handler.cc 185 KB
Newer Older
Sergei Golubchik's avatar
Sergei Golubchik committed
1 2
/* Copyright (c) 2000, 2013, Oracle and/or its affiliates.
   Copyright (c) 2009, 2013, Monty Program Ab.
unknown's avatar
unknown committed
3

unknown's avatar
unknown committed
4 5
   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
unknown's avatar
unknown committed
6
   the Free Software Foundation; version 2 of the License.
unknown's avatar
unknown committed
7

unknown's avatar
unknown 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.
unknown's avatar
unknown committed
12

unknown's avatar
unknown committed
13
   You should have received a copy of the GNU General Public License
14 15
   along with this program; if not, write to the Free Software
   Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301  USA */
unknown's avatar
unknown committed
16

17
/** @file handler.cc
unknown's avatar
unknown committed
18

19 20 21
    @brief
  Handler-calling-functions
*/
unknown's avatar
unknown committed
22

23
#include <my_global.h>
24 25
#include "sql_priv.h"
#include "unireg.h"
He Zhenxing's avatar
He Zhenxing committed
26
#include "rpl_handler.h"
27
#include "sql_cache.h"                   // query_cache, query_cache_*
Sergei Golubchik's avatar
Sergei Golubchik committed
28
#include "sql_connect.h"                 // global_table_stats
29 30 31 32 33
#include "key.h"     // key_copy, key_unpack, key_cmp_if_same, key_cmp
#include "sql_table.h"                   // build_table_filename
#include "sql_parse.h"                          // check_stack_overrun
#include "sql_acl.h"            // SUPER_ACL
#include "sql_base.h"           // free_io_cache
34
#include "discover.h"           // extension_based_table_discovery, etc
35
#include "log_event.h"          // *_rows_log_event
36
#include "create_options.h"
37
#include "rpl_filter.h"
38
#include <myisampack.h>
Konstantin Osipov's avatar
Konstantin Osipov committed
39
#include "transaction.h"
40
#include "myisam.h"
41
#include "probes_mysql.h"
Sergei Golubchik's avatar
Sergei Golubchik committed
42
#include <mysql/psi/mysql_table.h>
43
#include "debug_sync.h"         // DEBUG_SYNC
44
#include "sql_audit.h"
45

46 47
#ifdef WITH_PARTITION_STORAGE_ENGINE
#include "ha_partition.h"
48
#endif
49

50 51 52 53
#ifdef WITH_ARIA_STORAGE_ENGINE
#include "../storage/maria/ha_maria.h"
#endif

54 55 56
#include "wsrep_mysqld.h"
#include "wsrep.h"

57 58 59 60 61
/*
  While we have legacy_db_type, we have this array to
  check for dups and to find handlerton from legacy_db_type.
  Remove when legacy_db_type is finally gone
*/
unknown's avatar
unknown committed
62
st_plugin_int *hton2plugin[MAX_HA];
unknown's avatar
unknown committed
63

64
static handlerton *installed_htons[128];
65

unknown's avatar
unknown committed
66 67
#define BITMAP_STACKBUF_SIZE (128/8)

68
KEY_CREATE_INFO default_key_create_info=
69
{ HA_KEY_ALG_UNDEF, 0, {NullS, 0}, {NullS, 0}, true };
70

unknown's avatar
unknown committed
71
/* number of entries in handlertons[] */
unknown's avatar
unknown committed
72
ulong total_ha= 0;
unknown's avatar
unknown committed
73
/* number of storage engines (from handlertons[]) that support 2pc */
unknown's avatar
unknown committed
74
ulong total_ha_2pc= 0;
unknown's avatar
unknown committed
75
/* size of savepoint storage area (see ha_init) */
unknown's avatar
unknown committed
76
ulong savepoint_alloc_size= 0;
77

unknown's avatar
unknown committed
78
static const LEX_STRING sys_table_aliases[]=
79
{
unknown's avatar
unknown committed
80 81
  { C_STRING_WITH_LEN("INNOBASE") },  { C_STRING_WITH_LEN("INNODB") },
  { C_STRING_WITH_LEN("HEAP") },      { C_STRING_WITH_LEN("MEMORY") },
82
  { C_STRING_WITH_LEN("MERGE") },     { C_STRING_WITH_LEN("MRG_MYISAM") },
Sergei Golubchik's avatar
Sergei Golubchik committed
83
  { C_STRING_WITH_LEN("Maria") },      { C_STRING_WITH_LEN("Aria") },
unknown's avatar
unknown committed
84
  {NullS, 0}
unknown's avatar
unknown committed
85
};
86

unknown's avatar
unknown committed
87
const char *ha_row_type[] = {
Sergei Golubchik's avatar
Sergei Golubchik committed
88
  "", "FIXED", "DYNAMIC", "COMPRESSED", "REDUNDANT", "COMPACT", "PAGE"
unknown's avatar
unknown committed
89 90
};

unknown's avatar
unknown committed
91
const char *tx_isolation_names[] =
92 93 94
{ "READ-UNCOMMITTED", "READ-COMMITTED", "REPEATABLE-READ", "SERIALIZABLE",
  NullS};
TYPELIB tx_isolation_typelib= {array_elements(tx_isolation_names)-1,"",
95
			       tx_isolation_names, NULL};
unknown's avatar
unknown committed
96

97
static TYPELIB known_extensions= {0,"known_exts", NULL, NULL};
98
uint known_extensions_id= 0;
99

100 101
static int commit_one_phase_2(THD *thd, bool all, THD_TRANS *trans,
                              bool is_real_trans);
unknown's avatar
unknown committed
102

103

unknown's avatar
unknown committed
104 105 106 107
static plugin_ref ha_default_plugin(THD *thd)
{
  if (thd->variables.table_plugin)
    return thd->variables.table_plugin;
108
  return my_plugin_lock(thd, global_system_variables.table_plugin);
unknown's avatar
unknown committed
109 110
}

111 112 113 114 115 116 117 118
static plugin_ref ha_default_tmp_plugin(THD *thd)
{
  if (thd->variables.tmp_table_plugin)
    return thd->variables.tmp_table_plugin;
  if (global_system_variables.tmp_table_plugin)
    return my_plugin_lock(thd, global_system_variables.tmp_table_plugin);
  return ha_default_plugin(thd);
}
unknown's avatar
unknown committed
119

120

121
/** @brief
unknown's avatar
unknown committed
122
  Return the default storage engine handlerton for thread
123

unknown's avatar
unknown committed
124 125 126
  SYNOPSIS
    ha_default_handlerton(thd)
    thd         current thread
127

unknown's avatar
unknown committed
128 129 130 131
  RETURN
    pointer to handlerton
*/
handlerton *ha_default_handlerton(THD *thd)
132
{
unknown's avatar
unknown committed
133 134
  plugin_ref plugin= ha_default_plugin(thd);
  DBUG_ASSERT(plugin);
Sergei Golubchik's avatar
Sergei Golubchik committed
135
  handlerton *hton= plugin_hton(plugin);
unknown's avatar
unknown committed
136 137
  DBUG_ASSERT(hton);
  return hton;
unknown's avatar
unknown committed
138 139 140
}


141 142 143 144 145 146 147 148 149 150
handlerton *ha_default_tmp_handlerton(THD *thd)
{
  plugin_ref plugin= ha_default_tmp_plugin(thd);
  DBUG_ASSERT(plugin);
  handlerton *hton= plugin_hton(plugin);
  DBUG_ASSERT(hton);
  return hton;
}


151
/** @brief
unknown's avatar
unknown committed
152 153 154 155 156 157 158 159
  Return the storage engine handlerton for the supplied name
  
  SYNOPSIS
    ha_resolve_by_name(thd, name)
    thd         current thread
    name        name of storage engine
  
  RETURN
unknown's avatar
unknown committed
160
    pointer to storage engine plugin handle
unknown's avatar
unknown committed
161
*/
162
plugin_ref ha_resolve_by_name(THD *thd, const LEX_STRING *name, bool tmp_table)
unknown's avatar
unknown committed
163 164
{
  const LEX_STRING *table_alias;
unknown's avatar
unknown committed
165
  plugin_ref plugin;
166

unknown's avatar
unknown committed
167 168 169
redo:
  /* my_strnncoll is a macro and gcc doesn't do early expansion of macro */
  if (thd && !my_charset_latin1.coll->strnncoll(&my_charset_latin1,
unknown's avatar
unknown committed
170
                           (const uchar *)name->str, name->length,
unknown's avatar
unknown committed
171
                           (const uchar *)STRING_WITH_LEN("DEFAULT"), 0))
172
    return tmp_table ?  ha_default_tmp_plugin(thd) : ha_default_plugin(thd);
unknown's avatar
unknown committed
173

unknown's avatar
unknown committed
174
  if ((plugin= my_plugin_lock_by_name(thd, name, MYSQL_STORAGE_ENGINE_PLUGIN)))
175
  {
Sergei Golubchik's avatar
Sergei Golubchik committed
176
    handlerton *hton= plugin_hton(plugin);
177
    if (hton && !(hton->flags & HTON_NOT_USER_SELECTABLE))
unknown's avatar
unknown committed
178 179 180 181 182 183
      return plugin;
      
    /*
      unlocking plugin immediately after locking is relatively low cost.
    */
    plugin_unlock(thd, plugin);
184
  }
unknown's avatar
unknown committed
185

186
  /*
187
    We check for the historical aliases.
188
  */
unknown's avatar
unknown committed
189
  for (table_alias= sys_table_aliases; table_alias->str; table_alias+= 2)
190
  {
191
    if (!my_strnncoll(&my_charset_latin1,
unknown's avatar
unknown committed
192
                      (const uchar *)name->str, name->length,
unknown's avatar
unknown committed
193 194 195 196 197
                      (const uchar *)table_alias->str, table_alias->length))
    {
      name= table_alias + 1;
      goto redo;
    }
198
  }
199

unknown's avatar
unknown committed
200
  return NULL;
201
}
unknown's avatar
unknown committed
202 203


204
plugin_ref ha_lock_engine(THD *thd, const handlerton *hton)
205
{
unknown's avatar
unknown committed
206 207
  if (hton)
  {
208 209
    st_plugin_int *plugin= hton2plugin[hton->slot];
    return my_plugin_lock(thd, plugin_int_to_ref(plugin));
210
  }
unknown's avatar
unknown committed
211
  return NULL;
212 213
}

unknown's avatar
unknown committed
214

unknown's avatar
unknown committed
215 216
handlerton *ha_resolve_by_legacy_type(THD *thd, enum legacy_db_type db_type)
{
unknown's avatar
unknown committed
217
  plugin_ref plugin;
218
  switch (db_type) {
unknown's avatar
unknown committed
219
  case DB_TYPE_DEFAULT:
unknown's avatar
unknown committed
220
    return ha_default_handlerton(thd);
unknown's avatar
unknown committed
221
  default:
unknown's avatar
unknown committed
222 223
    if (db_type > DB_TYPE_UNKNOWN && db_type < DB_TYPE_DEFAULT &&
        (plugin= ha_lock_engine(thd, installed_htons[db_type])))
Sergei Golubchik's avatar
Sergei Golubchik committed
224
      return plugin_hton(plugin);
unknown's avatar
unknown committed
225 226
    /* fall through */
  case DB_TYPE_UNKNOWN:
unknown's avatar
unknown committed
227
    return NULL;
228
  }
229 230 231
}


unknown's avatar
unknown committed
232 233
/**
  Use other database handler if databasehandler is not compiled in.
234
*/
Sergei Golubchik's avatar
Sergei Golubchik committed
235
handlerton *ha_checktype(THD *thd, handlerton *hton, bool no_substitute)
unknown's avatar
unknown committed
236
{
unknown's avatar
unknown committed
237 238 239
  if (ha_storage_engine_is_enabled(hton))
    return hton;

240
  if (no_substitute)
unknown's avatar
unknown committed
241
    return NULL;
He Zhenxing's avatar
He Zhenxing committed
242

unknown's avatar
unknown committed
243
  return ha_default_handlerton(thd);
unknown's avatar
unknown committed
244 245 246
} /* ha_checktype */


unknown's avatar
unknown committed
247
handler *get_new_handler(TABLE_SHARE *share, MEM_ROOT *alloc,
unknown's avatar
unknown committed
248
                         handlerton *db_type)
unknown's avatar
unknown committed
249
{
250 251 252
  handler *file;
  DBUG_ENTER("get_new_handler");
  DBUG_PRINT("enter", ("alloc: 0x%lx", (long) alloc));
253

254
  if (db_type && db_type->state == SHOW_OPTION_YES && db_type->create)
255
  {
256
    if ((file= db_type->create(db_type, share, alloc)))
257 258
      file->init();
    DBUG_RETURN(file);
259
  }
260 261 262 263 264
  /*
    Try the default table type
    Here the call to current_thd() is ok as we call this function a lot of
    times but we enter this branch very seldom.
  */
265 266
  file= get_new_handler(share, alloc, ha_default_handlerton(current_thd));
  DBUG_RETURN(file);
unknown's avatar
unknown committed
267 268
}

269

270
#ifdef WITH_PARTITION_STORAGE_ENGINE
271 272 273 274
handler *get_ha_partition(partition_info *part_info)
{
  ha_partition *partition;
  DBUG_ENTER("get_ha_partition");
275
  if ((partition= new ha_partition(partition_hton, part_info)))
276
  {
277
    if (partition->initialize_partition(current_thd->mem_root))
278 279 280 281
    {
      delete partition;
      partition= 0;
    }
282 283
    else
      partition->init();
284 285 286
  }
  else
  {
287 288
    my_error(ER_OUTOFMEMORY, MYF(ME_FATALERROR), 
             static_cast<int>(sizeof(ha_partition)));
289 290 291 292 293 294
  }
  DBUG_RETURN(((handler*) partition));
}
#endif


295
static const char **handler_errmsgs;
296

297 298
C_MODE_START
static const char **get_handler_errmsgs()
299 300 301
{
  return handler_errmsgs;
}
302
C_MODE_END
303 304


unknown's avatar
unknown committed
305
/**
306 307
  Register handler error messages for use with my_error().

unknown's avatar
unknown committed
308
  @retval
309
    0           OK
unknown's avatar
unknown committed
310 311
  @retval
    !=0         Error
312
*/
313 314

int ha_init_errors(void)
315
{
316
#define SETMSG(nr, msg) handler_errmsgs[(nr) - HA_ERR_FIRST]= (msg)
317 318 319

  /* Allocate a pointer array for the error message strings. */
  /* Zerofill it to avoid uninitialized gaps. */
320 321
  if (! (handler_errmsgs= (const char**) my_malloc(HA_ERR_ERRORS * sizeof(char*),
                                                   MYF(MY_WME | MY_ZEROFILL))))
322 323 324
    return 1;

  /* Set the dedicated error messages. */
325 326
  SETMSG(HA_ERR_KEY_NOT_FOUND,          ER_DEFAULT(ER_KEY_NOT_FOUND));
  SETMSG(HA_ERR_FOUND_DUPP_KEY,         ER_DEFAULT(ER_DUP_KEY));
327 328
  SETMSG(HA_ERR_RECORD_CHANGED,         "Update wich is recoverable");
  SETMSG(HA_ERR_WRONG_INDEX,            "Wrong index given to function");
329 330
  SETMSG(HA_ERR_CRASHED,                ER_DEFAULT(ER_NOT_KEYFILE));
  SETMSG(HA_ERR_WRONG_IN_RECORD,        ER_DEFAULT(ER_CRASHED_ON_USAGE));
331 332 333
  SETMSG(HA_ERR_OUT_OF_MEM,             "Table handler out of memory");
  SETMSG(HA_ERR_NOT_A_TABLE,            "Incorrect file format '%.64s'");
  SETMSG(HA_ERR_WRONG_COMMAND,          "Command not supported");
334
  SETMSG(HA_ERR_OLD_FILE,               ER_DEFAULT(ER_OLD_KEYFILE));
335 336
  SETMSG(HA_ERR_NO_ACTIVE_RECORD,       "No record read in update");
  SETMSG(HA_ERR_RECORD_DELETED,         "Intern record deleted");
337
  SETMSG(HA_ERR_RECORD_FILE_FULL,       ER_DEFAULT(ER_RECORD_FILE_FULL));
338 339
  SETMSG(HA_ERR_INDEX_FILE_FULL,        "No more room in index file '%.64s'");
  SETMSG(HA_ERR_END_OF_FILE,            "End in next/prev/first/last");
340
  SETMSG(HA_ERR_UNSUPPORTED,            ER_DEFAULT(ER_ILLEGAL_HA));
341 342
  SETMSG(HA_ERR_TO_BIG_ROW,             "Too big row");
  SETMSG(HA_WRONG_CREATE_OPTION,        "Wrong create option");
343
  SETMSG(HA_ERR_FOUND_DUPP_UNIQUE,      ER_DEFAULT(ER_DUP_UNIQUE));
344
  SETMSG(HA_ERR_UNKNOWN_CHARSET,        "Can't open charset");
345 346 347 348 349 350 351 352 353 354
  SETMSG(HA_ERR_WRONG_MRG_TABLE_DEF,    ER_DEFAULT(ER_WRONG_MRG_TABLE));
  SETMSG(HA_ERR_CRASHED_ON_REPAIR,      ER_DEFAULT(ER_CRASHED_ON_REPAIR));
  SETMSG(HA_ERR_CRASHED_ON_USAGE,       ER_DEFAULT(ER_CRASHED_ON_USAGE));
  SETMSG(HA_ERR_LOCK_WAIT_TIMEOUT,      ER_DEFAULT(ER_LOCK_WAIT_TIMEOUT));
  SETMSG(HA_ERR_LOCK_TABLE_FULL,        ER_DEFAULT(ER_LOCK_TABLE_FULL));
  SETMSG(HA_ERR_READ_ONLY_TRANSACTION,  ER_DEFAULT(ER_READ_ONLY_TRANSACTION));
  SETMSG(HA_ERR_LOCK_DEADLOCK,          ER_DEFAULT(ER_LOCK_DEADLOCK));
  SETMSG(HA_ERR_CANNOT_ADD_FOREIGN,     ER_DEFAULT(ER_CANNOT_ADD_FOREIGN));
  SETMSG(HA_ERR_NO_REFERENCED_ROW,      ER_DEFAULT(ER_NO_REFERENCED_ROW_2));
  SETMSG(HA_ERR_ROW_IS_REFERENCED,      ER_DEFAULT(ER_ROW_IS_REFERENCED_2));
355 356 357
  SETMSG(HA_ERR_NO_SAVEPOINT,           "No savepoint with that name");
  SETMSG(HA_ERR_NON_UNIQUE_BLOCK_SIZE,  "Non unique key block size");
  SETMSG(HA_ERR_NO_SUCH_TABLE,          "No such table: '%.64s'");
358
  SETMSG(HA_ERR_TABLE_EXIST,            ER_DEFAULT(ER_TABLE_EXISTS_ERROR));
359
  SETMSG(HA_ERR_NO_CONNECTION,          "Could not connect to storage engine");
360
  SETMSG(HA_ERR_TABLE_DEF_CHANGED,      ER_DEFAULT(ER_TABLE_DEF_CHANGED));
361
  SETMSG(HA_ERR_FOREIGN_DUPLICATE_KEY,  "FK constraint would lead to duplicate key");
362 363 364 365 366
  SETMSG(HA_ERR_TABLE_NEEDS_UPGRADE,    ER_DEFAULT(ER_TABLE_NEEDS_UPGRADE));
  SETMSG(HA_ERR_TABLE_READONLY,         ER_DEFAULT(ER_OPEN_AS_READONLY));
  SETMSG(HA_ERR_AUTOINC_READ_FAILED,    ER_DEFAULT(ER_AUTOINC_READ_FAILED));
  SETMSG(HA_ERR_AUTOINC_ERANGE,         ER_DEFAULT(ER_WARN_DATA_OUT_OF_RANGE));
  SETMSG(HA_ERR_TOO_MANY_CONCURRENT_TRXS, ER_DEFAULT(ER_TOO_MANY_CONCURRENT_TRXS));
367
  SETMSG(HA_ERR_INDEX_COL_TOO_LONG,	ER_DEFAULT(ER_INDEX_COLUMN_TOO_LONG));
368
  SETMSG(HA_ERR_INDEX_CORRUPT,		ER_DEFAULT(ER_INDEX_CORRUPT));
Sergei Golubchik's avatar
Sergei Golubchik committed
369
  SETMSG(HA_FTS_INVALID_DOCID,		"Invalid InnoDB FTS Doc ID");
370
  SETMSG(HA_ERR_TABLE_IN_FK_CHECK,	ER_DEFAULT(ER_TABLE_IN_FK_CHECK));
Sergei Golubchik's avatar
Sergei Golubchik committed
371
  SETMSG(HA_ERR_DISK_FULL,              ER_DEFAULT(ER_DISK_FULL));
372
  SETMSG(HA_ERR_FTS_TOO_MANY_WORDS_IN_PHRASE,  "Too many words in a FTS phrase or proximity search");
373 374

  /* Register the error messages for use with my_error(). */
375
  return my_error_register(get_handler_errmsgs, HA_ERR_FIRST, HA_ERR_LAST);
376 377 378
}


unknown's avatar
unknown committed
379
/**
380 381
  Unregister handler error messages.

unknown's avatar
unknown committed
382
  @retval
383
    0           OK
unknown's avatar
unknown committed
384 385
  @retval
    !=0         Error
386 387 388 389 390 391 392 393
*/
static int ha_finish_errors(void)
{
  const char    **errmsgs;

  /* Allocate a pointer array for the error message strings. */
  if (! (errmsgs= my_error_unregister(HA_ERR_FIRST, HA_ERR_LAST)))
    return 1;
394
  my_free(errmsgs);
395 396
  return 0;
}
unknown's avatar
unknown committed
397

398 399
static volatile int32 need_full_discover_for_existence= 0;
static volatile int32 engines_with_discover_table_names= 0;
400
static volatile int32 engines_with_discover= 0;
401

402
static int full_discover_for_existence(handlerton *, const char *, const char *)
403
{ return 0; }
404

405
static int ext_based_existence(handlerton *, const char *, const char *)
406
{ return 0; }
407

408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425
static int hton_ext_based_table_discovery(handlerton *hton, LEX_STRING *db,
                             MY_DIR *dir, handlerton::discovered_list *result)
{
  /*
    tablefile_extensions[0] is the metadata file, see
    the comment above tablefile_extensions declaration
  */
  return extension_based_table_discovery(dir, hton->tablefile_extensions[0],
                                         result);
}

static void update_discovery_counters(handlerton *hton, int val)
{
  if (hton->discover_table_existence == full_discover_for_existence)
    my_atomic_add32(&need_full_discover_for_existence,  val);

  if (hton->discover_table_names)
    my_atomic_add32(&engines_with_discover_table_names, val);
426 427 428

  if (hton->discover_table)
    my_atomic_add32(&engines_with_discover, val);
429
}
430

unknown's avatar
unknown committed
431
int ha_finalize_handlerton(st_plugin_int *plugin)
432
{
unknown's avatar
unknown committed
433
  handlerton *hton= (handlerton *)plugin->data;
unknown's avatar
unknown committed
434 435
  DBUG_ENTER("ha_finalize_handlerton");

436 437 438 439
  /* hton can be NULL here, if ha_initialize_handlerton() failed. */
  if (!hton)
    goto end;

440
  switch (hton->state) {
unknown's avatar
unknown committed
441 442 443 444 445 446 447 448
  case SHOW_OPTION_NO:
  case SHOW_OPTION_DISABLED:
    break;
  case SHOW_OPTION_YES:
    if (installed_htons[hton->db_type] == hton)
      installed_htons[hton->db_type]= NULL;
    break;
  };
449

unknown's avatar
unknown committed
450 451 452
  if (hton->panic)
    hton->panic(hton, HA_PANIC_CLOSE);

453 454 455 456 457 458 459 460 461 462 463 464 465 466
  if (plugin->plugin->deinit)
  {
    /*
      Today we have no defined/special behavior for uninstalling
      engine plugins.
    */
    DBUG_PRINT("info", ("Deinitializing plugin: '%s'", plugin->name.str));
    if (plugin->plugin->deinit(NULL))
    {
      DBUG_PRINT("warning", ("Plugin '%s' deinit function returned error.",
                             plugin->name.str));
    }
  }

467 468 469
  free_sysvar_table_options(hton);
  update_discovery_counters(hton, -1);

470 471 472 473 474
  /*
    In case a plugin is uninstalled and re-installed later, it should
    reuse an array slot. Otherwise the number of uninstall/install
    cycles would be limited.
  */
475 476 477 478 479 480 481
  if (hton->slot != HA_SLOT_UNDEF)
  {
    /* Make sure we are not unpluging another plugin */
    DBUG_ASSERT(hton2plugin[hton->slot] == plugin);
    DBUG_ASSERT(hton->slot < MAX_HA);
    hton2plugin[hton->slot]= NULL;
  }
482

483
  my_free(hton);
484

485
 end:
unknown's avatar
unknown committed
486
  DBUG_RETURN(0);
487
}
488

unknown's avatar
unknown committed
489

unknown's avatar
unknown committed
490
int ha_initialize_handlerton(st_plugin_int *plugin)
unknown's avatar
unknown committed
491
{
492
  handlerton *hton;
493
  static const char *no_exts[]= { 0 };
unknown's avatar
unknown committed
494
  DBUG_ENTER("ha_initialize_handlerton");
495
  DBUG_PRINT("plugin", ("initialize plugin: '%s'", plugin->name.str));
unknown's avatar
unknown committed
496

497 498
  hton= (handlerton *)my_malloc(sizeof(handlerton),
                                MYF(MY_WME | MY_ZEROFILL));
499 500 501 502 503 504 505
  if (hton == NULL)
  {
    sql_print_error("Unable to allocate memory for plugin '%s' handlerton.",
                    plugin->name.str);
    goto err_no_hton_memory;
  }

506 507 508
  hton->tablefile_extensions= no_exts;
  hton->discover_table_names= hton_ext_based_table_discovery;

509
  hton->slot= HA_SLOT_UNDEF;
510
  /* Historical Requirement */
unknown's avatar
unknown committed
511
  plugin->data= hton; // shortcut for the future
512
  if (plugin->plugin->init && plugin->plugin->init(hton))
513
  {
514
    sql_print_error("Plugin '%s' init function returned error.",
515 516
		    plugin->name.str);
    goto err;
517
  }
unknown's avatar
unknown committed
518

519 520 521 522 523
  // hton_ext_based_table_discovery() works only when discovery
  // is supported and the engine if file-based.
  if (hton->discover_table_names == hton_ext_based_table_discovery &&
      (!hton->discover_table || !hton->tablefile_extensions[0]))
    hton->discover_table_names= NULL;
524

525
  // default discover_table_existence implementation
526
  if (!hton->discover_table_existence && hton->discover_table)
527 528 529 530 531 532 533
  {
    if (hton->tablefile_extensions[0])
      hton->discover_table_existence= ext_based_existence;
    else
      hton->discover_table_existence= full_discover_for_existence;
  }

unknown's avatar
unknown committed
534 535 536 537
  /*
    the switch below and hton->state should be removed when
    command-line options for plugins will be implemented
  */
unknown's avatar
unknown committed
538
  switch (hton->state) {
unknown's avatar
unknown committed
539 540 541 542
  case SHOW_OPTION_NO:
    break;
  case SHOW_OPTION_YES:
    {
unknown's avatar
unknown committed
543
      uint tmp;
544
      ulong fslot;
545 546 547 548 549 550

      DBUG_EXECUTE_IF("unstable_db_type", {
                        static int i= (int) DB_TYPE_FIRST_DYNAMIC;
                        hton->db_type= (enum legacy_db_type)++i;
                      });

unknown's avatar
unknown committed
551
      /* now check the db_type for conflict */
unknown's avatar
unknown committed
552
      if (hton->db_type <= DB_TYPE_UNKNOWN ||
unknown's avatar
unknown committed
553 554 555 556
          hton->db_type >= DB_TYPE_DEFAULT ||
          installed_htons[hton->db_type])
      {
        int idx= (int) DB_TYPE_FIRST_DYNAMIC;
unknown's avatar
unknown committed
557

unknown's avatar
unknown committed
558 559 560 561 562 563
        while (idx < (int) DB_TYPE_DEFAULT && installed_htons[idx])
          idx++;

        if (idx == (int) DB_TYPE_DEFAULT)
        {
          sql_print_warning("Too many storage engines!");
564
	  goto err_deinit;
unknown's avatar
unknown committed
565 566 567
        }
        if (hton->db_type != DB_TYPE_UNKNOWN)
          sql_print_warning("Storage engine '%s' has conflicting typecode. "
unknown's avatar
unknown committed
568
                            "Assigning value %d.", plugin->plugin->name, idx);
unknown's avatar
unknown committed
569 570
        hton->db_type= (enum legacy_db_type) idx;
      }
571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590

      /*
        In case a plugin is uninstalled and re-installed later, it should
        reuse an array slot. Otherwise the number of uninstall/install
        cycles would be limited. So look for a free slot.
      */
      DBUG_PRINT("plugin", ("total_ha: %lu", total_ha));
      for (fslot= 0; fslot < total_ha; fslot++)
      {
        if (!hton2plugin[fslot])
          break;
      }
      if (fslot < total_ha)
        hton->slot= fslot;
      else
      {
        if (total_ha >= MAX_HA)
        {
          sql_print_error("Too many plugins loaded. Limit is %lu. "
                          "Failed on '%s'", (ulong) MAX_HA, plugin->name.str);
591
          goto err_deinit;
592 593 594
        }
        hton->slot= total_ha++;
      }
595 596 597 598
      installed_htons[hton->db_type]= hton;
      tmp= hton->savepoint_offset;
      hton->savepoint_offset= savepoint_alloc_size;
      savepoint_alloc_size+= tmp;
unknown's avatar
unknown committed
599
      hton2plugin[hton->slot]=plugin;
unknown's avatar
unknown committed
600
      if (hton->prepare)
601
      {
unknown's avatar
unknown committed
602
        total_ha_2pc++;
603 604 605 606
        if (tc_log && tc_log != get_tc_log_implementation())
        {
          total_ha_2pc--;
          hton->prepare= 0;
Sergei Golubchik's avatar
Sergei Golubchik committed
607
          push_warning_printf(current_thd, Sql_condition::WARN_LEVEL_WARN,
608 609 610 611 612 613
                              ER_UNKNOWN_ERROR,
                              "Cannot enable tc-log at run-time. "
                              "XA features of %s are disabled",
                              plugin->name.str);
        }
      }
unknown's avatar
unknown committed
614 615 616 617 618 619 620
      break;
    }
    /* fall through */
  default:
    hton->state= SHOW_OPTION_DISABLED;
    break;
  }
621 622
  
  /* 
623 624 625
    This is entirely for legacy. We will create a new "disk based" hton and a 
    "memory" hton which will be configurable longterm. We should be able to 
    remove partition and myisammrg.
626 627 628 629 630 631 632 633 634 635 636 637 638 639 640
  */
  switch (hton->db_type) {
  case DB_TYPE_HEAP:
    heap_hton= hton;
    break;
  case DB_TYPE_MYISAM:
    myisam_hton= hton;
    break;
  case DB_TYPE_PARTITION_DB:
    partition_hton= hton;
    break;
  default:
    break;
  };

641
  resolve_sysvar_table_options(hton);
642
  update_discovery_counters(hton, 1);
643

unknown's avatar
unknown committed
644
  DBUG_RETURN(0);
645 646 647 648 649 650 651 652 653

err_deinit:
  /* 
    Let plugin do its inner deinitialization as plugin->init() 
    was successfully called before.
  */
  if (plugin->plugin->deinit)
    (void) plugin->plugin->deinit(NULL);
          
654
err:
655
  my_free(hton);
656
err_no_hton_memory:
657
  plugin->data= NULL;
658
  DBUG_RETURN(1);
unknown's avatar
unknown committed
659 660
}

unknown's avatar
unknown committed
661 662
int ha_init()
{
663
  int error= 0;
unknown's avatar
unknown committed
664 665
  DBUG_ENTER("ha_init");

666
  DBUG_ASSERT(total_ha < MAX_HA);
667 668 669 670 671 672
  /*
    Check if there is a transaction-capable storage engine besides the
    binary log (which is considered a transaction-capable storage engine in
    counting total_ha)
  */
  opt_using_transactions= total_ha>(ulong)opt_bin_log;
673
  savepoint_alloc_size+= sizeof(SAVEPOINT);
unknown's avatar
unknown committed
674
  DBUG_RETURN(error);
unknown's avatar
unknown committed
675 676
}

677
int ha_end()
unknown's avatar
unknown committed
678
{
679 680
  int error= 0;
  DBUG_ENTER("ha_end");
unknown's avatar
unknown committed
681

unknown's avatar
unknown committed
682

683 684 685 686 687 688 689
  /* 
    This should be eventualy based  on the graceful shutdown flag.
    So if flag is equal to HA_PANIC_CLOSE, the deallocate
    the errors.
  */
  if (ha_finish_errors())
    error= 1;
unknown's avatar
unknown committed
690

691 692
  DBUG_RETURN(error);
}
unknown's avatar
unknown committed
693

unknown's avatar
unknown committed
694
static my_bool dropdb_handlerton(THD *unused1, plugin_ref plugin,
unknown's avatar
unknown committed
695 696
                                 void *path)
{
Sergei Golubchik's avatar
Sergei Golubchik committed
697
  handlerton *hton= plugin_hton(plugin);
unknown's avatar
unknown committed
698
  if (hton->state == SHOW_OPTION_YES && hton->drop_database)
699
    hton->drop_database(hton, (char *)path);
unknown's avatar
unknown committed
700 701 702 703
  return FALSE;
}


704 705
void ha_drop_database(char* path)
{
unknown's avatar
unknown committed
706 707
  plugin_foreach(NULL, dropdb_handlerton, MYSQL_STORAGE_ENGINE_PLUGIN, path);
}
708

unknown's avatar
unknown committed
709

710 711 712
static my_bool checkpoint_state_handlerton(THD *unused1, plugin_ref plugin,
                                           void *disable)
{
Sergei Golubchik's avatar
Sergei Golubchik committed
713
  handlerton *hton= plugin_hton(plugin);
714 715 716 717 718 719 720 721 722 723 724 725
  if (hton->state == SHOW_OPTION_YES && hton->checkpoint_state)
    hton->checkpoint_state(hton, (int) *(bool*) disable);
  return FALSE;
}


void ha_checkpoint_state(bool disable)
{
  plugin_foreach(NULL, checkpoint_state_handlerton, MYSQL_STORAGE_ENGINE_PLUGIN, &disable);
}


726 727 728 729 730 731 732 733 734
struct st_commit_checkpoint_request {
  void *cookie;
  void (*pre_hook)(void *);
};

static my_bool commit_checkpoint_request_handlerton(THD *unused1, plugin_ref plugin,
                                           void *data)
{
  st_commit_checkpoint_request *st= (st_commit_checkpoint_request *)data;
Sergei Golubchik's avatar
Sergei Golubchik committed
735
  handlerton *hton= plugin_hton(plugin);
736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762
  if (hton->state == SHOW_OPTION_YES && hton->commit_checkpoint_request)
  {
    void *cookie= st->cookie;
    if (st->pre_hook)
      (*st->pre_hook)(cookie);
    (*hton->commit_checkpoint_request)(hton, cookie);
  }
  return FALSE;
}


/*
  Invoke commit_checkpoint_request() in all storage engines that implement it.

