's waiting
      pthread_mutex_unlock(&LOCK_sync);
      goto done;                             // we're done
    }
  }                                          // page was not synced! do it now
2697 2698 2699 2700
  DBUG_ASSERT(active == p && syncing == 0);
  pthread_mutex_lock(&LOCK_active);
  syncing=p;                                 // place is vacant - take it
  active=0;                                  // page is not active anymore
2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713
  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;

2714
  DBUG_ASSERT(syncing != active);
2715 2716 2717 2718 2719

  /*
    sit down and relax - this can take a while...
    note - no locks are held at this point
  */
2720
  err= my_msync(fd, syncing->start, 1, MS_SYNC);
2721

2722
  /* page is synced. let's move it to the pool */
2723
  pthread_mutex_lock(&LOCK_pool);
2724 2725 2726 2727 2728
  pool_last->next=syncing;
  pool_last=syncing;
  syncing->next=0;
  syncing->state= err ? ERROR : POOL;
  pthread_cond_broadcast(&syncing->cond);    // signal "sync done"
2729 2730 2731
  pthread_cond_signal(&COND_pool);           // in case somebody's waiting
  pthread_mutex_unlock(&LOCK_pool);

2732
  /* marking 'syncing' slot free */
2733
  pthread_mutex_lock(&LOCK_sync);
2734
  syncing=0;
2735 2736 2737 2738 2739
  pthread_cond_signal(&active->cond);        // wake up a new syncer
  pthread_mutex_unlock(&LOCK_sync);
  return err;
}

2740 2741 2742 2743
/*
  erase xid from the page, update page free space counters/pointers.
  cookie points directly to the memory where xid was logged
*/
2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756
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);
2757
  if (p->free == p->size)               // the page is completely empty
2758 2759
    statistic_decrement(tc_log_cur_pages_used, &LOCK_status);
  if (p->waiters == 0)                 // the page is in pool and ready to rock
2760
    pthread_cond_signal(&COND_pool);   // ping ... for overflow()
2761 2762 2763 2764 2765
  pthread_mutex_unlock(&p->lock);
}

void TC_LOG_MMAP::close()
{
serg@serg.mylan's avatar
serg@serg.mylan committed
2766
  uint i;
2767 2768 2769 2770 2771 2772 2773 2774 2775
  switch (inited) {
  case 6:
    pthread_mutex_destroy(&LOCK_sync);
    pthread_mutex_destroy(&LOCK_active);
    pthread_mutex_destroy(&LOCK_pool);
    pthread_cond_destroy(&COND_pool);
  case 5:
    data[0]='A'; // garble the first (signature) byte, in case my_delete fails
  case 4:
serg@serg.mylan's avatar
serg@serg.mylan committed
2776
    for (i=0; i < npages; i++)
2777 2778 2779 2780 2781 2782 2783 2784 2785
    {
      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:
2786
    my_munmap(data, (size_t)file_length);
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
  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)
  {
2812
    sql_print_error("Recovery failed! You must enable "
2813 2814 2815
                    "exactly %d storage engines that support "
                    "two-phase commit protocol",
                    data[sizeof(tc_log_magic)]);
2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833
    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);
2834
  bzero(data, (size_t)file_length);
2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845
  return 0;

err2:
  hash_free(&xids);
err1:
  sql_print_error("Crash recovery failed. Either correct the problem "
                  "(if it's, for example, out of memory error) and restart, "
                  "or delete tc log and start mysqld with "
                  "--tc-heuristic-recover={commit|rollback}");
  return 1;
}
serg@serg.mylan's avatar
serg@serg.mylan committed
2846 2847 2848 2849 2850
#endif

TC_LOG *tc_log;
TC_LOG_DUMMY tc_log_dummy;
TC_LOG_MMAP  tc_log_mmap;
2851

2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863
/*
  Perform heuristic recovery, if --tc-heuristic-recover was used

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

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

2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876
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 ******/
2877
#define TC_LOG_BINLOG MYSQL_LOG
2878 2879 2880 2881 2882 2883 2884 2885

/*
  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!
*/

2886
int TC_LOG_BINLOG::open(const char *opt_name)
2887
{
2888 2889
  LOG_INFO log_info;
  int      error= 1;
2890 2891

  DBUG_ASSERT(total_ha_2pc > 1);
2892
  DBUG_ASSERT(opt_name && opt_name[0]);
2893 2894 2895 2896

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

2897 2898 2899 2900 2901 2902 2903
  if (!my_b_inited(&index_file))
  {
    /* There was a failure to open the index file, can't open the binlog */
    cleanup();
    return 1;
  }

