ha_berkeley.cc 70.3 KB
Newer Older
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1
/* Copyright (C) 2000 MySQL AB & MySQL Finland AB & TCX DataKonsult AB
2

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

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

bk@work.mysql.com's avatar
bk@work.mysql.com committed
13 14 15 16 17 18 19 20 21 22 23
   You should have received a copy of the GNU General Public License
   along with this program; if not, write to the Free Software
   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA */


/*
  TODO:
  - Not compressed keys should use cmp_fix_length_key
  - Don't automaticly pack all string keys (To do this we need to modify
    CREATE TABLE so that one can use the pack_keys argument per key).
  - An argument to pack_key that we don't want compression.
24
  - DB_DBT_USERMEM should be used for fixed length tables
bk@work.mysql.com's avatar
bk@work.mysql.com committed
25 26 27
    We will need an updated Berkeley DB version for this.
  - Killing threads that has got a 'deadlock'
  - SHOW TABLE STATUS should give more information about the table.
28 29 30
  - Get a more accurate count of the number of rows (estimate_number_of_rows()).
    We could store the found number of rows when the table is scanned and
    then increment the counter for each attempted write.
31 32
  - We will need to extend the manager thread to makes checkpoints at
     given intervals.
bk@work.mysql.com's avatar
bk@work.mysql.com committed
33 34 35 36 37 38 39
  - When not using UPDATE IGNORE, don't make a sub transaction but abort
    the main transaction on errors.
  - Handling of drop table during autocommit=0 ?
    (Should we just give an error in this case if there is a pending
    transaction ?)
  - When using ALTER TABLE IGNORE, we should not start an transaction, but do
    everything wthout transactions.
40 41
  - When we do rollback, we need to subtract the number of changed rows
    from the updated tables.
bk@work.mysql.com's avatar
bk@work.mysql.com committed
42 43

  Testing of:
44 45
  - Mark tables that participate in a transaction so that they are not
    closed during the transaction.  We need to test what happens if
46
    MySQL closes a table that is updated by a not commited transaction.
bk@work.mysql.com's avatar
bk@work.mysql.com committed
47 48 49 50 51 52 53 54 55 56 57 58 59
*/


#ifdef __GNUC__
#pragma implementation				// gcc: Class implementation
#endif

#include "mysql_priv.h"
#ifdef HAVE_BERKELEY_DB
#include <m_ctype.h>
#include <myisampack.h>
#include <hash.h>
#include "ha_berkeley.h"
60
#include "sql_manager.h"
61
#include <stdarg.h>
bk@work.mysql.com's avatar
bk@work.mysql.com committed
62 63

#define HA_BERKELEY_ROWS_IN_TABLE 10000 /* to get optimization right */
64
#define HA_BERKELEY_RANGE_COUNT   100
65
#define HA_BERKELEY_MAX_ROWS	  10000000 /* Max rows in table */
66 67 68 69 70 71 72
/* extra rows for estimate_number_of_rows() */
#define HA_BERKELEY_EXTRA_ROWS	  100

/* Bits for share->status */
#define STATUS_PRIMARY_KEY_INIT 1
#define STATUS_ROW_COUNT_INIT	2
#define STATUS_BDB_ANALYZE	4
bk@work.mysql.com's avatar
bk@work.mysql.com committed
73 74

const char *ha_berkeley_ext=".db";
75
bool berkeley_skip=0,berkeley_shared_data=0;
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
76 77
u_int32_t berkeley_init_flags= DB_PRIVATE | DB_RECOVER, berkeley_env_flags=0,
          berkeley_lock_type=DB_LOCK_DEFAULT;
78
ulong berkeley_cache_size, berkeley_log_buffer_size, berkeley_log_file_size=0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
79 80
char *berkeley_home, *berkeley_tmpdir, *berkeley_logdir;
long berkeley_lock_scan_time=0;
81
ulong berkeley_trans_retry=1;
82
ulong berkeley_max_lock;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
83 84 85 86 87 88
pthread_mutex_t bdb_mutex;

static DB_ENV *db_env;
static HASH bdb_open_tables;

const char *berkeley_lock_names[] =
89
{ "DEFAULT", "OLDEST","RANDOM","YOUNGEST",0 };
bk@work.mysql.com's avatar
bk@work.mysql.com committed
90 91
u_int32_t berkeley_lock_types[]=
{ DB_LOCK_DEFAULT, DB_LOCK_OLDEST, DB_LOCK_RANDOM };
92
TYPELIB berkeley_lock_typelib= {array_elements(berkeley_lock_names)-1,"",
bk@work.mysql.com's avatar
bk@work.mysql.com committed
93 94 95 96 97
				berkeley_lock_names};

static void berkeley_print_error(const char *db_errpfx, char *buffer);
static byte* bdb_get_key(BDB_SHARE *share,uint *length,
			 my_bool not_used __attribute__((unused)));
98
static BDB_SHARE *get_share(const char *table_name, TABLE *table);
99 100
static int free_share(BDB_SHARE *share, TABLE *table, uint hidden_primary_key,
		      bool mutex_is_locked);
101
static int write_status(DB *status_block, char *buff, uint length);
102
static void update_status(BDB_SHARE *share, TABLE *table);
103
static void berkeley_noticecall(DB_ENV *db_env, db_notices notice);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
104 105


106

bk@work.mysql.com's avatar
bk@work.mysql.com committed
107 108 109 110 111 112 113 114 115 116
/* General functions */

bool berkeley_init(void)
{
  DBUG_ENTER("berkeley_init");

  if (!berkeley_tmpdir)
    berkeley_tmpdir=mysql_tmpdir;
  if (!berkeley_home)
    berkeley_home=mysql_real_data_home;
117 118
  DBUG_PRINT("bdb",("berkeley_home: %s",mysql_real_data_home));

119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135
  /*
    If we don't set set_lg_bsize() we will get into trouble when
    trying to use many open BDB tables.
    If log buffer is not set, assume that the we will need 512 byte per
    open table.  This is a number that we have reached by testing.
  */
  if (!berkeley_log_buffer_size)
  {
    berkeley_log_buffer_size= max(table_cache_size*512,32*1024);
  }
  /*
    Berkeley DB require that
    berkeley_log_file_size >= berkeley_log_buffer_size*4
  */
  berkeley_log_file_size= berkeley_log_buffer_size*4;
  berkeley_log_file_size= MY_ALIGN(berkeley_log_file_size,1024*1024L);
  berkeley_log_file_size= max(berkeley_log_file_size, 10*1024*1024L);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
136 137

  if (db_env_create(&db_env,0))
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
138
    DBUG_RETURN(1); /* purecov: inspected */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
139 140
  db_env->set_errcall(db_env,berkeley_print_error);
  db_env->set_errpfx(db_env,"bdb");
141
  db_env->set_noticecall(db_env, berkeley_noticecall);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
142 143
  db_env->set_tmp_dir(db_env, berkeley_tmpdir);
  db_env->set_data_dir(db_env, mysql_data_home);
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
144
  db_env->set_flags(db_env, berkeley_env_flags, 1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
145
  if (berkeley_logdir)
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
146
    db_env->set_lg_dir(db_env, berkeley_logdir); /* purecov: tested */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
147 148 149 150 151

  if (opt_endinfo)
    db_env->set_verbose(db_env,
			DB_VERB_CHKPOINT | DB_VERB_DEADLOCK | DB_VERB_RECOVERY,
			1);
152

bk@work.mysql.com's avatar
bk@work.mysql.com committed
153
  db_env->set_cachesize(db_env, 0, berkeley_cache_size, 0);
154 155
  db_env->set_lg_max(db_env, berkeley_log_file_size);
  db_env->set_lg_bsize(db_env, berkeley_log_buffer_size);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
156
  db_env->set_lk_detect(db_env, berkeley_lock_type);
157 158
  if (berkeley_max_lock)
    db_env->set_lk_max(db_env, berkeley_max_lock);
159

bk@work.mysql.com's avatar
bk@work.mysql.com committed
160 161
  if (db_env->open(db_env,
		   berkeley_home,
162
		   berkeley_init_flags |  DB_INIT_LOCK |
bk@work.mysql.com's avatar
bk@work.mysql.com committed
163
		   DB_INIT_LOG | DB_INIT_MPOOL | DB_INIT_TXN |
164
		   DB_CREATE | DB_THREAD, 0666))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
165
  {
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
166 167
    db_env->close(db_env,0); /* purecov: inspected */
    db_env=0; /* purecov: inspected */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
168
  }
169

170
  (void) hash_init(&bdb_open_tables,system_charset_info,32,0,0,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
171
		   (hash_get_key) bdb_get_key,0,0);
172
  pthread_mutex_init(&bdb_mutex,MY_MUTEX_INIT_FAST);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
173 174 175 176 177 178 179 180 181
  DBUG_RETURN(db_env == 0);
}


bool berkeley_end(void)
{
  int error;
  DBUG_ENTER("berkeley_end");
  if (!db_env)
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
182
    return 1; /* purecov: tested */
183
  berkeley_cleanup_log_files();
bk@work.mysql.com's avatar
bk@work.mysql.com committed
184 185 186 187 188 189 190 191 192 193 194 195
  error=db_env->close(db_env,0);		// Error is logged
  db_env=0;
  hash_free(&bdb_open_tables);
  pthread_mutex_destroy(&bdb_mutex);
  DBUG_RETURN(error != 0);
}

bool berkeley_flush_logs()
{
  int error;
  bool result=0;
  DBUG_ENTER("berkeley_flush_logs");
ram@mysql.r18.ru's avatar
ram@mysql.r18.ru committed
196
  if ((error=db_env->log_flush(db_env,0)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
197
  {
198 199
    my_error(ER_ERROR_DURING_FLUSH_LOGS,MYF(0),error); /* purecov: inspected */
    result=1; /* purecov: inspected */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
200
  }
ram@mysql.r18.ru's avatar
ram@mysql.r18.ru committed
201
  if ((error=db_env->txn_checkpoint(db_env,0,0,0)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
202
  {
203 204
    my_error(ER_ERROR_DURING_CHECKPOINT,MYF(0),error); /* purecov: inspected */
    result=1; /* purecov: inspected */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
205 206 207 208 209
  }
  DBUG_RETURN(result);
}


210
int berkeley_commit(THD *thd, void *trans)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
211 212
{
  DBUG_ENTER("berkeley_commit");
213 214
  DBUG_PRINT("trans",("ending transaction %s",
		      trans == thd->transaction.stmt.bdb_tid ? "stmt" : "all"));
215
  int error=txn_commit((DB_TXN*) trans,0);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
216 217
#ifndef DBUG_OFF
  if (error)
218
    DBUG_PRINT("error",("error: %d",error)); /* purecov: inspected */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
219 220 221 222
#endif
  DBUG_RETURN(error);
}

223
int berkeley_rollback(THD *thd, void *trans)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
224 225
{
  DBUG_ENTER("berkeley_rollback");
226 227
  DBUG_PRINT("trans",("aborting transaction %s",
		      trans == thd->transaction.stmt.bdb_tid ? "stmt" : "all"));
228
  int error=txn_abort((DB_TXN*) trans);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
229 230 231
  DBUG_RETURN(error);
}

tim@cane.mysql.fi's avatar
tim@cane.mysql.fi committed
232

233
int berkeley_show_logs(Protocol *protocol)
tim@cane.mysql.fi's avatar
tim@cane.mysql.fi committed
234
{
235 236 237 238
  char **all_logs, **free_logs, **a, **f;
  int error=1;
  MEM_ROOT show_logs_root;
  MEM_ROOT *old_root=my_pthread_getspecific_ptr(MEM_ROOT*,THR_MALLOC);
tim@cane.mysql.fi's avatar
tim@cane.mysql.fi committed
239 240 241
  DBUG_ENTER("berkeley_show_logs");

  init_alloc_root(&show_logs_root, 1024, 1024);
242 243
  my_pthread_setspecific_ptr(THR_MALLOC,&show_logs_root);

244 245 246
  if ((error= db_env->log_archive(db_env, &all_logs,
				  DB_ARCH_ABS | DB_ARCH_LOG)) ||
      (error= db_env->log_archive(db_env, &free_logs, DB_ARCH_ABS)))
tim@cane.mysql.fi's avatar
tim@cane.mysql.fi committed
247 248 249
  {
    DBUG_PRINT("error", ("log_archive failed (error %d)", error));
    db_env->err(db_env, error, "log_archive: DB_ARCH_ABS");
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
250 251
    if (error== DB_NOTFOUND)
      error=0;					// No log files
252
    goto err;
tim@cane.mysql.fi's avatar
tim@cane.mysql.fi committed
253
  }
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
254 255
  /* Error is 0 here */
  if (all_logs)
tim@cane.mysql.fi's avatar
tim@cane.mysql.fi committed
256
  {
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
257
    for (a = all_logs, f = free_logs; *a; ++a)
tim@cane.mysql.fi's avatar
tim@cane.mysql.fi committed
258
    {
259
      protocol->prepare_for_resend();
260 261
      protocol->store(*a, system_charset_info);
      protocol->store("BDB", 3, system_charset_info);
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
262 263
      if (f && *f && strcmp(*a, *f) == 0)
      {
264
	f++;
265
	protocol->store(SHOW_LOG_STATUS_FREE, system_charset_info);
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
266 267
      }
      else
268
	protocol->store(SHOW_LOG_STATUS_INUSE, system_charset_info);
tim@cane.mysql.fi's avatar
tim@cane.mysql.fi committed
269

