lock0lock.c 156 KB
Newer Older
Vadim Tkachenko's avatar
Vadim Tkachenko committed
1 2
/*****************************************************************************

3
Copyright (c) 1996, 2010, Innobase Oy. All Rights Reserved.
Vadim Tkachenko's avatar
Vadim Tkachenko committed
4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

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; version 2 of the License.

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.

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

*****************************************************************************/

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
19 20
/**************************************************//**
@file lock/lock0lock.c
21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216
The transaction lock system

Created 5/7/1996 Heikki Tuuri
*******************************************************/

#define LOCK_MODULE_IMPLEMENTATION

#include "lock0lock.h"
#include "lock0priv.h"

#ifdef UNIV_NONINL
#include "lock0lock.ic"
#include "lock0priv.ic"
#endif

#include "ha_prototypes.h"
#include "usr0sess.h"
#include "trx0purge.h"
#include "dict0mem.h"
#include "trx0sys.h"

/* Restricts the length of search we will do in the waits-for
graph of transactions */
#define LOCK_MAX_N_STEPS_IN_DEADLOCK_CHECK 1000000

/* Restricts the recursion depth of the search we will do in the waits-for
graph of transactions */
#define LOCK_MAX_DEPTH_IN_DEADLOCK_CHECK 200

/* When releasing transaction locks, this specifies how often we release
the kernel mutex for a moment to give also others access to it */

#define LOCK_RELEASE_KERNEL_INTERVAL	1000

/* Safety margin when creating a new record lock: this many extra records
can be inserted to the page without need to create a lock with a bigger
bitmap */

#define LOCK_PAGE_BITMAP_MARGIN		64

/* An explicit record lock affects both the record and the gap before it.
An implicit x-lock does not affect the gap, it only locks the index
record from read or update.

If a transaction has modified or inserted an index record, then
it owns an implicit x-lock on the record. On a secondary index record,
a transaction has an implicit x-lock also if it has modified the
clustered index record, the max trx id of the page where the secondary
index record resides is >= trx id of the transaction (or database recovery
is running), and there are no explicit non-gap lock requests on the
secondary index record.

This complicated definition for a secondary index comes from the
implementation: we want to be able to determine if a secondary index
record has an implicit x-lock, just by looking at the present clustered
index record, not at the historical versions of the record. The
complicated definition can be explained to the user so that there is
nondeterminism in the access path when a query is answered: we may,
or may not, access the clustered index record and thus may, or may not,
bump into an x-lock set there.

Different transaction can have conflicting locks set on the gap at the
same time. The locks on the gap are purely inhibitive: an insert cannot
be made, or a select cursor may have to wait if a different transaction
has a conflicting lock on the gap. An x-lock on the gap does not give
the right to insert into the gap.

An explicit lock can be placed on a user record or the supremum record of
a page. The locks on the supremum record are always thought to be of the gap
type, though the gap bit is not set. When we perform an update of a record
where the size of the record changes, we may temporarily store its explicit
locks on the infimum record of the page, though the infimum otherwise never
carries locks.

A waiting record lock can also be of the gap type. A waiting lock request
can be granted when there is no conflicting mode lock request by another
transaction ahead of it in the explicit lock queue.

In version 4.0.5 we added yet another explicit lock type: LOCK_REC_NOT_GAP.
It only locks the record it is placed on, not the gap before the record.
This lock type is necessary to emulate an Oracle-like READ COMMITTED isolation
level.

-------------------------------------------------------------------------
RULE 1: If there is an implicit x-lock on a record, and there are non-gap
-------
lock requests waiting in the queue, then the transaction holding the implicit
x-lock also has an explicit non-gap record x-lock. Therefore, as locks are
released, we can grant locks to waiting lock requests purely by looking at
the explicit lock requests in the queue.