  If pre_hook is non-NULL, the hook will be called prior to each invocation.
*/
void
ha_commit_checkpoint_request(void *cookie, void (*pre_hook)(void *))
{
  st_commit_checkpoint_request st;
  st.cookie= cookie;
  st.pre_hook= pre_hook;
  plugin_foreach(NULL, commit_checkpoint_request_handlerton,
                 MYSQL_STORAGE_ENGINE_PLUGIN, &st);
}


763

unknown's avatar
unknown committed
764
static my_bool closecon_handlerton(THD *thd, plugin_ref plugin,
unknown's avatar
unknown committed
765 766
                                   void *unused)
{
Sergei Golubchik's avatar
Sergei Golubchik committed
767
  handlerton *hton= plugin_hton(plugin);
unknown's avatar
unknown committed
768 769 770 771
  /*
    there's no need to rollback here as all transactions must
    be rolled back already
  */
772 773 774 775 776 777 778
  if (hton->state == SHOW_OPTION_YES && thd_get_ha_data(thd, hton))
  {
    if (hton->close_connection)
      hton->close_connection(hton, thd);
    /* make sure ha_data is reset and ha_data_lock is released */
    thd_set_ha_data(thd, hton, NULL);
  }
unknown's avatar
unknown committed
779
  return FALSE;
780
}
unknown's avatar
unknown committed
781

unknown's avatar
unknown committed
782 783 784
/**
  @note
    don't bother to rollback here, it's done already
785
*/
786 787
void ha_close_connection(THD* thd)
{
unknown's avatar
unknown committed
788
  plugin_foreach(thd, closecon_handlerton, MYSQL_STORAGE_ENGINE_PLUGIN, 0);
789 790
}

791
static my_bool kill_handlerton(THD *thd, plugin_ref plugin,
792
                               void *level)
793
{
Sergei Golubchik's avatar
Sergei Golubchik committed
794
  handlerton *hton= plugin_hton(plugin);
795 796 797

  if (hton->state == SHOW_OPTION_YES && hton->kill_query &&
      thd_get_ha_data(thd, hton))
798
    hton->kill_query(hton, thd, *(enum thd_kill_levels *) level);
799 800 801
  return FALSE;
}

802
void ha_kill_query(THD* thd, enum thd_kill_levels level)
803 804
{
  DBUG_ENTER("ha_kill_query");
805
  plugin_foreach(thd, kill_handlerton, MYSQL_STORAGE_ENGINE_PLUGIN, &level);
806 807 808 809
  DBUG_VOID_RETURN;
}


810 811 812
/* ========================================================================
 ======================= TRANSACTIONS ===================================*/

813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852
/**
  Transaction handling in the server
  ==================================

  In each client connection, MySQL maintains two transactional
  states:
  - a statement transaction,
  - a standard, also called normal transaction.

  Historical note
  ---------------
  "Statement transaction" is a non-standard term that comes
  from the times when MySQL supported BerkeleyDB storage engine.

  First of all, it should be said that in BerkeleyDB auto-commit
  mode auto-commits operations that are atomic to the storage
  engine itself, such as a write of a record, and are too
  high-granular to be atomic from the application perspective
  (MySQL). One SQL statement could involve many BerkeleyDB
  auto-committed operations and thus BerkeleyDB auto-commit was of
  little use to MySQL.

  Secondly, instead of SQL standard savepoints, BerkeleyDB
  provided the concept of "nested transactions". In a nutshell,
  transactions could be arbitrarily nested, but when the parent
  transaction was committed or aborted, all its child (nested)
  transactions were handled committed or aborted as well.
  Commit of a nested transaction, in turn, made its changes
  visible, but not durable: it destroyed the nested transaction,
  all its changes would become available to the parent and
  currently active nested transactions of this parent.

  So the mechanism of nested transactions was employed to
  provide "all or nothing" guarantee of SQL statements
  required by the standard.
  A nested transaction would be created at start of each SQL
  statement, and destroyed (committed or aborted) at statement
  end. Such nested transaction was internally referred to as
  a "statement transaction" and gave birth to the term.

Sergei Golubchik's avatar
Sergei Golubchik committed
853
  (Historical note ends)
854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051

  Since then a statement transaction is started for each statement
  that accesses transactional tables or uses the binary log.  If
  the statement succeeds, the statement transaction is committed.
  If the statement fails, the transaction is rolled back. Commits
  of statement transactions are not durable -- each such
  transaction is nested in the normal transaction, and if the
  normal transaction is rolled back, the effects of all enclosed
  statement transactions are undone as well.  Technically,
  a statement transaction can be viewed as a savepoint which is
  maintained automatically in order to make effects of one
  statement atomic.

  The normal transaction is started by the user and is ended
  usually upon a user request as well. The normal transaction
  encloses transactions of all statements issued between
  its beginning and its end.
  In autocommit mode, the normal transaction is equivalent
  to the statement transaction.

  Since MySQL supports PSEA (pluggable storage engine
  architecture), more than one transactional engine can be
  active at a time. Hence transactions, from the server
  point of view, are always distributed. In particular,
  transactional state is maintained independently for each
  engine. In order to commit a transaction the two phase
  commit protocol is employed.

  Not all statements are executed in context of a transaction.
  Administrative and status information statements do not modify
  engine data, and thus do not start a statement transaction and
  also have no effect on the normal transaction. Examples of such
  statements are SHOW STATUS and RESET SLAVE.

  Similarly DDL statements are not transactional,
  and therefore a transaction is [almost] never started for a DDL
  statement. The difference between a DDL statement and a purely
  administrative statement though is that a DDL statement always
  commits the current transaction before proceeding, if there is
  any.

  At last, SQL statements that work with non-transactional
  engines also have no effect on the transaction state of the
  connection. Even though they are written to the binary log,
  and the binary log is, overall, transactional, the writes
  are done in "write-through" mode, directly to the binlog
  file, followed with a OS cache sync, in other words,
  bypassing the binlog undo log (translog).
  They do not commit the current normal transaction.
  A failure of a statement that uses non-transactional tables
  would cause a rollback of the statement transaction, but
  in case there no non-transactional tables are used,
  no statement transaction is started.

  Data layout
  -----------

  The server stores its transaction-related data in
  thd->transaction. This structure has two members of type
  THD_TRANS. These members correspond to the statement and
  normal transactions respectively:

  - thd->transaction.stmt contains a list of engines
  that are participating in the given statement
  - thd->transaction.all contains a list of engines that
  have participated in any of the statement transactions started
  within the context of the normal transaction.
  Each element of the list contains a pointer to the storage
  engine, engine-specific transactional data, and engine-specific
  transaction flags.

  In autocommit mode thd->transaction.all is empty.
  Instead, data of thd->transaction.stmt is
  used to commit/rollback the normal transaction.

  The list of registered engines has a few important properties:
  - no engine is registered in the list twice
  - engines are present in the list a reverse temporal order --
  new participants are always added to the beginning of the list.

  Transaction life cycle
  ----------------------

  When a new connection is established, thd->transaction
  members are initialized to an empty state.
  If a statement uses any tables, all affected engines
  are registered in the statement engine list. In
  non-autocommit mode, the same engines are registered in
  the normal transaction list.
  At the end of the statement, the server issues a commit
  or a roll back for all engines in the statement list.
  At this point transaction flags of an engine, if any, are
  propagated from the statement list to the list of the normal
  transaction.
  When commit/rollback is finished, the statement list is
  cleared. It will be filled in again by the next statement,
  and emptied again at the next statement's end.

  The normal transaction is committed in a similar way
  (by going over all engines in thd->transaction.all list)
  but at different times:
  - upon COMMIT SQL statement is issued by the user
  - implicitly, by the server, at the beginning of a DDL statement
  or SET AUTOCOMMIT={0|1} statement.

  The normal transaction can be rolled back as well:
  - if the user has requested so, by issuing ROLLBACK SQL
  statement
  - if one of the storage engines requested a rollback
  by setting thd->transaction_rollback_request. This may
  happen in case, e.g., when the transaction in the engine was
  chosen a victim of the internal deadlock resolution algorithm
  and rolled back internally. When such a situation happens, there
  is little the server can do and the only option is to rollback
  transactions in all other participating engines.  In this case
  the rollback is accompanied by an error sent to the user.

  As follows from the use cases above, the normal transaction
  is never committed when there is an outstanding statement
  transaction. In most cases there is no conflict, since
  commits of the normal transaction are issued by a stand-alone
  administrative or DDL statement, thus no outstanding statement
  transaction of the previous statement exists. Besides,
  all statements that manipulate with the normal transaction
  are prohibited in stored functions and triggers, therefore
  no conflicting situation can occur in a sub-statement either.
  The remaining rare cases when the server explicitly has
  to commit the statement transaction prior to committing the normal
  one cover error-handling scenarios (see for example
  SQLCOM_LOCK_TABLES).

  When committing a statement or a normal transaction, the server
  either uses the two-phase commit protocol, or issues a commit
  in each engine independently. The two-phase commit protocol
  is used only if:
  - all participating engines support two-phase commit (provide
    handlerton::prepare PSEA API call) and
  - transactions in at least two engines modify data (i.e. are
  not read-only).

  Note that the two phase commit is used for
  statement transactions, even though they are not durable anyway.
  This is done to ensure logical consistency of data in a multiple-
  engine transaction.
  For example, imagine that some day MySQL supports unique
  constraint checks deferred till the end of statement. In such
  case a commit in one of the engines may yield ER_DUP_KEY,
  and MySQL should be able to gracefully abort statement
  transactions of other participants.

  After the normal transaction has been committed,
  thd->transaction.all list is cleared.

  When a connection is closed, the current normal transaction, if
  any, is rolled back.

  Roles and responsibilities
  --------------------------

  The server has no way to know that an engine participates in
  the statement and a transaction has been started
  in it unless the engine says so. Thus, in order to be
  a part of a transaction, the engine must "register" itself.
  This is done by invoking trans_register_ha() server call.
  Normally the engine registers itself whenever handler::external_lock()
  is called. trans_register_ha() can be invoked many times: if
  an engine is already registered, the call does nothing.
  In case autocommit is not set, the engine must register itself
  twice -- both in the statement list and in the normal transaction
  list.
  In which list to register is a parameter of trans_register_ha().

  Note, that although the registration interface in itself is
  fairly clear, the current usage practice often leads to undesired
  effects. E.g. since a call to trans_register_ha() in most engines
  is embedded into implementation of handler::external_lock(), some
  DDL statements start a transaction (at least from the server
  point of view) even though they are not expected to. E.g.
  CREATE TABLE does not start a transaction, since
  handler::external_lock() is never called during CREATE TABLE. But
  CREATE TABLE ... SELECT does, since handler::external_lock() is
  called for the table that is being selected from. This has no
  practical effects currently, but must be kept in mind
  nevertheless.

  Once an engine is registered, the server will do the rest
  of the work.

  During statement execution, whenever any of data-modifying
  PSEA API methods is used, e.g. handler::write_row() or
  handler::update_row(), the read-write flag is raised in the
  statement transaction for the involved engine.
  Currently All PSEA calls are "traced", and the data can not be
  changed in a way other than issuing a PSEA call. Important:
  unless this invariant is preserved the server will not know that
  a transaction in a given engine is read-write and will not
  involve the two-phase commit protocol!

Konstantin Osipov's avatar
Konstantin Osipov committed
1052 1053 1054 1055 1056 1057 1058 1059 1060 1061
  At the end of a statement, server call trans_commit_stmt is
  invoked. This call in turn invokes handlerton::prepare()
  for every involved engine. Prepare is followed by a call
  to handlerton::commit_one_phase() If a one-phase commit
  will suffice, handlerton::prepare() is not invoked and
  the server only calls handlerton::commit_one_phase().
  At statement commit, the statement-related read-write
  engine flag is propagated to the corresponding flag in the
  normal transaction.  When the commit is complete, the list
  of registered engines is cleared.
1062 1063 1064 1065 1066 1067 1068 1069 1070 1071

  Rollback is handled in a similar fashion.

  Additional notes on DDL and the normal transaction.
  ---------------------------------------------------

  DDLs and operations with non-transactional engines
  do not "register" in thd->transaction lists, and thus do not
  modify the transaction state. Besides, each DDL in
  MySQL is prefixed with an implicit normal transaction commit
Konstantin Osipov's avatar
Konstantin Osipov committed
1072
  (a call to trans_commit_implicit()), and thus leaves nothing
1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101
  to modify.
  However, as it has been pointed out with CREATE TABLE .. SELECT,
  some DDL statements can start a *new* transaction.

  Behaviour of the server in this case is currently badly
  defined.
  DDL statements use a form of "semantic" logging
  to maintain atomicity: if CREATE TABLE .. SELECT failed,
  the newly created table is deleted.
  In addition, some DDL statements issue interim transaction
  commits: e.g. ALTER TABLE issues a commit after data is copied
  from the original table to the internal temporary table. Other
  statements, e.g. CREATE TABLE ... SELECT do not always commit
  after itself.
  And finally there is a group of DDL statements such as
  RENAME/DROP TABLE that doesn't start a new transaction
  and doesn't commit.

  This diversity makes it hard to say what will happen if
  by chance a stored function is invoked during a DDL --
  whether any modifications it makes will be committed or not
  is not clear. Fortunately, SQL grammar of few DDLs allows
  invocation of a stored function.

  A consistent behaviour is perhaps to always commit the normal
  transaction after all DDLs, just like the statement transaction
  is always committed at the end of all statements.
*/

unknown's avatar
unknown committed
1102 1103
/**
  Register a storage engine for a transaction.
unknown's avatar
unknown committed
1104

unknown's avatar
unknown committed
1105 1106 1107 1108 1109
  Every storage engine MUST call this function when it starts
  a transaction or a statement (that is it must be called both for the
  "beginning of transaction" and "beginning of statement").
  Only storage engines registered for the transaction/statement
  will know when to commit/rollback it.
1110

unknown's avatar
unknown committed
1111
  @note
1112 1113 1114
    trans_register_ha is idempotent - storage engine may register many
    times per transaction.

unknown's avatar
unknown committed
1115
*/
1116 1117 1118
void trans_register_ha(THD *thd, bool all, handlerton *ht_arg)
{
  THD_TRANS *trans;
1119
  Ha_trx_info *ha_info;
unknown's avatar
unknown committed
1120 1121 1122
  DBUG_ENTER("trans_register_ha");
  DBUG_PRINT("enter",("%s", all ? "all" : "stmt"));

1123 1124 1125 1126
  if (all)
  {
    trans= &thd->transaction.all;
    thd->server_status|= SERVER_STATUS_IN_TRANS;
1127 1128 1129
    if (thd->tx_read_only)
      thd->server_status|= SERVER_STATUS_IN_TRANS_READONLY;
    DBUG_PRINT("info", ("setting SERVER_STATUS_IN_TRANS"));
1130 1131 1132 1133
  }
  else
    trans= &thd->transaction.stmt;

1134
  ha_info= thd->ha_data[ht_arg->slot].ha_info + (all ? 1 : 0);
1135 1136 1137 1138 1139

  if (ha_info->is_started())
    DBUG_VOID_RETURN; /* already registered, return */

  ha_info->register_ha(trans, ht_arg);
unknown's avatar
unknown committed
1140

1141
  trans->no_2pc|=(ht_arg->prepare==0);
1142 1143
  if (thd->transaction.xid_state.xid.is_null())
    thd->transaction.xid_state.xid.set(thd->query_id);
unknown's avatar
unknown committed
1144
  DBUG_VOID_RETURN;
1145 1146
}

1147 1148 1149 1150 1151 1152 1153 1154 1155

static int prepare_or_error(handlerton *ht, THD *thd, bool all)
{
  int err= ht->prepare(ht, thd, all);
  status_var_increment(thd->status_var.ha_prepare_count);
  if (err)
  {
    /* avoid sending error, if we're going to replay the transaction */
#ifdef WITH_WSREP
1156 1157
    if (ht != wsrep_hton ||
        err == EMSGSIZE || thd->wsrep_conflict_state != MUST_REPLAY)
1158 1159 1160 1161 1162 1163 1164
#endif
      my_error(ER_ERROR_DURING_COMMIT, MYF(0), err);
  }
  return err;
}


unknown's avatar
unknown committed
1165 1166 1167 1168 1169
/**
  @retval
    0   ok
  @retval
    1   error, transaction was rolled back
1170 1171 1172 1173 1174
*/
int ha_prepare(THD *thd)
{
  int error=0, all=1;
  THD_TRANS *trans=all ? &thd->transaction.all : &thd->transaction.stmt;
1175
  Ha_trx_info *ha_info= trans->ha_list;
1176
  DBUG_ENTER("ha_prepare");
Konstantin Osipov's avatar
Konstantin Osipov committed
1177

1178
  if (ha_info)
1179
  {
1180
    for (; ha_info; ha_info= ha_info->next())
1181
    {
1182 1183
      handlerton *ht= ha_info->ht();
      if (ht->prepare)
1184
      {
1185
        if (prepare_or_error(ht, thd, all))
1186
        {
1187 1188 1189
          ha_rollback_trans(thd, all);
          error=1;
          break;
1190 1191 1192 1193
        }
      }
      else
      {
1194
        push_warning_printf(thd, Sql_condition::WARN_LEVEL_WARN,
1195 1196
                            ER_GET_ERRNO, ER(ER_GET_ERRNO),
                            HA_ERR_WRONG_COMMAND,
1197
                            ha_resolve_storage_engine_name(ht));
Sergey Petrunya's avatar
Sergey Petrunya committed
1198

1199 1200 1201
      }
    }
  }
Konstantin Osipov's avatar
Konstantin Osipov committed
1202

1203 1204 1205
  DBUG_RETURN(error);
}

1206 1207 1208 1209 1210 1211
/**
  Check if we can skip the two-phase commit.

  A helper function to evaluate if two-phase commit is mandatory.
  As a side effect, propagates the read-only/read-write flags
  of the statement transaction to its enclosing normal transaction.
1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222
  
  If we have at least two engines with read-write changes we must
  run a two-phase commit. Otherwise we can run several independent
  commits as the only transactional engine has read-write changes
  and others are read-only.

  @retval   0   All engines are read-only.
  @retval   1   We have the only engine with read-write changes.
  @retval   >1  More than one engine have read-write changes.
                Note: return value might NOT be the exact number of
                engines with read-write changes.
1223 1224 1225
*/

static
1226
uint
1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262
ha_check_and_coalesce_trx_read_only(THD *thd, Ha_trx_info *ha_list,
                                    bool all)
{
  /* The number of storage engines that have actual changes. */
  unsigned rw_ha_count= 0;
  Ha_trx_info *ha_info;

  for (ha_info= ha_list; ha_info; ha_info= ha_info->next())
  {
    if (ha_info->is_trx_read_write())
      ++rw_ha_count;

    if (! all)
    {
      Ha_trx_info *ha_info_all= &thd->ha_data[ha_info->ht()->slot].ha_info[1];
      DBUG_ASSERT(ha_info != ha_info_all);
      /*
        Merge read-only/read-write information about statement
        transaction to its enclosing normal transaction. Do this
        only if in a real transaction -- that is, if we know
        that ha_info_all is registered in thd->transaction.all.
        Since otherwise we only clutter the normal transaction flags.
      */
      if (ha_info_all->is_started()) /* FALSE if autocommit. */
        ha_info_all->coalesce_trx_with(ha_info);
    }
    else if (rw_ha_count > 1)
    {
      /*
        It is a normal transaction, so we don't need to merge read/write
        information up, and the need for two-phase commit has been
        already established. Break the loop prematurely.
      */
      break;
    }
  }
1263
  return rw_ha_count;
1264 1265 1266
}


unknown's avatar
unknown committed
1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279
/**
  @retval
    0   ok
  @retval
    1   transaction was rolled back
  @retval
    2   error during commit, data may be inconsistent

  @todo
    Since we don't support nested statement transactions in 5.0,
    we can't commit or rollback stmt transactions while we are inside
    stored functions or triggers. So we simply do nothing now.
    TODO: This should be fixed in later ( >= 5.1) releases.
1280 1281 1282
*/
int ha_commit_trans(THD *thd, bool all)
{
1283
  int error= 0, cookie;
1284 1285 1286 1287
  /*
    'all' means that this is either an explicit commit issued by
    user, or an implicit commit issued by a DDL.
  */
1288
  THD_TRANS *trans= all ? &thd->transaction.all : &thd->transaction.stmt;
1289 1290 1291 1292 1293 1294 1295
  /*
    "real" is a nick name for a transaction for which a commit will
    make persistent changes. E.g. a 'stmt' transaction inside a 'all'
    transation is not 'real': even though it's possible to commit it,
    the changes are not durable as they might be rolled back if the
    enclosing 'all' transaction is rolled back.
  */
1296 1297
  bool is_real_trans= ((all || thd->transaction.all.ha_list == 0) &&
                       !(thd->variables.option_bits & OPTION_GTID_BEGIN));
1298
  Ha_trx_info *ha_info= trans->ha_list;
1299 1300
  bool need_prepare_ordered, need_commit_ordered;
  my_xid xid;
1301
  DBUG_ENTER("ha_commit_trans");
Michael Widenius's avatar
Michael Widenius committed
1302 1303
  DBUG_PRINT("info",("thd: %p  option_bits: %lu  all: %d",
                     thd, (ulong) thd->variables.option_bits, all));
1304

1305 1306
  /* Just a random warning to test warnings pushed during autocommit. */
  DBUG_EXECUTE_IF("warn_during_ha_commit_trans",
1307
    push_warning(thd, Sql_condition::WARN_LEVEL_WARN,
1308 1309 1310
                 ER_WARNING_NOT_COMPLETE_ROLLBACK,
                 ER(ER_WARNING_NOT_COMPLETE_ROLLBACK)););

1311
  DBUG_PRINT("info",
1312
             ("all: %d  thd->in_sub_stmt: %d  ha_info: %p  is_real_trans: %d",
1313
              all, thd->in_sub_stmt, ha_info, is_real_trans));
1314 1315 1316 1317 1318 1319 1320 1321 1322
  /*
    We must not commit the normal transaction if a statement
    transaction is pending. Otherwise statement transaction
    flags will not get propagated to its normal transaction's
    counterpart.
  */
  DBUG_ASSERT(thd->transaction.stmt.ha_list == NULL ||
              trans == &thd->transaction.stmt);

1323
  if (thd->in_sub_stmt)
1324
  {
1325
    DBUG_ASSERT(0);
1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342
    /*
      Since we don't support nested statement transactions in 5.0,
      we can't commit or rollback stmt transactions while we are inside
      stored functions or triggers. So we simply do nothing now.
      TODO: This should be fixed in later ( >= 5.1) releases.
    */
    if (!all)
      DBUG_RETURN(0);
    /*
      We assume that all statements which commit or rollback main transaction
      are prohibited inside of stored functions or triggers. So they should
      bail out with error even before ha_commit_trans() call. To be 100% safe
      let us throw error in non-debug builds.
    */
    my_error(ER_COMMIT_NOT_ALLOWED_IN_SF_OR_TRG, MYF(0));
    DBUG_RETURN(2);
  }
Konstantin Osipov's avatar
Konstantin Osipov committed
1343

1344
#ifdef WITH_ARIA_STORAGE_ENGINE
1345
    ha_maria::implicit_commit(thd, TRUE);
1346 1347
#endif

1348
  if (!ha_info)
1349
  {
1350 1351 1352
    /*
      Free resources and perform other cleanup even for 'empty' transactions.
    */
1353 1354 1355 1356
    if (is_real_trans)
      thd->transaction.cleanup();
    DBUG_RETURN(0);
  }
1357

1358
  DBUG_EXECUTE_IF("crash_commit_before", DBUG_SUICIDE(););
1359

1360 1361 1362
  /* Close all cursors that can not survive COMMIT */
  if (is_real_trans)                          /* not a statement commit */
    thd->stmt_map.close_transient_cursors();
1363

1364 1365 1366
  uint rw_ha_count= ha_check_and_coalesce_trx_read_only(thd, ha_info, all);
  /* rw_trans is TRUE when we in a transaction changing data */
  bool rw_trans= is_real_trans && (rw_ha_count > 0);
Sergei Golubchik's avatar
Sergei Golubchik committed
1367
  MDL_request mdl_request;
Michael Widenius's avatar
Michael Widenius committed
1368 1369
  DBUG_PRINT("info", ("is_real_trans: %d  rw_trans:  %d  rw_ha_count: %d",
                      is_real_trans, rw_trans, rw_ha_count));
1370

Sergei Golubchik's avatar
Sergei Golubchik committed
1371
  if (rw_trans)
1372
  {
Sergei Golubchik's avatar
Sergei Golubchik committed
1373 1374 1375 1376
    /*
      Acquire a metadata lock which will ensure that COMMIT is blocked
      by an active FLUSH TABLES WITH READ LOCK (and vice versa:
      COMMIT in progress blocks FTWRL).
1377

Sergei Golubchik's avatar
Sergei Golubchik committed
1378 1379 1380 1381 1382
      We allow the owner of FTWRL to COMMIT; we assume that it knows
      what it does.
    */
    mdl_request.init(MDL_key::COMMIT, "", "", MDL_INTENTION_EXCLUSIVE,
                     MDL_EXPLICIT);
unknown's avatar
unknown committed
1383

1384
    if (!WSREP(thd) &&
1385 1386
      thd->mdl_context.acquire_lock(&mdl_request,
                                    thd->variables.lock_wait_timeout))
unknown's avatar
unknown committed
1387 1388
    {
      ha_rollback_trans(thd, all);
Sergei Golubchik's avatar
Sergei Golubchik committed
1389
      DBUG_RETURN(1);
unknown's avatar
unknown committed
1390 1391
    }

Sergei Golubchik's avatar
Sergei Golubchik committed
1392
    DEBUG_SYNC(thd, "ha_commit_trans_after_acquire_commit_lock");
1393
  }
unknown's avatar
unknown committed
1394

1395 1396 1397 1398 1399 1400 1401 1402
  if (rw_trans &&
      opt_readonly &&
      !(thd->security_ctx->master_access & SUPER_ACL) &&
      !thd->slave_thread)
  {
    my_error(ER_OPTION_PREVENTS_STATEMENT, MYF(0), "--read-only");
    goto err;
  }
unknown's avatar
unknown committed
1403

1404 1405 1406
  if (trans->no_2pc || (rw_ha_count <= 1))
  {
    error= ha_commit_one_phase(thd, all);
Sergei Golubchik's avatar
Sergei Golubchik committed
1407
    goto done;
1408
  }
1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427

  need_prepare_ordered= FALSE;
  need_commit_ordered= FALSE;
  xid= thd->transaction.xid_state.xid.get_my_xid();

  for (Ha_trx_info *hi= ha_info; hi; hi= hi->next())
  {
    handlerton *ht= hi->ht();
    /*
      Do not call two-phase commit if this particular
      transaction is read-only. This allows for simpler
      implementation in engines that are always read-only.
    */
    if (! hi->is_trx_read_write())
      continue;
    /*
      Sic: we know that prepare() is not NULL since otherwise
      trans->no_2pc would have been set.
    */
1428
    if (prepare_or_error(ht, thd, all))
1429
      goto err;
1430

unknown's avatar
unknown committed
1431 1432
    need_prepare_ordered|= (ht->prepare_ordered != NULL);
    need_commit_ordered|= (ht->commit_ordered != NULL);
1433
  }
1434
  DEBUG_SYNC(thd, "ha_commit_trans_after_prepare");
1435
  DBUG_EXECUTE_IF("crash_commit_after_prepare", DBUG_SUICIDE(););
1436

1437 1438 1439 1440 1441 1442
  if (!error && WSREP_ON && wsrep_is_wsrep_xid(&thd->transaction.xid_state.xid))
  {
    // xid was rewritten by wsrep
    xid= wsrep_xid_seqno(&thd->transaction.xid_state.xid);
  }

1443 1444 1445
  if (!is_real_trans)
  {
    error= commit_one_phase_2(thd, all, trans, is_real_trans);
Sergei Golubchik's avatar
Sergei Golubchik committed
1446
    goto done;
1447 1448
  }

1449
  DEBUG_SYNC(thd, "ha_commit_trans_before_log_and_order");
1450 1451 1452 1453 1454
  cookie= tc_log->log_and_order(thd, xid, all, need_prepare_ordered,
                                need_commit_ordered);
  if (!cookie)
    goto err;

1455
  DEBUG_SYNC(thd, "ha_commit_trans_after_log_and_order");
1456
  DBUG_EXECUTE_IF("crash_commit_after_log", DBUG_SUICIDE(););
1457 1458 1459

  error= commit_one_phase_2(thd, all, trans, is_real_trans) ? 2 : 0;

1460
  DBUG_EXECUTE_IF("crash_commit_before_unlog", DBUG_SUICIDE(););
1461 1462 1463 1464 1465
  if (tc_log->unlog(cookie, xid))
  {
    error= 2;                                /* Error during commit */
    goto end;
  }
1466

Sergei Golubchik's avatar
Sergei Golubchik committed
1467
done:
1468
  DBUG_EXECUTE_IF("crash_commit_after", DBUG_SUICIDE(););
1469 1470

  mysql_mutex_assert_not_owner(&LOCK_prepare_ordered);
1471
  mysql_mutex_assert_not_owner(mysql_bin_log.get_log_lock());
1472 1473
  mysql_mutex_assert_not_owner(&LOCK_after_binlog_sync);
  mysql_mutex_assert_not_owner(&LOCK_commit_ordered);
Sergei Golubchik's avatar
Sergei Golubchik committed
1474
  RUN_HOOK(transaction, after_commit, (thd, FALSE));
1475 1476 1477 1478
  goto end;

  /* Come here if error and we need to rollback. */
err:
unknown's avatar
unknown committed
1479
  error= 1;                                  /* Transaction was rolled back */
1480 1481 1482 1483 1484 1485 1486
  /*
    In parallel replication, rollback is delayed, as there is extra replication
    book-keeping to be done before rolling back and allowing a conflicting
    transaction to continue (MDEV-7458).
  */
  if (!(thd->rgi_slave && thd->rgi_slave->is_parallel_exec))
    ha_rollback_trans(thd, all);
1487

1488
end:
Sergei Golubchik's avatar
Sergei Golubchik committed
1489 1490 1491 1492 1493 1494 1495 1496 1497
  if (rw_trans && mdl_request.ticket)
  {
    /*
      We do not always immediately release transactional locks
      after ha_commit_trans() (see uses of ha_enable_transaction()),
      thus we release the commit blocker lock as soon as it's
      not needed.
    */
    thd->mdl_context.release_lock(mdl_request.ticket);
1498 1499 1500 1501
  }
  DBUG_RETURN(error);
}

unknown's avatar
unknown committed
1502 1503 1504
/**
  @note
  This function does not care about global read lock. A caller should.
1505 1506 1507 1508 1509 1510