270
      if (protocol->write())
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
271 272 273 274 275
      {
	error=1;
	goto err;
      }
    }
tim@cane.mysql.fi's avatar
tim@cane.mysql.fi committed
276
  }
277
err:
tim@cane.mysql.fi's avatar
tim@cane.mysql.fi committed
278
  free_root(&show_logs_root,MYF(0));
279 280
  my_pthread_setspecific_ptr(THR_MALLOC,old_root);
  DBUG_RETURN(error);
tim@cane.mysql.fi's avatar
tim@cane.mysql.fi committed
281
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
282 283 284

static void berkeley_print_error(const char *db_errpfx, char *buffer)
{
285
  sql_print_error("%s:  %s",db_errpfx,buffer); /* purecov: tested */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
286 287
}

288 289 290 291
static void berkeley_noticecall(DB_ENV *db_env, db_notices notice)
{
  switch (notice)
  {
292
  case DB_NOTICE_LOGFILE_CHANGED: /* purecov: tested */
293 294 295 296 297 298 299 300 301 302 303 304 305 306
    pthread_mutex_lock(&LOCK_manager);
    manager_status |= MANAGER_BERKELEY_LOG_CLEANUP;
    pthread_mutex_unlock(&LOCK_manager);
    pthread_cond_signal(&COND_manager);
    break;
  }
}

void berkeley_cleanup_log_files(void)
{
  DBUG_ENTER("berkeley_cleanup_log_files");
  char **names;
  int error;

hf@deer.(none)'s avatar
hf@deer.(none) committed
307 308
// by HF. Sometimes it crashes. TODO - find out why
#ifndef EMBEDDED_LIBRARY
309 310
  /* XXX: Probably this should be done somewhere else, and
   * should be tunable by the user. */
ram@mysql.r18.ru's avatar
ram@mysql.r18.ru committed
311
  if ((error = db_env->txn_checkpoint(db_env, 0, 0, 0)))
312
    my_error(ER_ERROR_DURING_CHECKPOINT, MYF(0), error); /* purecov: inspected */
hf@deer.(none)'s avatar
hf@deer.(none) committed
313
#endif
ram@mysql.r18.ru's avatar
ram@mysql.r18.ru committed
314
  if ((error = db_env->log_archive(db_env, &names, DB_ARCH_ABS)) != 0)
315
  {
316 317 318
    DBUG_PRINT("error", ("log_archive failed (error %d)", error)); /* purecov: inspected */
    db_env->err(db_env, error, "log_archive: DB_ARCH_ABS"); /* purecov: inspected */
    DBUG_VOID_RETURN; /* purecov: inspected */
319 320 321
  }

  if (names)
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
322 323 324 325
  {						/* purecov: tested */
    char **np;					/* purecov: tested */
    for (np = names; *np; ++np)			/* purecov: tested */
      my_delete(*np, MYF(MY_WME));		/* purecov: tested */
326

monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
327
    free(names);				/* purecov: tested */
328 329 330 331
  }

  DBUG_VOID_RETURN;
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
332 333 334 335 336 337 338 339 340 341


/*****************************************************************************
** Berkeley DB tables
*****************************************************************************/

const char **ha_berkeley::bas_ext() const
{ static const char *ext[]= { ha_berkeley_ext, NullS }; return ext; }


342
static int
343
berkeley_cmp_hidden_key(DB* file, const DBT *new_key, const DBT *saved_key)
344 345 346
{
  ulonglong a=uint5korr((char*) new_key->data);
  ulonglong b=uint5korr((char*) saved_key->data);
347
  return  a < b ? -1 : (a > b ? 1 : 0);
348 349
}

bk@work.mysql.com's avatar
bk@work.mysql.com committed
350
static int
351
berkeley_cmp_packed_key(DB *file, const DBT *new_key, const DBT *saved_key)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
352
{
353 354
  KEY *key=	      (new_key->app_private ? (KEY*) new_key->app_private :
		       (KEY*) (file->app_private));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
355 356 357 358 359
  char *new_key_ptr=  (char*) new_key->data;
  char *saved_key_ptr=(char*) saved_key->data;
  KEY_PART_INFO *key_part= key->key_part, *end=key_part+key->key_parts;
  uint key_length=new_key->size;

360
  for (; key_part != end && (int) key_length > 0; key_part++)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
361 362 363 364
  {
    int cmp;
    if (key_part->null_bit)
    {
365 366
      if (*new_key_ptr != *saved_key_ptr++)
	return ((int) *new_key_ptr - (int) saved_key_ptr[-1]);
367
      key_length--;
368 369
      if (!*new_key_ptr++)
	continue;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
370 371 372 373
    }
    if ((cmp=key_part->field->pack_cmp(new_key_ptr,saved_key_ptr,
				       key_part->length)))
      return cmp;
374 375
    uint length=key_part->field->packed_col_length(new_key_ptr,
						   key_part->length);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
376 377
    new_key_ptr+=length;
    key_length-=length;
378 379
    saved_key_ptr+=key_part->field->packed_col_length(saved_key_ptr,
						      key_part->length);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
380
  }
381
  return key->handler.bdb_return_if_eq;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
382 383 384
}


385 386
/* The following is not yet used; Should be used for fixed length keys */

387
#ifdef NOT_YET
bk@work.mysql.com's avatar
bk@work.mysql.com committed
388
static int
389
berkeley_cmp_fix_length_key(DB *file, const DBT *new_key, const DBT *saved_key)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
390
{
391 392
  KEY *key=	      (new_key->app_private ? (KEY*) new_key->app_private :
		       (KEY*) (file->app_private));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
393 394 395 396 397
  char *new_key_ptr=  (char*) new_key->data;
  char *saved_key_ptr=(char*) saved_key->data;
  KEY_PART_INFO *key_part= key->key_part, *end=key_part+key->key_parts;
  uint key_length=new_key->size;

398
  for (; key_part != end && (int) key_length > 0 ; key_part++)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
399 400 401 402 403 404 405 406
  {
    int cmp;
    if ((cmp=key_part->field->pack_cmp(new_key_ptr,saved_key_ptr,0)))
      return cmp;
    new_key_ptr+=key_part->length;
    key_length-= key_part->length;
    saved_key_ptr+=key_part->length;
  }
407
  return key->handler.bdb_return_if_eq;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
408
}
409
#endif
bk@work.mysql.com's avatar
bk@work.mysql.com committed
410

411 412 413 414 415 416
/* Compare key against row */

static bool
berkeley_key_cmp(TABLE *table, KEY *key_info, const char *key, uint key_length)
{
  KEY_PART_INFO *key_part= key_info->key_part,
417
		*end=key_part+key_info->key_parts;
418

419
  for (; key_part != end && (int) key_length > 0; key_part++)
420 421 422 423 424
  {
    int cmp;
    if (key_part->null_bit)
    {
      key_length--;
425 426 427 428
      /*
	With the current usage, the following case will always be FALSE,
	because NULL keys are sorted before any other key
      */
429 430 431
      if (*key != (table->record[0][key_part->null_offset] &
		   key_part->null_bit) ? 0 : 1)
	return 1;
432
      if (!*key++)				// Null value
433 434 435 436
	continue;
    }
    if ((cmp=key_part->field->pack_cmp(key,key_part->length)))
      return cmp;
437
    uint length=key_part->field->packed_col_length(key,key_part->length);
438 439 440
    key+=length;
    key_length-=length;
  }
441
  return 0;					// Identical keys
442 443 444
}


445
int ha_berkeley::open(const char *name, int mode, uint test_if_locked)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
446 447 448 449 450 451
{
  char name_buff[FN_REFLEN];
  uint open_mode=(mode == O_RDONLY ? DB_RDONLY : 0) | DB_THREAD;
  int error;
  DBUG_ENTER("ha_berkeley::open");

452 453 454 455 456
  /* Open primary key */
  hidden_primary_key=0;
  if ((primary_key=table->primary_key) >= MAX_KEY)
  {						// No primary key
    primary_key=table->keys;
457
    key_used_on_scan=MAX_KEY;
458 459
    ref_length=hidden_primary_key=BDB_HIDDEN_PRIMARY_KEY_LENGTH;
  }
460 461
  else
    key_used_on_scan=primary_key;
462

bk@work.mysql.com's avatar
bk@work.mysql.com committed
463
  /* Need some extra memory in case of packed keys */
464
  uint max_key_length= table->max_key_length + MAX_REF_PARTS*3;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
465 466 467 468 469
  if (!(alloc_ptr=
	my_multi_malloc(MYF(MY_WME),
			&key_buff,  max_key_length,
			&key_buff2, max_key_length,
			&primary_key_buff,
470 471
			(hidden_primary_key ? 0 :
			 table->key_info[table->primary_key].key_length),
bk@work.mysql.com's avatar
bk@work.mysql.com committed
472
			NullS)))
473
    DBUG_RETURN(1); /* purecov: inspected */
474 475 476
  if (!(rec_buff= (byte*) my_malloc((alloced_rec_buff_length=
				     table->rec_buff_length),
				    MYF(MY_WME))))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
477
  {
478 479
    my_free(alloc_ptr,MYF(0)); /* purecov: inspected */
    DBUG_RETURN(1); /* purecov: inspected */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
480 481
  }

482
  /* Init shared structure */
483
  if (!(share=get_share(name,table)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
484
  {
485
    my_free((char*) rec_buff,MYF(0)); /* purecov: inspected */
486 487
    my_free(alloc_ptr,MYF(0)); /* purecov: inspected */
    DBUG_RETURN(1); /* purecov: inspected */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
488 489
  }
  thr_lock_data_init(&share->lock,&lock,(void*) 0);
490 491
  key_file = share->key_file;
  key_type = share->key_type;
492
  bzero((char*) &current_row,sizeof(current_row));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
493

494 495 496 497
  /* Fill in shared structure, if needed */
  pthread_mutex_lock(&share->mutex);
  file = share->file;
  if (!share->use_count++)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
498
  {
499 500
    if ((error=db_create(&file, db_env, 0)))
    {
501
      free_share(share,table, hidden_primary_key,1); /* purecov: inspected */
502
      my_free((char*) rec_buff,MYF(0)); /* purecov: inspected */
503 504 505
      my_free(alloc_ptr,MYF(0)); /* purecov: inspected */
      my_errno=error; /* purecov: inspected */
      DBUG_RETURN(1); /* purecov: inspected */
506 507 508 509 510 511 512 513
    }
    share->file = file;

    file->set_bt_compare(file,
			 (hidden_primary_key ? berkeley_cmp_hidden_key :
			  berkeley_cmp_packed_key));
    if (!hidden_primary_key)
      file->app_private= (void*) (table->key_info+table->primary_key);
ram@mysql.r18.ru's avatar
ram@mysql.r18.ru committed
514 515
    if ((error= txn_begin(db_env, 0, (DB_TXN**) &transaction, 0)) ||
	(error= (file->open(file, transaction,
516 517
			    fn_format(name_buff, name, "", ha_berkeley_ext,
				      2 | 4),
ram@mysql.r18.ru's avatar
ram@mysql.r18.ru committed
518 519
			    "main", DB_BTREE, open_mode, 0))) ||
	(error= transaction->commit(transaction, 0)))
520
    {
521
      free_share(share,table, hidden_primary_key,1); /* purecov: inspected */
522
      my_free((char*) rec_buff,MYF(0)); /* purecov: inspected */
523 524 525
      my_free(alloc_ptr,MYF(0)); /* purecov: inspected */
      my_errno=error; /* purecov: inspected */
      DBUG_RETURN(1); /* purecov: inspected */
526
    }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
527

528
    /* Open other keys;  These are part of the share structure */
529 530 531 532 533 534 535 536 537 538 539
    key_file[primary_key]=file;
    key_type[primary_key]=DB_NOOVERWRITE;