RULE 3: Different transactions cannot have conflicting granted non-gap locks
-------
on a record at the same time. However, they can have conflicting granted gap
locks.
RULE 4: If a there is a waiting lock request in a queue, no lock request,
-------
gap or not, can be inserted ahead of it in the queue. In record deletes
and page splits new gap type locks can be created by the database manager
for a transaction, and without rule 4, the waits-for graph of transactions
might become cyclic without the database noticing it, as the deadlock check
is only performed when a transaction itself requests a lock!
-------------------------------------------------------------------------

An insert is allowed to a gap if there are no explicit lock requests by
other transactions on the next record. It does not matter if these lock
requests are granted or waiting, gap bit set or not, with the exception
that a gap type request set by another transaction to wait for
its turn to do an insert is ignored. On the other hand, an
implicit x-lock by another transaction does not prevent an insert, which
allows for more concurrency when using an Oracle-style sequence number
generator for the primary key with many transactions doing inserts
concurrently.

A modify of a record is allowed if the transaction has an x-lock on the
record, or if other transactions do not have any non-gap lock requests on the
record.

A read of a single user record with a cursor is allowed if the transaction
has a non-gap explicit, or an implicit lock on the record, or if the other
transactions have no x-lock requests on the record. At a page supremum a
read is always allowed.

In summary, an implicit lock is seen as a granted x-lock only on the
record, not on the gap. An explicit lock with no gap bit set is a lock
both on the record and the gap. If the gap bit is set, the lock is only
on the gap. Different transaction cannot own conflicting locks on the
record at the same time, but they may own conflicting locks on the gap.
Granted locks on a record give an access right to the record, but gap type
locks just inhibit operations.

NOTE: Finding out if some transaction has an implicit x-lock on a secondary
index record can be cumbersome. We may have to look at previous versions of
the corresponding clustered index record to find out if a delete marked
secondary index record was delete marked by an active transaction, not by
a committed one.

FACT A: If a transaction has inserted a row, it can delete it any time
without need to wait for locks.

PROOF: The transaction has an implicit x-lock on every index record inserted
for the row, and can thus modify each record without the need to wait. Q.E.D.

FACT B: If a transaction has read some result set with a cursor, it can read
it again, and retrieves the same result set, if it has not modified the
result set in the meantime. Hence, there is no phantom problem. If the
biggest record, in the alphabetical order, touched by the cursor is removed,
a lock wait may occur, otherwise not.

PROOF: When a read cursor proceeds, it sets an s-lock on each user record
it passes, and a gap type s-lock on each page supremum. The cursor must
wait until it has these locks granted. Then no other transaction can
have a granted x-lock on any of the user records, and therefore cannot
modify the user records. Neither can any other transaction insert into
the gaps which were passed over by the cursor. Page splits and merges,
and removal of obsolete versions of records do not affect this, because
when a user record or a page supremum is removed, the next record inherits
its locks as gap type locks, and therefore blocks inserts to the same gap.
Also, if a page supremum is inserted, it inherits its locks from the successor
record. When the cursor is positioned again at the start of the result set,
the records it will touch on its course are either records it touched
during the last pass or new inserted page supremums. It can immediately
access all these records, and when it arrives at the biggest record, it
notices that the result set is complete. If the biggest record was removed,
lock wait can occur because the next record only inherits a gap type lock,
and a wait may be needed. Q.E.D. */

/* If an index record should be changed or a new inserted, we must check
the lock on the record or the next. When a read cursor starts reading,
we will set a record level s-lock on each record it passes, except on the
initial record on which the cursor is positioned before we start to fetch
records. Our index tree search has the convention that the B-tree
cursor is positioned BEFORE the first possibly matching record in
the search. Optimizations are possible here: if the record is searched
on an equality condition to a unique key, we could actually set a special
lock on the record, a lock which would not prevent any insert before
this record. In the next key locking an x-lock set on a record also
prevents inserts just before that record.
	There are special infimum and supremum records on each page.
A supremum record can be locked by a read cursor. This records cannot be
updated but the lock prevents insert of a user record to the end of
the page.
	Next key locks will prevent the phantom problem where new rows
could appear to SELECT result sets after the select operation has been
performed. Prevention of phantoms ensures the serilizability of
transactions.
	What should we check if an insert of a new record is wanted?
Only the lock on the next record on the same page, because also the
supremum record can carry a lock. An s-lock prevents insertion, but
what about an x-lock? If it was set by a searched update, then there
is implicitly an s-lock, too, and the insert should be prevented.
What if our transaction owns an x-lock to the next record, but there is
a waiting s-lock request on the next record? If this s-lock was placed
by a read cursor moving in the ascending order in the index, we cannot
do the insert immediately, because when we finally commit our transaction,
the read cursor should see also the new inserted record. So we should
217
move the read cursor backward from the next record for it to pass over
218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346
the new inserted record. This move backward may be too cumbersome to
implement. If we in this situation just enqueue a second x-lock request
for our transaction on the next record, then the deadlock mechanism
notices a deadlock between our transaction and the s-lock request
transaction. This seems to be an ok solution.
	We could have the convention that granted explicit record locks,
lock the corresponding records from changing, and also lock the gaps
before them from inserting. A waiting explicit lock request locks the gap
before from inserting. Implicit record x-locks, which we derive from the
transaction id in the clustered index record, only lock the record itself
from modification, not the gap before it from inserting.
	How should we store update locks? If the search is done by a unique
key, we could just modify the record trx id. Otherwise, we could put a record
x-lock on the record. If the update changes ordering fields of the
clustered index record, the inserted new record needs no record lock in
lock table, the trx id is enough. The same holds for a secondary index
record. Searched delete is similar to update.