2904
  if (using_heuristic_recover())
2905 2906 2907 2908
  {
    /* generate a new binlog to mask a corrupted one */
    open(opt_name, LOG_BIN, 0, WRITE_CACHE, 0, max_binlog_size, 0);
    cleanup();
2909
    return 1;
2910
  }
2911

2912
  if ((error= find_log_pos(&log_info, NullS, 1)))
2913
  {
2914 2915 2916 2917 2918
    if (error != LOG_INFO_EOF)
      sql_print_error("find_log_pos() failed (error: %d)", error);
    else
      error= 0;
    goto err;
2919
  }
2920 2921 2922 2923 2924

  {
    const char *errmsg;
    IO_CACHE    log;
    File        file;
2925 2926
    Log_event  *ev=0;
    Format_description_log_event fdle(BINLOG_VERSION);
2927
    char        log_name[FN_REFLEN];
2928 2929 2930 2931

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

2932
    do
2933
    {
2934 2935 2936 2937 2938 2939 2940
      strmake(log_name, log_info.log_file_name, sizeof(log_name)-1);
    } while (!(error= find_next_log(&log_info, 1)));

    if (error !=  LOG_INFO_EOF)
    {
      sql_print_error("find_log_pos() failed (error: %d)", error);
      goto err;
2941
    }
2942

2943
    if ((file= open_binlog(&log, log_name, &errmsg)) < 0)
2944 2945 2946 2947 2948
    {
      sql_print_error("%s", errmsg);
      goto err;
    }

2949 2950 2951
    if ((ev= Log_event::read_log_event(&log, 0, &fdle)) &&
        ev->get_type_code() == FORMAT_DESCRIPTION_EVENT &&
        ev->flags & LOG_EVENT_BINLOG_IN_USE_F)
serg@serg.mylan's avatar
serg@serg.mylan committed
2952 2953
    {
      sql_print_information("Recovering after a crash using %s", opt_name);
2954
      error= recover(&log, (Format_description_log_event *)ev);
serg@serg.mylan's avatar
serg@serg.mylan committed
2955
    }
2956 2957
    else
      error=0;
2958 2959 2960 2961 2962 2963 2964 2965 2966 2967

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

    if (error)
      goto err;
  }

err:
2968
  return error;
2969 2970
}

2971 2972
/* this is called on shutdown, after ha_panic */
void TC_LOG_BINLOG::close()
2973 2974 2975 2976 2977 2978
{
  DBUG_ASSERT(prepared_xids==0);
  pthread_mutex_destroy(&LOCK_prep_xids);
  pthread_cond_destroy (&COND_prep_xids);
}

2979 2980 2981 2982 2983 2984 2985 2986
/*
  TODO group commit

  RETURN
         0  - error
         1  - success
*/
int TC_LOG_BINLOG::log(THD *thd, my_xid xid)
2987 2988
{
  Xid_log_event xle(thd, xid);
2989
  IO_CACHE *trans_log= (IO_CACHE*)thd->ha_data[binlog_hton.slot];
2990
  return !binlog_end_trans(thd, trans_log, &xle);  // invert return value
2991 2992
}

2993
void TC_LOG_BINLOG::unlog(ulong cookie, my_xid xid)
2994
{
2995 2996
  pthread_mutex_lock(&LOCK_prep_xids);
  if (--prepared_xids == 0)
2997
    pthread_cond_signal(&COND_prep_xids);
2998
  pthread_mutex_unlock(&LOCK_prep_xids);
2999
  rotate_and_purge(0);     // as ::write() did not rotate
3000 3001
}

3002
int TC_LOG_BINLOG::recover(IO_CACHE *log, Format_description_log_event *fdle)
3003 3004 3005 3006 3007 3008
{
  Log_event  *ev;
  HASH xids;
  MEM_ROOT mem_root;

  if (! fdle->is_valid() ||
3009
      hash_init(&xids, &my_charset_bin, TC_LOG_PAGE_SIZE/3, 0,
3010 3011 3012
            sizeof(my_xid), 0, 0, MYF(0)))
    goto err1;

3013
  init_alloc_root(&mem_root, TC_LOG_PAGE_SIZE, TC_LOG_PAGE_SIZE);
3014

3015 3016
  fdle->flags&= ~LOG_EVENT_BINLOG_IN_USE_F; // abort on the first error

3017 3018 3019 3020 3021
  while ((ev= Log_event::read_log_event(log,0,fdle)) && ev->is_valid())
  {
    if (ev->get_type_code() == XID_EVENT)
    {
      Xid_log_event *xev=(Xid_log_event *)ev;
serg@serg.mylan's avatar
serg@serg.mylan committed
3022 3023
      byte *x=(byte *)memdup_root(&mem_root, (char *)& xev->xid,
                                  sizeof(xev->xid));
3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048
      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;
}