    DB **ptr=key_file;
    for (uint i=0, used_keys=0; i < table->keys ; i++, ptr++)
    {
      char part[7];
      if (i != primary_key)
      {
	if ((error=db_create(ptr, db_env, 0)))
	{
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
540 541 542
	  close();				/* purecov: inspected */
	  my_errno=error;			/* purecov: inspected */
	  DBUG_RETURN(1);			/* purecov: inspected */
543 544 545 546 547 548 549
	}
	sprintf(part,"key%02d",++used_keys);
	key_type[i]=table->key_info[i].flags & HA_NOSAME ? DB_NOOVERWRITE : 0;
	(*ptr)->set_bt_compare(*ptr, berkeley_cmp_packed_key);
	(*ptr)->app_private= (void*) (table->key_info+i);
	if (!(table->key_info[i].flags & HA_NOSAME))
	  (*ptr)->set_flags(*ptr, DB_DUP);
ram@mysql.r18.ru's avatar
ram@mysql.r18.ru committed
550 551 552 553
	if ((error= txn_begin(db_env, 0, (DB_TXN**) &transaction, 0)) ||
	    (error=((*ptr)->open(*ptr, transaction, name_buff, part, DB_BTREE,
				 open_mode, 0))) ||
	    (error= transaction->commit(transaction, 0)))
554
	{
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
555 556 557
	  close();				/* purecov: inspected */
	  my_errno=error;			/* purecov: inspected */
	  DBUG_RETURN(1);			/* purecov: inspected */
558 559 560 561
	}
      }
    }
    /* Calculate pack_length of primary key */
562
    share->fixed_length_primary_key=1;
563 564 565 566 567
    if (!hidden_primary_key)
    {
      ref_length=0;
      KEY_PART_INFO *key_part= table->key_info[primary_key].key_part;
      KEY_PART_INFO *end=key_part+table->key_info[primary_key].key_parts;
568
      for (; key_part != end ; key_part++)
569
	ref_length+= key_part->field->max_packed_col_length(key_part->length);
570
      share->fixed_length_primary_key=
571 572 573
	(ref_length == table->key_info[primary_key].key_length);
      share->status|=STATUS_PRIMARY_KEY_INIT;
    }    
574
    share->ref_length=ref_length;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
575
  }
576
  ref_length=share->ref_length;			// If second open
577
  pthread_mutex_unlock(&share->mutex);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
578 579 580

  transaction=0;
  cursor=0;
581
  key_read=0;
582
  block_size=8192;				// Berkeley DB block size
583
  share->fixed_length_row=!(table->db_create_options & HA_OPTION_PACK_RECORD);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
584

585
  get_status();
586
  info(HA_STATUS_NO_LOCK | HA_STATUS_VARIABLE | HA_STATUS_CONST);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
587 588 589 590 591 592 593 594
  DBUG_RETURN(0);
}


int ha_berkeley::close(void)
{
  DBUG_ENTER("ha_berkeley::close");

595
  my_free((char*) rec_buff,MYF(MY_ALLOW_ZERO_PTR));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
596
  my_free(alloc_ptr,MYF(MY_ALLOW_ZERO_PTR));
597
  ha_berkeley::extra(HA_EXTRA_RESET);		// current_row buffer
598
  DBUG_RETURN(free_share(share,table, hidden_primary_key,0));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
599 600 601 602 603 604 605 606 607 608 609 610
}


/* Reallocate buffer if needed */

bool ha_berkeley::fix_rec_buff_for_blob(ulong length)
{
  if (! rec_buff || length > alloced_rec_buff_length)
  {
    byte *newptr;
    if (!(newptr=(byte*) my_realloc((gptr) rec_buff, length,
				    MYF(MY_ALLOW_ZERO_PTR))))
611
      return 1; /* purecov: inspected */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
612 613 614 615 616 617 618 619 620 621 622 623 624
    rec_buff=newptr;
    alloced_rec_buff_length=length;
  }
  return 0;
}


/* Calculate max length needed for row */

ulong ha_berkeley::max_row_length(const byte *buf)
{
  ulong length=table->reclength + table->fields*2;
  for (Field_blob **ptr=table->blob_field ; *ptr ; ptr++)
625
    length+= (*ptr)->get_length((char*) buf+(*ptr)->offset())+2;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
626 627 628 629 630 631 632 633 634 635 636 637 638
  return length;
}


/*
  Pack a row for storage.  If the row is of fixed length, just store the
  row 'as is'.
  If not, we will generate a packed row suitable for storage.
  This will only fail if we don't have enough memory to pack the row, which;
  may only happen in rows with blobs,  as the default row length is
  pre-allocated.
*/

639
int ha_berkeley::pack_row(DBT *row, const byte *record, bool new_row)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
640 641
{
  bzero((char*) row,sizeof(*row));
642
  if (share->fixed_length_row)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
643 644
  {
    row->data=(void*) record;
645 646 647 648 649 650 651 652
    row->size=table->reclength+hidden_primary_key;
    if (hidden_primary_key)
    {
      if (new_row)
	get_auto_primary_key(current_ident);
      memcpy_fixed((char*) record+table->reclength, (char*) current_ident,
		   BDB_HIDDEN_PRIMARY_KEY_LENGTH);
    }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
653 654 655 656 657
    return 0;
  }
  if (table->blob_fields)
  {
    if (fix_rec_buff_for_blob(max_row_length(record)))
658
      return HA_ERR_OUT_OF_MEM; /* purecov: inspected */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
659 660 661 662 663 664 665
  }

  /* Copy null bits */
  memcpy(rec_buff, record, table->null_bytes);
  byte *ptr=rec_buff + table->null_bytes;

  for (Field **field=table->field ; *field ; field++)
666 667
    ptr=(byte*) (*field)->pack((char*) ptr,
			       (char*) record + (*field)->offset());
668 669 670 671 672 673 674 675 676

  if (hidden_primary_key)
  {
    if (new_row)
      get_auto_primary_key(current_ident);
    memcpy_fixed((char*) ptr, (char*) current_ident,
		 BDB_HIDDEN_PRIMARY_KEY_LENGTH);
    ptr+=BDB_HIDDEN_PRIMARY_KEY_LENGTH;
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
677 678 679 680 681 682 683 684
  row->data=rec_buff;
  row->size= (size_t) (ptr - rec_buff);
  return 0;
}


void ha_berkeley::unpack_row(char *record, DBT *row)
{
685
  if (share->fixed_length_row)
686
    memcpy(record,(char*) row->data,table->reclength+hidden_primary_key);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
687 688 689 690 691 692 693 694 695 696 697 698
  else
  {
    /* Copy null bits */
    const char *ptr= (const char*) row->data;
    memcpy(record, ptr, table->null_bytes);
    ptr+=table->null_bytes;
    for (Field **field=table->field ; *field ; field++)
      ptr= (*field)->unpack(record + (*field)->offset(), ptr);
  }
}


699 700 701 702 703 704
/* Store the key and the primary key into the row */

void ha_berkeley::unpack_key(char *record, DBT *key, uint index)
{
  KEY *key_info=table->key_info+index;
  KEY_PART_INFO *key_part= key_info->key_part,
705
		*end=key_part+key_info->key_parts;
706 707

  char *pos=(char*) key->data;
708
  for (; key_part != end; key_part++)
709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729
  {
    if (key_part->null_bit)
    {
      if (!*pos++)				// Null value
      {
	/*
	  We don't need to reset the record data as we will not access it
	  if the null data is set
	*/

	record[key_part->null_offset]|=key_part->null_bit;
	continue;
      }
      record[key_part->null_offset]&= ~key_part->null_bit;
    }
    pos= (char*) key_part->field->unpack(record + key_part->field->offset(),
					 pos);
  }
}


bk@work.mysql.com's avatar
bk@work.mysql.com committed
730 731 732 733 734
/*
  Create a packed key from from a row
  This will never fail as the key buffer is pre allocated.
*/

735 736
DBT *ha_berkeley::create_key(DBT *key, uint keynr, char *buff,
			     const byte *record, int key_length)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
737
{
738 739 740
  bzero((char*) key,sizeof(*key));
  if (hidden_primary_key && keynr == primary_key)
  {
741
    /* We don't need to set app_private here */
742 743 744 745 746
    key->data=current_ident;
    key->size=BDB_HIDDEN_PRIMARY_KEY_LENGTH;
    return key;
  }

bk@work.mysql.com's avatar
bk@work.mysql.com committed
747 748 749
  KEY *key_info=table->key_info+keynr;
  KEY_PART_INFO *key_part=key_info->key_part;
  KEY_PART_INFO *end=key_part+key_info->key_parts;
750
  DBUG_ENTER("create_key");
bk@work.mysql.com's avatar
bk@work.mysql.com committed
751 752

  key->data=buff;
753
  key->app_private= key_info;
754
  for (; key_part != end && key_length > 0; key_part++)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
755 756 757 758 759 760 761 762 763 764 765 766
  {
    if (key_part->null_bit)
    {
      /* Store 0 if the key part is a NULL part */
      if (record[key_part->null_offset] & key_part->null_bit)
      {
	*buff++ =0;
	key->flags|=DB_DBT_DUPOK;
	continue;
      }
      *buff++ = 1;				// Store NOT NULL marker
    }
767
    buff=key_part->field->pack_key(buff,(char*) (record + key_part->offset),
768
				   key_part->length);
769
    key_length-=key_part->length;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786
  }
  key->size= (buff  - (char*) key->data);
  DBUG_DUMP("key",(char*) key->data, key->size);
  DBUG_RETURN(key);
}


/*
  Create a packed key from from a MySQL unpacked key
*/

DBT *ha_berkeley::pack_key(DBT *key, uint keynr, char *buff,
			   const byte *key_ptr, uint key_length)
{
  KEY *key_info=table->key_info+keynr;
  KEY_PART_INFO *key_part=key_info->key_part;
  KEY_PART_INFO *end=key_part+key_info->key_parts;
787
  DBUG_ENTER("bdb:pack_key");
bk@work.mysql.com's avatar
bk@work.mysql.com committed
788 789 790

  bzero((char*) key,sizeof(*key));
  key->data=buff;
791
  key->app_private= (void*) key_info;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
792 793 794

  for (; key_part != end && (int) key_length > 0 ; key_part++)
  {
795
    uint offset=0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
796 797 798 799 800 801 802 803 804
    if (key_part->null_bit)
    {
      if (!(*buff++ = (*key_ptr == 0)))		// Store 0 if NULL
      {
	key_length-= key_part->store_length;
	key_ptr+=   key_part->store_length;
	key->flags|=DB_DBT_DUPOK;
	continue;
      }
805
      offset=1;					// Data is at key_ptr+1
806
    }
807
    buff=key_part->field->pack_key_from_key_image(buff,(char*) key_ptr+offset,
808
						  key_part->length);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828
    key_ptr+=key_part->store_length;
    key_length-=key_part->store_length;
  }
  key->size= (buff  - (char*) key->data);
  DBUG_DUMP("key",(char*) key->data, key->size);
  DBUG_RETURN(key);
}


int ha_berkeley::write_row(byte * record)
{
  DBT row,prim_key,key;
  int error;
  DBUG_ENTER("write_row");

  statistic_increment(ha_write_count,&LOCK_status);
  if (table->time_stamp)
    update_timestamp(record+table->time_stamp-1);
  if (table->next_number_field && record == table->record[0])
    update_auto_increment();
829
  if ((error=pack_row(&row, record,1)))
830
    DBUG_RETURN(error); /* purecov: inspected */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
831