PROBLEM:
What about waiting lock requests? If a transaction is waiting to make an
update to a record which another modified, how does the other transaction
know to send the end-lock-wait signal to the waiting transaction? If we have
the convention that a transaction may wait for just one lock at a time, how
do we preserve it if lock wait ends?

PROBLEM:
Checking the trx id label of a secondary index record. In the case of a
modification, not an insert, is this necessary? A secondary index record
is modified only by setting or resetting its deleted flag. A secondary index
record contains fields to uniquely determine the corresponding clustered
index record. A secondary index record is therefore only modified if we
also modify the clustered index record, and the trx id checking is done
on the clustered index record, before we come to modify the secondary index
record. So, in the case of delete marking or unmarking a secondary index
record, we do not have to care about trx ids, only the locks in the lock
table must be checked. In the case of a select from a secondary index, the
trx id is relevant, and in this case we may have to search the clustered
index record.

PROBLEM: How to update record locks when page is split or merged, or
--------------------------------------------------------------------
a record is deleted or updated?
If the size of fields in a record changes, we perform the update by
a delete followed by an insert. How can we retain the locks set or
waiting on the record? Because a record lock is indexed in the bitmap
by the heap number of the record, when we remove the record from the
record list, it is possible still to keep the lock bits. If the page
is reorganized, we could make a table of old and new heap numbers,
and permute the bitmaps in the locks accordingly. We can add to the
table a row telling where the updated record ended. If the update does
not require a reorganization of the page, we can simply move the lock
bits for the updated record to the position determined by its new heap
number (we may have to allocate a new lock, if we run out of the bitmap
in the old one).
	A more complicated case is the one where the reinsertion of the
updated record is done pessimistically, because the structure of the
tree may change.

PROBLEM: If a supremum record is removed in a page merge, or a record
---------------------------------------------------------------------
removed in a purge, what to do to the waiting lock requests? In a split to
the right, we just move the lock requests to the new supremum. If a record
is removed, we could move the waiting lock request to its inheritor, the
next record in the index. But, the next record may already have lock
requests on its own queue. A new deadlock check should be made then. Maybe
it is easier just to release the waiting transactions. They can then enqueue
new lock requests on appropriate records.

PROBLEM: When a record is inserted, what locks should it inherit from the
-------------------------------------------------------------------------
upper neighbor? An insert of a new supremum record in a page split is
always possible, but an insert of a new user record requires that the upper
neighbor does not have any lock requests by other transactions, granted or
waiting, in its lock queue. Solution: We can copy the locks as gap type
locks, so that also the waiting locks are transformed to granted gap type
locks on the inserted record. */

/* LOCK COMPATIBILITY MATRIX
 *    IS IX S  X  AI
 * IS +	 +  +  -  +
 * IX +	 +  -  -  +
 * S  +	 -  +  -  -
 * X  -	 -  -  -  -
 * AI +	 +  -  -  -
 *
 * Note that for rows, InnoDB only acquires S or X locks.
 * For tables, InnoDB normally acquires IS or IX locks.
 * S or X table locks are only acquired for LOCK TABLES.
 * Auto-increment (AI) locks are needed because of
 * statement-level MySQL binlog.
 * See also lock_mode_compatible().
 */
#define LK(a,b) (1 << ((a) * LOCK_NUM + (b)))
#define LKS(a,b) LK(a,b) | LK(b,a)