  @param[in]  all  Is set in case of explicit commit
                   (COMMIT statement), or implicit commit
                   issued by DDL. Is not set when called
                   at the end of statement, even if
                   autocommit=1.
1511
*/
1512

1513 1514 1515
int ha_commit_one_phase(THD *thd, bool all)
{
  THD_TRANS *trans=all ? &thd->transaction.all : &thd->transaction.stmt;
1516 1517 1518
  /*
    "real" is a nick name for a transaction for which a commit will
    make persistent changes. E.g. a 'stmt' transaction inside a 'all'
1519
    transaction is not 'real': even though it's possible to commit it,
1520 1521
    the changes are not durable as they might be rolled back if the
    enclosing 'all' transaction is rolled back.
1522 1523 1524 1525 1526 1527
    We establish the value of 'is_real_trans' by checking
    if it's an explicit COMMIT/BEGIN statement, or implicit
    commit issued by DDL (all == TRUE), or if we're running
    in autocommit mode (it's only in the autocommit mode
    ha_commit_one_phase() can be called with an empty
    transaction.all.ha_list, see why in trans_register_ha()).
1528
  */
1529 1530
  bool is_real_trans= ((all || thd->transaction.all.ha_list == 0) &&
                       !(thd->variables.option_bits & OPTION_GTID_BEGIN));
1531
  int res;
1532
  DBUG_ENTER("ha_commit_one_phase");
1533 1534 1535 1536 1537 1538
  if (is_real_trans)
  {
    DEBUG_SYNC(thd, "ha_commit_one_phase");
    if ((res= thd->wait_for_prior_commit()))
      DBUG_RETURN(res);
  }
1539
  res= commit_one_phase_2(thd, all, trans, is_real_trans);
Sergei Golubchik's avatar
Sergei Golubchik committed
1540
  DBUG_RETURN(res);
1541
}
Konstantin Osipov's avatar
Konstantin Osipov committed
1542

Sergei Golubchik's avatar
Sergei Golubchik committed
1543

1544 1545 1546 1547 1548 1549
static int
commit_one_phase_2(THD *thd, bool all, THD_TRANS *trans, bool is_real_trans)
{
  int error= 0;
  Ha_trx_info *ha_info= trans->ha_list, *ha_info_next;
  DBUG_ENTER("commit_one_phase_2");
1550 1551
  if (is_real_trans)
    DEBUG_SYNC(thd, "commit_one_phase_2");
1552
  if (ha_info)
1553
  {
1554
    for (; ha_info; ha_info= ha_info_next)
1555 1556
    {
      int err;
1557 1558
      handlerton *ht= ha_info->ht();
      if ((err= ht->commit(ht, thd, all)))
1559 1560 1561 1562
      {
        my_error(ER_ERROR_DURING_COMMIT, MYF(0), err);
        error=1;
      }
1563
      /* Should this be done only if is_real_trans is set ? */
1564
      status_var_increment(thd->status_var.ha_commit_count);
1565 1566
      ha_info_next= ha_info->next();
      ha_info->reset(); /* keep it conveniently zero-filled */
1567
    }
1568
    trans->ha_list= 0;
1569 1570 1571 1572 1573
    trans->no_2pc=0;
    if (all)
    {
#ifdef HAVE_QUERY_CACHE
      if (thd->transaction.changed_tables)
1574
        query_cache.invalidate(thd, thd->transaction.changed_tables);
unknown's avatar
unknown committed
1575
#endif
1576 1577
    }
  }
1578 1579 1580
  /* Free resources and perform other cleanup even for 'empty' transactions. */
  if (is_real_trans)
    thd->transaction.cleanup();
Konstantin Osipov's avatar
Konstantin Osipov committed
1581

1582 1583 1584 1585 1586 1587 1588 1589
  DBUG_RETURN(error);
}


int ha_rollback_trans(THD *thd, bool all)
{
  int error=0;
  THD_TRANS *trans=all ? &thd->transaction.all : &thd->transaction.stmt;
1590
  Ha_trx_info *ha_info= trans->ha_list, *ha_info_next;
1591 1592 1593
  /*
    "real" is a nick name for a transaction for which a commit will
    make persistent changes. E.g. a 'stmt' transaction inside a 'all'
1594
    transaction is not 'real': even though it's possible to commit it,
1595 1596
    the changes are not durable as they might be rolled back if the
    enclosing 'all' transaction is rolled back.
1597 1598 1599 1600 1601 1602
    We establish the value of 'is_real_trans' by checking
    if it's an explicit COMMIT or BEGIN statement, or implicit
    commit issued by DDL (in these cases all == TRUE),
    or if we're running in autocommit mode (it's only in the autocommit mode
    ha_commit_one_phase() is called with an empty
    transaction.all.ha_list, see why in trans_register_ha()).
1603
  */
1604
  bool is_real_trans=all || thd->transaction.all.ha_list == 0;
1605
  DBUG_ENTER("ha_rollback_trans");
1606 1607 1608 1609 1610 1611 1612 1613

  /*
    We must not rollback the normal transaction if a statement
    transaction is pending.
  */
  DBUG_ASSERT(thd->transaction.stmt.ha_list == NULL ||
              trans == &thd->transaction.stmt);

1614
  if (thd->in_sub_stmt)
1615
  {
1616
    DBUG_ASSERT(0);
1617 1618 1619 1620 1621 1622 1623 1624 1625 1626
    /*
      If we are inside stored function or trigger we should not commit or
      rollback current statement transaction. See comment in ha_commit_trans()
      call for more information.
    */
    if (!all)
      DBUG_RETURN(0);
    my_error(ER_COMMIT_NOT_ALLOWED_IN_SF_OR_TRG, MYF(0));
    DBUG_RETURN(1);
  }
Konstantin Osipov's avatar
Konstantin Osipov committed
1627

1628
  if (ha_info)
1629
  {
1630 1631 1632 1633
    /* Close all cursors that can not survive ROLLBACK */
    if (is_real_trans)                          /* not a statement commit */
      thd->stmt_map.close_transient_cursors();

1634
    for (; ha_info; ha_info= ha_info_next)
1635 1636
    {
      int err;
1637 1638
      handlerton *ht= ha_info->ht();
      if ((err= ht->rollback(ht, thd, all)))
1639 1640 1641 1642
      { // cannot happen
        my_error(ER_ERROR_DURING_ROLLBACK, MYF(0), err);
        error=1;
      }
1643
      status_var_increment(thd->status_var.ha_rollback_count);
1644 1645
      ha_info_next= ha_info->next();
      ha_info->reset(); /* keep it conveniently zero-filled */
1646
    }
1647
    trans->ha_list= 0;
1648 1649
    trans->no_2pc=0;
  }
1650 1651 1652 1653 1654 1655 1656

  /*
    Thanks to possibility of MDL deadlock rollback request can come even if
    transaction hasn't been started in any transactional storage engine.
  */
  if (is_real_trans && thd->transaction_rollback_request &&
      thd->transaction.xid_state.xa_state != XA_NOTR)
Sergei Golubchik's avatar
Sergei Golubchik committed
1657
    thd->transaction.xid_state.rm_error= thd->get_stmt_da()->sql_errno();
1658

1659
  /* Always cleanup. Even if nht==0. There may be savepoints. */
1660
  if (is_real_trans)
1661
    thd->transaction.cleanup();
1662 1663 1664
  if (all)
    thd->transaction_rollback_request= FALSE;

1665 1666 1667 1668 1669 1670 1671 1672
  /*
    If a non-transactional table was updated, warn; don't warn if this is a
    slave thread (because when a slave thread executes a ROLLBACK, it has
    been read from the binary log, so it's 100% sure and normal to produce
    error ER_WARNING_NOT_COMPLETE_ROLLBACK. If we sent the warning to the
    slave SQL thread, it would not stop the thread but just be printed in
    the error log; but we don't want users to wonder why they have this
    message in the error log, so we don't send it.
1673

1674
    We don't have to test for thd->killed == KILL_SYSTEM_THREAD as
1675 1676
    it doesn't matter if a warning is pushed to a system thread or not:
    No one will see it...
1677
  */
unknown's avatar
unknown committed
1678
  if (is_real_trans && thd->transaction.all.modified_non_trans_table &&
1679
      !thd->slave_thread && thd->killed < KILL_CONNECTION)
1680
    push_warning(thd, Sql_condition::WARN_LEVEL_WARN,
1681 1682
                 ER_WARNING_NOT_COMPLETE_ROLLBACK,
                 ER(ER_WARNING_NOT_COMPLETE_ROLLBACK));
1683
  (void) RUN_HOOK(transaction, after_rollback, (thd, FALSE));
1684
  DBUG_RETURN(error);
1685 1686
}

unknown's avatar
unknown committed
1687

unknown's avatar
unknown committed
1688 1689 1690 1691 1692
struct xahton_st {
  XID *xid;
  int result;
};

unknown's avatar
unknown committed
1693
static my_bool xacommit_handlerton(THD *unused1, plugin_ref plugin,
unknown's avatar
unknown committed
1694
                                   void *arg)
1695
{
Sergei Golubchik's avatar
Sergei Golubchik committed
1696
  handlerton *hton= plugin_hton(plugin);
unknown's avatar
unknown committed
1697 1698
  if (hton->state == SHOW_OPTION_YES && hton->recover)
  {
1699
    hton->commit_by_xid(hton, ((struct xahton_st *)arg)->xid);
unknown's avatar
unknown committed
1700 1701 1702 1703
    ((struct xahton_st *)arg)->result= 0;
  }
  return FALSE;
}
1704

unknown's avatar
unknown committed
1705
static my_bool xarollback_handlerton(THD *unused1, plugin_ref plugin,
unknown's avatar
unknown committed
1706 1707
                                     void *arg)
{
Sergei Golubchik's avatar
Sergei Golubchik committed
1708
  handlerton *hton= plugin_hton(plugin);
unknown's avatar
unknown committed
1709
  if (hton->state == SHOW_OPTION_YES && hton->recover)
unknown's avatar
unknown committed
1710
  {
1711
    hton->rollback_by_xid(hton, ((struct xahton_st *)arg)->xid);
unknown's avatar
unknown committed
1712
    ((struct xahton_st *)arg)->result= 0;
unknown's avatar
unknown committed
1713
  }
unknown's avatar
unknown committed
1714 1715 1716 1717 1718 1719 1720 1721 1722
  return FALSE;
}


int ha_commit_or_rollback_by_xid(XID *xid, bool commit)
{
  struct xahton_st xaop;
  xaop.xid= xid;
  xaop.result= 1;
unknown's avatar
unknown committed
1723

unknown's avatar
unknown committed
1724 1725 1726 1727
  plugin_foreach(NULL, commit ? xacommit_handlerton : xarollback_handlerton,
                 MYSQL_STORAGE_ENGINE_PLUGIN, &xaop);

  return xaop.result;
1728
}
unknown's avatar
unknown committed
1729

unknown's avatar
unknown committed
1730

unknown's avatar
unknown committed
1731
#ifndef DBUG_OFF
unknown's avatar
unknown committed
1732 1733 1734 1735
/**
  @note
    This does not need to be multi-byte safe or anything
*/
unknown's avatar
unknown committed
1736 1737 1738 1739 1740 1741 1742 1743
static char* xid_to_str(char *buf, XID *xid)
{
  int i;
  char *s=buf;
  *s++='\'';
  for (i=0; i < xid->gtrid_length+xid->bqual_length; i++)
  {
    uchar c=(uchar)xid->data[i];
unknown's avatar
unknown committed
1744 1745
    /* is_next_dig is set if next character is a number */
    bool is_next_dig= FALSE;
unknown's avatar
unknown committed
1746 1747
    if (i < XIDDATASIZE)
    {
unknown's avatar
unknown committed
1748 1749
      char ch= xid->data[i+1];
      is_next_dig= (ch >= '0' && ch <='9');
unknown's avatar
unknown committed
1750
    }
unknown's avatar
unknown committed
1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762
    if (i == xid->gtrid_length)
    {
      *s++='\'';
      if (xid->bqual_length)
      {
        *s++='.';
        *s++='\'';
      }
    }
    if (c < 32 || c > 126)
    {
      *s++='\\';
unknown's avatar
unknown committed
1763 1764 1765 1766 1767
      /*
        If next character is a number, write current character with
        3 octal numbers to ensure that the next number is not seen
        as part of the octal number
      */
unknown's avatar
unknown committed
1768 1769 1770 1771 1772
      if (c > 077 || is_next_dig)
        *s++=_dig_vec_lower[c >> 6];
      if (c > 007 || is_next_dig)
        *s++=_dig_vec_lower[(c >> 3) & 7];
      *s++=_dig_vec_lower[c & 7];
unknown's avatar
unknown committed
1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786
    }
    else
    {
      if (c == '\'' || c == '\\')
        *s++='\\';
      *s++=c;
    }
  }
  *s++='\'';
  *s=0;
  return buf;
}
#endif

unknown's avatar
unknown committed
1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801
/**
  recover() step of xa.

  @note
    there are three modes of operation:
    - automatic recover after a crash
    in this case commit_list != 0, tc_heuristic_recover==0
    all xids from commit_list are committed, others are rolled back
    - manual (heuristic) recover
    in this case commit_list==0, tc_heuristic_recover != 0
    DBA has explicitly specified that all prepared transactions should
    be committed (or rolled back).
    - no recovery (MySQL did not detect a crash)
    in this case commit_list==0, tc_heuristic_recover == 0
    there should be no prepared transactions in this case.
unknown's avatar
unknown committed
1802
*/
unknown's avatar
unknown committed
1803 1804 1805 1806 1807 1808 1809
struct xarecover_st
{
  int len, found_foreign_xids, found_my_xids;
  XID *list;
  HASH *commit_list;
  bool dry_run;
};
1810

unknown's avatar
unknown committed
1811
static my_bool xarecover_handlerton(THD *unused, plugin_ref plugin,
unknown's avatar
unknown committed
1812 1813
                                    void *arg)
{
Sergei Golubchik's avatar
Sergei Golubchik committed
1814
  handlerton *hton= plugin_hton(plugin);
unknown's avatar
unknown committed
1815 1816
  struct xarecover_st *info= (struct xarecover_st *) arg;
  int got;
unknown's avatar
unknown committed
1817

unknown's avatar
unknown committed
1818
  if (hton->state == SHOW_OPTION_YES && hton->recover)
1819
  {
1820
    while ((got= hton->recover(hton, info->list, info->len)) > 0 )
1821
    {
unknown's avatar
unknown committed
1822
      sql_print_information("Found %d prepared transaction(s) in %s",
1823
                            got, hton_name(hton)->str);
1824 1825
      for (int i=0; i < got; i ++)
      {
1826
        my_xid x= WSREP_ON && wsrep_is_wsrep_xid(&info->list[i]) ?
1827
                           wsrep_xid_seqno(&info->list[i]) :
1828
		           info->list[i].get_my_xid();
1829
        if (!x) // not "mine" - that is generated by external TM
unknown's avatar
unknown committed
1830
        {
unknown's avatar
unknown committed
1831 1832
#ifndef DBUG_OFF
          char buf[XIDDATASIZE*4+6]; // see xid_to_str
unknown's avatar
unknown committed
1833
          sql_print_information("ignore xid %s", xid_to_str(buf, info->list+i));
unknown's avatar
unknown committed
1834
#endif
unknown's avatar
unknown committed
1835 1836
          xid_cache_insert(info->list+i, XA_PREPARED);
          info->found_foreign_xids++;
unknown's avatar
unknown committed
1837 1838
          continue;
        }
unknown's avatar
unknown committed
1839
        if (info->dry_run)
unknown's avatar
unknown committed
1840
        {
unknown's avatar
unknown committed
1841
          info->found_my_xids++;
1842
          continue;
unknown's avatar
unknown committed
1843 1844
        }
        // recovery mode
unknown's avatar
unknown committed
1845
        if (info->commit_list ?
Konstantin Osipov's avatar
Konstantin Osipov committed
1846
            my_hash_search(info->commit_list, (uchar *)&x, sizeof(x)) != 0 :
1847
            tc_heuristic_recover == TC_HEURISTIC_RECOVER_COMMIT)
unknown's avatar
unknown committed
1848 1849 1850
        {
#ifndef DBUG_OFF
          char buf[XIDDATASIZE*4+6]; // see xid_to_str
unknown's avatar
unknown committed
1851
          sql_print_information("commit xid %s", xid_to_str(buf, info->list+i));
unknown's avatar
unknown committed
1852
#endif
1853
          hton->commit_by_xid(hton, info->list+i);
unknown's avatar
unknown committed
1854
        }
1855
        else
unknown's avatar
unknown committed
1856 1857 1858
        {
#ifndef DBUG_OFF
          char buf[XIDDATASIZE*4+6]; // see xid_to_str
unknown's avatar
unknown committed
1859 1860
          sql_print_information("rollback xid %s",
                                xid_to_str(buf, info->list+i));
unknown's avatar
unknown committed
1861
#endif
1862
          hton->rollback_by_xid(hton, info->list+i);
unknown's avatar
unknown committed
1863
        }
1864
      }
unknown's avatar
unknown committed
1865
      if (got < info->len)
1866
        break;
1867 1868
    }
  }
unknown's avatar
unknown committed
1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898
  return FALSE;
}

int ha_recover(HASH *commit_list)
{
  struct xarecover_st info;
  DBUG_ENTER("ha_recover");
  info.found_foreign_xids= info.found_my_xids= 0;
  info.commit_list= commit_list;
  info.dry_run= (info.commit_list==0 && tc_heuristic_recover==0);
  info.list= NULL;

  /* commit_list and tc_heuristic_recover cannot be set both */
  DBUG_ASSERT(info.commit_list==0 || tc_heuristic_recover==0);
  /* if either is set, total_ha_2pc must be set too */
  DBUG_ASSERT(info.dry_run || total_ha_2pc>(ulong)opt_bin_log);

  if (total_ha_2pc <= (ulong)opt_bin_log)
    DBUG_RETURN(0);

  if (info.commit_list)
    sql_print_information("Starting crash recovery...");

  for (info.len= MAX_XID_LIST_SIZE ; 
       info.list==0 && info.len > MIN_XID_LIST_SIZE; info.len/=2)
  {
    info.list=(XID *)my_malloc(info.len*sizeof(XID), MYF(0));
  }
  if (!info.list)
  {
1899 1900
    sql_print_error(ER(ER_OUTOFMEMORY),
                    static_cast<int>(info.len*sizeof(XID)));
unknown's avatar
unknown committed
1901 1902 1903 1904 1905 1906
    DBUG_RETURN(1);
  }

  plugin_foreach(NULL, xarecover_handlerton, 
                 MYSQL_STORAGE_ENGINE_PLUGIN, &info);

1907
  my_free(info.list);
unknown's avatar
unknown committed
1908 1909 1910 1911
  if (info.found_foreign_xids)
    sql_print_warning("Found %d prepared XA transactions", 
                      info.found_foreign_xids);
  if (info.dry_run && info.found_my_xids)
unknown's avatar
unknown committed
1912 1913 1914 1915 1916 1917 1918
  {
    sql_print_error("Found %d prepared transactions! It means that mysqld was "
                    "not shut down properly last time and critical recovery "
                    "information (last binlog or %s file) was manually deleted "
                    "after a crash. You have to start mysqld with "
                    "--tc-heuristic-recover switch to commit or rollback "
                    "pending transactions.",
unknown's avatar
unknown committed
1919
                    info.found_my_xids, opt_tc_log_file);
unknown's avatar
unknown committed
1920 1921
    DBUG_RETURN(1);
  }
unknown's avatar
unknown committed
1922
  if (info.commit_list)
unknown's avatar
unknown committed
1923
    sql_print_information("Crash recovery finished.");
1924
  DBUG_RETURN(0);
1925
}
1926

unknown's avatar
unknown committed
1927 1928
/**
  return the list of XID's to a client, the same way SHOW commands do.
unknown's avatar
unknown committed
1929

unknown's avatar
unknown committed
1930
  @note
1931 1932 1933
    I didn't find in XA specs that an RM cannot return the same XID twice,
    so mysql_xa_recover does not filter XID's to ensure uniqueness.
    It can be easily fixed later, if necessary.
unknown's avatar
unknown committed
1934
*/
1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952

static my_bool xa_recover_callback(XID_STATE *xs, Protocol *protocol)
{
  if (xs->xa_state == XA_PREPARED)
  {
    protocol->prepare_for_resend();
    protocol->store_longlong((longlong) xs->xid.formatID, FALSE);
    protocol->store_longlong((longlong) xs->xid.gtrid_length, FALSE);
    protocol->store_longlong((longlong) xs->xid.bqual_length, FALSE);
    protocol->store(xs->xid.data, xs->xid.gtrid_length + xs->xid.bqual_length,
                    &my_charset_bin);
    if (protocol->write())
      return TRUE;
  }
  return FALSE;
}


1953
bool mysql_xa_recover(THD *thd)
unknown's avatar
unknown committed
1954
{
1955 1956 1957 1958
  List<Item> field_list;
  Protocol *protocol= thd->protocol;
  DBUG_ENTER("mysql_xa_recover");

1959 1960 1961
  field_list.push_back(new Item_int("formatID", 0, MY_INT32_NUM_DECIMAL_DIGITS));
  field_list.push_back(new Item_int("gtrid_length", 0, MY_INT32_NUM_DECIMAL_DIGITS));
  field_list.push_back(new Item_int("bqual_length", 0, MY_INT32_NUM_DECIMAL_DIGITS));
1962 1963
  field_list.push_back(new Item_empty_string("data",XIDDATASIZE));

1964
  if (protocol->send_result_set_metadata(&field_list,
1965 1966
                            Protocol::SEND_NUM_ROWS | Protocol::SEND_EOF))
    DBUG_RETURN(1);
unknown's avatar
unknown committed
1967

1968 1969 1970
  if (xid_cache_iterate(thd, (my_hash_walk_action) xa_recover_callback,
                        protocol))
    DBUG_RETURN(1);
1971
  my_eof(thd);
1972
  DBUG_RETURN(0);
unknown's avatar
unknown committed
1973
}
1974

1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985
/*
  Called by engine to notify TC that a new commit checkpoint has been reached.
  See comments on handlerton method commit_checkpoint_request() for details.
*/
void
commit_checkpoint_notify_ha(handlerton *hton, void *cookie)
{
  tc_log->commit_checkpoint_notify(cookie);
}


unknown's avatar
unknown committed
1986 1987
/**
  @details
1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998
  This function should be called when MySQL sends rows of a SELECT result set
  or the EOF mark to the client. It releases a possible adaptive hash index
  S-latch held by thd in InnoDB and also releases a possible InnoDB query
  FIFO ticket to enter InnoDB. To save CPU time, InnoDB allows a thd to
  keep them over several calls of the InnoDB handler interface when a join
  is executed. But when we let the control to pass to the client they have
  to be released because if the application program uses mysql_use_result(),
  it may deadlock on the S-latch if the application on another connection
  performs another SQL query. In MySQL-4.1 this is even more important because
  there a connection can have several SELECT queries open at the same time.

unknown's avatar
unknown committed
1999 2000 2001 2002
  @param thd           the thread handle of the current connection

  @return
    always 0
2003
*/
2004

2005 2006
int ha_release_temporary_latches(THD *thd)
{
2007
  Ha_trx_info *info;
2008

2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020
  /*
    Note that below we assume that only transactional storage engines
    may need release_temporary_latches(). If this will ever become false,
    we could iterate on thd->open_tables instead (and remove duplicates
    as if (!seen[hton->slot]) { seen[hton->slot]=1; ... }).
  */
  for (info= thd->transaction.stmt.ha_list; info; info= info->next())
  {
    handlerton *hton= info->ht();
    if (hton && hton->release_temporary_latches)
        hton->release_temporary_latches(hton, thd);
  }
2021
  return 0;
2022 2023
}

2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058
/**
  Check if all storage engines used in transaction agree that after
  rollback to savepoint it is safe to release MDL locks acquired after
  savepoint creation.

  @param thd   The client thread that executes the transaction.

  @return true  - It is safe to release MDL locks.
          false - If it is not.
*/
bool ha_rollback_to_savepoint_can_release_mdl(THD *thd)
{
  Ha_trx_info *ha_info;
  THD_TRANS *trans= (thd->in_sub_stmt ? &thd->transaction.stmt :
                                        &thd->transaction.all);

  DBUG_ENTER("ha_rollback_to_savepoint_can_release_mdl");

  /**
    Checking whether it is safe to release metadata locks after rollback to
    savepoint in all the storage engines that are part of the transaction.
  */
  for (ha_info= trans->ha_list; ha_info; ha_info= ha_info->next())
  {
    handlerton *ht= ha_info->ht();
    DBUG_ASSERT(ht);

    if (ht->savepoint_rollback_can_release_mdl == 0 ||
        ht->savepoint_rollback_can_release_mdl(ht, thd) == false)
      DBUG_RETURN(false);
  }

  DBUG_RETURN(true);
}

2059
int ha_rollback_to_savepoint(THD *thd, SAVEPOINT *sv)
unknown's avatar
unknown committed
2060 2061
{
  int error=0;
2062 2063
  THD_TRANS *trans= (thd->in_sub_stmt ? &thd->transaction.stmt :
                                        &thd->transaction.all);
2064 2065
  Ha_trx_info *ha_info, *ha_info_next;

2066
  DBUG_ENTER("ha_rollback_to_savepoint");
2067

2068 2069 2070 2071 2072
  trans->no_2pc=0;
  /*
    rolling back to savepoint in all storage engines that were part of the
    transaction when the savepoint was set
  */
2073
  for (ha_info= sv->ha_list; ha_info; ha_info= ha_info->next())
2074 2075
  {
    int err;
2076 2077 2078 2079 2080
    handlerton *ht= ha_info->ht();
    DBUG_ASSERT(ht);
    DBUG_ASSERT(ht->savepoint_set != 0);
    if ((err= ht->savepoint_rollback(ht, thd,
                                     (uchar *)(sv+1)+ht->savepoint_offset)))
2081 2082 2083
    { // cannot happen
      my_error(ER_ERROR_DURING_ROLLBACK, MYF(0), err);
      error=1;
2084
    }
2085
    status_var_increment(thd->status_var.ha_savepoint_rollback_count);
2086
    trans->no_2pc|= ht->prepare == 0;
2087
  }
2088 2089 2090 2091
  /*
    rolling back the transaction in all storage engines that were not part of
    the transaction when the savepoint was set
  */
2092 2093
  for (ha_info= trans->ha_list; ha_info != sv->ha_list;
       ha_info= ha_info_next)
unknown's avatar
unknown committed
2094
  {
2095
    int err;
2096 2097
    handlerton *ht= ha_info->ht();
    if ((err= ht->rollback(ht, thd, !thd->in_sub_stmt)))
2098 2099 2100
    { // cannot happen
      my_error(ER_ERROR_DURING_ROLLBACK, MYF(0), err);
      error=1;
unknown's avatar
unknown committed
2101
    }
2102
    status_var_increment(thd->status_var.ha_rollback_count);
2103 2104
    ha_info_next= ha_info->next();
    ha_info->reset(); /* keep it conveniently zero-filled */
2105
  }
2106
  trans->ha_list= sv->ha_list;
unknown's avatar
unknown committed
2107 2108 2109
  DBUG_RETURN(error);
}

unknown's avatar
unknown committed
2110 2111 2112
/**
  @note
  according to the sql standard (ISO/IEC 9075-2:2003)
2113 2114
  section "4.33.4 SQL-statements and transaction states",
  SAVEPOINT is *not* transaction-initiating SQL-statement
unknown's avatar
unknown committed
2115
*/
2116
int ha_savepoint(THD *thd, SAVEPOINT *sv)
unknown's avatar
unknown committed
2117 2118
{
  int error=0;
2119 2120
  THD_TRANS *trans= (thd->in_sub_stmt ? &thd->transaction.stmt :
                                        &thd->transaction.all);
2121
  Ha_trx_info *ha_info= trans->ha_list;
2122
  DBUG_ENTER("ha_savepoint");
Konstantin Osipov's avatar
Konstantin Osipov committed
2123

2124
  for (; ha_info; ha_info= ha_info->next())
unknown's avatar
unknown committed
2125
  {
2126
    int err;
2127 2128 2129
    handlerton *ht= ha_info->ht();
    DBUG_ASSERT(ht);
    if (! ht->savepoint_set)
unknown's avatar
unknown committed
2130
    {
2131
      my_error(ER_CHECK_NOT_IMPLEMENTED, MYF(0), "SAVEPOINT");
unknown's avatar
unknown committed
2132
      error=1;
2133
      break;
unknown's avatar
unknown committed
2134
    }
2135
    if ((err= ht->savepoint_set(ht, thd, (uchar *)(sv+1)+ht->savepoint_offset)))
2136
    { // cannot happen
2137
      my_error(ER_GET_ERRNO, MYF(0), err, hton_name(ht)->str);
unknown's avatar
unknown committed
2138 2139
      error=1;
    }
2140
    status_var_increment(thd->status_var.ha_savepoint_count);
unknown's avatar
unknown committed
2141
  }
2142 2143 2144 2145 2146
  /*
    Remember the list of registered storage engines. All new
    engines are prepended to the beginning of the list.
  */
  sv->ha_list= trans->ha_list;
Konstantin Osipov's avatar
Konstantin Osipov committed
2147

unknown's avatar
unknown committed
2148 2149 2150
  DBUG_RETURN(error);
}

2151
int ha_release_savepoint(THD *thd, SAVEPOINT *sv)
unknown's avatar
unknown committed
2152 2153
{
  int error=0;
2154
  Ha_trx_info *ha_info= sv->ha_list;
2155 2156
  DBUG_ENTER("ha_release_savepoint");

2157
  for (; ha_info; ha_info= ha_info->next())
unknown's avatar
unknown committed
2158
  {
2159
    int err;
2160 2161 2162 2163
    handlerton *ht= ha_info->ht();
    /* Savepoint life time is enclosed into transaction life time. */
    DBUG_ASSERT(ht);
    if (!ht->savepoint_release)
2164
      continue;
2165 2166
    if ((err= ht->savepoint_release(ht, thd,
                                    (uchar *)(sv+1) + ht->savepoint_offset)))
2167
    { // cannot happen
2168
      my_error(ER_GET_ERRNO, MYF(0), err, hton_name(ht)->str);
2169
      error=1;
unknown's avatar
unknown committed
2170
    }
unknown's avatar
unknown committed
2171 2172 2173 2174
  }
  DBUG_RETURN(error);
}

2175

unknown's avatar
unknown committed
2176
static my_bool snapshot_handlerton(THD *thd, plugin_ref plugin,
unknown's avatar
unknown committed
2177 2178
                                   void *arg)
{
Sergei Golubchik's avatar
Sergei Golubchik committed
2179
  handlerton *hton= plugin_hton(plugin);
unknown's avatar
unknown committed
2180 2181 2182
  if (hton->state == SHOW_OPTION_YES &&
      hton->start_consistent_snapshot)
  {
2183
    hton->start_consistent_snapshot(hton, thd);
unknown's avatar
unknown committed
2184 2185 2186 2187 2188
    *((bool *)arg)= false;
  }
  return FALSE;
}

2189 2190
int ha_start_consistent_snapshot(THD *thd)
{
2191 2192
  bool warn= true;

unknown's avatar
unknown committed
2193
  /*
unknown's avatar
unknown committed
2194 2195 2196 2197 2198
    Holding the LOCK_commit_ordered mutex ensures that we get the same
    snapshot for all engines (including the binary log).  This allows us
    among other things to do backups with
    START TRANSACTION WITH CONSISTENT SNAPSHOT and
    have a consistent binlog position.
unknown's avatar
unknown committed
2199
  */
Sergei Golubchik's avatar
Sergei Golubchik committed
2200
  mysql_mutex_lock(&LOCK_commit_ordered);
unknown's avatar
unknown committed
2201
  plugin_foreach(thd, snapshot_handlerton, MYSQL_STORAGE_ENGINE_PLUGIN, &warn);
Sergei Golubchik's avatar
Sergei Golubchik committed
2202
  mysql_mutex_unlock(&LOCK_commit_ordered);
unknown's avatar
unknown committed
2203

2204 2205 2206 2207
  /*
    Same idea as when one wants to CREATE TABLE in one engine which does not
    exist:
  */
2208
  if (warn)
2209
    push_warning(thd, Sql_condition::WARN_LEVEL_WARN, ER_UNKNOWN_ERROR,
2210 2211
                 "This MySQL server does not support any "
                 "consistent-read capable storage engine");
2212 2213 2214 2215
  return 0;
}


unknown's avatar
unknown committed
2216
static my_bool flush_handlerton(THD *thd, plugin_ref plugin,
unknown's avatar
unknown committed
2217
                                void *arg)
unknown's avatar
unknown committed
2218
{
Sergei Golubchik's avatar
Sergei Golubchik committed
2219
  handlerton *hton= plugin_hton(plugin);
2220 2221
  if (hton->state == SHOW_OPTION_YES && hton->flush_logs && 
      hton->flush_logs(hton))
unknown's avatar
unknown committed
2222 2223 2224 2225
    return TRUE;
  return FALSE;
}

2226

unknown's avatar
unknown committed
2227 2228 2229
bool ha_flush_logs(handlerton *db_type)
{
  if (db_type == NULL)
2230
  {
unknown's avatar
unknown committed
2231 2232 2233
    if (plugin_foreach(NULL, flush_handlerton,
                          MYSQL_STORAGE_ENGINE_PLUGIN, 0))
      return TRUE;
2234
  }
unknown's avatar
unknown committed
2235 2236 2237
  else
  {
    if (db_type->state != SHOW_OPTION_YES ||
2238
        (db_type->flush_logs && db_type->flush_logs(db_type)))
unknown's avatar
unknown committed
2239 2240 2241
      return TRUE;
  }
  return FALSE;
unknown's avatar
unknown committed
2242 2243
}

2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266

/**
  @brief make canonical filename

  @param[in]  file     table handler
  @param[in]  path     original path
  @param[out] tmp_path buffer for canonized path

  @details Lower case db name and table name path parts for
           non file based tables when lower_case_table_names
           is 2 (store as is, compare in lower case).
           Filesystem path prefix (mysql_data_home or tmpdir)
           is left intact.