832
  if (table->keys + test(hidden_primary_key) == 1)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
833
  {
834 835
    error=file->put(file, transaction, create_key(&prim_key, primary_key,
						  key_buff, record),
bk@work.mysql.com's avatar
bk@work.mysql.com committed
836
		    &row, key_type[primary_key]);
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
837
    last_dup_key=primary_key;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
838 839 840
  }
  else
  {
841
    DB_TXN *sub_trans = transaction;
842 843
    /* Don't use sub transactions in temporary tables (in_use == 0) */
    ulong thd_options = table->in_use ? table->in_use->options : 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
844 845
    for (uint retry=0 ; retry < berkeley_trans_retry ; retry++)
    {
846 847 848
      key_map changed_keys = 0;
      if (using_ignore && (thd_options & OPTION_INTERNAL_SUBTRANSACTIONS))
      {
849 850 851
	if ((error=txn_begin(db_env, transaction, &sub_trans, 0))) /* purecov: deadcode */
	  break; /* purecov: deadcode */
	DBUG_PRINT("trans",("starting subtransaction")); /* purecov: deadcode */
852
      }
853 854
      if (!(error=file->put(file, sub_trans, create_key(&prim_key, primary_key,
							key_buff, record),
bk@work.mysql.com's avatar
bk@work.mysql.com committed
855 856
			    &row, key_type[primary_key])))
      {
857 858
	changed_keys |= (key_map) 1 << primary_key;
	for (uint keynr=0 ; keynr < table->keys ; keynr++)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
859 860 861 862
	{
	  if (keynr == primary_key)
	    continue;
	  if ((error=key_file[keynr]->put(key_file[keynr], sub_trans,
863 864
					  create_key(&key, keynr, key_buff2,
						     record),
bk@work.mysql.com's avatar
bk@work.mysql.com committed
865 866 867 868 869
					  &prim_key, key_type[keynr])))
	  {
	    last_dup_key=keynr;
	    break;
	  }
870
	  changed_keys |= (key_map) 1 << keynr;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
871 872
	}
      }
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
873 874
      else
	last_dup_key=primary_key;
875
      if (error)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
876 877 878
      {
	/* Remove inserted row */
	DBUG_PRINT("error",("Got error %d",error));
879
	if (using_ignore)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
880
	{
881 882 883
	  int new_error = 0;
	  if (thd_options & OPTION_INTERNAL_SUBTRANSACTIONS)
	  {
884 885
	    DBUG_PRINT("trans",("aborting subtransaction")); /* purecov: deadcode */
	    new_error=txn_abort(sub_trans); /* purecov: deadcode */
886 887 888 889 890 891 892 893 894
	  }
	  else if (changed_keys)
	  {
	    new_error = 0;
	    for (uint keynr=0; changed_keys; keynr++, changed_keys >>= 1)
	    {
	      if (changed_keys & 1)
	      {
		if ((new_error = remove_key(sub_trans, keynr, record,
895
					    &prim_key)))
896
		  break; /* purecov: inspected */
897 898 899 900 901
	      }
	    }
	  }
	  if (new_error)
	  {
902 903
	    error=new_error;			// This shouldn't happen /* purecov: inspected */
	    break; /* purecov: inspected */
904
	  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
905 906
	}
      }
907 908
      else if (using_ignore && (thd_options & OPTION_INTERNAL_SUBTRANSACTIONS))
      {
909 910
	DBUG_PRINT("trans",("committing subtransaction")); /* purecov: deadcode */
	error=txn_commit(sub_trans, 0); /* purecov: deadcode */
911
      }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
912 913 914 915 916 917
      if (error != DB_LOCK_DEADLOCK)
	break;
    }
  }
  if (error == DB_KEYEXIST)
    error=HA_ERR_FOUND_DUPP_KEY;
918 919
  else if (!error)
    changed_rows++;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
920 921 922 923 924 925 926 927 928 929 930 931
  DBUG_RETURN(error);
}


/* Compare if a key in a row has changed */

int ha_berkeley::key_cmp(uint keynr, const byte * old_row,
			 const byte * new_row)
{
  KEY_PART_INFO *key_part=table->key_info[keynr].key_part;
  KEY_PART_INFO *end=key_part+table->key_info[keynr].key_parts;

932
  for (; key_part != end ; key_part++)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
933 934 935 936 937 938 939 940 941
  {
    if (key_part->null_bit)
    {
      if ((old_row[key_part->null_offset] & key_part->null_bit) !=
	  (new_row[key_part->null_offset] & key_part->null_bit))
	return 1;
    }
    if (key_part->key_part_flag & (HA_BLOB_PART | HA_VAR_LENGTH))
    {
942

943 944
      if (key_part->field->cmp_binary((char*) (old_row + key_part->offset),
				      (char*) (new_row + key_part->offset),
bk@work.mysql.com's avatar
bk@work.mysql.com committed
945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960
				      (ulong) key_part->length))
	return 1;
    }
    else
    {
      if (memcmp(old_row+key_part->offset, new_row+key_part->offset,
		 key_part->length))
	return 1;
    }
  }
  return 0;
}


/*
  Update a row from one value to another.
961
  Clobbers key_buff2
bk@work.mysql.com's avatar
bk@work.mysql.com committed
962 963 964
*/

int ha_berkeley::update_primary_key(DB_TXN *trans, bool primary_key_changed,
965 966 967
				    const byte * old_row, DBT *old_key,
				    const byte * new_row, DBT *new_key,
				    ulong thd_options, bool local_using_ignore)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
968
{
969
  DBT row;
970
  int error;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
971 972 973 974 975 976
  DBUG_ENTER("update_primary_key");

  if (primary_key_changed)
  {
    // Primary key changed or we are updating a key that can have duplicates.
    // Delete the old row and add a new one
977
    if (!(error=remove_key(trans, primary_key, old_row, old_key)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
978
    {
979 980 981 982 983 984 985 986 987 988 989 990 991 992
      if (!(error=pack_row(&row, new_row, 0)))
      {
	if ((error=file->put(file, trans, new_key, &row,
			     key_type[primary_key])))
	{
	  // Probably a duplicated key; restore old key and row if needed
	  last_dup_key=primary_key;
	  if (local_using_ignore &&
	      !(thd_options & OPTION_INTERNAL_SUBTRANSACTIONS))
	  {
	    int new_error;
	    if ((new_error=pack_row(&row, old_row, 0)) ||
		(new_error=file->put(file, trans, old_key, &row,
				     key_type[primary_key])))
993
	      error=new_error;                  // fatal error /* purecov: inspected */
994 995 996
	  }
	}
      }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
997 998 999 1000 1001
    }
  }
  else
  {
    // Primary key didn't change;  just update the row data
1002 1003
    if (!(error=pack_row(&row, new_row, 0)))
      error=file->put(file, trans, new_key, &row, 0);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1004
  }
1005
  DBUG_RETURN(error);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1006 1007
}

1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021
/*
  Restore changed keys, when a non-fatal error aborts the insert/update
  of one row.
  Clobbers keybuff2
*/

int ha_berkeley::restore_keys(DB_TXN *trans, key_map changed_keys,
			      uint primary_key,
			      const byte *old_row, DBT *old_key,
			      const byte *new_row, DBT *new_key,
			      ulong thd_options)
{
  int error;
  DBT tmp_key;
1022
  uint keynr;
1023 1024 1025 1026 1027 1028
  DBUG_ENTER("restore_keys");

  /* Restore the old primary key, and the old row, but don't ignore
     duplicate key failure */
  if ((error=update_primary_key(trans, TRUE, new_row, new_key,
				old_row, old_key, thd_options, FALSE)))
1029
    goto err; /* purecov: inspected */
1030 1031 1032 1033 1034 1035

  /* Remove the new key, and put back the old key
     changed_keys is a map of all non-primary keys that need to be
     rolled back.  The last key set in changed_keys is the one that
     triggered the duplicate key error (it wasn't inserted), so for
     that one just put back the old value. */
1036
  for (keynr=0; changed_keys; keynr++, changed_keys >>= 1)
1037 1038 1039 1040
  {
    if (changed_keys & 1)
    {
      if (changed_keys != 1 &&
1041
	  (error = remove_key(trans, keynr, new_row, new_key)))
1042
	break; /* purecov: inspected */
1043 1044 1045 1046
      if ((error = key_file[keynr]->put(key_file[keynr], trans,
					create_key(&tmp_key, keynr, key_buff2,
						   old_row),
					old_key, key_type[keynr])))
1047
	break; /* purecov: inspected */
1048 1049
    }
  }
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
1050

1051
err:
monty@bitch.mysql.fi's avatar
monty@bitch.mysql.fi committed
1052
  DBUG_ASSERT(error != DB_KEYEXIST);
1053 1054
  DBUG_RETURN(error);
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1055 1056 1057 1058


int ha_berkeley::update_row(const byte * old_row, byte * new_row)
{
1059
  DBT prim_key, key, old_prim_key;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1060 1061
  int error;
  DB_TXN *sub_trans;
1062
  ulong thd_options = table->in_use ? table->in_use->options : 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1063 1064
  bool primary_key_changed;
  DBUG_ENTER("update_row");
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
1065
  LINT_INIT(error);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1066 1067 1068 1069

  statistic_increment(ha_update_count,&LOCK_status);
  if (table->time_stamp)
    update_timestamp(new_row+table->time_stamp-1);
1070 1071 1072 1073 1074 1075 1076

  if (hidden_primary_key)
  {
    primary_key_changed=0;
    bzero((char*) &prim_key,sizeof(prim_key));
    prim_key.data= (void*) current_ident;
    prim_key.size=BDB_HIDDEN_PRIMARY_KEY_LENGTH;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1077
    old_prim_key=prim_key;
1078 1079 1080
  }
  else
  {
1081
    create_key(&prim_key, primary_key, key_buff, new_row);
1082

1083
    if ((primary_key_changed=key_cmp(primary_key, old_row, new_row)))
1084
      create_key(&old_prim_key, primary_key, primary_key_buff, old_row);
1085 1086 1087
    else
      old_prim_key=prim_key;
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1088

1089
  sub_trans = transaction;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1090 1091
  for (uint retry=0 ; retry < berkeley_trans_retry ; retry++)
  {
1092 1093 1094
    key_map changed_keys = 0;
    if (using_ignore &&	(thd_options & OPTION_INTERNAL_SUBTRANSACTIONS))
    {
1095 1096 1097
      if ((error=txn_begin(db_env, transaction, &sub_trans, 0))) /* purecov: deadcode */
	break; /* purecov: deadcode */
      DBUG_PRINT("trans",("starting subtransaction")); /* purecov: deadcode */
1098
    }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1099 1100
    /* Start by updating the primary key */
    if (!(error=update_primary_key(sub_trans, primary_key_changed,
1101 1102 1103
				   old_row, &old_prim_key,
				   new_row, &prim_key,
				   thd_options, using_ignore)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1104 1105 1106 1107 1108 1109 1110 1111
    {
      // Update all other keys
      for (uint keynr=0 ; keynr < table->keys ; keynr++)
      {
	if (keynr == primary_key)
	  continue;
	if (key_cmp(keynr, old_row, new_row) || primary_key_changed)
	{
1112
	  if ((error=remove_key(sub_trans, keynr, old_row, &old_prim_key)))
1113
	  {
1114
	    if (using_ignore && /* purecov: inspected */
1115 1116 1117 1118 1119 1120 1121 1122
		(thd_options & OPTION_INTERNAL_SUBTRANSACTIONS))
            {
	      int new_error;
	      DBUG_PRINT("trans",("aborting subtransaction"));
	      new_error=txn_abort(sub_trans);
	      if (new_error)
		error = new_error;
	    }
1123
	    DBUG_RETURN(error);			// Fatal error /* purecov: inspected */
1124 1125 1126
	  }
	  changed_keys |= (key_map)1 << keynr;
	  if ((error=key_file[keynr]->put(key_file[keynr], sub_trans,
1127
					  create_key(&key, keynr, key_buff2,
1128
						     new_row),
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1129 1130 1131 1132 1133 1134 1135 1136
					  &prim_key, key_type[keynr])))
	  {
	    last_dup_key=keynr;
	    break;
	  }
	}
      }
    }
1137
    if (error)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1138 1139
    {
      /* Remove inserted row */
1140 1141 1142 1143 1144 1145
      DBUG_PRINT("error",("Got error %d",error));
      if (using_ignore)
      {
	int new_error = 0;
	if (thd_options & OPTION_INTERNAL_SUBTRANSACTIONS)
	{
1146 1147
	  DBUG_PRINT("trans",("aborting subtransaction")); /* purecov: deadcode */
	  new_error=txn_abort(sub_trans); /* purecov: deadcode */
1148 1149 1150 1151 1152 1153 1154
	}
	else if (changed_keys)
	  new_error=restore_keys(transaction, changed_keys, primary_key,
				 old_row, &old_prim_key, new_row, &prim_key,
				 thd_options);
	if (new_error)
	{
1155 1156
	  error=new_error;			// This shouldn't happen /* purecov: inspected */
	  break; /* purecov: inspected */
1157 1158 1159 1160 1161
	}
      }
    }
    else if (using_ignore && (thd_options & OPTION_INTERNAL_SUBTRANSACTIONS))
    {
1162 1163
      DBUG_PRINT("trans",("committing subtransaction")); /* purecov: deadcode */
      error=txn_commit(sub_trans, 0); /* purecov: deadcode */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179
    }
    if (error != DB_LOCK_DEADLOCK)
      break;
  }
  if (error == DB_KEYEXIST)
    error=HA_ERR_FOUND_DUPP_KEY;
  DBUG_RETURN(error);
}


/*
  Delete one key
  This uses key_buff2, when keynr != primary key, so it's important that
  a function that calls this doesn't use this buffer for anything else.
*/

1180
int ha_berkeley::remove_key(DB_TXN *trans, uint keynr, const byte *record,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1181 1182 1183 1184 1185 1186 1187
			    DBT *prim_key)
{
  int error;
  DBT key;
  DBUG_ENTER("remove_key");
  DBUG_PRINT("enter",("index: %d",keynr));