/* Define the lock compatibility matrix in a ulint.  The first line below
defines the diagonal entries.  The following lines define the compatibility
for LOCK_IX, LOCK_S, and LOCK_AUTO_INC using LKS(), since the matrix
is symmetric. */
#define LOCK_MODE_COMPATIBILITY 0					\
 | LK(LOCK_IS, LOCK_IS) | LK(LOCK_IX, LOCK_IX) | LK(LOCK_S, LOCK_S)	\
 | LKS(LOCK_IX, LOCK_IS) | LKS(LOCK_IS, LOCK_AUTO_INC)			\
 | LKS(LOCK_S, LOCK_IS)							\
 | LKS(LOCK_AUTO_INC, LOCK_IS) | LKS(LOCK_AUTO_INC, LOCK_IX)

/* STRONGER-OR-EQUAL RELATION (mode1=row, mode2=column)
 *    IS IX S  X  AI
 * IS +  -  -  -  -
 * IX +  +  -  -  -
 * S  +  -  +  -  -
 * X  +  +  +  +  +
 * AI -  -  -  -  +
 * See lock_mode_stronger_or_eq().
 */

/* Define the stronger-or-equal lock relation in a ulint.  This relation
contains all pairs LK(mode1, mode2) where mode1 is stronger than or
equal to mode2. */
#define LOCK_MODE_STRONGER_OR_EQ 0					\
 | LK(LOCK_IS, LOCK_IS)							\
 | LK(LOCK_IX, LOCK_IS) | LK(LOCK_IX, LOCK_IX)				\
 | LK(LOCK_S, LOCK_IS) | LK(LOCK_S, LOCK_S)				\
 | LK(LOCK_AUTO_INC, LOCK_AUTO_INC)					\
 | LK(LOCK_X, LOCK_IS) | LK(LOCK_X, LOCK_IX) | LK(LOCK_X, LOCK_S)	\
 | LK(LOCK_X, LOCK_AUTO_INC) | LK(LOCK_X, LOCK_X)

#ifdef UNIV_DEBUG
UNIV_INTERN ibool	lock_print_waits	= FALSE;

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
347 348 349
/*********************************************************************//**
Validates the lock system.
@return	TRUE if ok */
350 351 352 353 354
static
ibool
lock_validate(void);
/*===============*/

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
355 356 357
/*********************************************************************//**
Validates the record lock queues on a page.
@return	TRUE if ok */
358 359 360 361
static
ibool
lock_rec_validate_page(
/*===================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
362
	ulint	space,	/*!< in: space id */
363 364
	ulint	zip_size,/*!< in: compressed page size in bytes
			or 0 for uncompressed pages */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
365
	ulint	page_no);/*!< in: page number */
366 367 368 369 370 371 372 373 374 375 376 377 378
#endif /* UNIV_DEBUG */

/* The lock system */
UNIV_INTERN lock_sys_t*	lock_sys	= NULL;

/* We store info on the latest deadlock error to this buffer. InnoDB
Monitor will then fetch it and print */
UNIV_INTERN ibool	lock_deadlock_found = FALSE;
UNIV_INTERN FILE*	lock_latest_err_file;

/* Flags for recursive deadlock search */
#define LOCK_VICTIM_IS_START	1
#define LOCK_VICTIM_IS_OTHER	2
379
#define LOCK_EXCEED_MAX_DEPTH	3
380