  @note tmp_path may be left intact if no conversion was
        performed.

  @retval canonized path

  @todo This may be done more efficiently when table path
        gets built. Convert this function to something like
        ASSERT_CANONICAL_FILENAME.
*/
2267 2268
const char *get_canonical_filename(handler *file, const char *path,
                                   char *tmp_path)
2269
{
2270
  uint i;
2271 2272 2273
  if (lower_case_table_names != 2 || (file->ha_table_flags() & HA_FILE_BASED))
    return path;

2274 2275 2276 2277 2278 2279
  for (i= 0; i <= mysql_tmpdir_list.max; i++)
  {
    if (is_prefix(path, mysql_tmpdir_list.list[i]))
      return path;
  }

2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292
  /* Ensure that table handler get path in lower case */
  if (tmp_path != path)
    strmov(tmp_path, path);

  /*
    we only should turn into lowercase database/table part
    so start the process after homedirectory
  */
  my_casedn_str(files_charset_info, tmp_path + mysql_data_home_len);
  return tmp_path;
}


2293 2294 2295 2296 2297 2298 2299 2300 2301
/**
  An interceptor to hijack the text of the error message without
  setting an error in the thread. We need the text to present it
  in the form of a warning to the user.
*/

struct Ha_delete_table_error_handler: public Internal_error_handler
{
public:
Marc Alff's avatar
Marc Alff committed
2302 2303 2304
  virtual bool handle_condition(THD *thd,
                                uint sql_errno,
                                const char* sqlstate,
2305
                                Sql_condition::enum_warning_level level,
Marc Alff's avatar
Marc Alff committed
2306
                                const char* msg,
2307
                                Sql_condition ** cond_hdl);
2308 2309 2310 2311 2312 2313
  char buff[MYSQL_ERRMSG_SIZE];
};


bool
Ha_delete_table_error_handler::
Marc Alff's avatar
Marc Alff committed
2314 2315 2316
handle_condition(THD *,
                 uint,
                 const char*,
2317
                 Sql_condition::enum_warning_level,
Marc Alff's avatar
Marc Alff committed
2318
                 const char* msg,
2319
                 Sql_condition ** cond_hdl)
2320
{
Marc Alff's avatar
Marc Alff committed
2321
  *cond_hdl= NULL;
2322
  /* Grab the error message */
2323
  strmake_buf(buff, msg);
2324 2325 2326 2327
  return TRUE;
}


2328
/** @brief
unknown's avatar
unknown committed
2329 2330 2331
  This should return ENOENT if the file doesn't exists.
  The .frm file will be deleted only if we return 0 or ENOENT
*/
unknown's avatar
unknown committed
2332
int ha_delete_table(THD *thd, handlerton *table_type, const char *path,
unknown's avatar
unknown committed
2333
                    const char *db, const char *alias, bool generate_warning)
unknown's avatar
unknown committed
2334
{
2335
  handler *file;
unknown's avatar
unknown committed
2336
  char tmp_path[FN_REFLEN];
2337 2338 2339 2340 2341
  int error;
  TABLE dummy_table;
  TABLE_SHARE dummy_share;
  DBUG_ENTER("ha_delete_table");

2342 2343 2344 2345 2346
  /* table_type is NULL in ALTER TABLE when renaming only .frm files */
  if (table_type == NULL || table_type == view_pseudo_hton ||
      ! (file=get_new_handler((TABLE_SHARE*)0, thd->mem_root, table_type)))
    DBUG_RETURN(HA_ERR_NO_SUCH_TABLE);

2347 2348 2349
  bzero((char*) &dummy_table, sizeof(dummy_table));
  bzero((char*) &dummy_share, sizeof(dummy_share));
  dummy_table.s= &dummy_share;
2350

2351
  path= get_canonical_filename(file, path, tmp_path);
2352
  if ((error= file->ha_delete_table(path)) && generate_warning)
2353 2354 2355
  {
    /*
      Because file->print_error() use my_error() to generate the error message
2356 2357 2358
      we use an internal error handler to intercept it and store the text
      in a temporary buffer. Later the message will be presented to user
      as a warning.
2359
    */
2360
    Ha_delete_table_error_handler ha_delete_table_error_handler;
2361 2362

    /* Fill up strucutures that print_error may need */
unknown's avatar
unknown committed
2363 2364
    dummy_share.path.str= (char*) path;
    dummy_share.path.length= strlen(path);
2365
    dummy_share.normalized_path= dummy_share.path;
unknown's avatar
unknown committed
2366 2367 2368 2369
    dummy_share.db.str= (char*) db;
    dummy_share.db.length= strlen(db);
    dummy_share.table_name.str= (char*) alias;
    dummy_share.table_name.length= strlen(alias);
2370 2371
    dummy_table.alias.set(alias, dummy_share.table_name.length,
                          table_alias_charset);
2372

2373
    file->change_table_ptr(&dummy_table, &dummy_share);
2374 2375

    thd->push_internal_handler(&ha_delete_table_error_handler);
2376
    file->print_error(error, 0);
2377 2378 2379 2380 2381 2382 2383

    thd->pop_internal_handler();

    /*
      XXX: should we convert *all* errors to warnings here?
      What if the error is fatal?
    */
2384
    push_warning(thd, Sql_condition::WARN_LEVEL_WARN, error,
2385
                ha_delete_table_error_handler.buff);
2386
  }
unknown's avatar
unknown committed
2387
  delete file;
2388

2389
  DBUG_RETURN(error);
unknown's avatar
unknown committed
2390
}
unknown's avatar
unknown committed
2391

unknown's avatar
unknown committed
2392 2393 2394
/****************************************************************************
** General handler functions
****************************************************************************/
2395
handler *handler::clone(const char *name, MEM_ROOT *mem_root)
unknown's avatar
unknown committed
2396
{
2397
  handler *new_handler= get_new_handler(table->s, mem_root, ht);
2398

2399
  if (!new_handler)
2400
    return NULL;
2401 2402
  if (new_handler->set_ha_share_ref(ha_share))
    goto err;
2403

2404 2405 2406 2407 2408
  /*
    Allocate handler->ref here because otherwise ha_open will allocate it
    on this->table->mem_root and we will not be able to reclaim that memory 
    when the clone handler object is destroyed.
  */
2409 2410 2411

  if (!(new_handler->ref= (uchar*) alloc_root(mem_root,
                                              ALIGN_SIZE(ref_length)*2)))
2412
    goto err;
2413

2414 2415 2416
  /*
    TODO: Implement a more efficient way to have more than one index open for
    the same table instance. The ha_open call is not cachable for clone.
2417 2418 2419

    This is not critical as the engines already have the table open
    and should be able to use the original instance of the table.
2420
  */
2421
  if (new_handler->ha_open(table, name, table->db_stat,
2422
                           HA_OPEN_IGNORE_IF_LOCKED))
2423
    goto err;
2424

2425
  return new_handler;
2426 2427 2428 2429

err:
  delete new_handler;
  return NULL;
unknown's avatar
unknown committed
2430 2431
}

unknown's avatar
unknown committed
2432

Sergei Golubchik's avatar
Sergei Golubchik committed
2433
double handler::keyread_time(uint index, uint ranges, ha_rows rows)
2434
{
Sergei Golubchik's avatar
Sergei Golubchik committed
2435 2436 2437 2438 2439 2440 2441
  /*
    It is assumed that we will read trough the whole key range and that all
    key blocks are half full (normally things are much better). It is also
    assumed that each time we read the next key from the index, the handler
    performs a random seek, thus the cost is proportional to the number of
    blocks read. This model does not take into account clustered indexes -
    engines that support that (e.g. InnoDB) may want to overwrite this method.
2442
    The model counts in the time to read index entries from cache.
Sergei Golubchik's avatar
Sergei Golubchik committed
2443
  */
2444
  ulong len= table->key_info[index].key_length + ref_length;
2445 2446
  if (index == table->s->primary_key && table->file->primary_key_is_clustered())
    len= table->s->stored_rec_length;
2447 2448 2449
  double keys_per_block= (stats.block_size/2.0/len+1);
  return (rows + keys_per_block-1)/ keys_per_block +
         len*rows/(stats.block_size+1)/TIME_FOR_COMPARE ;
2450 2451
}

2452
void **handler::ha_data(THD *thd) const
unknown's avatar
unknown committed
2453
{
2454
  return thd_ha_data(thd, ht);
unknown's avatar
unknown committed
2455 2456 2457 2458
}

THD *handler::ha_thd(void) const
{
unknown's avatar
unknown committed
2459
  DBUG_ASSERT(!table || !table->in_use || table->in_use == current_thd);
unknown's avatar
unknown committed
2460 2461 2462
  return (table && table->in_use) ? table->in_use : current_thd;
}

2463 2464 2465 2466 2467 2468
void handler::unbind_psi()
{
  /*
    Notify the instrumentation that this table is not owned
    by this thread any more.
  */
2469
  PSI_CALL_unbind_table(m_psi);
2470 2471 2472 2473 2474 2475 2476 2477
}

void handler::rebind_psi()
{
  /*
    Notify the instrumentation that this table is now owned
    by this thread.
  */
2478
  m_psi= PSI_CALL_rebind_table(ha_table_share_psi(), this, m_psi);
2479 2480 2481
}


2482
PSI_table_share *handler::ha_table_share_psi() const
Marc Alff's avatar
Marc Alff committed
2483
{
2484
  return table_share->m_psi;
Marc Alff's avatar
Marc Alff committed
2485
}
2486

2487
/** @brief
unknown's avatar
unknown committed
2488 2489 2490 2491 2492 2493 2494
  Open database-handler.

  IMPLEMENTATION
    Try O_RDONLY if cannot open as O_RDWR
    Don't wait for locks if not HA_OPEN_WAIT_IF_LOCKED is set
*/
int handler::ha_open(TABLE *table_arg, const char *name, int mode,
2495
                     uint test_if_locked)
unknown's avatar
unknown committed
2496 2497
{
  int error;
2498
  DBUG_ENTER("handler::ha_open");
unknown's avatar
unknown committed
2499 2500
  DBUG_PRINT("enter",
             ("name: %s  db_type: %d  db_stat: %d  mode: %d  lock_test: %d",
unknown's avatar
unknown committed
2501
              name, ht->db_type, table_arg->db_stat, mode,
unknown's avatar
unknown committed
2502 2503 2504 2505
              test_if_locked));

  table= table_arg;
  DBUG_ASSERT(table->s == table_share);
2506 2507
  DBUG_ASSERT(m_lock_type == F_UNLCK);
  DBUG_PRINT("info", ("old m_lock_type: %d F_UNLCK %d", m_lock_type, F_UNLCK));
2508
  DBUG_ASSERT(alloc_root_inited(&table->mem_root));
unknown's avatar
unknown committed
2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520

  if ((error=open(name,mode,test_if_locked)))
  {
    if ((error == EACCES || error == EROFS) && mode == O_RDWR &&
	(table->db_stat & HA_TRY_READ_ONLY))
    {
      table->db_stat|=HA_READ_ONLY;
      error=open(name,O_RDONLY,test_if_locked);
    }
  }
  if (error)
  {
unknown's avatar
unknown committed
2521
    my_errno= error;                            /* Safeguard */
unknown's avatar
unknown committed
2522 2523 2524 2525
    DBUG_PRINT("error",("error: %d  errno: %d",error,errno));
  }
  else
  {
2526 2527
    DBUG_ASSERT(m_psi == NULL);
    DBUG_ASSERT(table_share != NULL);
2528 2529 2530 2531 2532 2533 2534
    /*
      Do not call this for partitions handlers, since it may take too much
      resources.
      So only use the m_psi on table level, not for individual partitions.
    */
    if (!(test_if_locked & HA_OPEN_NO_PSI_CALL))
    {
Sergei Golubchik's avatar
Sergei Golubchik committed
2535
      m_psi= PSI_CALL_open_table(ha_table_share_psi(), this);
2536
    }
2537

2538
    if (table->s->db_options_in_use & HA_OPTION_READ_ONLY_DATA)
unknown's avatar
unknown committed
2539
      table->db_stat|=HA_READ_ONLY;
2540 2541
    (void) extra(HA_EXTRA_NO_READCHECK);	// Not needed in SQL

2542
    /* ref is already allocated for us if we're called from handler::clone() */
2543
    if (!ref && !(ref= (uchar*) alloc_root(&table->mem_root, 
2544
                                          ALIGN_SIZE(ref_length)*2)))
unknown's avatar
unknown committed
2545
    {
2546
      ha_close();
unknown's avatar
unknown committed
2547 2548 2549
      error=HA_ERR_OUT_OF_MEM;
    }
    else
2550 2551
      dup_ref=ref+ALIGN_SIZE(ref_length);
    cached_table_flags= table_flags();
unknown's avatar
unknown committed
2552
  }
2553
  reset_statistics();
2554
  internal_tmp_table= MY_TEST(test_if_locked & HA_OPEN_INTERNAL_TABLE);
unknown's avatar
unknown committed
2555 2556 2557
  DBUG_RETURN(error);
}

2558
int handler::ha_close(void)
2559 2560 2561 2562 2563 2564 2565 2566
{
  DBUG_ENTER("ha_close");
  /*
    Increment global statistics for temporary tables.
    In_use is 0 for tables that was closed from the table cache.
  */
  if (table->in_use)
    status_var_add(table->in_use->status_var.rows_tmp_read, rows_tmp_read);
2567
  PSI_CALL_close_table(m_psi);
2568
  m_psi= NULL; /* instrumentation handle, invalid after close_table() */
2569 2570 2571

  /* Detach from ANALYZE tracker */
  tracker= NULL;
2572
  
2573 2574
  DBUG_ASSERT(m_lock_type == F_UNLCK);
  DBUG_ASSERT(inited == NONE);
2575 2576
  DBUG_RETURN(close());
}
unknown's avatar
unknown committed
2577

2578

Sergei Golubchik's avatar
Sergei Golubchik committed
2579 2580 2581
int handler::ha_rnd_next(uchar *buf)
{
  int result;
2582 2583 2584 2585
  DBUG_ENTER("handler::ha_rnd_next");
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type != F_UNLCK);
  DBUG_ASSERT(inited == RND);
Sergei Golubchik's avatar
Sergei Golubchik committed
2586

2587
  TABLE_IO_WAIT(tracker, m_psi, PSI_TABLE_FETCH_ROW, MAX_KEY, 0,
Sergei Golubchik's avatar
Sergei Golubchik committed
2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599
    { result= rnd_next(buf); })
  if (!result)
  {
    update_rows_read();
    increment_statistics(&SSV::ha_read_rnd_next_count);
  }
  else if (result == HA_ERR_RECORD_DELETED)
    increment_statistics(&SSV::ha_read_rnd_deleted_count);
  else
    increment_statistics(&SSV::ha_read_rnd_next_count);

  table->status=result ? STATUS_NOT_FOUND: 0;
2600
  DBUG_RETURN(result);
Sergei Golubchik's avatar
Sergei Golubchik committed
2601 2602 2603 2604 2605
}

int handler::ha_rnd_pos(uchar *buf, uchar *pos)
{
  int result;
2606 2607 2608 2609 2610
  DBUG_ENTER("handler::ha_rnd_pos");
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type != F_UNLCK);
  /* TODO: Find out how to solve ha_rnd_pos when finding duplicate update. */
  /* DBUG_ASSERT(inited == RND); */
Sergei Golubchik's avatar
Sergei Golubchik committed
2611

2612
  TABLE_IO_WAIT(tracker, m_psi, PSI_TABLE_FETCH_ROW, MAX_KEY, 0,
Sergei Golubchik's avatar
Sergei Golubchik committed
2613 2614 2615 2616 2617
    { result= rnd_pos(buf, pos); })
  increment_statistics(&SSV::ha_read_rnd_count);
  if (!result)
    update_rows_read();
  table->status=result ? STATUS_NOT_FOUND: 0;
2618
  DBUG_RETURN(result);
Sergei Golubchik's avatar
Sergei Golubchik committed
2619 2620 2621 2622 2623 2624 2625
}

int handler::ha_index_read_map(uchar *buf, const uchar *key,
                                      key_part_map keypart_map,
                                      enum ha_rkey_function find_flag)
{
  int result;
2626 2627 2628
  DBUG_ENTER("handler::ha_index_read_map");
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type != F_UNLCK);
Sergei Golubchik's avatar
Sergei Golubchik committed
2629 2630
  DBUG_ASSERT(inited==INDEX);

2631
  TABLE_IO_WAIT(tracker, m_psi, PSI_TABLE_FETCH_ROW, active_index, 0,
Sergei Golubchik's avatar
Sergei Golubchik committed
2632 2633 2634 2635 2636
    { result= index_read_map(buf, key, keypart_map, find_flag); })
  increment_statistics(&SSV::ha_read_key_count);
  if (!result)
    update_index_statistics();
  table->status=result ? STATUS_NOT_FOUND: 0;
2637
  DBUG_RETURN(result);
Sergei Golubchik's avatar
Sergei Golubchik committed
2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651
}

/*
  @note: Other index lookup/navigation functions require prior
  handler->index_init() call. This function is different, it requires
  that the scan is not initialized, and accepts "uint index" as an argument.
*/

int handler::ha_index_read_idx_map(uchar *buf, uint index, const uchar *key,
                                          key_part_map keypart_map,
                                          enum ha_rkey_function find_flag)
{
  int result;
  DBUG_ASSERT(inited==NONE);
2652 2653
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type != F_UNLCK);
Sergei Golubchik's avatar
Sergei Golubchik committed
2654
  DBUG_ASSERT(end_range == NULL);
2655
  TABLE_IO_WAIT(tracker, m_psi, PSI_TABLE_FETCH_ROW, index, 0,
Sergei Golubchik's avatar
Sergei Golubchik committed
2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669
    { result= index_read_idx_map(buf, index, key, keypart_map, find_flag); })
  increment_statistics(&SSV::ha_read_key_count);
  if (!result)
  {
    update_rows_read();
    index_rows_read[index]++;
  }
  table->status=result ? STATUS_NOT_FOUND: 0;
  return result;
}

int handler::ha_index_next(uchar * buf)
{
  int result;
2670 2671 2672
  DBUG_ENTER("handler::ha_index_next");
 DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type != F_UNLCK);
Sergei Golubchik's avatar
Sergei Golubchik committed
2673 2674
  DBUG_ASSERT(inited==INDEX);

2675
  TABLE_IO_WAIT(tracker, m_psi, PSI_TABLE_FETCH_ROW, active_index, 0,
Sergei Golubchik's avatar
Sergei Golubchik committed
2676 2677 2678 2679 2680
    { result= index_next(buf); })
  increment_statistics(&SSV::ha_read_next_count);
  if (!result)
    update_index_statistics();
  table->status=result ? STATUS_NOT_FOUND: 0;
2681
  DBUG_RETURN(result);
Sergei Golubchik's avatar
Sergei Golubchik committed
2682 2683 2684 2685 2686
}

int handler::ha_index_prev(uchar * buf)
{
  int result;
2687 2688 2689
  DBUG_ENTER("handler::ha_index_prev");
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type != F_UNLCK);
Sergei Golubchik's avatar
Sergei Golubchik committed
2690 2691
  DBUG_ASSERT(inited==INDEX);

2692
  TABLE_IO_WAIT(tracker, m_psi, PSI_TABLE_FETCH_ROW, active_index, 0,
Sergei Golubchik's avatar
Sergei Golubchik committed
2693 2694 2695 2696 2697
    { result= index_prev(buf); })
  increment_statistics(&SSV::ha_read_prev_count);
  if (!result)
    update_index_statistics();
  table->status=result ? STATUS_NOT_FOUND: 0;
2698
  DBUG_RETURN(result);
Sergei Golubchik's avatar
Sergei Golubchik committed
2699 2700 2701 2702 2703
}

int handler::ha_index_first(uchar * buf)
{
  int result;
2704 2705
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type != F_UNLCK);
Sergei Golubchik's avatar
Sergei Golubchik committed
2706 2707
  DBUG_ASSERT(inited==INDEX);

2708
  TABLE_IO_WAIT(tracker, m_psi, PSI_TABLE_FETCH_ROW, active_index, 0,
Sergei Golubchik's avatar
Sergei Golubchik committed
2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719
    { result= index_first(buf); })
  increment_statistics(&SSV::ha_read_first_count);
  if (!result)
    update_index_statistics();
  table->status=result ? STATUS_NOT_FOUND: 0;
  return result;
}

int handler::ha_index_last(uchar * buf)
{
  int result;
2720 2721
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type != F_UNLCK);
Sergei Golubchik's avatar
Sergei Golubchik committed
2722 2723
  DBUG_ASSERT(inited==INDEX);

2724
  TABLE_IO_WAIT(tracker, m_psi, PSI_TABLE_FETCH_ROW, active_index, 0,
Sergei Golubchik's avatar
Sergei Golubchik committed
2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735
    { result= index_last(buf); })
  increment_statistics(&SSV::ha_read_last_count);
  if (!result)
    update_index_statistics();
  table->status=result ? STATUS_NOT_FOUND: 0;
  return result;
}

int handler::ha_index_next_same(uchar *buf, const uchar *key, uint keylen)
{
  int result;
2736 2737
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type != F_UNLCK);
Sergei Golubchik's avatar
Sergei Golubchik committed
2738 2739
  DBUG_ASSERT(inited==INDEX);

2740
  TABLE_IO_WAIT(tracker, m_psi, PSI_TABLE_FETCH_ROW, active_index, 0,
Sergei Golubchik's avatar
Sergei Golubchik committed
2741 2742 2743 2744 2745 2746 2747 2748
    { result= index_next_same(buf, key, keylen); })
  increment_statistics(&SSV::ha_read_next_count);
  if (!result)
    update_index_statistics();
  table->status=result ? STATUS_NOT_FOUND: 0;
  return result;
}

2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760
/* Initialize handler for random reading, with error handling */

int handler::ha_rnd_init_with_error(bool scan)
{
  int error;
  if (!(error= ha_rnd_init(scan)))
    return 0;
  table->file->print_error(error, MYF(0));
  return error;
}


unknown's avatar
unknown committed
2761 2762 2763
/**
  Read first row (only) from a table.

2764
  This is never called for InnoDB tables, as these table types
2765
  has the HA_STATS_RECORDS_IS_EXACT set.
2766
*/
2767
int handler::read_first_row(uchar * buf, uint primary_key)
unknown's avatar
unknown committed
2768 2769
{
  register int error;
2770
  DBUG_ENTER("handler::read_first_row");
unknown's avatar
unknown committed
2771

2772 2773 2774
  /*
    If there is very few deleted rows in the table, find the first row by
    scanning the table.
2775
    TODO remove the test for HA_READ_ORDER
2776
  */
2777
  if (stats.deleted < 10 || primary_key >= MAX_KEY ||
2778
      !(index_flags(primary_key, 0, 0) & HA_READ_ORDER))
2779
  {
2780
    if (!(error= ha_rnd_init(1)))
2781
    {
Sergei Golubchik's avatar
Sergei Golubchik committed
2782
      while ((error= ha_rnd_next(buf)) == HA_ERR_RECORD_DELETED)
2783 2784 2785 2786
        /* skip deleted row */;
      const int end_error= ha_rnd_end();
      if (!error)
        error= end_error;
2787
    }
2788 2789 2790 2791
  }
  else
  {
    /* Find the first row through the primary key */
2792 2793
    if (!(error= ha_index_init(primary_key, 0)))
    {
2794
      error= ha_index_first(buf);
2795 2796 2797 2798
      const int end_error= ha_index_end();
      if (!error)
        error= end_error;
    }
2799
  }
unknown's avatar
unknown committed
2800 2801 2802
  DBUG_RETURN(error);
}

unknown's avatar
unknown committed
2803 2804
/**
  Generate the next auto-increment number based on increment and offset.
2805 2806 2807
  computes the lowest number
  - strictly greater than "nr"
  - of the form: auto_increment_offset + N * auto_increment_increment
2808 2809
  If overflow happened then return MAX_ULONGLONG value as an
  indication of overflow.
2810
  In most cases increment= offset= 1, in which case we get:
unknown's avatar
unknown committed
2811 2812 2813
  @verbatim 1,2,3,4,5,... @endverbatim
    If increment=10 and offset=5 and previous number is 1, we get:
  @verbatim 1,5,15,25,35,... @endverbatim
2814 2815
*/
inline ulonglong
2816
compute_next_insert_id(ulonglong nr,struct system_variables *variables)
2817
{
2818 2819
  const ulonglong save_nr= nr;

2820
  if (variables->auto_increment_increment == 1)
2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834
    nr= nr + 1; // optimization of the formula below
  else
  {
    nr= (((nr+ variables->auto_increment_increment -
           variables->auto_increment_offset)) /
         (ulonglong) variables->auto_increment_increment);
    nr= (nr* (ulonglong) variables->auto_increment_increment +
         variables->auto_increment_offset);
  }

  if (unlikely(nr <= save_nr))
    return ULONGLONG_MAX;

  return nr;
2835 2836 2837
}


2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849
void handler::adjust_next_insert_id_after_explicit_value(ulonglong nr)
{
  /*
    If we have set THD::next_insert_id previously and plan to insert an
    explicitely-specified value larger than this, we need to increase
    THD::next_insert_id to be greater than the explicit value.
  */
  if ((next_insert_id > 0) && (nr >= next_insert_id))
    set_next_insert_id(compute_next_insert_id(nr, &table->in_use->variables));
}


2850
/** @brief
2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876
  Computes the largest number X:
  - smaller than or equal to "nr"
  - of the form: auto_increment_offset + N * auto_increment_increment
  where N>=0.

  SYNOPSIS
    prev_insert_id
      nr            Number to "round down"
      variables     variables struct containing auto_increment_increment and
                    auto_increment_offset

  RETURN
    The number X if it exists, "nr" otherwise.
*/
inline ulonglong
prev_insert_id(ulonglong nr, struct system_variables *variables)
{
  if (unlikely(nr < variables->auto_increment_offset))
  {
    /*
      There's nothing good we can do here. That is a pathological case, where
      the offset is larger than the column's max possible value, i.e. not even
      the first sequence value may be inserted. User will receive warning.
    */
    DBUG_PRINT("info",("auto_increment: nr: %lu cannot honour "
                       "auto_increment_offset: %lu",
unknown's avatar
unknown committed
2877
                       (ulong) nr, variables->auto_increment_offset));
2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888
    return nr;
  }
  if (variables->auto_increment_increment == 1)
    return nr; // optimization of the formula below
  nr= (((nr - variables->auto_increment_offset)) /
       (ulonglong) variables->auto_increment_increment);
  return (nr * (ulonglong) variables->auto_increment_increment +
          variables->auto_increment_offset);
}


unknown's avatar
unknown committed
2889 2890
/**
  Update the auto_increment field if necessary.
2891

unknown's avatar
unknown committed
2892
  Updates columns with type NEXT_NUMBER if:
2893 2894 2895 2896 2897 2898 2899

  - If column value is set to NULL (in which case
    auto_increment_field_not_null is 0)
  - If column is set to 0 and (sql_mode & MODE_NO_AUTO_VALUE_ON_ZERO) is not
    set. In the future we will only set NEXT_NUMBER fields if one sets them
    to NULL (or they are not included in the insert list).

2900 2901 2902
    In those cases, we check if the currently reserved interval still has
    values we have not used. If yes, we pick the smallest one and use it.
    Otherwise:
2903

2904 2905 2906
  - If a list of intervals has been provided to the statement via SET
    INSERT_ID or via an Intvar_log_event (in a replication slave), we pick the
    first unused interval from this list, consider it as reserved.
2907

2908 2909 2910 2911 2912 2913
  - Otherwise we set the column for the first row to the value
    next_insert_id(get_auto_increment(column))) which is usually
    max-used-column-value+1.
    We call get_auto_increment() for the first row in a multi-row
    statement. get_auto_increment() will tell us the interval of values it
    reserved for us.
2914

2915 2916 2917 2918 2919
  - In both cases, for the following rows we use those reserved values without
    calling the handler again (we just progress in the interval, computing
    each new value from the previous one). Until we have exhausted them, then
    we either take the next provided interval or call get_auto_increment()
    again to reserve a new interval.
2920

2921 2922 2923
  - In both cases, the reserved intervals are remembered in
    thd->auto_inc_intervals_in_cur_stmt_for_binlog if statement-based
    binlogging; the last reserved interval is remembered in
Guilhem Bichot's avatar
Guilhem Bichot committed
2924 2925 2926 2927 2928 2929
    auto_inc_interval_for_cur_row. The number of reserved intervals is
    remembered in auto_inc_intervals_count. It differs from the number of
    elements in thd->auto_inc_intervals_in_cur_stmt_for_binlog() because the
    latter list is cumulative over all statements forming one binlog event
    (when stored functions and triggers are used), and collapses two
    contiguous intervals in one (see its append() method).
2930 2931 2932 2933 2934 2935 2936 2937 2938 2939

    The idea is that generated auto_increment values are predictable and
    independent of the column values in the table.  This is needed to be
    able to replicate into a table that already has rows with a higher
    auto-increment value than the one that is inserted.

    After we have already generated an auto-increment number and the user
    inserts a column with a higher value than the last used one, we will
    start counting from the inserted value.