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
1188 1189 1190 1191 1192
  if (keynr == active_index && cursor)
    error=cursor->c_del(cursor,0);
  else if (keynr == primary_key ||
	   ((table->key_info[keynr].flags & (HA_NOSAME | HA_NULL_PART_KEY)) ==
	    HA_NOSAME))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1193
  {						// Unique key
monty@bitch.mysql.fi's avatar
monty@bitch.mysql.fi committed
1194
    DBUG_ASSERT(keynr == primary_key || prim_key->data != key_buff2);
1195
    error=key_file[keynr]->del(key_file[keynr], trans,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1196 1197
			       keynr == primary_key ?
			       prim_key :
1198
			       create_key(&key, keynr, key_buff2, record),
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1199 1200 1201 1202 1203 1204 1205 1206 1207
			       0);
  }
  else
  {
    /*
      To delete the not duplicated key, we need to open an cursor on the
      row to find the key to be delete and delete it.
      We will never come here with keynr = primary_key
    */
monty@bitch.mysql.fi's avatar
monty@bitch.mysql.fi committed
1208
    DBUG_ASSERT(keynr != primary_key && prim_key->data != key_buff2);
1209
    DBC *tmp_cursor;
1210 1211
    if (!(error=key_file[keynr]->cursor(key_file[keynr], trans,
					&tmp_cursor, 0)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1212
    {
1213
      if (!(error=tmp_cursor->c_get(tmp_cursor,
1214 1215
                                    create_key(&key, keynr, key_buff2, record),
                                    prim_key, DB_GET_BOTH | DB_RMW)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1216
      {					// This shouldn't happen
1217
	error=tmp_cursor->c_del(tmp_cursor,0);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1218
      }
1219
      int result=tmp_cursor->c_close(tmp_cursor);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230
      if (!error)
	error=result;
    }
  }
  DBUG_RETURN(error);
}


/* Delete all keys for new_record */

int ha_berkeley::remove_keys(DB_TXN *trans, const byte *record,
1231
			     DBT *new_record, DBT *prim_key, key_map keys)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1232
{
1233 1234
  int result = 0;
  for (uint keynr=0; keys; keynr++, keys>>=1)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1235 1236 1237
  {
    if (keys & 1)
    {
1238
      int new_error=remove_key(trans, keynr, record, prim_key);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1239 1240
      if (new_error)
      {
1241 1242
	result=new_error;			// Return last error /* purecov: inspected */
	break;					// Let rollback correct things /* purecov: inspected */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254
      }
    }
  }
  return result;
}


int ha_berkeley::delete_row(const byte * record)
{
  int error;
  DBT row, prim_key;
  key_map keys=table->keys_in_use;
1255
  ulong thd_options = table->in_use ? table->in_use->options : 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1256 1257
  DBUG_ENTER("delete_row");
  statistic_increment(ha_delete_count,&LOCK_status);
1258

1259
  if ((error=pack_row(&row, record, 0)))
1260
    DBUG_RETURN((error)); /* purecov: inspected */
1261
  create_key(&prim_key, primary_key, key_buff, record);
1262 1263 1264
  if (hidden_primary_key)
    keys|= (key_map) 1 << primary_key;

1265 1266 1267
  /* Subtransactions may be used in order to retry the delete in
     case we get a DB_LOCK_DEADLOCK error. */
  DB_TXN *sub_trans = transaction;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1268 1269
  for (uint retry=0 ; retry < berkeley_trans_retry ; retry++)
  {
1270 1271
    if (thd_options & OPTION_INTERNAL_SUBTRANSACTIONS)
    {
1272 1273 1274
      if ((error=txn_begin(db_env, transaction, &sub_trans, 0))) /* purecov: deadcode */
	break; /* purecov: deadcode */
      DBUG_PRINT("trans",("starting sub transaction")); /* purecov: deadcode */
1275 1276 1277
    }
    error=remove_keys(sub_trans, record, &row, &prim_key, keys);
    if (!error && (thd_options & OPTION_INTERNAL_SUBTRANSACTIONS))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1278
    {
1279 1280
      DBUG_PRINT("trans",("ending sub transaction")); /* purecov: deadcode */
      error=txn_commit(sub_trans, 0); /* purecov: deadcode */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1281 1282
    }
    if (error)
1283
    { /* purecov: inspected */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1284
      DBUG_PRINT("error",("Got error %d",error));
1285
      if (thd_options & OPTION_INTERNAL_SUBTRANSACTIONS)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1286
      {
1287 1288 1289 1290 1291 1292 1293 1294
	/* retry */
	int new_error;
	DBUG_PRINT("trans",("aborting subtransaction"));
	if ((new_error=txn_abort(sub_trans)))
	{
	  error=new_error;			// This shouldn't happen
	  break;
	}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1295
      }
1296 1297
      else
	break;					// No retry - return error
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1298 1299 1300 1301
    }
    if (error != DB_LOCK_DEADLOCK)
      break;
  }
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
1302
#ifdef CANT_COUNT_DELETED_ROWS
1303 1304
  if (!error)
    changed_rows--;
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
1305
#endif
1306
  DBUG_RETURN(error);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1307 1308 1309 1310 1311 1312
}


int ha_berkeley::index_init(uint keynr)
{
  int error;
1313
  DBUG_ENTER("ha_berkeley::index_init");
1314
  DBUG_PRINT("enter",("table: '%s'  key: %d", table->real_name, keynr));
1315 1316 1317 1318 1319 1320

  /*
    Under some very rare conditions (like full joins) we may already have
    an active cursor at this point
  */
  if (cursor)
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
1321 1322
  {
    DBUG_PRINT("note",("Closing active cursor"));
1323
    cursor->c_close(cursor);
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
1324
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1325
  active_index=keynr;
1326 1327 1328 1329
  if ((error=key_file[keynr]->cursor(key_file[keynr], transaction, &cursor,
				     table->reginfo.lock_type >
				     TL_WRITE_ALLOW_READ ?
				     0 : 0)))
1330
    cursor=0;				// Safety /* purecov: inspected */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1331 1332 1333 1334 1335 1336 1337 1338 1339 1340
  bzero((char*) &last_key,sizeof(last_key));
  DBUG_RETURN(error);
}

int ha_berkeley::index_end()
{
  int error=0;
  DBUG_ENTER("index_end");
  if (cursor)
  {
1341
    DBUG_PRINT("enter",("table: '%s'", table->real_name));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1342 1343 1344 1345 1346 1347 1348 1349 1350 1351
    error=cursor->c_close(cursor);
    cursor=0;
  }
  DBUG_RETURN(error);
}


/* What to do after we have read a row based on an index */

int ha_berkeley::read_row(int error, char *buf, uint keynr, DBT *row,
1352
			  DBT *found_key, bool read_next)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1353
{
1354
  DBUG_ENTER("ha_berkeley::read_row");
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1355 1356 1357 1358 1359 1360 1361
  if (error)
  {
    if (error == DB_NOTFOUND || error == DB_KEYEMPTY)
      error=read_next ? HA_ERR_END_OF_FILE : HA_ERR_KEY_NOT_FOUND;
    table->status=STATUS_NOT_FOUND;
    DBUG_RETURN(error);
  }
1362 1363 1364 1365 1366
  if (hidden_primary_key)
    memcpy_fixed(current_ident,
		 (char*) row->data+row->size-BDB_HIDDEN_PRIMARY_KEY_LENGTH,
		 BDB_HIDDEN_PRIMARY_KEY_LENGTH);
  table->status=0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1367 1368
  if (keynr != primary_key)
  {
1369 1370 1371 1372 1373 1374 1375 1376 1377
    /* We only found the primary key.  Now we have to use this to find
       the row data */
    if (key_read && found_key)
    {
      unpack_key(buf,found_key,keynr);
      if (!hidden_primary_key)
	unpack_key(buf,row,primary_key);
      DBUG_RETURN(0);
    }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1378 1379
    DBT key;
    bzero((char*) &key,sizeof(key));
1380
    key.data=key_buff;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1381
    key.size=row->size;
1382
    key.app_private= (void*) (table->key_info+primary_key);
1383
    memcpy(key_buff,row->data,row->size);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1384 1385 1386 1387
    /* Read the data into current_row */
    current_row.flags=DB_DBT_REALLOC;
    if ((error=file->get(file, transaction, &key, &current_row, 0)))
    {
1388 1389
      table->status=STATUS_NOT_FOUND; /* purecov: inspected */
      DBUG_RETURN(error == DB_NOTFOUND ? HA_ERR_CRASHED : error); /* purecov: inspected */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1390 1391 1392 1393 1394 1395 1396 1397
    }
    row= &current_row;
  }
  unpack_row(buf,row);
  DBUG_RETURN(0);
}


1398 1399
/* This is only used to read whole keys */

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1400 1401 1402 1403 1404 1405
int ha_berkeley::index_read_idx(byte * buf, uint keynr, const byte * key,
				uint key_len, enum ha_rkey_function find_flag)
{
  statistic_increment(ha_read_key_count,&LOCK_status);
  DBUG_ENTER("index_read_idx");
  current_row.flags=DB_DBT_REALLOC;
1406
  active_index= (uint) -1;
1407
  DBUG_RETURN(read_row(key_file[keynr]->get(key_file[keynr], transaction,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1408 1409 1410
				 pack_key(&last_key, keynr, key_buff, key,
					  key_len),
				 &current_row,0),
1411
		       (char*) buf, keynr, &current_row, &last_key, 0));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1412 1413 1414 1415 1416 1417 1418
}


int ha_berkeley::index_read(byte * buf, const byte * key,
			    uint key_len, enum ha_rkey_function find_flag)
{
  DBT row;
1419
  int error;
1420
  KEY *key_info= &table->key_info[active_index];
1421
  int do_prev= 0;
1422
  DBUG_ENTER("ha_berkeley::index_read");
1423

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1424 1425
  statistic_increment(ha_read_key_count,&LOCK_status);
  bzero((char*) &row,sizeof(row));
1426 1427 1428 1429 1430 1431 1432 1433 1434 1435
  if (find_flag == HA_READ_BEFORE_KEY)
  {
    find_flag= HA_READ_KEY_OR_NEXT;
    do_prev= 1;
  }
  else if (find_flag == HA_READ_PREFIX_LAST_OR_PREV)
  {
    find_flag= HA_READ_AFTER_KEY;
    do_prev= 1;
  }
1436
  if (key_len == key_info->key_length)
1437
  {
1438 1439
    if (find_flag == HA_READ_AFTER_KEY)
      key_info->handler.bdb_return_if_eq= 1;
1440 1441 1442 1443
    error=read_row(cursor->c_get(cursor, pack_key(&last_key,
						  active_index,
						  key_buff,
						  key, key_len),
1444 1445 1446
				 &row,
				 (find_flag == HA_READ_KEY_EXACT ?
				  DB_SET : DB_SET_RANGE)),
1447
		   (char*) buf, active_index, &row, (DBT*) 0, 0);
1448
    key_info->handler.bdb_return_if_eq= 0;
1449 1450 1451 1452 1453 1454
  }
  else
  {
    /* read of partial key */
    pack_key(&last_key, active_index, key_buff, key, key_len);
    /* Store for compare */
1455
    memcpy(key_buff2, key_buff, (key_len=last_key.size));
1456 1457 1458 1459 1460 1461
    /*
      If HA_READ_AFTER_KEY is set, return next key, else return first
      matching key.
    */
    key_info->handler.bdb_return_if_eq= (find_flag == HA_READ_AFTER_KEY ?
					 1 : -1);
1462
    error=read_row(cursor->c_get(cursor, &last_key, &row, DB_SET_RANGE),
1463
		   (char*) buf, active_index, &row, (DBT*) 0, 0);
1464
    key_info->handler.bdb_return_if_eq= 0;
1465 1466
    if (!error && find_flag == HA_READ_KEY_EXACT)
    {
1467 1468
      /* Ensure that we found a key that is equal to the current one */
      if (!error && berkeley_key_cmp(table, key_info, key_buff2, key_len))
1469 1470 1471
	error=HA_ERR_KEY_NOT_FOUND;
    }
  }
1472 1473 1474 1475 1476 1477
  if (do_prev)
  {
    bzero((char*) &row, sizeof(row));
    error= read_row(cursor->c_get(cursor, &last_key, &row, DB_PREV),
                         (char*) buf, active_index, &row, &last_key, 1);
  }   
1478
  DBUG_RETURN(error);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1479 1480
}

1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511
/*
  Read last key is solved by reading the next key and then reading
  the previous key
*/

int ha_berkeley::index_read_last(byte * buf, const byte * key, uint key_len)
{
  DBT row;
  int error;
  KEY *key_info= &table->key_info[active_index];
  DBUG_ENTER("ha_berkeley::index_read");

  statistic_increment(ha_read_key_count,&LOCK_status);
  bzero((char*) &row,sizeof(row));