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
381 382 383 384 385
/********************************************************************//**
Checks if a lock request results in a deadlock.
@return TRUE if a deadlock was detected and we chose trx as a victim;
FALSE if no deadlock, or there was a deadlock, but we chose other
transaction(s) as victim(s) */
386 387 388 389
static
ibool
lock_deadlock_occurs(
/*=================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
390 391 392 393 394 395 396 397
	lock_t*	lock,	/*!< in: lock the transaction is requesting */
	trx_t*	trx);	/*!< in: transaction */
/********************************************************************//**
Looks recursively for a deadlock.
@return 0 if no deadlock found, LOCK_VICTIM_IS_START if there was a
deadlock and we chose 'start' as the victim, LOCK_VICTIM_IS_OTHER if a
deadlock was found and we chose some other trx as a victim: we must do
the search again in this last case because there may be another
398 399
deadlock!
LOCK_EXCEED_MAX_DEPTH if the lock search exceeds max steps or max depth. */
400 401 402 403
static
ulint
lock_deadlock_recursive(
/*====================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
404 405
	trx_t*	start,		/*!< in: recursion starting point */
	trx_t*	trx,		/*!< in: a transaction waiting for a lock */
406
	lock_t*	wait_lock,	/*!< in:  lock that is waiting to be granted */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
407
	ulint*	cost,		/*!< in/out: number of calculation steps thus
408
				far: if this exceeds LOCK_MAX_N_STEPS_...
409
				we return LOCK_EXCEED_MAX_DEPTH */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
410
	ulint	depth);		/*!< in: recursion depth: if this exceeds
411
				LOCK_MAX_DEPTH_IN_DEADLOCK_CHECK, we
412
				return LOCK_EXCEED_MAX_DEPTH */
413

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
414 415
/*********************************************************************//**
Gets the nth bit of a record lock.
416
@return	TRUE if bit set also if i == ULINT_UNDEFINED return FALSE*/
417 418 419 420
UNIV_INLINE
ibool
lock_rec_get_nth_bit(
/*=================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
421 422
	const lock_t*	lock,	/*!< in: record lock */
	ulint		i)	/*!< in: index of the bit */
423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445
{
	ulint	byte_index;
	ulint	bit_index;

	ut_ad(lock);
	ut_ad(lock_get_type_low(lock) == LOCK_REC);

	if (i >= lock->un_member.rec_lock.n_bits) {

		return(FALSE);
	}

	byte_index = i / 8;
	bit_index = i % 8;

	return(1 & ((const byte*) &lock[1])[byte_index] >> bit_index);
}

/*************************************************************************/

#define lock_mutex_enter_kernel()	mutex_enter(&kernel_mutex)
#define lock_mutex_exit_kernel()	mutex_exit(&kernel_mutex)

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
446 447 448
/*********************************************************************//**
Checks that a transaction id is sensible, i.e., not in the future.
@return	TRUE if ok */
449 450 451 452
UNIV_INTERN
ibool
lock_check_trx_id_sanity(
/*=====================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
453 454 455 456 457
	trx_id_t	trx_id,		/*!< in: trx id */
	const rec_t*	rec,		/*!< in: user record */
	dict_index_t*	index,		/*!< in: index */
	const ulint*	offsets,	/*!< in: rec_get_offsets(rec, index) */
	ibool		has_kernel_mutex)/*!< in: TRUE if the caller owns the
458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496
					kernel mutex */
{
	ibool	is_ok		= TRUE;

	ut_ad(rec_offs_validate(rec, index, offsets));

	if (!has_kernel_mutex) {
		mutex_enter(&kernel_mutex);
	}

	/* A sanity check: the trx_id in rec must be smaller than the global
	trx id counter */

	if (ut_dulint_cmp(trx_id, trx_sys->max_trx_id) >= 0) {
		ut_print_timestamp(stderr);
		fputs("  InnoDB: Error: transaction id associated"
		      " with record\n",
		      stderr);
		rec_print_new(stderr, rec, offsets);
		fputs("InnoDB: in ", stderr);
		dict_index_name_print(stderr, NULL, index);
		fprintf(stderr, "\n"
			"InnoDB: is " TRX_ID_FMT " which is higher than the"
			" global trx id counter " TRX_ID_FMT "!\n"
			"InnoDB: The table is corrupt. You have to do"
			" dump + drop + reimport.\n",
			TRX_ID_PREP_PRINTF(trx_id),
			TRX_ID_PREP_PRINTF(trx_sys->max_trx_id));