2940 2941 2942 2943 2944 2945 2946
    This function's "outputs" are: the table's auto_increment field is filled
    with a value, thd->next_insert_id is filled with the value to use for the
    next row, if a value was autogenerated for the current row it is stored in
    thd->insert_id_for_cur_row, if get_auto_increment() was called
    thd->auto_inc_interval_for_cur_row is modified, if that interval is not
    present in thd->auto_inc_intervals_in_cur_stmt_for_binlog it is added to
    this list.
2947

unknown's avatar
unknown committed
2948
  @todo
2949 2950 2951 2952 2953
    Replace all references to "next number" or NEXT_NUMBER to
    "auto_increment", everywhere (see below: there is
    table->auto_increment_field_not_null, and there also exists
    table->next_number_field, it's not consistent).

unknown's avatar
unknown committed
2954 2955 2956 2957 2958 2959 2960 2961
  @retval
    0	ok
  @retval
    HA_ERR_AUTOINC_READ_FAILED  get_auto_increment() was called and
    returned ~(ulonglong) 0
  @retval
    HA_ERR_AUTOINC_ERANGE storing value in field caused strict mode
    failure.
2962
*/
unknown's avatar
unknown committed
2963

2964 2965 2966 2967
#define AUTO_INC_DEFAULT_NB_ROWS 1 // Some prefer 1024 here
#define AUTO_INC_DEFAULT_NB_MAX_BITS 16
#define AUTO_INC_DEFAULT_NB_MAX ((1 << AUTO_INC_DEFAULT_NB_MAX_BITS) - 1)

unknown's avatar
unknown committed
2968
int handler::update_auto_increment()
unknown's avatar
unknown committed
2969
{
2970 2971
  ulonglong nr, nb_reserved_values;
  bool append= FALSE;
2972 2973
  THD *thd= table->in_use;
  struct system_variables *variables= &thd->variables;
2974 2975
  int result=0, tmp;
  enum enum_check_fields save_count_cuted_fields;
2976
  DBUG_ENTER("handler::update_auto_increment");
2977 2978

  /*
2979 2980
    next_insert_id is a "cursor" into the reserved interval, it may go greater
    than the interval, but not smaller.
2981
  */
2982
  DBUG_ASSERT(next_insert_id >= auto_inc_interval_for_cur_row.minimum());
2983 2984

  if ((nr= table->next_number_field->val_int()) != 0 ||
2985
      (table->auto_increment_field_not_null &&
2986
       thd->variables.sql_mode & MODE_NO_AUTO_VALUE_ON_ZERO))
2987
  {
2988 2989 2990 2991 2992
    /*
      Update next_insert_id if we had already generated a value in this
      statement (case of INSERT VALUES(null),(3763),(null):
      the last NULL needs to insert 3764, not the value of the first NULL plus
      1).
2993
      Ignore negative values.
2994
    */
2995 2996
    if ((longlong) nr > 0 || (table->next_number_field->flags & UNSIGNED_FLAG))
      adjust_next_insert_id_after_explicit_value(nr);
2997
    insert_id_for_cur_row= 0; // didn't generate anything
2998
    DBUG_RETURN(0);
2999
  }
3000 3001

  if ((nr= next_insert_id) >= auto_inc_interval_for_cur_row.maximum())
3002
  {
3003 3004 3005 3006 3007 3008 3009 3010 3011
    /* next_insert_id is beyond what is reserved, so we reserve more. */
    const Discrete_interval *forced=
      thd->auto_inc_intervals_forced.get_next();
    if (forced != NULL)
    {
      nr= forced->minimum();
      nb_reserved_values= forced->values();
    }
    else
3012
    {
3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027
      /*
        handler::estimation_rows_to_insert was set by
        handler::ha_start_bulk_insert(); if 0 it means "unknown".
      */
      ulonglong nb_desired_values;
      /*
        If an estimation was given to the engine:
        - use it.
        - if we already reserved numbers, it means the estimation was
        not accurate, then we'll reserve 2*AUTO_INC_DEFAULT_NB_ROWS the 2nd
        time, twice that the 3rd time etc.
        If no estimation was given, use those increasing defaults from the
        start, starting from AUTO_INC_DEFAULT_NB_ROWS.
        Don't go beyond a max to not reserve "way too much" (because
        reservation means potentially losing unused values).
Guilhem Bichot's avatar
Guilhem Bichot committed
3028
        Note that in prelocked mode no estimation is given.
3029
      */
3030

Guilhem Bichot's avatar
Guilhem Bichot committed
3031
      if ((auto_inc_intervals_count == 0) && (estimation_rows_to_insert > 0))
3032
        nb_desired_values= estimation_rows_to_insert;
3033 3034 3035 3036 3037 3038 3039 3040 3041 3042
      else if ((auto_inc_intervals_count == 0) &&
               (thd->lex->many_values.elements > 0))
      {
        /*
          For multi-row inserts, if the bulk inserts cannot be started, the
          handler::estimation_rows_to_insert will not be set. But we still
          want to reserve the autoinc values.
        */
        nb_desired_values= thd->lex->many_values.elements;
      }
3043
      else /* go with the increasing defaults */
3044
      {
3045
        /* avoid overflow in formula, with this if() */
Guilhem Bichot's avatar
Guilhem Bichot committed
3046
        if (auto_inc_intervals_count <= AUTO_INC_DEFAULT_NB_MAX_BITS)
3047
        {
Guilhem Bichot's avatar
Guilhem Bichot committed
3048 3049
          nb_desired_values= AUTO_INC_DEFAULT_NB_ROWS *
            (1 << auto_inc_intervals_count);
3050 3051 3052 3053
          set_if_smaller(nb_desired_values, AUTO_INC_DEFAULT_NB_MAX);
        }
        else
          nb_desired_values= AUTO_INC_DEFAULT_NB_MAX;
3054
      }
3055 3056 3057 3058
      get_auto_increment(variables->auto_increment_offset,
                         variables->auto_increment_increment,
                         nb_desired_values, &nr,
                         &nb_reserved_values);
3059
      if (nr == ULONGLONG_MAX)
3060
        DBUG_RETURN(HA_ERR_AUTOINC_READ_FAILED);  // Mark failure
Guilhem Bichot's avatar
Guilhem Bichot committed
3061

3062 3063 3064 3065 3066 3067 3068 3069 3070 3071
      /*
        That rounding below should not be needed when all engines actually
        respect offset and increment in get_auto_increment(). But they don't
        so we still do it. Wonder if for the not-first-in-index we should do
        it. Hope that this rounding didn't push us out of the interval; even
        if it did we cannot do anything about it (calling the engine again
        will not help as we inserted no row).
      */
      nr= compute_next_insert_id(nr-1, variables);
    }
Guilhem Bichot's avatar
Guilhem Bichot committed
3072

3073
    if (table->s->next_number_keypart == 0)
3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086
    {
      /* We must defer the appending until "nr" has been possibly truncated */
      append= TRUE;
    }
    else
    {
      /*
        For such auto_increment there is no notion of interval, just a
        singleton. The interval is not even stored in
        thd->auto_inc_interval_for_cur_row, so we are sure to call the engine
        for next row.
      */
      DBUG_PRINT("info",("auto_increment: special not-first-in-index"));
3087
    }
3088 3089
  }

3090
  if (unlikely(nr == ULONGLONG_MAX))
3091
      DBUG_RETURN(HA_ERR_AUTOINC_ERANGE);
3092

3093
  DBUG_PRINT("info",("auto_increment: %llu  nb_reserved_values: %llu",
3094
                     nr, append ? nb_reserved_values : 0));
3095

3096 3097 3098 3099 3100
  /* Store field without warning (Warning will be printed by insert) */
  save_count_cuted_fields= thd->count_cuted_fields;
  thd->count_cuted_fields= CHECK_FIELD_IGNORE;
  tmp= table->next_number_field->store((longlong) nr, TRUE);
  thd->count_cuted_fields= save_count_cuted_fields;
3101

3102 3103
  if (unlikely(tmp))                            // Out of range value in store
  {
3104
    /*
3105 3106
      It's better to return an error here than getting a confusing
      'duplicate key error' later.
3107
    */
3108
    result= HA_ERR_AUTOINC_ERANGE;
3109 3110 3111 3112 3113
  }
  if (append)
  {
    auto_inc_interval_for_cur_row.replace(nr, nb_reserved_values,
                                          variables->auto_increment_increment);
Guilhem Bichot's avatar
Guilhem Bichot committed
3114
    auto_inc_intervals_count++;
3115
    /* Row-based replication does not need to store intervals in binlog */
3116
    if (((WSREP(thd) && wsrep_emulate_bin_log ) || mysql_bin_log.is_open())
3117 3118 3119 3120 3121
        && !thd->is_current_stmt_binlog_format_row())
      thd->auto_inc_intervals_in_cur_stmt_for_binlog.
        append(auto_inc_interval_for_cur_row.minimum(),
               auto_inc_interval_for_cur_row.values(),
               variables->auto_increment_increment);
unknown's avatar
unknown committed
3122
  }
3123

3124 3125 3126 3127 3128 3129 3130 3131
  /*
    Record this autogenerated value. If the caller then
    succeeds to insert this value, it will call
    record_first_successful_insert_id_in_cur_stmt()
    which will set first_successful_insert_id_in_cur_stmt if it's not
    already set.
  */
  insert_id_for_cur_row= nr;
3132 3133 3134 3135

  if (result)                                   // overflow
    DBUG_RETURN(result);

3136 3137 3138 3139 3140
  /*
    Set next insert id to point to next auto-increment value to be able to
    handle multi-row statements.
  */
  set_next_insert_id(compute_next_insert_id(nr, variables));
3141

unknown's avatar
unknown committed
3142
  DBUG_RETURN(0);
3143 3144
}

unknown's avatar
unknown committed
3145

3146
/** @brief
3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161
  MySQL signal that it changed the column bitmap

  USAGE
    This is for handlers that needs to setup their own column bitmaps.
    Normally the handler should set up their own column bitmaps in
    index_init() or rnd_init() and in any column_bitmaps_signal() call after
    this.

    The handler is allowd to do changes to the bitmap after a index_init or
    rnd_init() call is made as after this, MySQL will not use the bitmap
    for any program logic checking.
*/
void handler::column_bitmaps_signal()
{
  DBUG_ENTER("column_bitmaps_signal");
3162 3163 3164
  if (table)
    DBUG_PRINT("info", ("read_set: 0x%lx  write_set: 0x%lx",
                        (long) table->read_set, (long) table->write_set));
3165 3166 3167 3168
  DBUG_VOID_RETURN;
}


3169
/** @brief
3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189
  Reserves an interval of auto_increment values from the handler.

  SYNOPSIS
    get_auto_increment()
    offset              
    increment
    nb_desired_values   how many values we want
    first_value         (OUT) the first value reserved by the handler
    nb_reserved_values  (OUT) how many values the handler reserved

  offset and increment means that we want values to be of the form
  offset + N * increment, where N>=0 is integer.
  If the function sets *first_value to ~(ulonglong)0 it means an error.
  If the function sets *nb_reserved_values to ULONGLONG_MAX it means it has
  reserved to "positive infinite".
*/
void handler::get_auto_increment(ulonglong offset, ulonglong increment,
                                 ulonglong nb_desired_values,
                                 ulonglong *first_value,
                                 ulonglong *nb_reserved_values)
unknown's avatar
unknown committed
3190
{
3191
  ulonglong nr;
unknown's avatar
unknown committed
3192
  int error;
unknown's avatar
unknown committed
3193

unknown's avatar
unknown committed
3194
  (void) extra(HA_EXTRA_KEYREAD);
3195 3196 3197
  table->mark_columns_used_by_index_no_reset(table->s->next_number_index,
                                        table->read_set);
  column_bitmaps_signal();
3198 3199 3200 3201 3202

  if (ha_index_init(table->s->next_number_index, 1))
  {
    /* This should never happen, assert in debug, and fail in release build */
    DBUG_ASSERT(0);
3203
    (void) extra(HA_EXTRA_NO_KEYREAD);
3204 3205 3206 3207
    *first_value= ULONGLONG_MAX;
    return;
  }

3208
  if (table->s->next_number_keypart == 0)
unknown's avatar
unknown committed
3209
  {						// Autoincrement at key-start
3210
    error= ha_index_last(table->record[1]);
3211 3212 3213 3214 3215 3216
    /*
      MySQL implicitely assumes such method does locking (as MySQL decides to
      use nr+increment without checking again with the handler, in
      handler::update_auto_increment()), so reserves to infinite.
    */
    *nb_reserved_values= ULONGLONG_MAX;
unknown's avatar
unknown committed
3217 3218 3219
  }
  else
  {
3220
    uchar key[MAX_KEY_LENGTH];
3221
    key_copy(key, table->record[0],
3222 3223
             table->key_info + table->s->next_number_index,
             table->s->next_number_key_offset);
3224 3225 3226 3227
    error= ha_index_read_map(table->record[1], key,
                             make_prev_keypart_map(table->s->
                                                   next_number_keypart),
                             HA_READ_PREFIX_LAST);
3228 3229 3230 3231 3232 3233 3234
    /*
      MySQL needs to call us for next row: assume we are inserting ("a",null)
      here, we return 3, and next this statement will want to insert
      ("b",null): there is no reason why ("b",3+1) would be the good row to
      insert: maybe it already exists, maybe 3+1 is too large...
    */
    *nb_reserved_values= 1;
unknown's avatar
unknown committed
3235 3236
  }

unknown's avatar
unknown committed
3237
  if (error)
3238 3239
  {
    if (error == HA_ERR_END_OF_FILE || error == HA_ERR_KEY_NOT_FOUND)
3240
      /* No entry found, that's fine */;
3241
    else
3242 3243
      print_error(error, MYF(0));
    nr= 1;
3244
  }
unknown's avatar
unknown committed
3245
  else
3246 3247
    nr= ((ulonglong) table->next_number_field->
         val_int_offset(table->s->rec_buff_length)+1);
3248
  ha_index_end();
unknown's avatar
unknown committed
3249
  (void) extra(HA_EXTRA_NO_KEYREAD);
3250
  *first_value= nr;
3251
  return;
unknown's avatar
unknown committed
3252 3253
}

3254

3255 3256
void handler::ha_release_auto_increment()
{
3257
  DBUG_ENTER("ha_release_auto_increment");
3258 3259 3260
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type != F_UNLCK ||
              (!next_insert_id && !insert_id_for_cur_row));
3261 3262 3263
  release_auto_increment();
  insert_id_for_cur_row= 0;
  auto_inc_interval_for_cur_row.replace(0, 0, 0);
Guilhem Bichot's avatar
Guilhem Bichot committed
3264
  auto_inc_intervals_count= 0;
3265 3266 3267 3268 3269 3270 3271 3272 3273
  if (next_insert_id > 0)
  {
    next_insert_id= 0;
    /*
      this statement used forced auto_increment values if there were some,
      wipe them away for other statements.
    */
    table->in_use->auto_inc_intervals_forced.empty();
  }
3274
  DBUG_VOID_RETURN;
3275 3276 3277
}


3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290
/**
  Construct and emit duplicate key error message using information
  from table's record buffer.

  @param table    TABLE object which record buffer should be used as
                  source for column values.
  @param key      Key description.
  @param msg      Error message template to which key value should be
                  added.
  @param errflag  Flags for my_error() call.
*/

void print_keydup_error(TABLE *table, KEY *key, const char *msg, myf errflag)
3291 3292
{
  /* Write the duplicated key in the error message */
3293 3294
  char key_buff[MAX_KEY_LENGTH];
  String str(key_buff,sizeof(key_buff),system_charset_info);
3295

3296
  if (key == NULL)
3297 3298
  {
    /* Key is unknown */
3299
    str.copy("", 0, system_charset_info);
3300
    my_printf_error(ER_DUP_ENTRY, msg, errflag, str.c_ptr(), "*UNKNOWN*");
3301 3302
  }
  else
3303
  {
3304
    /* Table is opened and defined at this point */
3305
    key_unpack(&str,table, key);
3306 3307 3308 3309 3310 3311
    uint max_length=MYSQL_ERRMSG_SIZE-(uint) strlen(msg);
    if (str.length() >= max_length)
    {
      str.length(max_length-4);
      str.append(STRING_WITH_LEN("..."));
    }
3312
    my_printf_error(ER_DUP_ENTRY, msg, errflag, str.c_ptr_safe(), key->name);
3313 3314 3315
  }
}

3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327
/**
  Construct and emit duplicate key error message using information
  from table's record buffer.

  @sa print_keydup_error(table, key, msg, errflag).
*/

void print_keydup_error(TABLE *table, KEY *key, myf errflag)
{
  print_keydup_error(table, key, ER(ER_DUP_ENTRY_WITH_KEY_NAME), errflag);
}

3328

unknown's avatar
unknown committed
3329 3330
/**
  Print error that we got from handler function.
3331

unknown's avatar
unknown committed
3332 3333 3334 3335 3336
  @note
    In case of delete table it's only safe to use the following parts of
    the 'table' structure:
    - table->s->path
    - table->alias
3337
*/
3338 3339 3340

#define SET_FATAL_ERROR fatal_error=1

unknown's avatar
unknown committed
3341 3342
void handler::print_error(int error, myf errflag)
{
3343
  bool fatal_error= 0;
3344
  DBUG_ENTER("handler::print_error");
unknown's avatar
unknown committed
3345 3346
  DBUG_PRINT("enter",("error: %d",error));

3347
  int textno= -1; // impossible value
unknown's avatar
unknown committed
3348
  switch (error) {
3349 3350 3351
  case EACCES:
    textno=ER_OPEN_AS_READONLY;
    break;
unknown's avatar
unknown committed
3352 3353 3354 3355 3356 3357
  case EAGAIN:
    textno=ER_FILE_USED;
    break;
  case ENOENT:
    textno=ER_FILE_NOT_FOUND;
    break;
3358 3359 3360 3361 3362
  case ENOSPC:
  case HA_ERR_DISK_FULL:
    textno= ER_DISK_FULL;
    SET_FATAL_ERROR;                            // Ensure error is logged
    break;
unknown's avatar
unknown committed
3363 3364
  case HA_ERR_KEY_NOT_FOUND:
  case HA_ERR_NO_ACTIVE_RECORD:
3365
  case HA_ERR_RECORD_DELETED:
unknown's avatar
unknown committed
3366
  case HA_ERR_END_OF_FILE:
3367 3368 3369 3370 3371 3372 3373
    /*
      This errors is not not normally fatal (for example for reads). However
      if you get it during an update or delete, then its fatal.
      As the user is calling print_error() (which is not done on read), we
      assume something when wrong with the update or delete.
    */
    SET_FATAL_ERROR;
unknown's avatar
unknown committed
3374 3375
    textno=ER_KEY_NOT_FOUND;
    break;
3376 3377 3378 3379 3380 3381
  case HA_ERR_ABORTED_BY_USER:
  {
    DBUG_ASSERT(table->in_use->killed);
    table->in_use->send_kill_message();
    DBUG_VOID_RETURN;
  }
unknown's avatar
unknown committed
3382
  case HA_ERR_WRONG_MRG_TABLE_DEF:
unknown's avatar
unknown committed
3383 3384 3385 3386
    textno=ER_WRONG_MRG_TABLE;
    break;
  case HA_ERR_FOUND_DUPP_KEY:
  {
3387
    if (table)
unknown's avatar
unknown committed
3388
    {
3389 3390 3391
      uint key_nr=get_dup_key(error);
      if ((int) key_nr >= 0)
      {
3392 3393 3394
        print_keydup_error(table,
                           key_nr == MAX_KEY ? NULL : &table->key_info[key_nr],
                           errflag);
3395 3396
        DBUG_VOID_RETURN;
      }
unknown's avatar
unknown committed
3397 3398 3399 3400
    }
    textno=ER_DUP_KEY;
    break;
  }
3401 3402
  case HA_ERR_FOREIGN_DUPLICATE_KEY:
  {
Sergei Golubchik's avatar
Sergei Golubchik committed
3403 3404 3405
    char rec_buf[MAX_KEY_LENGTH];
    String rec(rec_buf, sizeof(rec_buf), system_charset_info);
    /* Table is opened and defined at this point */
3406 3407 3408 3409 3410 3411

    /*
      Just print the subset of fields that are part of the first index,
      printing the whole row from there is not easy.
    */
    key_unpack(&rec, table, &table->key_info[0]);
Sergei Golubchik's avatar
Sergei Golubchik committed
3412 3413 3414 3415 3416

    char child_table_name[NAME_LEN + 1];
    char child_key_name[NAME_LEN + 1];
    if (get_foreign_dup_key(child_table_name, sizeof(child_table_name),
                            child_key_name, sizeof(child_key_name)))
3417
    {
Sergei Golubchik's avatar
Sergei Golubchik committed
3418 3419 3420
      my_error(ER_FOREIGN_DUPLICATE_KEY_WITH_CHILD_INFO, errflag,
               table_share->table_name.str, rec.c_ptr_safe(),
               child_table_name, child_key_name);
3421
      }
Sergei Golubchik's avatar
Sergei Golubchik committed
3422 3423 3424 3425
    else
    {
      my_error(ER_FOREIGN_DUPLICATE_KEY_WITHOUT_CHILD_INFO, errflag,
               table_share->table_name.str, rec.c_ptr_safe());
3426
    }
Sergei Golubchik's avatar
Sergei Golubchik committed
3427
    DBUG_VOID_RETURN;
3428
  }
3429
  case HA_ERR_NULL_IN_SPATIAL:
Michael Widenius's avatar
Michael Widenius committed
3430
    my_error(ER_CANT_CREATE_GEOMETRY_OBJECT, errflag);
3431
    DBUG_VOID_RETURN;
unknown's avatar
unknown committed
3432 3433 3434 3435
  case HA_ERR_FOUND_DUPP_UNIQUE:
    textno=ER_DUP_UNIQUE;
    break;
  case HA_ERR_RECORD_CHANGED:
3436 3437 3438 3439
    /*
      This is not fatal error when using HANDLER interface
      SET_FATAL_ERROR;
    */
unknown's avatar
unknown committed
3440 3441 3442
    textno=ER_CHECKREAD;
    break;
  case HA_ERR_CRASHED:
3443
    SET_FATAL_ERROR;
unknown's avatar
unknown committed
3444 3445
    textno=ER_NOT_KEYFILE;
    break;
3446
  case HA_ERR_WRONG_IN_RECORD:
3447
    SET_FATAL_ERROR;
3448 3449
    textno= ER_CRASHED_ON_USAGE;
    break;
3450
  case HA_ERR_CRASHED_ON_USAGE:
3451
    SET_FATAL_ERROR;
3452 3453
    textno=ER_CRASHED_ON_USAGE;
    break;
3454 3455 3456
  case HA_ERR_NOT_A_TABLE:
    textno= error;
    break;
3457
  case HA_ERR_CRASHED_ON_REPAIR:
3458
    SET_FATAL_ERROR;
3459 3460
    textno=ER_CRASHED_ON_REPAIR;
    break;
unknown's avatar
unknown committed
3461
  case HA_ERR_OUT_OF_MEM:
3462 3463
    textno=ER_OUT_OF_RESOURCES;
    break;
unknown's avatar
unknown committed
3464
  case HA_ERR_WRONG_COMMAND:
3465 3466 3467
    my_error(ER_ILLEGAL_HA, MYF(0), table_type(), table_share->db.str,
             table_share->table_name.str);
    DBUG_VOID_RETURN;
unknown's avatar
unknown committed
3468 3469 3470 3471 3472 3473 3474 3475
    break;
  case HA_ERR_OLD_FILE:
    textno=ER_OLD_KEYFILE;
    break;
  case HA_ERR_UNSUPPORTED:
    textno=ER_UNSUPPORTED_EXTENSION;
    break;
  case HA_ERR_RECORD_FILE_FULL:
3476
  case HA_ERR_INDEX_FILE_FULL:
3477
  {
3478
    textno=ER_RECORD_FILE_FULL;
3479 3480
    /* Write the error message to error log */
    errflag|= ME_NOREFRESH;
3481
    break;
3482
  }
3483 3484 3485 3486 3487 3488
  case HA_ERR_LOCK_WAIT_TIMEOUT:
    textno=ER_LOCK_WAIT_TIMEOUT;
    break;
  case HA_ERR_LOCK_TABLE_FULL:
    textno=ER_LOCK_TABLE_FULL;
    break;
3489 3490
  case HA_ERR_LOCK_DEADLOCK:
    textno=ER_LOCK_DEADLOCK;
3491 3492
    /* cannot continue. the statement was already aborted in the engine */
    SET_FATAL_ERROR;
3493
    break;
3494 3495 3496
  case HA_ERR_READ_ONLY_TRANSACTION:
    textno=ER_READ_ONLY_TRANSACTION;
    break;
unknown's avatar
Merge  
unknown committed
3497 3498 3499 3500
  case HA_ERR_CANNOT_ADD_FOREIGN:
    textno=ER_CANNOT_ADD_FOREIGN;
    break;
  case HA_ERR_ROW_IS_REFERENCED:
3501 3502 3503
  {
    String str;
    get_error_message(error, &str);
Michael Widenius's avatar
Michael Widenius committed
3504
    my_error(ER_ROW_IS_REFERENCED_2, errflag, str.c_ptr_safe());
3505 3506
    DBUG_VOID_RETURN;
  }
unknown's avatar
Merge  
unknown committed
3507
  case HA_ERR_NO_REFERENCED_ROW:
3508 3509 3510
  {
    String str;
    get_error_message(error, &str);
Michael Widenius's avatar
Michael Widenius committed
3511
    my_error(ER_NO_REFERENCED_ROW_2, errflag, str.c_ptr_safe());
3512 3513
    DBUG_VOID_RETURN;
  }
3514 3515 3516
  case HA_ERR_TABLE_DEF_CHANGED:
    textno=ER_TABLE_DEF_CHANGED;
    break;
3517
  case HA_ERR_NO_SUCH_TABLE:
3518
    my_error(ER_NO_SUCH_TABLE_IN_ENGINE, errflag, table_share->db.str,
unknown's avatar
unknown committed
3519
             table_share->table_name.str);
3520
    DBUG_VOID_RETURN;
3521 3522 3523
  case HA_ERR_RBR_LOGGING_FAILED:
    textno= ER_BINLOG_ROW_LOGGING_FAILED;
    break;
3524 3525 3526 3527 3528 3529
  case HA_ERR_DROP_INDEX_FK:
  {
    const char *ptr= "???";
    uint key_nr= get_dup_key(error);
    if ((int) key_nr >= 0)
      ptr= table->key_info[key_nr].name;
Michael Widenius's avatar
Michael Widenius committed
3530
    my_error(ER_DROP_INDEX_FK, errflag, ptr);
3531 3532
    DBUG_VOID_RETURN;
  }
unknown's avatar
unknown committed
3533 3534 3535
  case HA_ERR_TABLE_NEEDS_UPGRADE:
    textno=ER_TABLE_NEEDS_UPGRADE;
    break;
3536 3537 3538
  case HA_ERR_NO_PARTITION_FOUND:
    textno=ER_WRONG_PARTITION_NAME;
    break;
3539 3540 3541
  case HA_ERR_TABLE_READONLY:
    textno= ER_OPEN_AS_READONLY;
    break;
unknown's avatar
unknown committed
3542 3543 3544 3545
  case HA_ERR_AUTOINC_READ_FAILED:
    textno= ER_AUTOINC_READ_FAILED;
    break;
  case HA_ERR_AUTOINC_ERANGE:
3546 3547
    textno= error;
    my_error(textno, errflag, table->next_number_field->field_name,
3548
             table->in_use->get_stmt_da()->current_row_for_warning());
3549
    DBUG_VOID_RETURN;
unknown's avatar
unknown committed
3550
    break;
3551 3552 3553
  case HA_ERR_TOO_MANY_CONCURRENT_TRXS:
    textno= ER_TOO_MANY_CONCURRENT_TRXS;
    break;
3554 3555 3556
  case HA_ERR_INDEX_COL_TOO_LONG:
    textno= ER_INDEX_COLUMN_TOO_LONG;
    break;
3557 3558 3559
  case HA_ERR_NOT_IN_LOCK_PARTITIONS:
    textno=ER_ROW_DOES_NOT_MATCH_GIVEN_PARTITION_SET;
    break;
3560 3561 3562
  case HA_ERR_INDEX_CORRUPT:
    textno= ER_INDEX_CORRUPT;
    break;
3563 3564 3565
  case HA_ERR_UNDO_REC_TOO_BIG:
    textno= ER_UNDO_RECORD_TOO_BIG;
    break;
3566 3567 3568
  case HA_ERR_TABLE_IN_FK_CHECK:
    textno= ER_TABLE_IN_FK_CHECK;
    break;
unknown's avatar
unknown committed
3569 3570
  default:
    {
3571 3572 3573
      /* The error was "unknown" to this function.
	 Ask handler if it has got a message for this error */
      bool temporary= FALSE;
3574 3575 3576
      String str;
      temporary= get_error_message(error, &str);
      if (!str.is_empty())
3577
      {
3578
	const char* engine= table_type();
3579
	if (temporary)
Michael Widenius's avatar
Michael Widenius committed
3580
	  my_error(ER_GET_TEMPORARY_ERRMSG, errflag, error, str.c_ptr(),
3581
                   engine);
3582
	else
3583 3584
        {
          SET_FATAL_ERROR;
Michael Widenius's avatar
Michael Widenius committed
3585
	  my_error(ER_GET_ERRMSG, errflag, error, str.c_ptr(), engine);
3586
        }
3587
      }
3588
      else
Sergei Golubchik's avatar
Sergei Golubchik committed
3589
        my_error(ER_GET_ERRNO, errflag, error, table_type());
unknown's avatar
unknown committed
3590 3591 3592
      DBUG_VOID_RETURN;
    }
  }
3593
  DBUG_ASSERT(textno > 0);
Michael Widenius's avatar
Michael Widenius committed
3594
  if (fatal_error)
3595
  {
Michael Widenius's avatar
Michael Widenius committed
3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606
    /* Ensure this becomes a true error */
    errflag&= ~(ME_JUST_WARNING | ME_JUST_INFO);
    if ((debug_assert_if_crashed_table ||
                      global_system_variables.log_warnings > 1))
    {
      /*
        Log error to log before we crash or if extended warnings are requested
      */
      errflag|= ME_NOREFRESH;
    }
  }    
3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617

  /* if we got an OS error from a file-based engine, specify a path of error */
  if (error < HA_ERR_FIRST && bas_ext()[0])
  {
    char buff[FN_REFLEN];
    strxnmov(buff, sizeof(buff),
             table_share->normalized_path.str, bas_ext()[0], NULL);
    my_error(textno, errflag, buff, error);
  }
  else
    my_error(textno, errflag, table_share->table_name.str, error);
unknown's avatar
unknown committed
3618 3619 3620
  DBUG_VOID_RETURN;
}

unknown's avatar
unknown committed
3621

unknown's avatar
unknown committed
3622 3623
/**
  Return an error message specific to this handler.
3624

unknown's avatar
unknown committed
3625 3626
  @param error  error code previously returned by handler
  @param buf    pointer to String where to add error message
unknown's avatar
unknown committed
3627

unknown's avatar
unknown committed
3628 3629
  @return
    Returns true if this is a temporary error
3630
*/
3631
bool handler::get_error_message(int error, String* buf)
3632
{
unknown's avatar
unknown committed
3633
  return FALSE;
3634 3635
}

3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648
/**
  Check for incompatible collation changes.
   