  /* read of partial key */
  pack_key(&last_key, active_index, key_buff, key, key_len);
  /* Store for compare */
  memcpy(key_buff2, key_buff, (key_len=last_key.size));
  key_info->handler.bdb_return_if_eq= 1;
  error=read_row(cursor->c_get(cursor, &last_key, &row, DB_SET_RANGE),
		 (char*) buf, active_index, &row, (DBT*) 0, 0);
  key_info->handler.bdb_return_if_eq= 0;
  bzero((char*) &row,sizeof(row));
  if (read_row(cursor->c_get(cursor, &last_key, &row, DB_PREV),
	       (char*) buf, active_index, &row, &last_key, 1) ||
      berkeley_key_cmp(table, key_info, key_buff2, key_len))
    error=HA_ERR_KEY_NOT_FOUND;
  DBUG_RETURN(error);
}

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1512 1513 1514 1515 1516 1517 1518 1519

int ha_berkeley::index_next(byte * buf)
{
  DBT row;
  DBUG_ENTER("index_next");
  statistic_increment(ha_read_next_count,&LOCK_status);
  bzero((char*) &row,sizeof(row));
  DBUG_RETURN(read_row(cursor->c_get(cursor, &last_key, &row, DB_NEXT),
1520
		       (char*) buf, active_index, &row, &last_key, 1));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531
}

int ha_berkeley::index_next_same(byte * buf, const byte *key, uint keylen)
{
  DBT row;
  int error;
  DBUG_ENTER("index_next_same");
  statistic_increment(ha_read_next_count,&LOCK_status);
  bzero((char*) &row,sizeof(row));
  if (keylen == table->key_info[active_index].key_length)
    error=read_row(cursor->c_get(cursor, &last_key, &row, DB_NEXT_DUP),
1532
		   (char*) buf, active_index, &row, &last_key, 1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1533 1534 1535
  else
  {
    error=read_row(cursor->c_get(cursor, &last_key, &row, DB_NEXT),
1536
		   (char*) buf, active_index, &row, &last_key, 1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550
    if (!error && ::key_cmp(table, key, active_index, keylen))
      error=HA_ERR_END_OF_FILE;
  }
  DBUG_RETURN(error);
}


int ha_berkeley::index_prev(byte * buf)
{
  DBT row;
  DBUG_ENTER("index_prev");
  statistic_increment(ha_read_prev_count,&LOCK_status);
  bzero((char*) &row,sizeof(row));
  DBUG_RETURN(read_row(cursor->c_get(cursor, &last_key, &row, DB_PREV),
1551
		       (char*) buf, active_index, &row, &last_key, 1));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1552
}
1553

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1554 1555 1556 1557 1558 1559 1560 1561

int ha_berkeley::index_first(byte * buf)
{
  DBT row;
  DBUG_ENTER("index_first");
  statistic_increment(ha_read_first_count,&LOCK_status);
  bzero((char*) &row,sizeof(row));
  DBUG_RETURN(read_row(cursor->c_get(cursor, &last_key, &row, DB_FIRST),
1562
		       (char*) buf, active_index, &row, &last_key, 1));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1563 1564 1565 1566 1567 1568 1569 1570 1571
}

int ha_berkeley::index_last(byte * buf)
{
  DBT row;
  DBUG_ENTER("index_last");
  statistic_increment(ha_read_last_count,&LOCK_status);
  bzero((char*) &row,sizeof(row));
  DBUG_RETURN(read_row(cursor->c_get(cursor, &last_key, &row, DB_LAST),
1572
		       (char*) buf, active_index, &row, &last_key, 0));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1573 1574 1575 1576
}

int ha_berkeley::rnd_init(bool scan)
{
1577
  DBUG_ENTER("rnd_init");
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1578
  current_row.flags=DB_DBT_REALLOC;
1579
  DBUG_RETURN(index_init(primary_key));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593
}

int ha_berkeley::rnd_end()
{
  return index_end();
}

int ha_berkeley::rnd_next(byte *buf)
{
  DBT row;
  DBUG_ENTER("rnd_next");
  statistic_increment(ha_read_rnd_next_count,&LOCK_status);
  bzero((char*) &row,sizeof(row));
  DBUG_RETURN(read_row(cursor->c_get(cursor, &last_key, &row, DB_NEXT),
1594
		       (char*) buf, primary_key, &row, &last_key, 1));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1595 1596 1597 1598 1599
}


DBT *ha_berkeley::get_pos(DBT *to, byte *pos)
{
1600
  /* We don't need to set app_private here */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1601 1602 1603
  bzero((char*) to,sizeof(*to));

  to->data=pos;
1604
  if (share->fixed_length_primary_key)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1605 1606 1607 1608 1609 1610
    to->size=ref_length;
  else
  {
    KEY_PART_INFO *key_part=table->key_info[primary_key].key_part;
    KEY_PART_INFO *end=key_part+table->key_info[primary_key].key_parts;

1611
    for (; key_part != end ; key_part++)
1612
      pos+=key_part->field->packed_col_length((char*) pos,key_part->length);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1613 1614 1615 1616 1617 1618 1619 1620 1621 1622
    to->size= (uint) (pos- (byte*) to->data);
  }
  return to;
}


int ha_berkeley::rnd_pos(byte * buf, byte *pos)
{
  DBT db_pos;
  statistic_increment(ha_read_rnd_count,&LOCK_status);
1623
  DBUG_ENTER("ha_berkeley::rnd_pos");
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1624

1625
  active_index= (uint) -1;			// Don't delete via cursor
1626 1627 1628 1629
  DBUG_RETURN(read_row(file->get(file, transaction,
				 get_pos(&db_pos, pos),
				 &current_row, 0),
		       (char*) buf, primary_key, &current_row, (DBT*) 0, 0));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1630 1631 1632 1633 1634
}

void ha_berkeley::position(const byte *record)
{
  DBT key;
1635 1636 1637
  if (hidden_primary_key)
    memcpy_fixed(ref, (char*) current_ident, BDB_HIDDEN_PRIMARY_KEY_LENGTH);
  else
1638
    create_key(&key, primary_key, (char*) ref, record);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1639 1640 1641 1642 1643
}


void ha_berkeley::info(uint flag)
{
1644
  DBUG_ENTER("ha_berkeley::info");
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1645 1646
  if (flag & HA_STATUS_VARIABLE)
  {
1647
    records = share->rows + changed_rows; // Just to get optimisations right
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1648 1649
    deleted = 0;
  }
1650 1651 1652 1653 1654 1655 1656 1657 1658
  if ((flag & HA_STATUS_CONST) || version != share->version)
  {
    version=share->version;
    for (uint i=0 ; i < table->keys ; i++)
    {
      table->key_info[i].rec_per_key[table->key_info[i].key_parts-1]=
	share->rec_per_key[i];
    }
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1659 1660 1661 1662 1663 1664 1665 1666
  else if (flag & HA_STATUS_ERRKEY)
    errkey=last_dup_key;
  DBUG_VOID_RETURN;
}


int ha_berkeley::extra(enum ha_extra_function operation)
{
1667 1668 1669 1670
  switch (operation) {
  case HA_EXTRA_RESET:
  case HA_EXTRA_RESET_STATE:
    key_read=0;
1671
    using_ignore=0;
1672 1673 1674 1675 1676 1677 1678 1679 1680
    if (current_row.flags & (DB_DBT_MALLOC | DB_DBT_REALLOC))
    {
      current_row.flags=0;
      if (current_row.data)
      {
	free(current_row.data);
	current_row.data=0;
      }
    }
1681 1682 1683 1684 1685 1686 1687
    break;
  case HA_EXTRA_KEYREAD:
    key_read=1;					// Query satisfied with key
    break;
  case HA_EXTRA_NO_KEYREAD:
    key_read=0;
    break;
1688 1689 1690 1691 1692 1693
  case HA_EXTRA_IGNORE_DUP_KEY:
    using_ignore=1;
    break;
  case HA_EXTRA_NO_IGNORE_DUP_KEY:
    using_ignore=0;
    break;
1694 1695 1696
  default:
    break;
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1697 1698 1699
  return 0;
}

1700

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1701 1702
int ha_berkeley::reset(void)
{
1703
  key_read=0;					// Reset to state after open
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1704 1705 1706 1707 1708 1709 1710
  return 0;
}


/*
  As MySQL will execute an external lock for every new table it uses
  we can use this to start the transactions.
1711 1712 1713 1714
  If we are in auto_commit mode we just need to start a transaction
  for the statement to be able to rollback the statement.
  If not, we have to start a master transaction if there doesn't exist
  one from before.
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1715 1716 1717 1718 1719 1720 1721 1722
*/

int ha_berkeley::external_lock(THD *thd, int lock_type)
{
  int error=0;
  DBUG_ENTER("ha_berkeley::external_lock");
  if (lock_type != F_UNLCK)
  {
1723
    if (!thd->transaction.bdb_lock_count++)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1724
    {
1725
      DBUG_ASSERT(thd->transaction.stmt.bdb_tid == 0);
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
1726
      transaction=0;				// Safety
1727
      /* First table lock, start transaction */
1728
      if ((thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN |
1729
			   OPTION_TABLE_LOCK)) &&
1730 1731 1732
	  !thd->transaction.all.bdb_tid)
      {
	/* We have to start a master transaction */
1733
	DBUG_PRINT("trans",("starting transaction all"));
1734 1735 1736 1737
	if ((error=txn_begin(db_env, 0,
			     (DB_TXN**) &thd->transaction.all.bdb_tid,
			     0)))
	{
1738 1739
	  thd->transaction.bdb_lock_count--;	// We didn't get the lock /* purecov: inspected */
	  DBUG_RETURN(error); /* purecov: inspected */
1740
	}
1741 1742
	if (thd->in_lock_tables)
	  DBUG_RETURN(0);			// Don't create stmt trans
1743
      }
1744
      DBUG_PRINT("trans",("starting transaction stmt"));
1745 1746 1747
      if ((error=txn_begin(db_env,
			   (DB_TXN*) thd->transaction.all.bdb_tid,
			   (DB_TXN**) &thd->transaction.stmt.bdb_tid,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1748
			   0)))
1749 1750
      {
	/* We leave the possible master transaction open */
1751 1752
	thd->transaction.bdb_lock_count--;	// We didn't get the lock /* purecov: inspected */
	DBUG_RETURN(error); /* purecov: inspected */
1753
      }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1754
    }
1755
    transaction= (DB_TXN*) thd->transaction.stmt.bdb_tid;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1756 1757 1758 1759
  }
  else
  {
    lock.type=TL_UNLOCK;			// Unlocked
1760
    thread_safe_add(share->rows, changed_rows, &share->mutex);
1761
    changed_rows=0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1762 1763
    if (!--thd->transaction.bdb_lock_count)
    {
1764
      if (thd->transaction.stmt.bdb_tid)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1765
      {
1766
	/*
1767
	   F_UNLOCK is done without a transaction commit / rollback.
1768 1769
	   This happens if the thread didn't update any rows
	   We must in this case commit the work to keep the row locks
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1770
	*/
1771 1772 1773
	DBUG_PRINT("trans",("commiting non-updating transaction"));
	error=txn_commit((DB_TXN*) thd->transaction.stmt.bdb_tid,0);
	thd->transaction.stmt.bdb_tid=0;
1774
	transaction=0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1775 1776 1777 1778
      }
    }
  }
  DBUG_RETURN(error);
1779
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1780

1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793

/*
  When using LOCK TABLE's external_lock is only called when the actual
  TABLE LOCK is done.
  Under LOCK TABLES, each used tables will force a call to start_stmt.
*/

int ha_berkeley::start_stmt(THD *thd)
{
  int error=0;
  DBUG_ENTER("ha_berkeley::start_stmt");
  if (!thd->transaction.stmt.bdb_tid)
  {
1794
    DBUG_PRINT("trans",("starting transaction stmt"));
1795 1796 1797 1798
    error=txn_begin(db_env, (DB_TXN*) thd->transaction.all.bdb_tid,
		    (DB_TXN**) &thd->transaction.stmt.bdb_tid,
		    0);
  }
1799
  transaction= (DB_TXN*) thd->transaction.stmt.bdb_tid;
1800 1801 1802 1803
  DBUG_RETURN(error);
}


1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831
/*
  The idea with handler::store_lock() is the following:

  The statement decided which locks we should need for the table
  for updates/deletes/inserts we get WRITE locks, for SELECT... we get
  read locks.

  Before adding the lock into the table lock handler (see thr_lock.c)
  mysqld calls store lock with the requested locks.  Store lock can now
  modify a write lock to a read lock (or some other lock), ignore the
  lock (if we don't want to use MySQL table locks at all) or add locks
  for many tables (like we do when we are using a MERGE handler).

  Berkeley DB changes all WRITE locks to TL_WRITE_ALLOW_WRITE (which
  signals that we are doing WRITES, but we are still allowing other
  reader's and writer's.

  When releasing locks, store_lock() are also called. In this case one
  usually doesn't have to do anything.