		is_ok = FALSE;
	}

	if (!has_kernel_mutex) {
		mutex_exit(&kernel_mutex);
	}

	return(is_ok);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
497 498 499 500
/*********************************************************************//**
Checks that a record is seen in a consistent read.
@return TRUE if sees, or FALSE if an earlier version of the record
should be retrieved */
501 502 503 504
UNIV_INTERN
ibool
lock_clust_rec_cons_read_sees(
/*==========================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
505
	const rec_t*	rec,	/*!< in: user record which should be read or
506
				passed over by a read cursor */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
507 508 509
	dict_index_t*	index,	/*!< in: clustered index */
	const ulint*	offsets,/*!< in: rec_get_offsets(rec, index) */
	read_view_t*	view)	/*!< in: consistent read view */
510
{
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
511
	trx_id_t	trx_id;
512 513 514 515 516 517 518 519 520 521 522 523 524 525

	ut_ad(dict_index_is_clust(index));
	ut_ad(page_rec_is_user_rec(rec));
	ut_ad(rec_offs_validate(rec, index, offsets));

	/* NOTE that we call this function while holding the search
	system latch. To obey the latching order we must NOT reserve the
	kernel mutex here! */

	trx_id = row_get_rec_trx_id(rec, index, offsets);

	return(read_view_sees_trx_id(view, trx_id));
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
526 527 528 529 530 531 532 533 534 535
/*********************************************************************//**
Checks that a non-clustered index record is seen in a consistent read.

NOTE that a non-clustered index page contains so little information on
its modifications that also in the case FALSE, the present version of
rec may be the right, but we must check this from the clustered index
record.

@return TRUE if certainly sees, or FALSE if an earlier version of the
clustered index record might be needed */
536 537 538 539
UNIV_INTERN
ulint
lock_sec_rec_cons_read_sees(
/*========================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
540
	const rec_t*		rec,	/*!< in: user record which
541 542
					should be read or passed over
					by a read cursor */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
543
	const read_view_t*	view)	/*!< in: consistent read view */
544
{
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
545
	trx_id_t	max_trx_id;
546 547 548 549 550 551 552 553 554 555 556 557 558

	ut_ad(page_rec_is_user_rec(rec));

	/* NOTE that we might call this function while holding the search
	system latch. To obey the latching order we must NOT reserve the
	kernel mutex here! */

	if (recv_recovery_is_on()) {

		return(FALSE);
	}

	max_trx_id = page_get_max_trx_id(page_align(rec));
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
559
	ut_ad(!ut_dulint_is_zero(max_trx_id));
560 561 562 563

	return(ut_dulint_cmp(max_trx_id, view->up_limit_id) < 0);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
564
/*********************************************************************//**
565 566 567 568 569
Creates the lock system at database start. */
UNIV_INTERN
void
lock_sys_create(
/*============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
570
	ulint	n_cells)	/*!< in: number of slots in lock hash table */
571 572 573 574 575 576 577 578 579 580 581
{
	lock_sys = mem_alloc(sizeof(lock_sys_t));

	lock_sys->rec_hash = hash_create(n_cells);

	/* hash_create_mutexes(lock_sys->rec_hash, 2, SYNC_REC_LOCK); */

	lock_latest_err_file = os_file_create_tmpfile();
	ut_a(lock_latest_err_file);
}

582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598
/*********************************************************************//**
Closes the lock system at database shutdown. */
UNIV_INTERN
void
lock_sys_close(void)
/*================*/
{
	if (lock_latest_err_file != NULL) {
		fclose(lock_latest_err_file);
		lock_latest_err_file = NULL;
	}

	hash_table_free(lock_sys->rec_hash);
	mem_free(lock_sys);
	lock_sys = NULL;
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
599 600 601
/*********************************************************************//**
Gets the size of a lock struct.
@return	size in bytes */
602 603 604 605 606 607 608 609
UNIV_INTERN
ulint
lock_get_size(void)
/*===============*/
{
	return((ulint)sizeof(lock_t));
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
610 611 612
/*********************************************************************//**
Gets the mode of a lock.
@return	mode */
613 614 615 616
UNIV_INLINE
enum lock_mode
lock_get_mode(
/*==========*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
617
	const lock_t*	lock)	/*!< in: lock */
618 619 620 621 622 623
{
	ut_ad(lock);

	return(lock->type_mode & LOCK_MODE_MASK);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
624 625 626
/*********************************************************************//**
Gets the wait flag of a lock.
@return	TRUE if waiting */
627 628 629 630
UNIV_INLINE
ibool
lock_get_wait(
/*==========*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
631
	const lock_t*	lock)	/*!< in: lock */
632 633 634 635 636 637 638 639 640 641 642
{
	ut_ad(lock);