  @retval
    HA_ADMIN_NEEDS_UPGRADE   Table may have data requiring upgrade.
  @retval
    0                        No upgrade required.
*/

int handler::check_collation_compatibility()
{
  ulong mysql_version= table->s->mysql_version;

3649
  if (mysql_version < 50124)
3650 3651 3652 3653 3654 3655
  {
    KEY *key= table->key_info;
    KEY *key_end= key + table->s->keys;
    for (; key < key_end; key++)
    {
      KEY_PART_INFO *key_part= key->key_part;
3656
      KEY_PART_INFO *key_part_end= key_part + key->user_defined_key_parts;
3657 3658 3659 3660 3661 3662
      for (; key_part < key_part_end; key_part++)
      {
        if (!key_part->fieldnr)
          continue;
        Field *field= table->field[key_part->fieldnr - 1];
        uint cs_number= field->charset()->number;
3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673 3674
        if ((mysql_version < 50048 &&
             (cs_number == 11 || /* ascii_general_ci - bug #29499, bug #27562 */
              cs_number == 41 || /* latin7_general_ci - bug #29461 */
              cs_number == 42 || /* latin7_general_cs - bug #29461 */
              cs_number == 20 || /* latin7_estonian_cs - bug #29461 */
              cs_number == 21 || /* latin2_hungarian_ci - bug #29461 */
              cs_number == 22 || /* koi8u_general_ci - bug #29461 */
              cs_number == 23 || /* cp1251_ukrainian_ci - bug #29461 */
              cs_number == 26)) || /* cp1250_general_ci - bug #29461 */
             (mysql_version < 50124 &&
             (cs_number == 33 || /* utf8_general_ci - bug #27877 */
              cs_number == 35))) /* ucs2_general_ci - bug #27877 */
3675
          return HA_ADMIN_NEEDS_UPGRADE;
3676 3677 3678 3679
      }
    }
  }

3680 3681 3682 3683
  return 0;
}


unknown's avatar
unknown committed
3684 3685
int handler::ha_check_for_upgrade(HA_CHECK_OPT *check_opt)
{
3686
  int error;
unknown's avatar
unknown committed
3687 3688 3689
  KEY *keyinfo, *keyend;
  KEY_PART_INFO *keypart, *keypartend;

3690 3691 3692
  if (table->s->incompatible_version)
    return HA_ADMIN_NEEDS_ALTER;

unknown's avatar
unknown committed
3693 3694 3695 3696 3697 3698 3699 3700
  if (!table->s->mysql_version)
  {
    /* check for blob-in-key error */
    keyinfo= table->key_info;
    keyend= table->key_info + table->s->keys;
    for (; keyinfo < keyend; keyinfo++)
    {
      keypart= keyinfo->key_part;
3701
      keypartend= keypart + keyinfo->user_defined_key_parts;
unknown's avatar
unknown committed
3702 3703 3704 3705 3706
      for (; keypart < keypartend; keypart++)
      {
        if (!keypart->fieldnr)
          continue;
        Field *field= table->field[keypart->fieldnr-1];
3707
        if (field->type() == MYSQL_TYPE_BLOB)
unknown's avatar
unknown committed
3708 3709 3710 3711 3712 3713 3714 3715
        {
          if (check_opt->sql_flags & TT_FOR_UPGRADE)
            check_opt->flags= T_MEDIUM;
          return HA_ADMIN_NEEDS_CHECK;
        }
      }
    }
  }
3716 3717
  if (table->s->frm_version != FRM_VER_TRUE_VARCHAR)
    return HA_ADMIN_NEEDS_ALTER;
3718 3719 3720 3721

  if ((error= check_collation_compatibility()))
    return error;
    
unknown's avatar
unknown committed
3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734
  return check_for_upgrade(check_opt);
}


int handler::check_old_types()
{
  Field** field;

  if (!table->s->mysql_version)
  {
    /* check for bad DECIMAL field */
    for (field= table->field; (*field); field++)
    {
3735
      if ((*field)->type() == MYSQL_TYPE_NEWDECIMAL)
unknown's avatar
unknown committed
3736 3737 3738
      {
        return HA_ADMIN_NEEDS_ALTER;
      }
3739 3740 3741 3742
      if ((*field)->type() == MYSQL_TYPE_VAR_STRING)
      {
        return HA_ADMIN_NEEDS_ALTER;
      }
unknown's avatar
unknown committed
3743 3744 3745 3746 3747 3748
    }
  }
  return 0;
}


3749
static bool update_frm_version(TABLE *table)
unknown's avatar
unknown committed
3750 3751 3752 3753 3754 3755
{
  char path[FN_REFLEN];
  File file;
  int result= 1;
  DBUG_ENTER("update_frm_version");

3756 3757 3758 3759 3760 3761 3762
  /*
    No need to update frm version in case table was created or checked
    by server with the same version. This also ensures that we do not
    update frm version for temporary tables as this code doesn't support
    temporary tables.
  */
  if (table->s->mysql_version == MYSQL_VERSION_ID)
unknown's avatar
unknown committed
3763 3764
    DBUG_RETURN(0);

unknown's avatar
unknown committed
3765
  strxmov(path, table->s->normalized_path.str, reg_ext, NullS);
unknown's avatar
unknown committed
3766

Marc Alff's avatar
Marc Alff committed
3767 3768
  if ((file= mysql_file_open(key_file_frm,
                             path, O_RDWR|O_BINARY, MYF(MY_WME))) >= 0)
unknown's avatar
unknown committed
3769 3770 3771 3772 3773
  {
    uchar version[4];

    int4store(version, MYSQL_VERSION_ID);

Marc Alff's avatar
Marc Alff committed
3774
    if ((result= mysql_file_pwrite(file, (uchar*) version, 4, 51L, MYF_RW)))
unknown's avatar
unknown committed
3775 3776
      goto err;

3777
    table->s->mysql_version= MYSQL_VERSION_ID;
unknown's avatar
unknown committed
3778 3779 3780
  }
err:
  if (file >= 0)
Marc Alff's avatar
Marc Alff committed
3781
    (void) mysql_file_close(file, MYF(MY_WME));
unknown's avatar
unknown committed
3782 3783 3784 3785 3786
  DBUG_RETURN(result);
}



unknown's avatar
unknown committed
3787 3788 3789 3790
/**
  @return
    key if error because of duplicated keys
*/
unknown's avatar
unknown committed
3791 3792
uint handler::get_dup_key(int error)
{
3793 3794
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type != F_UNLCK);
3795
  DBUG_ENTER("handler::get_dup_key");
unknown's avatar
unknown committed
3796
  table->file->errkey  = (uint) -1;
3797 3798 3799
  if (error == HA_ERR_FOUND_DUPP_KEY || error == HA_ERR_FOREIGN_DUPLICATE_KEY ||
      error == HA_ERR_FOUND_DUPP_UNIQUE || error == HA_ERR_NULL_IN_SPATIAL ||
      error == HA_ERR_DROP_INDEX_FK)
3800
    table->file->info(HA_STATUS_ERRKEY | HA_STATUS_NO_LOCK);
unknown's avatar
unknown committed
3801 3802 3803
  DBUG_RETURN(table->file->errkey);
}

unknown's avatar
unknown committed
3804

unknown's avatar
unknown committed
3805 3806
/**
  Delete all files with extension from bas_ext().
3807

unknown's avatar
unknown committed
3808
  @param name		Base name of table
3809

unknown's avatar
unknown committed
3810
  @note
3811 3812 3813
    We assume that the handler may return more extensions than
    was actually used for the file.

unknown's avatar
unknown committed
3814
  @retval
3815
    0   If we successfully deleted at least one file from base_ext and
unknown's avatar
unknown committed
3816 3817 3818
    didn't get any other errors than ENOENT
  @retval
    !0  Error
3819
*/
unknown's avatar
unknown committed
3820 3821
int handler::delete_table(const char *name)
{
3822
  int saved_error= 0;
3823
  int error= 0;
3824
  int enoent_or_zero;
3825
  char buff[FN_REFLEN];
3826

3827 3828 3829 3830 3831
  if (ht->discover_table)
    enoent_or_zero= 0; // the table may not exist in the engine, it's ok
  else
    enoent_or_zero= ENOENT;  // the first file of bas_ext() *must* exist

unknown's avatar
unknown committed
3832 3833
  for (const char **ext=bas_ext(); *ext ; ext++)
  {
3834
    fn_format(buff, name, "", *ext, MY_UNPACK_FILENAME|MY_APPEND_EXT);
Marc Alff's avatar
Marc Alff committed
3835
    if (mysql_file_delete_with_symlink(key_file_misc, buff, MYF(0)))
unknown's avatar
unknown committed
3836
    {
3837 3838 3839 3840 3841 3842 3843 3844 3845 3846
      if (my_errno != ENOENT)
      {
        /*
          If error on the first existing file, return the error.
          Otherwise delete as much as possible.
        */
        if (enoent_or_zero)
          return my_errno;
	saved_error= my_errno;
      }
unknown's avatar
unknown committed
3847
    }
3848
    else
3849
      enoent_or_zero= 0;                        // No error for ENOENT
3850
    error= enoent_or_zero;
unknown's avatar
unknown committed
3851
  }
3852
  return saved_error ? saved_error : error;
unknown's avatar
unknown committed
3853 3854 3855 3856 3857
}


int handler::rename_table(const char * from, const char * to)
{
unknown's avatar
unknown committed
3858
  int error= 0;
3859 3860 3861
  const char **ext, **start_ext;
  start_ext= bas_ext();
  for (ext= start_ext; *ext ; ext++)
unknown's avatar
unknown committed
3862
  {
unknown's avatar
unknown committed
3863 3864 3865 3866 3867 3868
    if (rename_file_ext(from, to, *ext))
    {
      if ((error=my_errno) != ENOENT)
	break;
      error= 0;
    }
unknown's avatar
unknown committed
3869
  }
3870 3871 3872 3873 3874 3875
  if (error)
  {
    /* Try to revert the rename. Ignore errors. */
    for (; ext >= start_ext; ext--)
      rename_file_ext(to, from, *ext);
  }
unknown's avatar
unknown committed
3876
  return error;
unknown's avatar
unknown committed
3877 3878
}

unknown's avatar
unknown committed
3879 3880 3881

void handler::drop_table(const char *name)
{
3882
  ha_close();
unknown's avatar
unknown committed
3883 3884 3885 3886
  delete_table(name);
}


unknown's avatar
unknown committed
3887 3888
/**
  Performs checks upon the table.
unknown's avatar
unknown committed
3889

unknown's avatar
unknown committed
3890 3891
  @param thd                thread doing CHECK TABLE operation
  @param check_opt          options from the parser
unknown's avatar
unknown committed
3892

unknown's avatar
unknown committed
3893 3894 3895 3896 3897 3898 3899 3900
  @retval
    HA_ADMIN_OK               Successful upgrade
  @retval
    HA_ADMIN_NEEDS_UPGRADE    Table has structures requiring upgrade
  @retval
    HA_ADMIN_NEEDS_ALTER      Table has structures requiring ALTER TABLE
  @retval
    HA_ADMIN_NOT_IMPLEMENTED
unknown's avatar
unknown committed
3901 3902 3903 3904
*/
int handler::ha_check(THD *thd, HA_CHECK_OPT *check_opt)
{
  int error;
3905 3906
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type != F_UNLCK);
unknown's avatar
unknown committed
3907 3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923

  if ((table->s->mysql_version >= MYSQL_VERSION_ID) &&
      (check_opt->sql_flags & TT_FOR_UPGRADE))
    return 0;

  if (table->s->mysql_version < MYSQL_VERSION_ID)
  {
    if ((error= check_old_types()))
      return error;
    error= ha_check_for_upgrade(check_opt);
    if (error && (error != HA_ADMIN_NEEDS_CHECK))
      return error;
    if (!error && (check_opt->sql_flags & TT_FOR_UPGRADE))
      return 0;
  }
  if ((error= check(thd, check_opt)))
    return error;
3924 3925 3926
  /* Skip updating frm version if not main handler. */
  if (table->file != this)
    return error;
3927
  return update_frm_version(table);
unknown's avatar
unknown committed
3928 3929
}

3930 3931 3932 3933 3934
/**
  A helper function to mark a transaction read-write,
  if it is started.
*/

3935
inline
3936
void
3937
handler::mark_trx_read_write()
3938 3939 3940 3941 3942 3943 3944 3945 3946 3947 3948 3949 3950 3951 3952 3953 3954 3955 3956 3957 3958 3959
{
  Ha_trx_info *ha_info= &ha_thd()->ha_data[ht->slot].ha_info[0];
  /*
    When a storage engine method is called, the transaction must
    have been started, unless it's a DDL call, for which the
    storage engine starts the transaction internally, and commits
    it internally, without registering in the ha_list.
    Unfortunately here we can't know know for sure if the engine
    has registered the transaction or not, so we must check.
  */
  if (ha_info->is_started())
  {
    DBUG_ASSERT(has_transactions());
    /*
      table_share can be NULL in ha_delete_table(). See implementation
      of standalone function ha_delete_table() in sql_base.cc.
    */
    if (table_share == NULL || table_share->tmp_table == NO_TMP_TABLE)
      ha_info->set_trx_read_write();
  }
}

unknown's avatar
unknown committed
3960

3961 3962 3963 3964 3965 3966
/**
  Repair table: public interface.

  @sa handler::repair()
*/

unknown's avatar
unknown committed
3967 3968 3969
int handler::ha_repair(THD* thd, HA_CHECK_OPT* check_opt)
{
  int result;
3970 3971 3972

  mark_trx_read_write();

3973 3974 3975 3976 3977 3978 3979
  result= repair(thd, check_opt);
  DBUG_ASSERT(result == HA_ADMIN_NOT_IMPLEMENTED ||
              ha_table_flags() & HA_CAN_REPAIR);

  if (result == HA_ADMIN_OK)
    result= update_frm_version(table);
  return result;
unknown's avatar
unknown committed
3980 3981 3982
}


3983 3984 3985 3986 3987 3988 3989 3990 3991 3992
/**
  Bulk update row: public interface.

  @sa handler::bulk_update_row()
*/

int
handler::ha_bulk_update_row(const uchar *old_data, uchar *new_data,
                            uint *dup_key_found)
{
3993 3994
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type == F_WRLCK);
3995 3996
  mark_trx_read_write();

3997 3998 3999 4000 4001 4002 4003 4004 4005 4006 4007 4008 4009
  return bulk_update_row(old_data, new_data, dup_key_found);
}


/**
  Delete all rows: public interface.

  @sa handler::delete_all_rows()
*/

int
handler::ha_delete_all_rows()
{
4010 4011
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type == F_WRLCK);
4012 4013
  mark_trx_read_write();

4014 4015 4016 4017
  return delete_all_rows();
}


4018 4019 4020 4021 4022 4023 4024 4025 4026
/**
  Truncate table: public interface.

  @sa handler::truncate()
*/

int
handler::ha_truncate()
{
4027 4028
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type == F_WRLCK);
4029 4030 4031 4032 4033 4034
  mark_trx_read_write();

  return truncate();
}


4035 4036 4037 4038 4039 4040 4041 4042 4043
/**
  Reset auto increment: public interface.

  @sa handler::reset_auto_increment()
*/

int
handler::ha_reset_auto_increment(ulonglong value)
{
4044 4045
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type == F_WRLCK);
4046 4047
  mark_trx_read_write();

4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059 4060
  return reset_auto_increment(value);
}


/**
  Optimize table: public interface.

  @sa handler::optimize()
*/

int
handler::ha_optimize(THD* thd, HA_CHECK_OPT* check_opt)
{
4061 4062
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type == F_WRLCK);
4063 4064
  mark_trx_read_write();

4065 4066 4067 4068 4069 4070 4071 4072 4073 4074 4075 4076 4077
  return optimize(thd, check_opt);
}


/**
  Analyze table: public interface.

  @sa handler::analyze()
*/

int
handler::ha_analyze(THD* thd, HA_CHECK_OPT* check_opt)
{
4078 4079
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type != F_UNLCK);
4080 4081
  mark_trx_read_write();

4082 4083 4084 4085 4086 4087 4088 4089 4090 4091 4092 4093 4094
  return analyze(thd, check_opt);
}


/**
  Check and repair table: public interface.

  @sa handler::check_and_repair()
*/

bool
handler::ha_check_and_repair(THD *thd)
{
4095 4096
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type == F_UNLCK);
4097 4098
  mark_trx_read_write();

4099 4100 4101 4102 4103 4104 4105 4106 4107 4108 4109 4110 4111
  return check_and_repair(thd);
}


/**
  Disable indexes: public interface.

  @sa handler::disable_indexes()
*/

int
handler::ha_disable_indexes(uint mode)
{
4112 4113
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type != F_UNLCK);
4114 4115
  mark_trx_read_write();

4116 4117 4118 4119 4120 4121 4122 4123 4124 4125 4126 4127 4128
  return disable_indexes(mode);
}


/**
  Enable indexes: public interface.

  @sa handler::enable_indexes()
*/

int
handler::ha_enable_indexes(uint mode)
{
4129 4130
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type != F_UNLCK);
4131 4132
  mark_trx_read_write();

4133 4134 4135 4136 4137 4138 4139 4140 4141 4142 4143 4144 4145
  return enable_indexes(mode);
}


/**
  Discard or import tablespace: public interface.

  @sa handler::discard_or_import_tablespace()
*/

int
handler::ha_discard_or_import_tablespace(my_bool discard)
{
4146 4147
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type == F_WRLCK);
4148 4149
  mark_trx_read_write();

4150 4151 4152 4153
  return discard_or_import_tablespace(discard);
}


4154 4155 4156 4157 4158 4159 4160 4161 4162
bool handler::ha_prepare_inplace_alter_table(TABLE *altered_table,
                                             Alter_inplace_info *ha_alter_info)
{
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type != F_UNLCK);
  mark_trx_read_write();

  return prepare_inplace_alter_table(altered_table, ha_alter_info);
}
4163 4164


4165 4166 4167 4168 4169 4170 4171 4172 4173 4174 4175 4176 4177 4178 4179
bool handler::ha_commit_inplace_alter_table(TABLE *altered_table,
                                            Alter_inplace_info *ha_alter_info,
                                            bool commit)
{
   /*
     At this point we should have an exclusive metadata lock on the table.
     The exception is if we're about to roll back changes (commit= false).
     In this case, we might be rolling back after a failed lock upgrade,
     so we could be holding the same lock level as for inplace_alter_table().
   */
   DBUG_ASSERT(ha_thd()->mdl_context.is_lock_owner(MDL_key::TABLE,
                                                   table->s->db.str,
                                                   table->s->table_name.str,
                                                   MDL_EXCLUSIVE) ||
               !commit);
4180

4181 4182 4183 4184 4185 4186 4187
   return commit_inplace_alter_table(altered_table, ha_alter_info, commit);
}


/*
   Default implementation to support in-place alter table
   and old online add/drop index API
4188 4189
*/

4190 4191 4192
enum_alter_inplace_result
handler::check_if_supported_inplace_alter(TABLE *altered_table,
                                          Alter_inplace_info *ha_alter_info)
4193
{
4194 4195 4196
  DBUG_ENTER("check_if_supported_alter");

  HA_CREATE_INFO *create_info= ha_alter_info->create_info;
4197

4198 4199 4200 4201
  Alter_inplace_info::HA_ALTER_FLAGS inplace_offline_operations=
    Alter_inplace_info::ALTER_COLUMN_EQUAL_PACK_LENGTH |
    Alter_inplace_info::ALTER_COLUMN_NAME |
    Alter_inplace_info::ALTER_COLUMN_DEFAULT |
4202
    Alter_inplace_info::ALTER_COLUMN_OPTION |
4203 4204 4205 4206 4207 4208 4209 4210 4211 4212 4213 4214 4215 4216 4217 4218 4219 4220 4221 4222 4223 4224 4225 4226 4227 4228 4229 4230 4231
    Alter_inplace_info::CHANGE_CREATE_OPTION |
    Alter_inplace_info::ALTER_RENAME;

  /* Is there at least one operation that requires copy algorithm? */
  if (ha_alter_info->handler_flags & ~inplace_offline_operations)
    DBUG_RETURN(HA_ALTER_INPLACE_NOT_SUPPORTED);

  /*
    ALTER TABLE tbl_name CONVERT TO CHARACTER SET .. and
    ALTER TABLE table_name DEFAULT CHARSET = .. most likely
    change column charsets and so not supported in-place through
    old API.

    Changing of PACK_KEYS, MAX_ROWS and ROW_FORMAT options were
    not supported as in-place operations in old API either.
  */
  if (create_info->used_fields & (HA_CREATE_USED_CHARSET |
                                  HA_CREATE_USED_DEFAULT_CHARSET |
                                  HA_CREATE_USED_PACK_KEYS |
                                  HA_CREATE_USED_MAX_ROWS) ||
      (table->s->row_type != create_info->row_type))
    DBUG_RETURN(HA_ALTER_INPLACE_NOT_SUPPORTED);

  uint table_changes= (ha_alter_info->handler_flags &
                       Alter_inplace_info::ALTER_COLUMN_EQUAL_PACK_LENGTH) ?
    IS_EQUAL_PACK_LENGTH : IS_EQUAL_YES;
  if (table->file->check_if_incompatible_data(create_info, table_changes)
      == COMPATIBLE_DATA_YES)
    DBUG_RETURN(HA_ALTER_INPLACE_EXCLUSIVE_LOCK);
4232

4233 4234 4235 4236 4237 4238 4239 4240 4241 4242 4243
  DBUG_RETURN(HA_ALTER_INPLACE_NOT_SUPPORTED);
}


/*
   Default implementation to support in-place alter table
   and old online add/drop index API
*/

void handler::notify_table_changed()
{
Sergei Golubchik's avatar
Sergei Golubchik committed
4244
  ha_create_partitioning_metadata(table->s->path.str, NULL, CHF_INDEX_FLAG);
4245 4246 4247 4248 4249 4250 4251 4252 4253 4254 4255 4256
}


void Alter_inplace_info::report_unsupported_error(const char *not_supported,
                                                  const char *try_instead)
{
  if (unsupported_reason == NULL)
    my_error(ER_ALTER_OPERATION_NOT_SUPPORTED, MYF(0),
             not_supported, try_instead);
  else
    my_error(ER_ALTER_OPERATION_NOT_SUPPORTED_REASON, MYF(0),
             not_supported, unsupported_reason, try_instead);
4257 4258 4259 4260 4261 4262 4263 4264 4265 4266 4267 4268
}


/**
  Rename table: public interface.

  @sa handler::rename_table()
*/

int
handler::ha_rename_table(const char *from, const char *to)
{
4269
  DBUG_ASSERT(m_lock_type == F_UNLCK);
4270 4271
  mark_trx_read_write();

4272 4273 4274 4275 4276 4277 4278 4279 4280 4281 4282 4283 4284
  return rename_table(from, to);
}


/**
  Delete table: public interface.

  @sa handler::delete_table()
*/

int
handler::ha_delete_table(const char *name)
{
4285
  mark_trx_read_write();
4286 4287 4288 4289 4290 4291 4292 4293
  return delete_table(name);
}


/**
  Drop table in the engine: public interface.

  @sa handler::drop_table()
4294 4295 4296

  The difference between this and delete_table() is that the table is open in
  drop_table().
4297 4298 4299 4300 4301
*/

void
handler::ha_drop_table(const char *name)
{
4302
  DBUG_ASSERT(m_lock_type == F_UNLCK);
4303 4304
  mark_trx_read_write();

4305 4306 4307 4308 4309 4310 4311 4312 4313 4314 4315 4316 4317
  return drop_table(name);
}


/**
  Create a table in the engine: public interface.

  @sa handler::create()
*/

int
handler::ha_create(const char *name, TABLE *form, HA_CREATE_INFO *info)
{
4318
  DBUG_ASSERT(m_lock_type == F_UNLCK);
4319
  mark_trx_read_write();
4320 4321 4322 4323 4324
  int error= create(name, form, info);
  if (!error &&
      !(info->options & (HA_LEX_CREATE_TMP_TABLE | HA_CREATE_TMP_ALTER)))
    mysql_audit_create_table(form);
  return error;
4325 4326 4327 4328 4329 4330
}


/**
  Create handler files for CREATE TABLE: public interface.

4331
  @sa handler::create_partitioning_metadata()
4332 4333 4334
*/

int
4335
handler::ha_create_partitioning_metadata(const char *name, const char *old_name,
4336
                                         int action_flag)
4337
{
4338 4339 4340 4341 4342 4343 4344
  /*
    Normally this is done when unlocked, but in fast_alter_partition_table,
    it is done on an already locked handler when preparing to alter/rename
    partitions.
  */
  DBUG_ASSERT(m_lock_type == F_UNLCK ||
              (!old_name && strcmp(name, table_share->path.str)));
4345 4346
  mark_trx_read_write();

4347
  return create_partitioning_metadata(name, old_name, action_flag);
4348 4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359
}


/**
  Change partitions: public interface.

  @sa handler::change_partitions()
*/

int
handler::ha_change_partitions(HA_CREATE_INFO *create_info,
                     const char *path,
4360 4361
                     ulonglong * const copied,
                     ulonglong * const deleted,
4362 4363
                     const uchar *pack_frm_data,
                     size_t pack_frm_len)
4364 4365 4366 4367 4368 4369 4370
{  /*
    Must have at least RDLCK or be a TMP table. Read lock is needed to read
    from current partitions and write lock will be taken on new partitions.
  */
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type != F_UNLCK);

4371 4372
  mark_trx_read_write();

4373 4374 4375 4376 4377 4378 4379 4380 4381 4382 4383 4384 4385 4386
  return change_partitions(create_info, path, copied, deleted,
                           pack_frm_data, pack_frm_len);
}


/**
  Drop partitions: public interface.

  @sa handler::drop_partitions()
*/

int
handler::ha_drop_partitions(const char *path)
{
4387 4388
  DBUG_ASSERT(!table->db_stat);

4389 4390
  mark_trx_read_write();

4391 4392 4393 4394 4395 4396 4397 4398 4399 4400 4401 4402 4403
  return drop_partitions(path);
}


/**
  Rename partitions: public interface.

  @sa handler::rename_partitions()
*/

int
handler::ha_rename_partitions(const char *path)
{
4404 4405
  DBUG_ASSERT(!table->db_stat);

4406 4407
  mark_trx_read_write();

4408 4409 4410 4411
  return rename_partitions(path);
}


unknown's avatar
unknown committed
4412
/**
4413
  Tell the storage engine that it is allowed to "disable transaction" in the
4414
  handler. It is a hint that ACID is not required - it was used in NDB for
4415 4416
  ALTER TABLE, for example, when data are copied to temporary table.
  A storage engine may treat this hint any way it likes. NDB for example
4417
  started to commit every now and then automatically.
4418
  This hint can be safely ignored.
4419
*/
4420
int ha_enable_transaction(THD *thd, bool on)
4421 4422
{
  int error=0;
4423
  DBUG_ENTER("ha_enable_transaction");
unknown's avatar
unknown committed
4424 4425
  DBUG_PRINT("enter", ("on: %d", (int) on));

4426
  if ((thd->transaction.on= on))
4427 4428 4429 4430 4431 4432 4433
  {
    /*
      Now all storage engines should have transaction handling enabled.
      But some may have it enabled all the time - "disabling" transactions
      is an optimization hint that storage engine is free to ignore.
      So, let's commit an open transaction (if any) now.
    */
4434
    if (!(error= ha_commit_trans(thd, 0)))
Konstantin Osipov's avatar
Konstantin Osipov committed
4435
      error= trans_commit_implicit(thd);
4436
  }
4437 4438 4439
  DBUG_RETURN(error);
}

4440
int handler::index_next_same(uchar *buf, const uchar *key, uint keylen)
unknown's avatar
unknown committed
4441 4442
{
  int error;
4443
  DBUG_ENTER("handler::index_next_same");
unknown's avatar
unknown committed
4444 4445
  if (!(error=index_next(buf)))
  {
4446
    my_ptrdiff_t ptrdiff= buf - table->record[0];
4447 4448 4449 4450
    uchar *UNINIT_VAR(save_record_0);
    KEY *UNINIT_VAR(key_info);
    KEY_PART_INFO *UNINIT_VAR(key_part);
    KEY_PART_INFO *UNINIT_VAR(key_part_end);
4451 4452 4453 4454 4455 4456 4457 4458 4459 4460 4461 4462 4463 4464 4465

    /*
      key_cmp_if_same() compares table->record[0] against 'key'.
      In parts it uses table->record[0] directly, in parts it uses
      field objects with their local pointers into table->record[0].
      If 'buf' is distinct from table->record[0], we need to move
      all record references. This is table->record[0] itself and
      the field pointers of the fields used in this key.
    */
    if (ptrdiff)
    {
      save_record_0= table->record[0];
      table->record[0]= buf;
      key_info= table->key_info + active_index;
      key_part= key_info->key_part;
4466
      key_part_end= key_part + key_info->user_defined_key_parts;
4467 4468 4469 4470 4471 4472 4473
      for (; key_part < key_part_end; key_part++)
      {
        DBUG_ASSERT(key_part->field);
        key_part->field->move_field_offset(ptrdiff);
      }
    }

unknown's avatar
unknown committed
4474
    if (key_cmp_if_same(table, key, active_index, keylen))
unknown's avatar
unknown committed
4475 4476 4477 4478
    {
      table->status=STATUS_NOT_FOUND;
      error=HA_ERR_END_OF_FILE;
    }
4479 4480 4481 4482 4483 4484 4485 4486

    /* Move back if necessary. */
    if (ptrdiff)
    {
      table->record[0]= save_record_0;
      for (key_part= key_info->key_part; key_part < key_part_end; key_part++)
        key_part->field->move_field_offset(-ptrdiff);
    }
unknown's avatar
unknown committed
4487
  }
4488
  DBUG_PRINT("return",("%i", error));
4489
  DBUG_RETURN(error);
unknown's avatar
unknown committed
4490 4491 4492
}


4493
void handler::get_dynamic_partition_info(PARTITION_STATS *stat_info,
4494
                                         uint part_id)
4495 4496 4497
{
  info(HA_STATUS_CONST | HA_STATUS_TIME | HA_STATUS_VARIABLE |
       HA_STATUS_NO_LOCK);
4498 4499 4500 4501 4502 4503 4504 4505 4506 4507
  stat_info->records=              stats.records;
  stat_info->mean_rec_length=      stats.mean_rec_length;
  stat_info->data_file_length=     stats.data_file_length;
  stat_info->max_data_file_length= stats.max_data_file_length;
  stat_info->index_file_length=    stats.index_file_length;
  stat_info->delete_length=        stats.delete_length;
  stat_info->create_time=          stats.create_time;
  stat_info->update_time=          stats.update_time;
  stat_info->check_time=           stats.check_time;
  stat_info->check_sum=            0;
4508
  if (table_flags() & (HA_HAS_OLD_CHECKSUM | HA_HAS_OLD_CHECKSUM))
4509
    stat_info->check_sum= checksum();
4510 4511 4512 4513
  return;
}


4514 4515 4516 4517 4518 4519 4520 4521 4522
/*
  Updates the global table stats with the TABLE this handler represents
*/

void handler::update_global_table_stats()
{
  TABLE_STATS * table_stats;

  status_var_add(table->in_use->status_var.rows_read, rows_read);
4523
  DBUG_ASSERT(rows_tmp_read == 0);
4524 4525 4526 4527 4528 4529 4530 4531 4532 4533 4534 4535

  if (!table->in_use->userstat_running)
  {
    rows_read= rows_changed= 0;
    return;
  }

  if (rows_read + rows_changed == 0)
    return;                                     // Nothing to update.

  DBUG_ASSERT(table->s && table->s->table_cache_key.str);

4536
  mysql_mutex_lock(&LOCK_global_table_stats);
4537 4538
  /* Gets the global table stats, creating one if necessary. */
  if (!(table_stats= (TABLE_STATS*)
4539
        my_hash_search(&global_table_stats,
4540 4541 4542 4543 4544 4545 4546 4547 4548 4549 4550 4551 4552 4553 4554 4555 4556 4557 4558
                    (uchar*) table->s->table_cache_key.str,
                    table->s->table_cache_key.length)))
  {
    if (!(table_stats = ((TABLE_STATS*)
                         my_malloc(sizeof(TABLE_STATS),
                                   MYF(MY_WME | MY_ZEROFILL)))))
    {
      /* Out of memory error already given */
      goto end;
    }
    memcpy(table_stats->table, table->s->table_cache_key.str,
           table->s->table_cache_key.length);
    table_stats->table_name_length= table->s->table_cache_key.length;
    table_stats->engine_type= ht->db_type;
    /* No need to set variables to 0, as we use MY_ZEROFILL above */

    if (my_hash_insert(&global_table_stats, (uchar*) table_stats))
    {
      /* Out of memory error is already given */
4559
      my_free(table_stats);
4560 4561 4562 4563 4564 4565 4566 4567 4568 4569 4570
      goto end;
    }
  }
  // Updates the global table stats.
  table_stats->rows_read+=    rows_read;
  table_stats->rows_changed+= rows_changed;
  table_stats->rows_changed_x_indexes+= (rows_changed *
                                         (table->s->keys ? table->s->keys :
                                          1));
  rows_read= rows_changed= 0;
end:
4571
  mysql_mutex_unlock(&LOCK_global_table_stats);
4572 4573 4574 4575 4576 4577 4578 4579 4580 4581 4582 4583 4584 4585 4586 4587 4588 4589 4590 4591 4592 4593 4594 4595 4596 4597 4598 4599 4600 4601
}


/*
  Updates the global index stats with this handler's accumulated index reads.
*/

void handler::update_global_index_stats()
{
  DBUG_ASSERT(table->s);

  if (!table->in_use->userstat_running)
  {
    /* Reset all index read values */
    bzero(index_rows_read, sizeof(index_rows_read[0]) * table->s->keys);
    return;
  }

  for (uint index = 0; index < table->s->keys; index++)
  {
    if (index_rows_read[index])
    {
      INDEX_STATS* index_stats;
      uint key_length;
      KEY *key_info = &table->key_info[index];  // Rows were read using this

      DBUG_ASSERT(key_info->cache_name);
      if (!key_info->cache_name)
        continue;
      key_length= table->s->table_cache_key.length + key_info->name_length + 1;
4602
      mysql_mutex_lock(&LOCK_global_index_stats);
4603
      // Gets the global index stats, creating one if necessary.
4604
      if (!(index_stats= (INDEX_STATS*) my_hash_search(&global_index_stats,
4605 4606 4607 4608 4609 4610 4611 4612 4613 4614 4615 4616
                                                    key_info->cache_name,
                                                    key_length)))
      {
        if (!(index_stats = ((INDEX_STATS*)
                             my_malloc(sizeof(INDEX_STATS),
                                       MYF(MY_WME | MY_ZEROFILL)))))
          goto end;                             // Error is already given

        memcpy(index_stats->index, key_info->cache_name, key_length);
        index_stats->index_name_length= key_length;
        if (my_hash_insert(&global_index_stats, (uchar*) index_stats))
        {
4617
          my_free(index_stats);
4618 4619 4620 4621 4622 4623 4624
          goto end;
        }
      }
      /* Updates the global index stats. */
      index_stats->rows_read+= index_rows_read[index];
      index_rows_read[index]= 0;
end:
4625
      mysql_mutex_unlock(&LOCK_global_index_stats);
4626 4627 4628 4629 4630
    }
  }
}


unknown's avatar
unknown committed
4631 4632 4633 4634
/****************************************************************************
** Some general functions that isn't in the handler class
****************************************************************************/

unknown's avatar
unknown committed
4635 4636
/**
  Initiates table-file and calls appropriate database-creator.
unknown's avatar
unknown committed
4637

unknown's avatar
unknown committed
4638
  @retval
unknown's avatar
unknown committed
4639
   0  ok
unknown's avatar
unknown committed
4640
  @retval
unknown's avatar
unknown committed
4641
   1  error
unknown's avatar
unknown committed
4642
*/
unknown's avatar
unknown committed
4643 4644
int ha_create_table(THD *thd, const char *path,
                    const char *db, const char *table_name,
4645
                    HA_CREATE_INFO *create_info, LEX_CUSTRING *frm)
unknown's avatar
unknown committed
4646
{
unknown's avatar
unknown committed
4647
  int error= 1;
unknown's avatar
unknown committed
4648
  TABLE table;
unknown's avatar
unknown committed
4649
  char name_buff[FN_REFLEN];
unknown's avatar
unknown committed
4650 4651
  const char *name;
  TABLE_SHARE share;
4652 4653 4654
  bool temp_table __attribute__((unused)) =
    create_info->options & (HA_LEX_CREATE_TMP_TABLE | HA_CREATE_TMP_ALTER);
                                 
unknown's avatar
unknown committed
4655
  DBUG_ENTER("ha_create_table");
4656

4657
  init_tmp_table_share(thd, &share, db, 0, table_name, path);
4658 4659 4660 4661 4662 4663 4664 4665 4666 4667 4668 4669 4670 4671 4672 4673

  if (frm)
  {
    bool write_frm_now= !create_info->db_type->discover_table &&
                        !create_info->tmp_table();

    share.frm_image= frm;

    // open an frm image
    if (share.init_from_binary_frm_image(thd, write_frm_now,
                                         frm->str, frm->length))
      goto err;
  }
  else
  {
    // open an frm file
4674 4675
    share.db_plugin= ha_lock_engine(thd, create_info->db_type);

4676 4677 4678
    if (open_table_def(thd, &share))
      goto err;
  }
4679

4680
  share.m_psi= PSI_CALL_get_table_share(temp_table, &share);
4681 4682

  if (open_table_from_share(thd, &share, "", 0, READ_ALL, 0, &table, true))
unknown's avatar
unknown committed
4683
    goto err;
unknown's avatar
unknown committed
4684

4685
  update_create_info_from_table(create_info, &table);
unknown's avatar
unknown committed
4686

4687
  name= get_canonical_filename(table.file, share.path.str, name_buff);
unknown's avatar
unknown committed
4688

4689
  error= table.file->ha_create(name, &table, create_info);
4690

unknown's avatar
unknown committed
4691
  if (error)
unknown's avatar
unknown committed
4692
  {
4693 4694 4695
    if (!thd->is_error())
      my_error(ER_CANT_CREATE_TABLE, MYF(0), db, table_name, error);
    table.file->print_error(error, MYF(ME_JUST_WARNING));
4696 4697
    PSI_CALL_drop_table_share(temp_table, share.db.str, share.db.length,
                              share.table_name.str, share.table_name.length);
unknown's avatar
unknown committed
4698
  }
4699 4700

  (void) closefrm(&table, 0);
4701
 
unknown's avatar
unknown committed
4702 4703
err:
  free_table_share(&share);
unknown's avatar
unknown committed
4704 4705 4706
  DBUG_RETURN(error != 0);
}

4707 4708 4709
void st_ha_check_opt::init()
{
  flags= sql_flags= 0;
4710
  start_time= my_time(0);
4711 4712 4713 4714 4715 4716 4717 4718 4719 4720 4721 4722 4723
}


/*****************************************************************************
  Key cache handling.