  In some exceptional cases MySQL may send a request for a TL_IGNORE;
  This means that we are requesting the same lock as last time and this
  should also be ignored. (This may happen when someone does a flush
  table when we have opened a part of the tables, in which case mysqld
  closes and reopens the tables and tries to get the same locks at last
  time).  In the future we will probably try to remove this.
*/

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843

THR_LOCK_DATA **ha_berkeley::store_lock(THD *thd, THR_LOCK_DATA **to,
					enum thr_lock_type lock_type)
{
  if (lock_type != TL_IGNORE && lock.type == TL_UNLOCK)
  {
    /* If we are not doing a LOCK TABLE, then allow multiple writers */
    if ((lock_type >= TL_WRITE_CONCURRENT_INSERT &&
	 lock_type <= TL_WRITE) &&
	!thd->in_lock_tables)
      lock_type = TL_WRITE_ALLOW_WRITE;
    lock.type=lock_type;
1844 1845
    lock_on_read= ((table->reginfo.lock_type > TL_WRITE_ALLOW_READ) ? DB_RMW :
		   0);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1846 1847 1848 1849 1850 1851 1852 1853 1854
  }
  *to++= &lock;
  return to;
}


static int create_sub_table(const char *table_name, const char *sub_name,
			    DBTYPE type, int flags)
{
1855
  int error;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1856 1857 1858 1859 1860 1861 1862
  DB *file;
  DBUG_ENTER("create_sub_table");
  DBUG_PRINT("enter",("sub_name: %s",sub_name));

  if (!(error=db_create(&file, db_env, 0)))
  {
    file->set_flags(file, flags);
ram@mysql.r18.ru's avatar
ram@mysql.r18.ru committed
1863
    error=(file->open(file, NULL, table_name, sub_name, type,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1864 1865 1866
		      DB_THREAD | DB_CREATE, my_umask));
    if (error)
    {
1867 1868 1869
      DBUG_PRINT("error",("Got error: %d when opening table '%s'",error, /* purecov: inspected */
			  table_name)); /* purecov: inspected */
      (void) file->remove(file,table_name,NULL,0); /* purecov: inspected */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1870 1871 1872 1873 1874 1875
    }
    else
      (void) file->close(file,0);
  }
  else
  {
1876
    DBUG_PRINT("error",("Got error: %d when creting table",error)); /* purecov: inspected */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1877 1878
  }
  if (error)
1879
    my_errno=error; /* purecov: inspected */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1880 1881 1882 1883 1884 1885 1886 1887 1888
  DBUG_RETURN(error);
}


int ha_berkeley::create(const char *name, register TABLE *form,
			HA_CREATE_INFO *create_info)
{
  char name_buff[FN_REFLEN];
  char part[7];
1889
  uint index=1;
1890
  int error=1;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1891 1892 1893 1894 1895 1896
  DBUG_ENTER("ha_berkeley::create");

  fn_format(name_buff,name,"", ha_berkeley_ext,2 | 4);

  /* Create the main table that will hold the real rows */
  if (create_sub_table(name_buff,"main",DB_BTREE,0))
1897
    DBUG_RETURN(1); /* purecov: inspected */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1898

1899
  primary_key=table->primary_key;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1900
  /* Create the keys */
1901
  for (uint i=0; i < form->keys; i++)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1902
  {
1903 1904 1905 1906 1907 1908
    if (i != primary_key)
    {
      sprintf(part,"key%02d",index++);
      if (create_sub_table(name_buff, part, DB_BTREE,
			   (table->key_info[i].flags & HA_NOSAME) ? 0 :
			   DB_DUP))
1909
	DBUG_RETURN(1); /* purecov: inspected */
1910
    }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1911 1912 1913 1914
  }

  /* Create the status block to save information from last status command */
  /* Is DB_BTREE the best option here ? (QUEUE can't be used in sub tables) */
1915 1916 1917 1918

  DB *status_block;
  if (!db_create(&status_block, db_env, 0))
  {
ram@mysql.r18.ru's avatar
ram@mysql.r18.ru committed
1919
    if (!status_block->open(status_block, NULL, name_buff,
1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930
			    "status", DB_BTREE, DB_CREATE, 0))
    {
      char rec_buff[4+MAX_KEY*4];
      uint length= 4+ table->keys*4;
      bzero(rec_buff, length);
      if (!write_status(status_block, rec_buff, length))
	error=0;
      status_block->close(status_block,0);
    }
  }
  DBUG_RETURN(error);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1931 1932 1933 1934 1935 1936 1937
}


int ha_berkeley::delete_table(const char *name)
{
  int error;
  char name_buff[FN_REFLEN];
1938
  DBUG_ENTER("delete_table");
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1939
  if ((error=db_create(&file, db_env, 0)))
1940
    my_errno=error; /* purecov: inspected */
1941 1942 1943
  else
    error=file->remove(file,fn_format(name_buff,name,"",ha_berkeley_ext,2 | 4),
		       NULL,0);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1944
  file=0;					// Safety
1945
  DBUG_RETURN(error);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1946 1947
}

1948 1949 1950 1951 1952 1953 1954 1955 1956 1957

int ha_berkeley::rename_table(const char * from, const char * to)
{
  int error;
  char from_buff[FN_REFLEN];
  char to_buff[FN_REFLEN];

  if ((error= db_create(&file, db_env, 0)))
    my_errno= error;
  else
1958 1959
  {
    /* On should not do a file->close() after rename returns */
1960 1961 1962 1963
    error= file->rename(file, 
			fn_format(from_buff, from, "", ha_berkeley_ext, 2 | 4),
			NULL, fn_format(to_buff, to, "", ha_berkeley_ext,
					2 | 4), 0);
1964
  }
1965 1966 1967 1968
  return error;
}


bk@work.mysql.com's avatar
bk@work.mysql.com committed
1969 1970 1971 1972 1973 1974 1975 1976
/*
  How many seeks it will take to read through the table
  This is to be comparable to the number returned by records_in_range so
  that we can decide if we should scan the table or use keys.
*/

double ha_berkeley::scan_time()
{
1977
  return rows2double(records/3);
1978
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1979 1980 1981 1982 1983 1984 1985 1986 1987

ha_rows ha_berkeley::records_in_range(int keynr,
				      const byte *start_key,uint start_key_len,
				      enum ha_rkey_function start_search_flag,
				      const byte *end_key,uint end_key_len,
				      enum ha_rkey_function end_search_flag)
{
  DBT key;
  DB_KEY_RANGE start_range, end_range;
1988
  DB *kfile=key_file[keynr];
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1989 1990
  double start_pos,end_pos,rows;
  DBUG_ENTER("records_in_range");
1991 1992 1993 1994 1995 1996 1997 1998 1999

  if ((start_key && kfile->key_range(kfile,transaction,
				     pack_key(&key, keynr, key_buff, start_key,
					      start_key_len),
				     &start_range,0)) ||
      (end_key && kfile->key_range(kfile,transaction,
				   pack_key(&key, keynr, key_buff, end_key,
					    end_key_len),
				   &end_range,0)))
2000
    DBUG_RETURN(HA_BERKELEY_RANGE_COUNT); // Better than returning an error /* purecov: inspected */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019

  if (!start_key)
    start_pos=0.0;
  else if (start_search_flag == HA_READ_KEY_EXACT)
    start_pos=start_range.less;
  else
    start_pos=start_range.less+start_range.equal;

  if (!end_key)
    end_pos=1.0;
  else if (end_search_flag == HA_READ_BEFORE_KEY)
    end_pos=end_range.less;
  else
    end_pos=end_range.less+end_range.equal;
  rows=(end_pos-start_pos)*records;
  DBUG_PRINT("exit",("rows: %g",rows));
  DBUG_RETURN(rows <= 1.0 ? (ha_rows) 1 : (ha_rows) rows);
}

2020 2021 2022 2023 2024 2025

longlong ha_berkeley::get_auto_increment()
{
  longlong nr=1;				// Default if error or new key
  int error;
  (void) ha_berkeley::extra(HA_EXTRA_KEYREAD);
2026 2027

  /* Set 'active_index' */
2028 2029 2030 2031 2032 2033 2034 2035
  ha_berkeley::index_init(table->next_number_index);

  if (!table->next_number_key_offset)
  {						// Autoincrement at key-start
    error=ha_berkeley::index_last(table->record[1]);
  }
  else
  {
2036
    DBT row,old_key;
2037 2038 2039 2040 2041 2042 2043 2044
    bzero((char*) &row,sizeof(row));
    KEY *key_info= &table->key_info[active_index];

    /* Reading next available number for a sub key */
    ha_berkeley::create_key(&last_key, active_index,
			    key_buff, table->record[0],
			    table->next_number_key_offset);
    /* Store for compare */
2045
    memcpy(old_key.data=key_buff2, key_buff, (old_key.size=last_key.size));
2046
    old_key.app_private=(void*) key_info;
2047
    error=1;
2048
    {
2049 2050 2051
      /* Modify the compare so that we will find the next key */
      key_info->handler.bdb_return_if_eq= 1;
      /* We lock the next key as the new key will probl. be on the same page */
2052
      error=cursor->c_get(cursor, &last_key, &row, DB_SET_RANGE | DB_RMW);
2053 2054 2055 2056 2057 2058 2059 2060
      key_info->handler.bdb_return_if_eq= 0;
      if (!error || error == DB_NOTFOUND)
      {
	/*
	  Now search go one step back and then we should have found the
	  biggest key with the given prefix
	  */
	error=1;
2061 2062 2063
	if (!cursor->c_get(cursor, &last_key, &row, DB_PREV | DB_RMW) &&
	    !berkeley_cmp_packed_key(key_file[active_index], &old_key,
				     &last_key))
2064 2065
	{
	  error=0;				// Found value
2066
	  unpack_key((char*) table->record[1], &last_key, active_index);
2067 2068
	}
      }
2069 2070
    }
  }
2071 2072 2073
  if (!error)
    nr=(longlong)
      table->next_number_field->val_int_offset(table->rec_buff_length)+1;
2074 2075 2076 2077 2078
  ha_berkeley::index_end();
  (void) ha_berkeley::extra(HA_EXTRA_NO_KEYREAD);
  return nr;
}

2079 2080 2081 2082 2083 2084
void ha_berkeley::print_error(int error, myf errflag)
{
  if (error == DB_LOCK_DEADLOCK)
    error=HA_ERR_LOCK_DEADLOCK;
  handler::print_error(error,errflag);
}
2085 2086 2087 2088 2089

/****************************************************************************
	 Analyzing, checking, and optimizing tables
****************************************************************************/

2090
#ifdef NOT_YET
2091 2092 2093
static void print_msg(THD *thd, const char *table_name, const char *op_name,
		      const char *msg_type, const char *fmt, ...)
{
2094
  Protocol *protocol= thd->protocol;
2095 2096 2097 2098 2099 2100 2101 2102 2103
  char msgbuf[256];
  msgbuf[0] = 0;
  va_list args;
  va_start(args,fmt);

  my_vsnprintf(msgbuf, sizeof(msgbuf), fmt, args);
  msgbuf[sizeof(msgbuf) - 1] = 0; // healthy paranoia
  DBUG_PRINT(msg_type,("message: %s",msgbuf));

hf@deer.mysql.r18.ru's avatar
hf@deer.mysql.r18.ru committed
2104
  protocol->set_nfields(4);
2105 2106 2107 2108 2109 2110
  protocol->prepare_for_resend();
  protocol->store(table_name);
  protocol->store(op_name);
  protocol->store(msg_type);
  protocol->store(msgbuf);
  if (protocol->write())
2111 2112
    thd->killed=1;
}
2113
#endif
2114 2115 2116

int ha_berkeley::analyze(THD* thd, HA_CHECK_OPT* check_opt)
{
2117
  DB_BTREE_STAT *stat=0;
2118 2119 2120 2121
  uint i;

  for (i=0 ; i < table->keys ; i++)
  {
2122 2123 2124 2125 2126
    if (stat)
    {
      free(stat);
      stat=0;
    }
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
2127
    if ((key_file[i]->stat)(key_file[i], (void*) &stat, 0))
2128
      goto err; /* purecov: inspected */
2129 2130
    share->rec_per_key[i]= (stat->bt_ndata /
			    (stat->bt_nkeys ? stat->bt_nkeys : 1));
2131
  }
2132
  /* A hidden primary key is not in key_file[] */
2133
  if (hidden_primary_key)
2134 2135 2136 2137 2138 2139
  {
    if (stat)
    {
      free(stat);
      stat=0;
    }
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
2140
    if ((file->stat)(file, (void*) &stat, 0))
2141
      goto err; /* purecov: inspected */
2142
  }
2143
  pthread_mutex_lock(&share->mutex);
2144
  share->rows=stat->bt_ndata;
2145 2146 2147
  share->status|=STATUS_BDB_ANALYZE;		// Save status on close
  share->version++;				// Update stat in table
  pthread_mutex_unlock(&share->mutex);
2148 2149 2150
  update_status(share,table);			// Write status to file
  if (stat)
    free(stat);
2151 2152
  return ((share->status & STATUS_BDB_ANALYZE) ? HA_ADMIN_FAILED :
	  HA_ADMIN_OK);
2153 2154

err:
2155 2156 2157
  if (stat) /* purecov: inspected */
    free(stat); /* purecov: inspected */
  return HA_ADMIN_FAILED; /* purecov: inspected */
2158 2159 2160 2161 2162 2163 2164 2165 2166 2167
}

int ha_berkeley::optimize(THD* thd, HA_CHECK_OPT* check_opt)
{
  return ha_berkeley::analyze(thd,check_opt);
}


int ha_berkeley::check(THD* thd, HA_CHECK_OPT* check_opt)
{
2168 2169 2170 2171 2172
  DBUG_ENTER("ha_berkeley::check");