	if (UNIV_UNLIKELY(lock->type_mode & LOCK_WAIT)) {

		return(TRUE);
	}

	return(FALSE);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
643
/*********************************************************************//**
644
Gets the source table of an ALTER TABLE transaction.  The table must be
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
645 646 647 648 649
covered by an IX or IS table lock.
@return the source table of transaction, if it is covered by an IX or
IS table lock; dest if there is no source table, and NULL if the
transaction is locking more than two tables or an inconsistency is
found */
650 651 652 653
UNIV_INTERN
dict_table_t*
lock_get_src_table(
/*===============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
654 655 656
	trx_t*		trx,	/*!< in: transaction */
	dict_table_t*	dest,	/*!< in: destination of ALTER TABLE */
	enum lock_mode*	mode)	/*!< out: lock mode of the source table */
657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711
{
	dict_table_t*	src;
	lock_t*		lock;

	src = NULL;
	*mode = LOCK_NONE;

	for (lock = UT_LIST_GET_FIRST(trx->trx_locks);
	     lock;
	     lock = UT_LIST_GET_NEXT(trx_locks, lock)) {
		lock_table_t*	tab_lock;
		enum lock_mode	lock_mode;
		if (!(lock_get_type_low(lock) & LOCK_TABLE)) {
			/* We are only interested in table locks. */
			continue;
		}
		tab_lock = &lock->un_member.tab_lock;
		if (dest == tab_lock->table) {
			/* We are not interested in the destination table. */
			continue;
		} else if (!src) {
			/* This presumably is the source table. */
			src = tab_lock->table;
			if (UT_LIST_GET_LEN(src->locks) != 1
			    || UT_LIST_GET_FIRST(src->locks) != lock) {
				/* We only support the case when
				there is only one lock on this table. */
				return(NULL);
			}
		} else if (src != tab_lock->table) {
			/* The transaction is locking more than
			two tables (src and dest): abort */
			return(NULL);
		}

		/* Check that the source table is locked by
		LOCK_IX or LOCK_IS. */
		lock_mode = lock_get_mode(lock);
		if (lock_mode == LOCK_IX || lock_mode == LOCK_IS) {
			if (*mode != LOCK_NONE && *mode != lock_mode) {
				/* There are multiple locks on src. */
				return(NULL);
			}
			*mode = lock_mode;
		}
	}

	if (!src) {
		/* No source table lock found: flag the situation to caller */
		src = dest;
	}

	return(src);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
712
/*********************************************************************//**
713 714
Determine if the given table is exclusively "owned" by the given
transaction, i.e., transaction holds LOCK_IX and possibly LOCK_AUTO_INC
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
715 716 717
on the table.
@return TRUE if table is only locked by trx, with LOCK_IX, and
possibly LOCK_AUTO_INC */
718 719 720 721
UNIV_INTERN
ibool
lock_is_table_exclusive(
/*====================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
722 723
	dict_table_t*	table,	/*!< in: table */
	trx_t*		trx)	/*!< in: transaction */
724 725 726 727
{
	const lock_t*	lock;
	ibool		ok	= FALSE;

Vadim Tkachenko's avatar
Vadim Tkachenko committed
728 729 730 731
	ut_ad(table);
	ut_ad(trx);

	lock_mutex_enter_kernel();
732 733 734 735 736 737 738

	for (lock = UT_LIST_GET_FIRST(table->locks);
	     lock;
	     lock = UT_LIST_GET_NEXT(locks, &lock->un_member.tab_lock)) {
		if (lock->trx != trx) {
			/* A lock on the table is held
			by some other transaction. */
Vadim Tkachenko's avatar
Vadim Tkachenko committed
739
			goto not_ok;
740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755
		}

		if (!(lock_get_type_low(lock) & LOCK_TABLE)) {
			/* We are interested in table locks only. */
			continue;
		}

		switch (lock_get_mode(lock)) {
		case LOCK_IX:
			ok = TRUE;
			break;
		case LOCK_AUTO_INC:
			/* It is allowed for trx to hold an
			auto_increment lock. */
			break;
		default:
Vadim Tkachenko's avatar
Vadim Tkachenko committed
756
not_ok:
757
			/* Other table locks than LOCK_IX are not allowed. */
Vadim Tkachenko's avatar
Vadim Tkachenko committed
758 759
			ok = FALSE;
			goto func_exit;
760 761 762
		}
	}