  This code is only relevant for ISAM/MyISAM tables

  key_cache->cache may be 0 only in the case where a key cache is not
  initialized or when we where not able to init the key cache in a previous
  call to ha_init_key_cache() (probably out of memory)
*****************************************************************************/

unknown's avatar
unknown committed
4724 4725
/**
  Init a key cache if it has not been initied before.
4726
*/
4727 4728
int ha_init_key_cache(const char *name, KEY_CACHE *key_cache, void *unused
                      __attribute__((unused)))
unknown's avatar
unknown committed
4729
{
4730 4731
  DBUG_ENTER("ha_init_key_cache");

unknown's avatar
unknown committed
4732
  if (!key_cache->key_cache_inited)
4733
  {
Marc Alff's avatar
Marc Alff committed
4734
    mysql_mutex_lock(&LOCK_global_system_variables);
4735
    size_t tmp_buff_size= (size_t) key_cache->param_buff_size;
4736
    uint tmp_block_size= (uint) key_cache->param_block_size;
4737 4738 4739
    uint division_limit= (uint)key_cache->param_division_limit;
    uint age_threshold=  (uint)key_cache->param_age_threshold;
    uint partitions=     (uint)key_cache->param_partitions;
4740
    uint changed_blocks_hash_size=  (uint)key_cache->changed_blocks_hash_size;
Marc Alff's avatar
Marc Alff committed
4741
    mysql_mutex_unlock(&LOCK_global_system_variables);
unknown's avatar
unknown committed
4742
    DBUG_RETURN(!init_key_cache(key_cache,
4743 4744
				tmp_block_size,
				tmp_buff_size,
4745
				division_limit, age_threshold,
4746
                                changed_blocks_hash_size,
4747
                                partitions));
4748
  }
4749
  DBUG_RETURN(0);
unknown's avatar
unknown committed
4750 4751
}

4752

unknown's avatar
unknown committed
4753 4754
/**
  Resize key cache.
4755
*/
unknown's avatar
unknown committed
4756
int ha_resize_key_cache(KEY_CACHE *key_cache)
unknown's avatar
unknown committed
4757
{
4758 4759
  DBUG_ENTER("ha_resize_key_cache");

unknown's avatar
unknown committed
4760
  if (key_cache->key_cache_inited)
4761
  {
Marc Alff's avatar
Marc Alff committed
4762
    mysql_mutex_lock(&LOCK_global_system_variables);
4763
    size_t tmp_buff_size= (size_t) key_cache->param_buff_size;
unknown's avatar
unknown committed
4764
    long tmp_block_size= (long) key_cache->param_block_size;
4765 4766
    uint division_limit= (uint)key_cache->param_division_limit;
    uint age_threshold=  (uint)key_cache->param_age_threshold;
4767
    uint changed_blocks_hash_size=  (uint)key_cache->changed_blocks_hash_size;
Marc Alff's avatar
Marc Alff committed
4768
    mysql_mutex_unlock(&LOCK_global_system_variables);
unknown's avatar
unknown committed
4769 4770
    DBUG_RETURN(!resize_key_cache(key_cache, tmp_block_size,
				  tmp_buff_size,
4771 4772
				  division_limit, age_threshold,
                                  changed_blocks_hash_size));
4773
  }
4774
  DBUG_RETURN(0);
4775 4776
}

4777

unknown's avatar
unknown committed
4778
/**
4779
  Change parameters for key cache (like division_limit)
4780
*/
unknown's avatar
unknown committed
4781
int ha_change_key_cache_param(KEY_CACHE *key_cache)
4782
{
4783 4784
  DBUG_ENTER("ha_change_key_cache_param");

unknown's avatar
unknown committed
4785 4786
  if (key_cache->key_cache_inited)
  {
Marc Alff's avatar
Marc Alff committed
4787
    mysql_mutex_lock(&LOCK_global_system_variables);
4788 4789
    uint division_limit= (uint)key_cache->param_division_limit;
    uint age_threshold=  (uint)key_cache->param_age_threshold;
Marc Alff's avatar
Marc Alff committed
4790
    mysql_mutex_unlock(&LOCK_global_system_variables);
unknown's avatar
unknown committed
4791 4792
    change_key_cache_param(key_cache, division_limit, age_threshold);
  }
4793
  DBUG_RETURN(0);
4794
}
unknown's avatar
unknown committed
4795

4796 4797 4798 4799 4800 4801 4802 4803 4804 4805

/**
  Repartition key cache 
*/
int ha_repartition_key_cache(KEY_CACHE *key_cache)
{
  DBUG_ENTER("ha_repartition_key_cache");

  if (key_cache->key_cache_inited)
  {
4806
    mysql_mutex_lock(&LOCK_global_system_variables);
4807 4808
    size_t tmp_buff_size= (size_t) key_cache->param_buff_size;
    long tmp_block_size= (long) key_cache->param_block_size;
4809 4810 4811
    uint division_limit= (uint)key_cache->param_division_limit;
    uint age_threshold=  (uint)key_cache->param_age_threshold;
    uint partitions=     (uint)key_cache->param_partitions;
4812
    uint changed_blocks_hash_size=  (uint)key_cache->changed_blocks_hash_size;
4813
    mysql_mutex_unlock(&LOCK_global_system_variables);
4814 4815 4816
    DBUG_RETURN(!repartition_key_cache(key_cache, tmp_block_size,
				       tmp_buff_size,
				       division_limit, age_threshold,
4817
                                       changed_blocks_hash_size,
4818 4819 4820
                                       partitions));
  }
  DBUG_RETURN(0);
4821
}
unknown's avatar
unknown committed
4822

4823

unknown's avatar
unknown committed
4824 4825
/**
  Move all tables from one key cache to another one.
4826
*/
unknown's avatar
unknown committed
4827 4828
int ha_change_key_cache(KEY_CACHE *old_key_cache,
			KEY_CACHE *new_key_cache)
unknown's avatar
unknown committed
4829
{
4830 4831
  mi_change_key_cache(old_key_cache, new_key_cache);
  return 0;
unknown's avatar
unknown committed
4832
}
4833 4834


unknown's avatar
unknown committed
4835
static my_bool discover_handlerton(THD *thd, plugin_ref plugin,
unknown's avatar
unknown committed
4836 4837
                                   void *arg)
{
4838
  TABLE_SHARE *share= (TABLE_SHARE *)arg;
Sergei Golubchik's avatar
Sergei Golubchik committed
4839
  handlerton *hton= plugin_hton(plugin);
4840 4841
  if (hton->state == SHOW_OPTION_YES && hton->discover_table)
  {
4842
    share->db_plugin= plugin;
4843 4844 4845 4846 4847
    int error= hton->discover_table(hton, thd, share);
    if (error != HA_ERR_NO_SUCH_TABLE)
    {
      if (error)
      {
4848
        DBUG_ASSERT(share->error); // tdc_lock_share needs that
4849 4850 4851 4852 4853 4854
        /*
          report an error, unless it is "generic" and a more
          specific one was already reported
        */
        if (error != HA_ERR_GENERIC || !thd->is_error())
          my_error(ER_GET_ERRNO, MYF(0), error, plugin_name(plugin)->str);
4855
        share->db_plugin= 0;
4856 4857 4858
      }
      else
        share->error= OPEN_FRM_OK;
unknown's avatar
unknown committed
4859

4860 4861 4862
      status_var_increment(thd->status_var.ha_discover_count);
      return TRUE; // abort the search
    }
4863
    share->db_plugin= 0;
4864 4865 4866 4867
  }

  DBUG_ASSERT(share->error == OPEN_FRM_OPEN_ERROR);
  return FALSE;    // continue with the next engine
unknown's avatar
unknown committed
4868 4869
}

4870
int ha_discover_table(THD *thd, TABLE_SHARE *share)
unknown's avatar
unknown committed
4871
{
4872
  DBUG_ENTER("ha_discover_table");
4873
  int found;
unknown's avatar
unknown committed
4874

4875
  DBUG_ASSERT(share->error == OPEN_FRM_OPEN_ERROR);   // share is not OK yet
unknown's avatar
unknown committed
4876

4877 4878 4879
  if (!engines_with_discover)
    found= FALSE;
  else if (share->db_plugin)
4880 4881 4882 4883 4884 4885
    found= discover_handlerton(thd, share->db_plugin, share);
  else
    found= plugin_foreach(thd, discover_handlerton,
                        MYSQL_STORAGE_ENGINE_PLUGIN, share);
  
  if (!found)
4886
    open_table_error(share, OPEN_FRM_OPEN_ERROR, ENOENT); // not found
unknown's avatar
unknown committed
4887

4888
  DBUG_RETURN(share->error != OPEN_FRM_OK);
unknown's avatar
unknown committed
4889 4890
}

4891 4892 4893 4894 4895
static my_bool file_ext_exists(char *path, size_t path_len, const char *ext)
{
  strmake(path + path_len, ext, FN_REFLEN - path_len);
  return !access(path, F_OK);
}
unknown's avatar
unknown committed
4896

4897
struct st_discover_existence_args
4898
{
4899 4900 4901
  char *path;
  size_t  path_len;
  const char *db, *table_name;
4902
  handlerton *hton;
4903
  bool frm_exists;
4904 4905
};

4906 4907
static my_bool discover_existence(THD *thd, plugin_ref plugin,
                                  void *arg)
4908
{
4909
  st_discover_existence_args *args= (st_discover_existence_args*)arg;
Sergei Golubchik's avatar
Sergei Golubchik committed
4910
  handlerton *ht= plugin_hton(plugin);
4911
  if (ht->state != SHOW_OPTION_YES || !ht->discover_table_existence)
4912
    return args->frm_exists;
4913

4914
  args->hton= ht;
4915

4916 4917 4918
  if (ht->discover_table_existence == ext_based_existence)
    return file_ext_exists(args->path, args->path_len,
                           ht->tablefile_extensions[0]);
4919

4920
  return ht->discover_table_existence(ht, args->db, args->table_name);
4921
}
4922

4923
class Table_exists_error_handler : public Internal_error_handler
4924
{
4925 4926 4927 4928 4929 4930 4931 4932
public:
  Table_exists_error_handler()
    : m_handled_errors(0), m_unhandled_errors(0)
  {}

  bool handle_condition(THD *thd,
                        uint sql_errno,
                        const char* sqlstate,
Sergei Golubchik's avatar
Sergei Golubchik committed
4933
                        Sql_condition::enum_warning_level level,
4934
                        const char* msg,
Sergei Golubchik's avatar
Sergei Golubchik committed
4935
                        Sql_condition ** cond_hdl)
4936 4937 4938 4939
  {
    *cond_hdl= NULL;
    if (sql_errno == ER_NO_SUCH_TABLE ||
        sql_errno == ER_NO_SUCH_TABLE_IN_ENGINE ||
4940
        sql_errno == ER_WRONG_OBJECT)
4941 4942 4943 4944 4945
    {
      m_handled_errors++;
      return TRUE;
    }

Sergei Golubchik's avatar
Sergei Golubchik committed
4946
    if (level == Sql_condition::WARN_LEVEL_ERROR)
4947 4948 4949 4950 4951 4952 4953 4954 4955 4956 4957 4958 4959 4960
      m_unhandled_errors++;
    return FALSE;
  }

  bool safely_trapped_errors()
  {
    return ((m_handled_errors > 0) && (m_unhandled_errors == 0));
  }

private:
  int m_handled_errors;
  int m_unhandled_errors;
};

4961 4962 4963 4964 4965 4966
/**
  Check if a given table exists, without doing a full discover, if possible

  If the 'hton' is not NULL, it's set to the handlerton of the storage engine
  of this table, or to view_pseudo_hton if the frm belongs to a view.

4967 4968 4969 4970 4971 4972 4973 4974 4975
  This function takes discovery correctly into account. If frm is found,
  it discovers the table to make sure it really exists in the engine.
  If no frm is found it discovers the table, in case it still exists in
  the engine.

  While it tries to cut corners (don't open .frm if no discovering engine is
  enabled, no full discovery if all discovering engines support
  discover_table_existence, etc), it still *may* be quite expensive
  and must be used sparingly.
4976 4977 4978

  @retval true    Table exists (even if the error occurred, like bad frm)
  @retval false   Table does not exist (one can do CREATE TABLE table_name)
4979 4980 4981 4982 4983 4984 4985

  @note if frm exists and the table in engine doesn't, *hton will be set,
        but the return value will be false.

  @note if frm file exists, but the table cannot be opened (engine not
        loaded, frm is invalid), the return value will be true, but
        *hton will be NULL.
4986 4987 4988
*/
bool ha_table_exists(THD *thd, const char *db, const char *table_name,
                     handlerton **hton)
4989
{
4990
  handlerton *dummy;
4991 4992 4993 4994
  DBUG_ENTER("ha_table_exists");

  if (hton)
    *hton= 0;
4995 4996
  else if (engines_with_discover)
    hton= &dummy;
4997

4998 4999
  TDC_element *element= tdc_lock_share(thd, db, table_name);
  if (element && element != MY_ERRPTR)
5000 5001
  {
    if (hton)
5002 5003
      *hton= element->share->db_type();
    tdc_unlock_share(element);
5004
    DBUG_RETURN(TRUE);
5005
  }
5006 5007 5008 5009

  char path[FN_REFLEN + 1];
  size_t path_len = build_table_filename(path, sizeof(path) - 1,
                                         db, table_name, "", 0);
5010
  st_discover_existence_args args= {path, path_len, db, table_name, 0, true};
5011 5012

  if (file_ext_exists(path, path_len, reg_ext))
5013
  {
5014
    bool exists= true;
5015 5016 5017 5018
    if (hton)
    {
      enum legacy_db_type db_type;
      if (dd_frm_type(thd, path, &db_type) != FRMTYPE_VIEW)
5019 5020 5021 5022 5023 5024 5025
      {
        handlerton *ht= ha_resolve_by_legacy_type(thd, db_type);
        if ((*hton= ht))
          // verify that the table really exists
          exists= discover_existence(thd,
                             plugin_int_to_ref(hton2plugin[ht->slot]), &args);
      }
5026 5027 5028
      else
        *hton= view_pseudo_hton;
    }
5029
    DBUG_RETURN(exists);
5030
  }
5031

5032
  args.frm_exists= false;
5033 5034
  if (plugin_foreach(thd, discover_existence, MYSQL_STORAGE_ENGINE_PLUGIN,
                     &args))
5035 5036 5037
  {
    if (hton)
      *hton= args.hton;
5038
    DBUG_RETURN(TRUE);
5039
  }
5040

5041 5042 5043 5044 5045 5046 5047 5048 5049 5050 5051 5052 5053 5054 5055 5056 5057 5058 5059 5060 5061 5062 5063 5064

  if (need_full_discover_for_existence)
  {
    TABLE_LIST table;
    uint flags = GTS_TABLE | GTS_VIEW;

    if (!hton)
      flags|= GTS_NOLOCK;

    Table_exists_error_handler no_such_table_handler;
    thd->push_internal_handler(&no_such_table_handler);
    TABLE_SHARE *share= tdc_acquire_share(thd, db, table_name, flags);
    thd->pop_internal_handler();

    if (hton && share)
    {
      *hton= share->db_type();
      tdc_release_share(share);
    }

    // the table doesn't exist if we've caught ER_NO_SUCH_TABLE and nothing else
    DBUG_RETURN(!no_such_table_handler.safely_trapped_errors());
  }

5065
  DBUG_RETURN(FALSE);
5066 5067
}

unknown's avatar
unknown committed
5068
/**
5069
  Discover all table names in a given database
unknown's avatar
unknown committed
5070
*/
5071 5072 5073
extern "C" {

static int cmp_file_names(const void *a, const void *b)
5074
{
5075 5076 5077 5078 5079
  CHARSET_INFO *cs= character_set_filesystem;
  char *aa= ((FILEINFO *)a)->name;
  char *bb= ((FILEINFO *)b)->name;
  return my_strnncoll(cs, (uchar*)aa, strlen(aa), (uchar*)bb, strlen(bb));
}
5080

5081
static int cmp_table_names(LEX_STRING * const *a, LEX_STRING * const *b)
5082
{
5083 5084 5085
  return my_strnncoll(&my_charset_bin, (uchar*)((*a)->str), (*a)->length,
                                       (uchar*)((*b)->str), (*b)->length);
}
5086

5087
}
5088

5089 5090
Discovered_table_list::Discovered_table_list(THD *thd_arg,
                 Dynamic_array<LEX_STRING*> *tables_arg,
5091 5092
                 const LEX_STRING *wild_arg) :
  thd(thd_arg), with_temps(false), tables(tables_arg)
5093 5094 5095 5096 5097 5098 5099 5100 5101
{
  if (wild_arg->str && wild_arg->str[0])
  {
    wild= wild_arg->str;
    wend= wild + wild_arg->length;
  }
  else
    wild= 0;
}
5102

5103 5104
bool Discovered_table_list::add_table(const char *tname, size_t tlen)
{
5105 5106 5107 5108 5109 5110
  /*
    TODO Check with_temps and filter out temp tables.
    Implement the check, when we'll have at least one affected engine (with
    custom discover_table_names() method, that calls add_table() directly).
    Note: avoid comparing the same name twice (here and in add_file).
  */
5111 5112 5113
  if (wild && my_wildcmp(files_charset_info, tname, tname + tlen, wild, wend,
                         wild_prefix, wild_one, wild_many))
      return 0;
5114

5115 5116 5117 5118 5119 5120 5121 5122
  LEX_STRING *name= thd->make_lex_string(tname, tlen);
  if (!name || tables->append(name))
    return 1;
  return 0;
}

bool Discovered_table_list::add_file(const char *fname)
{
5123 5124 5125 5126 5127
  bool is_temp= strncmp(fname, STRING_WITH_LEN(tmp_file_prefix)) == 0;

  if (is_temp && !with_temps)
    return 0;

5128
  char tname[SAFE_NAME_LEN + 1];
5129
  size_t tlen= filename_to_tablename(fname, tname, sizeof(tname), is_temp);
5130 5131 5132 5133 5134 5135 5136 5137 5138 5139 5140 5141 5142
  return add_table(tname, tlen);
}


void Discovered_table_list::sort()
{
  tables->sort(cmp_table_names);
}

void Discovered_table_list::remove_duplicates()
{
  LEX_STRING **src= tables->front();
  LEX_STRING **dst= src;
Sergei Golubchik's avatar
Sergei Golubchik committed
5143
  while (++dst <= tables->back())
5144 5145
  {
    LEX_STRING *s= *src, *d= *dst;
Sergei Golubchik's avatar
Sergei Golubchik committed
5146
    DBUG_ASSERT(strncmp(s->str, d->str, MY_MIN(s->length, d->length)) <= 0);
5147 5148 5149 5150 5151 5152 5153
    if ((s->length != d->length || strncmp(s->str, d->str, d->length)))
    {
      src++;
      if (src != dst)
        *src= *dst;
    }
  }
Sergei Golubchik's avatar
Sergei Golubchik committed
5154
  tables->elements(src - tables->front() + 1);
5155
}
5156 5157 5158 5159 5160

struct st_discover_names_args
{
  LEX_STRING *db;
  MY_DIR *dirp;
5161 5162
  Discovered_table_list *result;
  uint possible_duplicates;
5163 5164 5165 5166 5167 5168
};

static my_bool discover_names(THD *thd, plugin_ref plugin,
                              void *arg)
{
  st_discover_names_args *args= (st_discover_names_args *)arg;
Sergei Golubchik's avatar
Sergei Golubchik committed
5169
  handlerton *ht= plugin_hton(plugin);
5170 5171 5172 5173 5174 5175 5176 5177 5178 5179 5180 5181 5182 5183

  if (ht->state == SHOW_OPTION_YES && ht->discover_table_names)
  {
    uint old_elements= args->result->tables->elements();
    if (ht->discover_table_names(ht, args->db, args->dirp, args->result))
      return 1;

    /*
      hton_ext_based_table_discovery never discovers a table that has
      a corresponding .frm file; but custom engine discover methods might
    */
    if (ht->discover_table_names != hton_ext_based_table_discovery)
      args->possible_duplicates+= args->result->tables->elements() - old_elements;
  }
5184 5185

  return 0;
5186 5187
}

5188 5189 5190 5191 5192 5193 5194 5195 5196 5197 5198 5199 5200 5201 5202 5203
/**
  Return the list of tables

  @param thd
  @param db         database to look into
  @param dirp       list of files in this database (as returned by my_dir())
  @param result     the object to return the list of files in
  @param reusable   if true, on return, 'dirp' will be a valid list of all
                    non-table files. If false, discovery will work much faster,
                    but it will leave 'dirp' corrupted and completely unusable,
                    only good for my_dirend().

  Normally, reusable=false for SHOW and INFORMATION_SCHEMA, and reusable=true
  for DROP DATABASE (as it needs to know and delete non-table files).
*/

5204
int ha_discover_table_names(THD *thd, LEX_STRING *db, MY_DIR *dirp,
5205
                            Discovered_table_list *result, bool reusable)
5206
{
5207 5208 5209
  int error;
  DBUG_ENTER("ha_discover_table_names");

5210
  if (engines_with_discover_table_names == 0 && !reusable)
5211 5212 5213 5214 5215 5216 5217
  {
    error= ext_table_discovery_simple(dirp, result);
    result->sort();
  }
  else
  {
    st_discover_names_args args= {db, dirp, result, 0};
5218

5219 5220 5221
    /* extension_based_table_discovery relies on dirp being sorted */
    my_qsort(dirp->dir_entry, dirp->number_of_files,
             sizeof(FILEINFO), cmp_file_names);
5222

5223 5224 5225 5226 5227 5228 5229 5230
    error= extension_based_table_discovery(dirp, reg_ext, result) ||
           plugin_foreach(thd, discover_names,
                            MYSQL_STORAGE_ENGINE_PLUGIN, &args);
    result->sort();

    if (args.possible_duplicates > 0)
      result->remove_duplicates();
  }
5231 5232

  DBUG_RETURN(error);
5233 5234
}

unknown's avatar
unknown committed
5235

unknown's avatar
unknown committed
5236
/**
5237
  Read first row between two ranges.
unknown's avatar
unknown committed
5238
  Store ranges for future calls to read_range_next.
5239

unknown's avatar
unknown committed
5240 5241 5242 5243 5244 5245
  @param start_key		Start key. Is 0 if no min range
  @param end_key		End key.  Is 0 if no max range
  @param eq_range_arg	        Set to 1 if start_key == end_key
  @param sorted		Set to 1 if result should be sorted per key

  @note
5246 5247
    Record is read into table->record[0]

unknown's avatar
unknown committed
5248
  @retval
5249
    0			Found row
unknown's avatar
unknown committed
5250
  @retval
5251
    HA_ERR_END_OF_FILE	No rows in range
unknown's avatar
unknown committed
5252 5253
  @retval
    \#			Error code
5254 5255 5256
*/
int handler::read_range_first(const key_range *start_key,
			      const key_range *end_key,
unknown's avatar
unknown committed
5257
			      bool eq_range_arg, bool sorted)
5258 5259 5260 5261
{
  int result;
  DBUG_ENTER("handler::read_range_first");

unknown's avatar
unknown committed
5262
  eq_range= eq_range_arg;
5263
  set_end_range(end_key);
5264 5265 5266
  range_key_part= table->key_info[active_index].key_part;

  if (!start_key)			// Read first record
5267
    result= ha_index_first(table->record[0]);
5268
  else
5269 5270 5271 5272
    result= ha_index_read_map(table->record[0],
                              start_key->key,
                              start_key->keypart_map,
                              start_key->flag);
5273
  if (result)
unknown's avatar
unknown committed
5274 5275 5276
    DBUG_RETURN((result == HA_ERR_KEY_NOT_FOUND) 
		? HA_ERR_END_OF_FILE
		: result);
5277

5278 5279 5280 5281 5282 5283 5284 5285 5286 5287 5288 5289 5290
  if (compare_key(end_range) <= 0)
  {
    DBUG_RETURN(0);
  }
  else
  {
    /*
      The last read row does not fall in the range. So request
      storage engine to release row lock if possible.
    */
    unlock_row();
    DBUG_RETURN(HA_ERR_END_OF_FILE);
  }
5291 5292 5293
}


unknown's avatar
unknown committed
5294
/**
5295 5296
  Read next row between two ranges.

unknown's avatar
unknown committed
5297
  @note
5298 5299
    Record is read into table->record[0]

unknown's avatar
unknown committed
5300
  @retval
5301
    0			Found row
unknown's avatar
unknown committed
5302
  @retval
5303
    HA_ERR_END_OF_FILE	No rows in range
unknown's avatar
unknown committed
5304 5305
  @retval
    \#			Error code
5306
*/
unknown's avatar
unknown committed
5307
int handler::read_range_next()
5308 5309 5310 5311 5312
{
  int result;
  DBUG_ENTER("handler::read_range_next");

  if (eq_range)
unknown's avatar
unknown committed
5313 5314
  {
    /* We trust that index_next_same always gives a row in range */
5315 5316 5317
    DBUG_RETURN(ha_index_next_same(table->record[0],
                                   end_range->key,
                                   end_range->length));
unknown's avatar
unknown committed
5318
  }
5319
  result= ha_index_next(table->record[0]);
5320 5321
  if (result)
    DBUG_RETURN(result);
5322 5323 5324 5325 5326 5327 5328 5329 5330 5331 5332 5333 5334 5335

  if (compare_key(end_range) <= 0)
  {
    DBUG_RETURN(0);
  }
  else
  {
    /*
      The last read row does not fall in the range. So request
      storage engine to release row lock if possible.
    */
    unlock_row();
    DBUG_RETURN(HA_ERR_END_OF_FILE);
  }
5336 5337 5338
}


5339 5340 5341 5342 5343 5344 5345 5346 5347 5348 5349 5350 5351 5352
void handler::set_end_range(const key_range *end_key)
{
  end_range= 0;
  if (end_key)
  {
    end_range= &save_end_range;
    save_end_range= *end_key;
    key_compare_result_on_equal=
      ((end_key->flag == HA_READ_BEFORE_KEY) ? 1 :
       (end_key->flag == HA_READ_AFTER_KEY) ? -1 : 0);
  }
}


unknown's avatar
unknown committed
5353 5354
/**
  Compare if found key (in row) is over max-value.
5355

unknown's avatar
unknown committed
5356
  @param range		range to compare to row. May be 0 for no range
5357

5358
  @see also
unknown's avatar
unknown committed
5359 5360 5361
    key.cc::key_cmp()

  @return
unknown's avatar
unknown committed
5362 5363
    The return value is SIGN(key_in_row - range_key):

unknown's avatar
unknown committed
5364 5365 5366
    - 0   : Key is equal to range or 'range' == 0 (no range)
    - -1  : Key is less than range
    - 1   : Key is larger than range
5367 5368 5369
*/
int handler::compare_key(key_range *range)
{
unknown's avatar
unknown committed
5370
  int cmp;
5371
  if (!range || in_range_check_pushed_down)
5372
    return 0;					// No max range
unknown's avatar
unknown committed
5373 5374 5375 5376
  cmp= key_cmp(range_key_part, range->key, range->length);
  if (!cmp)
    cmp= key_compare_result_on_equal;
  return cmp;
5377
}
unknown's avatar
unknown committed
5378

5379

5380 5381 5382 5383 5384 5385 5386 5387 5388 5389 5390 5391 5392 5393 5394 5395 5396
/*
  Same as compare_key() but doesn't check have in_range_check_pushed_down.
  This is used by index condition pushdown implementation.
*/

int handler::compare_key2(key_range *range)
{
  int cmp;
  if (!range)
    return 0;					// no max range
  cmp= key_cmp(range_key_part, range->key, range->length);
  if (!cmp)
    cmp= key_compare_result_on_equal;
  return cmp;
}


5397 5398 5399 5400 5401 5402 5403 5404 5405
/**
  ICP callback - to be called by an engine to check the pushed condition
*/
extern "C" enum icp_result handler_index_cond_check(void* h_arg)
{
  handler *h= (handler*)h_arg;
  THD *thd= h->table->in_use;
  enum icp_result res;

5406 5407 5408
  enum thd_kill_levels abort_at= h->has_transactions() ?
    THD_ABORT_SOFTLY : THD_ABORT_ASAP;
  if (thd_kill_level(thd) > abort_at)
5409 5410 5411 5412
    return ICP_ABORTED_BY_USER;

  if (h->end_range && h->compare_key2(h->end_range) > 0)
    return ICP_OUT_OF_RANGE;
5413
  h->increment_statistics(&SSV::ha_icp_attempts);
5414
  if ((res= h->pushed_idx_cond->val_int()? ICP_MATCH : ICP_NO_MATCH) ==
5415 5416
      ICP_MATCH)
    h->increment_statistics(&SSV::ha_icp_match);
5417 5418 5419
  return res;
}

5420 5421 5422
int handler::index_read_idx_map(uchar * buf, uint index, const uchar * key,
                                key_part_map keypart_map,
                                enum ha_rkey_function find_flag)
unknown's avatar
unknown committed
5423
{
5424
  int error, UNINIT_VAR(error1);
Michael Widenius's avatar
Michael Widenius committed
5425

5426
  error= ha_index_init(index, 0);
unknown's avatar
unknown committed
5427
  if (!error)
5428
  {
5429
    error= index_read_map(buf, key, keypart_map, find_flag);
5430
    error1= ha_index_end();
5431 5432
  }
  return error ?  error : error1;
unknown's avatar
unknown committed
5433 5434
}

5435

unknown's avatar
unknown committed
5436
/**
5437 5438 5439
  Returns a list of all known extensions.