  DBUG_RETURN(HA_ADMIN_NOT_IMPLEMENTED);

#ifdef NOT_YET
2173 2174 2175
  char name_buff[FN_REFLEN];
  int error;
  DB *tmp_file;
2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193
  /*
    To get this to work we need to ensure that no running transaction is
    using the table. We also need to create a new environment without
    locking for this.
  */

  /* We must open the file again to be able to check it! */
  if ((error=db_create(&tmp_file, db_env, 0)))
  {
    print_msg(thd, table->real_name, "check", "error",
	      "Got error %d creating environment",error);
    DBUG_RETURN(HA_ADMIN_FAILED);
  }

  /* Compare the overall structure */
  tmp_file->set_bt_compare(tmp_file,
			   (hidden_primary_key ? berkeley_cmp_hidden_key :
			    berkeley_cmp_packed_key));
2194
  tmp_file->app_private= (void*) (table->key_info+table->primary_key);
2195
  fn_format(name_buff,share->table_name,"", ha_berkeley_ext, 2 | 4);
2196 2197
  if ((error=tmp_file->verify(tmp_file, name_buff, NullS, (FILE*) 0,
			      hidden_primary_key ? 0 : DB_NOORDERCHK)))
2198 2199 2200
  {
    print_msg(thd, table->real_name, "check", "error",
	      "Got error %d checking file structure",error);
2201 2202
    tmp_file->close(tmp_file,0);
    DBUG_RETURN(HA_ADMIN_CORRUPT);
2203
  }
2204 2205 2206 2207

  /* Check each index */
  tmp_file->set_bt_compare(tmp_file, berkeley_cmp_packed_key);
  for (uint index=0,i=0 ; i < table->keys ; i++)
2208
  {
2209 2210 2211 2212 2213 2214 2215 2216
    char part[7];
    if (i == primary_key)
      strmov(part,"main");
    else
      sprintf(part,"key%02d",++index);
    tmp_file->app_private= (void*) (table->key_info+i);
    if ((error=tmp_file->verify(tmp_file, name_buff, part, (FILE*) 0,
				DB_ORDERCHKONLY)))
2217 2218
    {
      print_msg(thd, table->real_name, "check", "error",
2219 2220 2221 2222 2223
		"Key %d was not in order (Error: %d)",
		index+ test(i >= primary_key),
		error);
      tmp_file->close(tmp_file,0);
      DBUG_RETURN(HA_ADMIN_CORRUPT);
2224 2225
    }
  }
2226 2227 2228
  tmp_file->close(tmp_file,0);
  DBUG_RETURN(HA_ADMIN_OK);
#endif
2229 2230
}

bk@work.mysql.com's avatar
bk@work.mysql.com committed
2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242
/****************************************************************************
 Handling the shared BDB_SHARE structure that is needed to provide table
 locking.
****************************************************************************/

static byte* bdb_get_key(BDB_SHARE *share,uint *length,
			 my_bool not_used __attribute__((unused)))
{
  *length=share->table_name_length;
  return (byte*) share->table_name;
}

2243
static BDB_SHARE *get_share(const char *table_name, TABLE *table)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2244 2245 2246
{
  BDB_SHARE *share;
  pthread_mutex_lock(&bdb_mutex);
2247
  uint length=(uint) strlen(table_name);
2248 2249
  if (!(share=(BDB_SHARE*) hash_search(&bdb_open_tables, (byte*) table_name,
				       length)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2250
  {
2251
    ulong *rec_per_key;
2252 2253 2254
    char *tmp_name;
    DB **key_file;
    u_int32_t *key_type;
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2255

2256 2257 2258 2259 2260 2261 2262 2263
    if ((share=(BDB_SHARE *)
	 my_multi_malloc(MYF(MY_WME | MY_ZEROFILL),
			 &share, sizeof(*share),
			 &rec_per_key, table->keys * sizeof(ha_rows),
			 &tmp_name, length+1,
			 &key_file, (table->keys+1) * sizeof(*key_file),
			 &key_type, (table->keys+1) * sizeof(u_int32_t),
			 NullS)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2264
    {
2265 2266
      share->rec_per_key = rec_per_key;
      share->table_name = tmp_name;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2267 2268
      share->table_name_length=length;
      strmov(share->table_name,table_name);
2269 2270
      share->key_file = key_file;
      share->key_type = key_type;
hf@deer.(none)'s avatar
SCRUM  
hf@deer.(none) committed
2271
      if (my_hash_insert(&bdb_open_tables, (byte*) share))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2272
      {
2273 2274 2275
	pthread_mutex_unlock(&bdb_mutex); /* purecov: inspected */
	my_free((gptr) share,0); /* purecov: inspected */
	return 0; /* purecov: inspected */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2276 2277
      }
      thr_lock_init(&share->lock);
2278
      pthread_mutex_init(&share->mutex,MY_MUTEX_INIT_FAST);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2279 2280 2281 2282 2283 2284
    }
  }
  pthread_mutex_unlock(&bdb_mutex);
  return share;
}

2285 2286
static int free_share(BDB_SHARE *share, TABLE *table, uint hidden_primary_key,
		      bool mutex_is_locked)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2287
{
2288
  int error, result = 0;
2289
  uint keys=table->keys + test(hidden_primary_key);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2290
  pthread_mutex_lock(&bdb_mutex);
2291
  if (mutex_is_locked)
2292
    pthread_mutex_unlock(&share->mutex); /* purecov: inspected */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2293 2294
  if (!--share->use_count)
  {
2295
    DB **key_file = share->key_file;
2296
    update_status(share,table);
2297
    /* this does share->file->close() implicitly */
2298
    for (uint i=0; i < keys; i++)
2299 2300
    {
      if (key_file[i] && (error=key_file[i]->close(key_file[i],0)))
2301
	result=error; /* purecov: inspected */
2302 2303 2304
    }
    if (share->status_block &&
	(error = share->status_block->close(share->status_block,0)))
2305
      result = error; /* purecov: inspected */
2306
    hash_delete(&bdb_open_tables, (byte*) share);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2307
    thr_lock_delete(&share->lock);
2308
    pthread_mutex_destroy(&share->mutex);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2309 2310 2311
    my_free((gptr) share, MYF(0));
  }
  pthread_mutex_unlock(&bdb_mutex);
2312
  return result;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2313 2314
}

2315 2316 2317 2318
/*
  Get status information that is stored in the 'status' sub database
  and the max used value for the hidden primary key.
*/
2319

2320
void ha_berkeley::get_status()
2321
{
2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342
  if (!test_all_bits(share->status,(STATUS_PRIMARY_KEY_INIT |
				    STATUS_ROW_COUNT_INIT)))
  {
    pthread_mutex_lock(&share->mutex);
    if (!(share->status & STATUS_PRIMARY_KEY_INIT))
    {
      (void) extra(HA_EXTRA_KEYREAD);
      index_init(primary_key);
      if (!index_last(table->record[1]))
	share->auto_ident=uint5korr(current_ident);
      index_end();
      (void) extra(HA_EXTRA_NO_KEYREAD);
    }
    if (! share->status_block)
    {
      char name_buff[FN_REFLEN];
      uint open_mode= (((table->db_stat & HA_READ_ONLY) ? DB_RDONLY : 0)
		       | DB_THREAD);
      fn_format(name_buff, share->table_name,"", ha_berkeley_ext, 2 | 4);
      if (!db_create(&share->status_block, db_env, 0))
      {
ram@mysql.r18.ru's avatar
ram@mysql.r18.ru committed
2343
	if (share->status_block->open(share->status_block, NULL, name_buff,
2344
				      "status", DB_BTREE, open_mode, 0))
2345
	{
2346 2347
	  share->status_block->close(share->status_block, 0); /* purecov: inspected */
	  share->status_block=0; /* purecov: inspected */
2348 2349 2350 2351 2352 2353 2354
	}
      }
    }
    if (!(share->status & STATUS_ROW_COUNT_INIT) && share->status_block)
    {
      share->org_rows=share->rows=
	table->max_rows ? table->max_rows : HA_BERKELEY_MAX_ROWS;
2355
      if (!share->status_block->cursor(share->status_block, 0, &cursor, 0))
2356 2357
      {
	DBT row;
2358
	char rec_buff[64];
2359 2360 2361
	bzero((char*) &row,sizeof(row));
	bzero((char*) &last_key,sizeof(last_key));
	row.data=rec_buff;
2362
	row.ulen=sizeof(rec_buff);
2363 2364 2365 2366
	row.flags=DB_DBT_USERMEM;
	if (!cursor->c_get(cursor, &last_key, &row, DB_FIRST))
	{
	  uint i;
2367
	  uchar *pos=(uchar*) row.data;
2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383
	  share->org_rows=share->rows=uint4korr(pos); pos+=4;
	  for (i=0 ; i < table->keys ; i++)
	  {
	    share->rec_per_key[i]=uint4korr(pos); pos+=4;
	  }
	}
	cursor->c_close(cursor);
      }
      cursor=0;					// Safety
    }
    share->status|= STATUS_PRIMARY_KEY_INIT | STATUS_ROW_COUNT_INIT;
    pthread_mutex_unlock(&share->mutex);
  }
}


2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400
static int write_status(DB *status_block, char *buff, uint length)
{
  DBT row,key;
  int error;
  const char *key_buff="status";

  bzero((char*) &row,sizeof(row));
  bzero((char*) &key,sizeof(key));
  row.data=buff;
  key.data=(void*) key_buff;
  key.size=sizeof(key_buff);
  row.size=length;
  error=status_block->put(status_block, 0, &key, &row, 0);
  return error;
}


2401 2402 2403 2404 2405
static void update_status(BDB_SHARE *share, TABLE *table)
{
  DBUG_ENTER("update_status");
  if (share->rows != share->org_rows ||
      (share->status & STATUS_BDB_ANALYZE))
2406
  {
2407 2408 2409 2410 2411 2412 2413 2414
    pthread_mutex_lock(&share->mutex);
    if (!share->status_block)
    {
      /*
	Create sub database 'status' if it doesn't exist from before
	(This '*should*' always exist for table created with MySQL)
      */

2415 2416 2417 2418
      char name_buff[FN_REFLEN]; /* purecov: inspected */
      if (db_create(&share->status_block, db_env, 0)) /* purecov: inspected */
	goto end; /* purecov: inspected */
      share->status_block->set_flags(share->status_block,0); /* purecov: inspected */
ram@mysql.r18.ru's avatar
ram@mysql.r18.ru committed
2419
      if (share->status_block->open(share->status_block, NULL,
2420 2421 2422
				    fn_format(name_buff,share->table_name,"",
					      ha_berkeley_ext,2 | 4),
				    "status", DB_BTREE,
2423 2424
				    DB_THREAD | DB_CREATE, my_umask)) /* purecov: inspected */
	goto end; /* purecov: inspected */
2425 2426
    }
    {
2427
      char rec_buff[4+MAX_KEY*4], *pos=rec_buff;
2428
      int4store(pos,share->rows); pos+=4;
2429
      for (uint i=0 ; i < table->keys ; i++)
2430 2431 2432
      {
	int4store(pos,share->rec_per_key[i]); pos+=4;
      }
2433 2434 2435
      DBUG_PRINT("info",("updating status for %s",share->table_name));
      (void) write_status(share->status_block, rec_buff,
			  (uint) (pos-rec_buff));
2436
      share->status&= ~STATUS_BDB_ANALYZE;
2437
      share->org_rows=share->rows;
2438 2439 2440
    }
end:
    pthread_mutex_unlock(&share->mutex);
2441
  }
2442
  DBUG_VOID_RETURN;
2443 2444
}

2445 2446 2447 2448 2449 2450 2451
/*
  Return an estimated of the number of rows in the table.
  Used when sorting to allocate buffers and by the optimizer.
*/

ha_rows ha_berkeley::estimate_number_of_rows()
{
2452
  return share->rows + HA_BERKELEY_EXTRA_ROWS;
2453 2454
}

bk@work.mysql.com's avatar
bk@work.mysql.com committed
2455
#endif /* HAVE_BERKELEY_DB */