    No mutexes, worst case race is a minor surplus memory allocation
5440 5441
    We have to recreate the extension map if mysqld is restarted (for example
    within libmysqld)
5442

unknown's avatar
unknown committed
5443
  @retval
5444 5445
    pointer		pointer to TYPELIB structure
*/
unknown's avatar
unknown committed
5446
static my_bool exts_handlerton(THD *unused, plugin_ref plugin,
unknown's avatar
unknown committed
5447 5448 5449
                               void *arg)
{
  List<char> *found_exts= (List<char> *) arg;
Sergei Golubchik's avatar
Sergei Golubchik committed
5450
  handlerton *hton= plugin_hton(plugin);
5451 5452
  List_iterator_fast<char> it(*found_exts);
  const char **ext, *old_ext;
unknown's avatar
unknown committed
5453

5454 5455 5456
  for (ext= hton->tablefile_extensions; *ext; ext++)
  {
    while ((old_ext= it++))
unknown's avatar
unknown committed
5457
    {
5458 5459
      if (!strcmp(old_ext, *ext))
        break;
unknown's avatar
unknown committed
5460
    }
5461 5462 5463 5464
    if (!old_ext)
      found_exts->push_back((char *) *ext);

    it.rewind();
unknown's avatar
unknown committed
5465 5466 5467 5468
  }
  return FALSE;
}

5469 5470
TYPELIB *ha_known_exts(void)
{
5471
  if (!known_extensions.type_names || mysys_usage_id != known_extensions_id)
5472 5473
  {
    List<char> found_exts;
5474 5475 5476
    const char **ext, *old_ext;

    known_extensions_id= mysys_usage_id;
5477 5478
    found_exts.push_back((char*) TRG_EXT);
    found_exts.push_back((char*) TRN_EXT);
unknown's avatar
unknown committed
5479 5480

    plugin_foreach(NULL, exts_handlerton,
unknown's avatar
unknown committed
5481 5482
                   MYSQL_STORAGE_ENGINE_PLUGIN, &found_exts);

5483 5484 5485
    ext= (const char **) my_once_alloc(sizeof(char *)*
                                       (found_exts.elements+1),
                                       MYF(MY_WME | MY_FAE));
unknown's avatar
unknown committed
5486

5487
    DBUG_ASSERT(ext != 0);
5488 5489
    known_extensions.count= found_exts.elements;
    known_extensions.type_names= ext;
5490

unknown's avatar
unknown committed
5491
    List_iterator_fast<char> it(found_exts);
5492 5493 5494
    while ((old_ext= it++))
      *ext++= old_ext;
    *ext= 0;
5495 5496 5497
  }
  return &known_extensions;
}
unknown's avatar
unknown committed
5498

5499

unknown's avatar
unknown committed
5500 5501 5502
static bool stat_print(THD *thd, const char *type, uint type_len,
                       const char *file, uint file_len,
                       const char *status, uint status_len)
5503 5504 5505
{
  Protocol *protocol= thd->protocol;
  protocol->prepare_for_resend();
unknown's avatar
unknown committed
5506 5507 5508
  protocol->store(type, type_len, system_charset_info);
  protocol->store(file, file_len, system_charset_info);
  protocol->store(status, status_len, system_charset_info);
5509 5510 5511 5512 5513
  if (protocol->write())
    return TRUE;
  return FALSE;
}

unknown's avatar
unknown committed
5514

unknown's avatar
unknown committed
5515
static my_bool showstat_handlerton(THD *thd, plugin_ref plugin,
unknown's avatar
unknown committed
5516 5517 5518
                                   void *arg)
{
  enum ha_stat_type stat= *(enum ha_stat_type *) arg;
Sergei Golubchik's avatar
Sergei Golubchik committed
5519
  handlerton *hton= plugin_hton(plugin);
unknown's avatar
unknown committed
5520
  if (hton->state == SHOW_OPTION_YES && hton->show_status &&
5521
      hton->show_status(hton, thd, stat_print, stat))
unknown's avatar
unknown committed
5522 5523 5524 5525 5526
    return TRUE;
  return FALSE;
}

bool ha_show_status(THD *thd, handlerton *db_type, enum ha_stat_type stat)
5527 5528 5529
{
  List<Item> field_list;
  Protocol *protocol= thd->protocol;
unknown's avatar
unknown committed
5530
  bool result;
5531 5532 5533 5534 5535

  field_list.push_back(new Item_empty_string("Type",10));
  field_list.push_back(new Item_empty_string("Name",FN_REFLEN));
  field_list.push_back(new Item_empty_string("Status",10));

5536
  if (protocol->send_result_set_metadata(&field_list,
5537 5538 5539
                            Protocol::SEND_NUM_ROWS | Protocol::SEND_EOF))
    return TRUE;

unknown's avatar
unknown committed
5540
  if (db_type == NULL)
5541
  {
unknown's avatar
unknown committed
5542
    result= plugin_foreach(thd, showstat_handlerton,
unknown's avatar
unknown committed
5543 5544 5545 5546 5547
                           MYSQL_STORAGE_ENGINE_PLUGIN, &stat);
  }
  else
  {
    if (db_type->state != SHOW_OPTION_YES)
unknown's avatar
unknown committed
5548
    {
5549
      const LEX_STRING *name= hton_name(db_type);
unknown's avatar
unknown committed
5550
      result= stat_print(thd, name->str, name->length,
unknown's avatar
unknown committed
5551
                         "", 0, "DISABLED", 8) ? 1 : 0;
unknown's avatar
unknown committed
5552
    }
unknown's avatar
unknown committed
5553
    else
5554
    {
unknown's avatar
unknown committed
5555
      result= db_type->show_status &&
5556
              db_type->show_status(db_type, thd, stat_print, stat) ? 1 : 0;
5557
    }
5558 5559
  }

5560 5561 5562 5563 5564
  /*
    We also check thd->is_error() as Innodb may return 0 even if
    there was an error.
  */
  if (!result && !thd->is_error())
5565
    my_eof(thd);
5566
  else if (!thd->is_error())
5567
    my_error(ER_GET_ERRNO, MYF(0), errno, hton_name(db_type)->str);
unknown's avatar
unknown committed
5568
  return result;
5569 5570
}

5571 5572 5573 5574 5575
/*
  Function to check if the conditions for row-based binlogging is
  correct for the table.

  A row in the given table should be replicated if:
5576
  - Row-based replication is enabled in the current thread
5577
  - The binlog is enabled
5578 5579 5580
  - It is not a temporary table
  - The binary log is open
  - The database the table resides in shall be binlogged (binlog_*_db rules)
5581
  - table is not mysql.event
5582 5583
*/

5584 5585 5586
static bool check_table_binlog_row_based(THD *thd, TABLE *table)
{
  if (table->s->cached_row_logging_check == -1)
5587
  {
5588
    int const check(table->s->tmp_table == NO_TMP_TABLE &&
5589
                    ! table->no_replicate &&
5590 5591 5592
                    binlog_filter->db_ok(table->s->db.str));
    table->s->cached_row_logging_check= check;
  }
5593

5594 5595
  DBUG_ASSERT(table->s->cached_row_logging_check == 0 ||
              table->s->cached_row_logging_check == 1);
5596

5597
  return thd->is_current_stmt_binlog_format_row() &&
5598
          table->s->cached_row_logging_check &&
5599
          (thd->variables.option_bits & OPTION_BIN_LOG) &&
5600
          /* applier and replayer should not binlog */
5601 5602 5603
          ((IF_WSREP(WSREP_EMULATE_BINLOG(thd) &&
                     thd->wsrep_exec_mode != REPL_RECV, 0)) ||
           mysql_bin_log.is_open());
5604 5605 5606
}


5607
/** @brief
5608
   Write table maps for all (manually or automatically) locked tables
5609
   to the binary log. Also, if binlog_annotate_row_events is ON,
5610
   write Annotate_rows event before the first table map.
5611

5612 5613 5614 5615 5616 5617
   SYNOPSIS
     write_locked_table_maps()
       thd     Pointer to THD structure

   DESCRIPTION
       This function will generate and write table maps for all tables
Konstantin Osipov's avatar
Konstantin Osipov committed
5618
       that are locked by the thread 'thd'.
5619

5620 5621 5622 5623 5624 5625
   RETURN VALUE
       0   All OK
       1   Failed to write all table maps

   SEE ALSO
       THD::lock
5626
*/
5627

5628
static int write_locked_table_maps(THD *thd)
5629
{
5630
  DBUG_ENTER("write_locked_table_maps");
Konstantin Osipov's avatar
Konstantin Osipov committed
5631
  DBUG_PRINT("enter", ("thd: 0x%lx  thd->lock: 0x%lx "
5632
                       "thd->extra_lock: 0x%lx",
Konstantin Osipov's avatar
Konstantin Osipov committed
5633
                       (long) thd, (long) thd->lock, (long) thd->extra_lock));
5634

5635 5636
  DBUG_PRINT("debug", ("get_binlog_table_maps(): %d", thd->get_binlog_table_maps()));

5637 5638
  if (thd->get_binlog_table_maps() == 0)
  {
Konstantin Osipov's avatar
Konstantin Osipov committed
5639
    MYSQL_LOCK *locks[2];
5640 5641
    locks[0]= thd->extra_lock;
    locks[1]= thd->lock;
5642
    my_bool with_annotate= thd->variables.binlog_annotate_row_events &&
5643 5644
                           thd->query() && thd->query_length();

5645
    for (uint i= 0 ; i < sizeof(locks)/sizeof(*locks) ; ++i )
5646
    {
5647 5648 5649 5650 5651 5652 5653 5654
      MYSQL_LOCK const *const lock= locks[i];
      if (lock == NULL)
        continue;

      TABLE **const end_ptr= lock->table + lock->table_count;
      for (TABLE **table_ptr= lock->table ; 
           table_ptr != end_ptr ;
           ++table_ptr)
5655
      {
5656 5657 5658 5659
        TABLE *const table= *table_ptr;
        DBUG_PRINT("info", ("Checking table %s", table->s->table_name.str));
        if (table->current_lock == F_WRLCK &&
            check_table_binlog_row_based(thd, table))
5660
        {
5661 5662 5663 5664 5665 5666 5667 5668 5669 5670 5671 5672 5673 5674 5675
          /*
            We need to have a transactional behavior for SQLCOM_CREATE_TABLE
            (e.g. CREATE TABLE... SELECT * FROM TABLE) in order to keep a
            compatible behavior with the STMT based replication even when
            the table is not transactional. In other words, if the operation
            fails while executing the insert phase nothing is written to the
            binlog.

            Note that at this point, we check the type of a set of tables to
            create the table map events. In the function binlog_log_row(),
            which calls the current function, we check the type of the table
            of the current row.
          */
          bool const has_trans= thd->lex->sql_command == SQLCOM_CREATE_TABLE ||
                                table->file->has_transactions();
5676 5677
          int const error= thd->binlog_write_table_map(table, has_trans,
                                                       &with_annotate);
5678 5679 5680 5681 5682 5683
          /*
            If an error occurs, it is the responsibility of the caller to
            roll back the transaction.
          */
          if (unlikely(error))
            DBUG_RETURN(1);
5684
        }
5685 5686 5687
      }
    }
  }
5688 5689
  DBUG_RETURN(0);
}
5690

5691 5692 5693 5694

typedef bool Log_func(THD*, TABLE*, bool, MY_BITMAP*,
                      uint, const uchar*, const uchar*);

5695 5696 5697 5698 5699 5700 5701 5702
static int binlog_log_row(TABLE* table,
                          const uchar *before_record,
                          const uchar *after_record,
                          Log_func *log_func)
{
  bool error= 0;
  THD *const thd= table->in_use;

5703 5704 5705 5706
  /* only InnoDB tables will be replicated through binlog emulation */
  if (WSREP_EMULATE_BINLOG(thd) &&
      table->file->partition_ht()->db_type != DB_TYPE_INNODB)
    return 0;
5707

5708
  if (check_table_binlog_row_based(thd, table))
5709
  {
5710 5711 5712 5713 5714
    MY_BITMAP cols;
    /* Potential buffer on the stack for the bitmap */
    uint32 bitbuf[BITMAP_STACKBUF_SIZE/sizeof(uint32)];
    uint n_fields= table->s->fields;
    my_bool use_bitbuf= n_fields <= sizeof(bitbuf)*8;
5715

5716 5717 5718 5719 5720
    /*
      If there are no table maps written to the binary log, this is
      the first row handled in this statement. In that case, we need
      to write table maps for all locked tables to the binary log.
    */
5721
    if (likely(!(error= my_bitmap_init(&cols,
5722 5723 5724
                                    use_bitbuf ? bitbuf : NULL,
                                    (n_fields + 7) & ~7UL,
                                    FALSE))))
5725
    {
5726 5727
      bitmap_set_all(&cols);
      if (likely(!(error= write_locked_table_maps(thd))))
5728 5729 5730 5731 5732 5733 5734 5735 5736 5737 5738 5739
      {
        /*
          We need to have a transactional behavior for SQLCOM_CREATE_TABLE
          (i.e. CREATE TABLE... SELECT * FROM TABLE) in order to keep a
          compatible behavior with the STMT based replication even when
          the table is not transactional. In other words, if the operation
          fails while executing the insert phase nothing is written to the
          binlog.
        */
        bool const has_trans= thd->lex->sql_command == SQLCOM_CREATE_TABLE ||
                             table->file->has_transactions();
        error= (*log_func)(thd, table, has_trans, &cols, table->s->fields,
5740
                           before_record, after_record);
5741
      }
5742
      if (!use_bitbuf)
5743
        my_bitmap_free(&cols);
5744 5745
    }
  }
5746
  return error ? HA_ERR_RBR_LOGGING_FAILED : 0;
5747
}
5748

5749
int handler::ha_external_lock(THD *thd, int lock_type)
5750
{
5751
  int error;
5752
  DBUG_ENTER("handler::ha_external_lock");
5753 5754 5755 5756 5757 5758
  /*
    Whether this is lock or unlock, this should be true, and is to verify that
    if get_auto_increment() was called (thus may have reserved intervals or
    taken a table lock), ha_release_auto_increment() was too.
  */
  DBUG_ASSERT(next_insert_id == 0);
5759 5760 5761 5762 5763 5764
  /* Consecutive calls for lock without unlocking in between is not allowed */
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              ((lock_type != F_UNLCK && m_lock_type == F_UNLCK) ||
               lock_type == F_UNLCK));
  /* SQL HANDLER call locks/unlock while scanning (RND/INDEX). */
  DBUG_ASSERT(inited == NONE || table->open_by_handler);
5765

5766 5767 5768 5769 5770 5771 5772 5773 5774 5775 5776 5777 5778 5779 5780 5781 5782 5783 5784 5785 5786
  if (MYSQL_HANDLER_RDLOCK_START_ENABLED() ||
      MYSQL_HANDLER_WRLOCK_START_ENABLED() ||
      MYSQL_HANDLER_UNLOCK_START_ENABLED())
  {
    if (lock_type == F_RDLCK)
    {
      MYSQL_HANDLER_RDLOCK_START(table_share->db.str,
                                 table_share->table_name.str);
    }
    else if (lock_type == F_WRLCK)
    {
      MYSQL_HANDLER_WRLOCK_START(table_share->db.str,
                                 table_share->table_name.str);
    }
    else if (lock_type == F_UNLCK)
    {
      MYSQL_HANDLER_UNLOCK_START(table_share->db.str,
                                 table_share->table_name.str);
    }
  }

5787 5788
  ha_statistic_increment(&SSV::ha_external_lock_count);

5789 5790 5791 5792
  /*
    We cache the table flags if the locking succeeded. Otherwise, we
    keep them as they were when they were fetched in ha_open().
  */
5793 5794
  MYSQL_TABLE_LOCK_WAIT(m_psi, PSI_TABLE_EXTERNAL_LOCK, lock_type,
    { error= external_lock(thd, lock_type); })
5795

5796
  if (error == 0)
5797
  {
5798
    m_lock_type= lock_type;
5799
    cached_table_flags= table_flags();
5800 5801 5802
    if (table_share->tmp_table == NO_TMP_TABLE)
      mysql_audit_external_lock(thd, table_share, lock_type);
  }
5803

5804 5805 5806 5807 5808 5809 5810 5811 5812 5813 5814 5815 5816 5817 5818 5819 5820
  if (MYSQL_HANDLER_RDLOCK_DONE_ENABLED() ||
      MYSQL_HANDLER_WRLOCK_DONE_ENABLED() ||
      MYSQL_HANDLER_UNLOCK_DONE_ENABLED())
  {
    if (lock_type == F_RDLCK)
    {
      MYSQL_HANDLER_RDLOCK_DONE(error);
    }
    else if (lock_type == F_WRLCK)
    {
      MYSQL_HANDLER_WRLOCK_DONE(error);
    }
    else if (lock_type == F_UNLCK)
    {
      MYSQL_HANDLER_UNLOCK_DONE(error);
    }
  }
5821
  DBUG_RETURN(error);
5822 5823
}

5824

5825
/** @brief
5826 5827 5828 5829 5830
  Check handler usage and reset state of file to after 'open'
*/
int handler::ha_reset()
{
  DBUG_ENTER("ha_reset");
unknown's avatar
unknown committed
5831
  /* Check that we have called all proper deallocation functions */
5832
  DBUG_ASSERT((uchar*) table->def_read_set.bitmap +
5833
              table->s->column_bitmap_size ==
5834
              (uchar*) table->def_write_set.bitmap);
5835 5836 5837 5838 5839 5840
  DBUG_ASSERT(bitmap_is_set_all(&table->s->all_set));
  DBUG_ASSERT(table->key_read == 0);
  /* ensure that ha_index_end / ha_rnd_end has been called */
  DBUG_ASSERT(inited == NONE);
  /* Free cache used by filesort */
  free_io_cache(table);
5841 5842
  /* reset the bitmaps to point to defaults */
  table->default_column_bitmaps();
5843
  pushed_cond= NULL;
5844
  tracker= NULL;
5845 5846 5847
  /* Reset information about pushed engine conditions */
  cancel_pushed_idx_cond();
  /* Reset information about pushed index conditions */
5848 5849 5850 5851
  DBUG_RETURN(reset());
}


5852
int handler::ha_write_row(uchar *buf)
5853 5854
{
  int error;
5855
  Log_func *log_func= Write_rows_log_event::binlog_row_logging_function;
5856 5857
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type == F_WRLCK);
5858
  DBUG_ENTER("handler::ha_write_row");
5859
  DEBUG_SYNC_C("ha_write_row_start");
5860

5861
  MYSQL_INSERT_ROW_START(table_share->db.str, table_share->table_name.str);
5862
  mark_trx_read_write();
5863
  increment_statistics(&SSV::ha_write_count);
5864

5865
  TABLE_IO_WAIT(tracker, m_psi, PSI_TABLE_WRITE_ROW, MAX_KEY, 0,
5866 5867
                      { error= write_row(buf); })

5868 5869
  MYSQL_INSERT_ROW_DONE(error);
  if (unlikely(error))
5870
    DBUG_RETURN(error);
5871
  rows_changed++;
5872
  if (unlikely(error= binlog_log_row(table, 0, buf, log_func)))
5873
    DBUG_RETURN(error); /* purecov: inspected */
5874 5875

  DEBUG_SYNC_C("ha_write_row_end");
5876
  DBUG_RETURN(0);
5877 5878
}

5879

5880
int handler::ha_update_row(const uchar *old_data, uchar *new_data)
5881 5882
{
  int error;
5883
  Log_func *log_func= Update_rows_log_event::binlog_row_logging_function;
5884 5885
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type == F_WRLCK);
5886 5887 5888 5889 5890 5891

  /*
    Some storage engines require that the new record is in record[0]
    (and the old record is in record[1]).
   */
  DBUG_ASSERT(new_data == table->record[0]);
5892
  DBUG_ASSERT(old_data == table->record[1]);
5893

5894
  MYSQL_UPDATE_ROW_START(table_share->db.str, table_share->table_name.str);
5895
  mark_trx_read_write();
5896
  increment_statistics(&SSV::ha_update_count);
5897

5898
  TABLE_IO_WAIT(tracker, m_psi, PSI_TABLE_UPDATE_ROW, active_index, 0,
5899 5900
                      { error= update_row(old_data, new_data);})

5901 5902
  MYSQL_UPDATE_ROW_DONE(error);
  if (unlikely(error))
5903
    return error;
5904
  rows_changed++;
5905
  if (unlikely(error= binlog_log_row(table, old_data, new_data, log_func)))
5906 5907
    return error;
  return 0;
5908 5909
}

5910
int handler::ha_delete_row(const uchar *buf)
5911 5912
{
  int error;
5913
  Log_func *log_func= Delete_rows_log_event::binlog_row_logging_function;
5914 5915
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE ||
              m_lock_type == F_WRLCK);
5916 5917 5918 5919 5920
  /*
    Normally table->record[0] is used, but sometimes table->record[1] is used.
  */
  DBUG_ASSERT(buf == table->record[0] ||
              buf == table->record[1]);
5921

5922
  MYSQL_DELETE_ROW_START(table_share->db.str, table_share->table_name.str);
5923
  mark_trx_read_write();
5924
  increment_statistics(&SSV::ha_delete_count);
5925

5926
  TABLE_IO_WAIT(tracker, m_psi, PSI_TABLE_DELETE_ROW, active_index, 0,
5927
    { error= delete_row(buf);})
5928 5929
  MYSQL_DELETE_ROW_DONE(error);
  if (unlikely(error))
5930
    return error;
5931
  rows_changed++;
5932
  if (unlikely(error= binlog_log_row(table, buf, 0, log_func)))
5933 5934
    return error;
  return 0;
5935
}
5936

5937

5938

5939
/** @brief
5940 5941 5942 5943 5944 5945 5946
  use_hidden_primary_key() is called in case of an update/delete when
  (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
  but we don't have a primary key
*/
void handler::use_hidden_primary_key()
{
  /* fallback to use all columns in the table to identify row */
5947
  table->column_bitmaps_set(&table->s->all_set, table->write_set);
5948
}
5949 5950


5951 5952 5953 5954 5955 5956 5957 5958 5959 5960 5961 5962 5963 5964 5965 5966 5967 5968 5969 5970 5971 5972 5973 5974 5975 5976 5977 5978 5979 5980 5981 5982 5983 5984 5985 5986 5987 5988 5989 5990 5991 5992 5993 5994 5995 5996 5997 5998 5999 6000 6001 6002 6003 6004 6005 6006 6007 6008 6009 6010 6011 6012 6013 6014 6015 6016 6017 6018 6019 6020 6021
/**
  Get an initialized ha_share.

  @return Initialized ha_share
    @retval NULL    ha_share is not yet initialized.
    @retval != NULL previous initialized ha_share.

  @note
  If not a temp table, then LOCK_ha_data must be held.
*/

Handler_share *handler::get_ha_share_ptr()
{
  DBUG_ENTER("handler::get_ha_share_ptr");
  DBUG_ASSERT(ha_share && table_share);

#ifndef DBUG_OFF
  if (table_share->tmp_table == NO_TMP_TABLE)
    mysql_mutex_assert_owner(&table_share->LOCK_ha_data);
#endif

  DBUG_RETURN(*ha_share);
}


/**
  Set ha_share to be used by all instances of the same table/partition.

  @param ha_share    Handler_share to be shared.

  @note
  If not a temp table, then LOCK_ha_data must be held.
*/

void handler::set_ha_share_ptr(Handler_share *arg_ha_share)
{
  DBUG_ENTER("handler::set_ha_share_ptr");
  DBUG_ASSERT(ha_share);
#ifndef DBUG_OFF
  if (table_share->tmp_table == NO_TMP_TABLE)
    mysql_mutex_assert_owner(&table_share->LOCK_ha_data);
#endif

  *ha_share= arg_ha_share;
  DBUG_VOID_RETURN;
}


/**
  Take a lock for protecting shared handler data.
*/

void handler::lock_shared_ha_data()
{
  DBUG_ASSERT(table_share);
  if (table_share->tmp_table == NO_TMP_TABLE)
    mysql_mutex_lock(&table_share->LOCK_ha_data);
}


/**
  Release lock for protecting ha_share.
*/

void handler::unlock_shared_ha_data()
{
  DBUG_ASSERT(table_share);
  if (table_share->tmp_table == NO_TMP_TABLE)
    mysql_mutex_unlock(&table_share->LOCK_ha_data);
}

6022
/** @brief
6023 6024 6025 6026 6027 6028 6029 6030 6031 6032
  Dummy function which accept information about log files which is not need
  by handlers
*/
void signal_log_not_needed(struct handlerton, char *log_file)
{
  DBUG_ENTER("signal_log_not_needed");
  DBUG_PRINT("enter", ("logfile '%s'", log_file));
  DBUG_VOID_RETURN;
}

6033
void handler::set_lock_type(enum thr_lock_type lock)
6034 6035 6036
{
  table->reginfo.lock_type= lock;
}
6037

6038 6039 6040 6041 6042 6043 6044 6045
#ifdef WITH_WSREP
/**
  @details
  This function makes the storage engine to force the victim transaction
  to abort. Currently, only innodb has this functionality, but any SE
  implementing the wsrep API should provide this service to support
  multi-master operation.

6046 6047 6048
  @note Aborting the transaction does NOT end it, it still has to
  be rolled back with hton->rollback().

6049 6050 6051 6052 6053 6054 6055 6056 6057 6058 6059 6060 6061 6062 6063 6064 6065 6066 6067 6068 6069 6070 6071
  @param bf_thd       brute force THD asking for the abort
  @param victim_thd   victim THD to be aborted

  @return
    always 0
*/

int ha_abort_transaction(THD *bf_thd, THD *victim_thd, my_bool signal)
{
  DBUG_ENTER("ha_abort_transaction");
  if (!WSREP(bf_thd) &&
      !(wsrep_OSU_method_options == WSREP_OSU_RSU &&
        bf_thd->wsrep_exec_mode == TOTAL_ORDER)) {
    DBUG_RETURN(0);
  }

  THD_TRANS *trans= &victim_thd->transaction.all;
  Ha_trx_info *ha_info= trans->ha_list, *ha_info_next;

  for (; ha_info; ha_info= ha_info_next)
  {
    handlerton *hton= ha_info->ht();
    if (!hton->abort_transaction)
6072
      WSREP_WARN("cannot abort transaction");
6073
    else
6074
      hton->abort_transaction(hton, bf_thd, victim_thd, signal);
6075 6076 6077 6078 6079 6080 6081 6082 6083 6084 6085 6086 6087 6088 6089 6090 6091 6092 6093 6094 6095 6096 6097 6098 6099 6100 6101 6102 6103 6104 6105 6106 6107
    ha_info_next= ha_info->next();
  }
  DBUG_RETURN(0);
}

void ha_fake_trx_id(THD *thd)
{
  DBUG_ENTER("ha_fake_trx_id");
  if (!WSREP(thd))
  {
    DBUG_VOID_RETURN;
  }

  THD_TRANS *trans= &thd->transaction.all;
  Ha_trx_info *ha_info= trans->ha_list, *ha_info_next;

  for (; ha_info; ha_info= ha_info_next)
  {
    handlerton *hton= ha_info->ht();
    if (!hton->fake_trx_id)
    {
      WSREP_WARN("cannot get fake InnoDB transaction ID");
    }
    else
      hton->fake_trx_id(hton, thd);
    ha_info_next= ha_info->next();
    ha_info->reset(); /* keep it conveniently zero-filled */
  }
  DBUG_VOID_RETURN;
}
#endif /* WITH_WSREP */


6108 6109 6110 6111 6112 6113 6114 6115 6116 6117 6118 6119 6120 6121 6122 6123 6124
#ifdef TRANS_LOG_MGM_EXAMPLE_CODE
/*
  Example of transaction log management functions based on assumption that logs
  placed into a directory
*/
#include <my_dir.h>
#include <my_sys.h>
int example_of_iterator_using_for_logs_cleanup(handlerton *hton)
{
  void *buffer;
  int res= 1;
  struct handler_iterator iterator;
  struct handler_log_file_data data;

  if (!hton->create_iterator)
    return 1; /* iterator creator is not supported */

6125
  if ((*hton->create_iterator)(hton, HA_TRANSACTLOG_ITERATOR, &iterator) !=
6126 6127 6128 6129 6130 6131 6132 6133 6134
      HA_ITERATOR_OK)
  {
    /* error during creation of log iterator or iterator is not supported */
    return 1;
  }
  while((*iterator.next)(&iterator, (void*)&data) == 0)
  {
    printf("%s\n", data.filename.str);
    if (data.status == HA_LOG_STATUS_FREE &&
Marc Alff's avatar
Marc Alff committed
6135 6136
        mysql_file_delete(INSTRUMENT_ME,
                          data.filename.str, MYF(MY_WME)))
6137 6138 6139 6140 6141 6142 6143 6144 6145 6146 6147 6148 6149 6150 6151 6152 6153 6154 6155
      goto err;
  }
  res= 0;
err:
  (*iterator.destroy)(&iterator);
  return res;
}


/*
  Here we should get info from handler where it save logs but here is
  just example, so we use constant.
  IMHO FN_ROOTDIR ("/") is safe enough for example, because nobody has
  rights on it except root and it consist of directories only at lest for
  *nix (sorry, can't find windows-safe solution here, but it is only example).
*/
#define fl_dir FN_ROOTDIR


6156
/** @brief
6157 6158 6159 6160 6161 6162 6163
  Dummy function to return log status should be replaced by function which
  really detect the log status and check that the file is a log of this
  handler.
*/
enum log_status fl_get_log_status(char *log)
{
  MY_STAT stat_buff;
Marc Alff's avatar
Marc Alff committed
6164
  if (mysql_file_stat(INSTRUMENT_ME, log, &stat_buff, MYF(0)))
6165 6166 6167 6168 6169 6170 6171 6172 6173 6174 6175 6176 6177 6178 6179 6180 6181 6182 6183 6184 6185 6186 6187 6188 6189 6190 6191 6192 6193 6194 6195
    return HA_LOG_STATUS_INUSE;
  return HA_LOG_STATUS_NOSUCHLOG;
}


struct fl_buff
{
  LEX_STRING *names;
  enum log_status *statuses;
  uint32 entries;
  uint32 current;
};


int fl_log_iterator_next(struct handler_iterator *iterator,
                          void *iterator_object)
{
  struct fl_buff *buff= (struct fl_buff *)iterator->buffer;
  struct handler_log_file_data *data=
    (struct handler_log_file_data *) iterator_object;
  if (buff->current >= buff->entries)
    return 1;
  data->filename= buff->names[buff->current];
  data->status= buff->statuses[buff->current];
  buff->current++;
  return 0;
}


void fl_log_iterator_destroy(struct handler_iterator *iterator)
{
6196
  my_free(iterator->buffer);
6197 6198 6199
}


6200
/** @brief
6201 6202 6203 6204 6205 6206 6207 6208
  returns buffer, to be assigned in handler_iterator struct
*/
enum handler_create_iterator_result
fl_log_iterator_buffer_init(struct handler_iterator *iterator)
{
  MY_DIR *dirp;
  struct fl_buff *buff;
  char *name_ptr;
6209
  uchar *ptr;
6210 6211 6212 6213 6214 6215
  FILEINFO *file;
  uint32 i;

  /* to be able to make my_free without crash in case of error */
  iterator->buffer= 0;

6216
  if (!(dirp = my_dir(fl_dir, MYF(MY_THREAD_SPECIFIC))))
6217 6218 6219
  {
    return HA_ITERATOR_ERROR;
  }
6220
  if ((ptr= (uchar*)my_malloc(ALIGN_SIZE(sizeof(fl_buff)) +
6221 6222
                             ((ALIGN_SIZE(sizeof(LEX_STRING)) +
                               sizeof(enum log_status) +
6223
                               + FN_REFLEN + 1) *
6224
                              (uint) dirp->number_off_files),
6225
                             MYF(MY_THREAD_SPECIFIC))) == 0)
6226 6227 6228 6229 6230 6231 6232 6233 6234 6235 6236 6237 6238 6239 6240 6241 6242 6243 6244 6245 6246 6247 6248 6249 6250
  {
    return HA_ITERATOR_ERROR;
  }
  buff= (struct fl_buff *)ptr;
  buff->entries= buff->current= 0;
  ptr= ptr + (ALIGN_SIZE(sizeof(fl_buff)));
  buff->names= (LEX_STRING*) (ptr);
  ptr= ptr + ((ALIGN_SIZE(sizeof(LEX_STRING)) *
               (uint) dirp->number_off_files));
  buff->statuses= (enum log_status *)(ptr);
  name_ptr= (char *)(ptr + (sizeof(enum log_status) *
                            (uint) dirp->number_off_files));
  for (i=0 ; i < (uint) dirp->number_off_files  ; i++)
  {
    enum log_status st;
    file= dirp->dir_entry + i;
    if ((file->name[0] == '.' &&
         ((file->name[1] == '.' && file->name[2] == '\0') ||
            file->name[1] == '\0')))
      continue;
    if ((st= fl_get_log_status(file->name)) == HA_LOG_STATUS_NOSUCHLOG)
      continue;
    name_ptr= strxnmov(buff->names[buff->entries].str= name_ptr,
                       FN_REFLEN, fl_dir, file->name, NullS);
    buff->names[buff->entries].length= (name_ptr -
6251
                                        buff->names[buff->entries].str);
6252 6253 6254 6255 6256 6257 6258
    buff->statuses[buff->entries]= st;
    buff->entries++;
  }

  iterator->buffer= buff;
  iterator->next= &fl_log_iterator_next;
  iterator->destroy= &fl_log_iterator_destroy;
6259
  my_dirend(dirp);
6260 6261 6262 6263 6264 6265 6266 6267 6268
  return HA_ITERATOR_OK;
}


/* An example of a iterator creator */
enum handler_create_iterator_result
fl_create_iterator(enum handler_iterator_type type,
                   struct handler_iterator *iterator)
{
6269
  switch(type) {
6270 6271 6272 6273 6274 6275 6276
  case HA_TRANSACTLOG_ITERATOR:
    return fl_log_iterator_buffer_init(iterator);
  default:
    return HA_ITERATOR_UNSUPPORTED;
  }
}
#endif /*TRANS_LOG_MGM_EXAMPLE_CODE*/