Vadim Tkachenko's avatar
Vadim Tkachenko committed
763 764 765
func_exit:
	lock_mutex_exit_kernel();

766 767 768
	return(ok);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
769
/*********************************************************************//**
770 771 772 773 774
Sets the wait flag of a lock and the back pointer in trx to lock. */
UNIV_INLINE
void
lock_set_lock_and_trx_wait(
/*=======================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
775 776
	lock_t*	lock,	/*!< in: lock */
	trx_t*	trx)	/*!< in: trx */
777 778 779 780 781 782 783 784
{
	ut_ad(lock);
	ut_ad(trx->wait_lock == NULL);

	trx->wait_lock = lock;
	lock->type_mode |= LOCK_WAIT;
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
785
/**********************************************************************//**
786 787 788 789 790 791
The back pointer to a waiting lock request in the transaction is set to NULL
and the wait bit in lock type_mode is reset. */
UNIV_INLINE
void
lock_reset_lock_and_trx_wait(
/*=========================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
792
	lock_t*	lock)	/*!< in: record lock */
793 794 795 796 797 798 799 800 801 802
{
	ut_ad((lock->trx)->wait_lock == lock);
	ut_ad(lock_get_wait(lock));

	/* Reset the back pointer in trx to this waiting lock request */

	(lock->trx)->wait_lock = NULL;
	lock->type_mode &= ~LOCK_WAIT;
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
803 804 805
/*********************************************************************//**
Gets the gap flag of a record lock.
@return	TRUE if gap flag set */
806 807 808 809
UNIV_INLINE
ibool
lock_rec_get_gap(
/*=============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
810
	const lock_t*	lock)	/*!< in: record lock */
811 812 813 814 815 816 817 818 819 820 821 822
{
	ut_ad(lock);
	ut_ad(lock_get_type_low(lock) == LOCK_REC);

	if (lock->type_mode & LOCK_GAP) {

		return(TRUE);
	}

	return(FALSE);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
823 824 825
/*********************************************************************//**
Gets the LOCK_REC_NOT_GAP flag of a record lock.
@return	TRUE if LOCK_REC_NOT_GAP flag set */
826 827 828 829
UNIV_INLINE
ibool
lock_rec_get_rec_not_gap(
/*=====================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
830
	const lock_t*	lock)	/*!< in: record lock */
831 832 833 834 835 836 837 838 839 840 841 842
{
	ut_ad(lock);
	ut_ad(lock_get_type_low(lock) == LOCK_REC);

	if (lock->type_mode & LOCK_REC_NOT_GAP) {

		return(TRUE);
	}

	return(FALSE);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
843 844 845
/*********************************************************************//**
Gets the waiting insert flag of a record lock.
@return	TRUE if gap flag set */
846 847 848 849
UNIV_INLINE
ibool
lock_rec_get_insert_intention(
/*==========================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
850
	const lock_t*	lock)	/*!< in: record lock */
851 852 853 854 855 856 857 858 859 860 861 862
{
	ut_ad(lock);
	ut_ad(lock_get_type_low(lock) == LOCK_REC);

	if (lock->type_mode & LOCK_INSERT_INTENTION) {

		return(TRUE);
	}

	return(FALSE);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
863 864 865
/*********************************************************************//**
Calculates if lock mode 1 is stronger or equal to lock mode 2.
@return	nonzero if mode1 stronger or equal to mode2 */
866 867 868 869
UNIV_INLINE
ulint
lock_mode_stronger_or_eq(
/*=====================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
870 871
	enum lock_mode	mode1,	/*!< in: lock mode */
	enum lock_mode	mode2)	/*!< in: lock mode */
872 873 874 875 876 877 878 879 880
{
	ut_ad(mode1 == LOCK_X || mode1 == LOCK_S || mode1 == LOCK_IX
	      || mode1 == LOCK_IS || mode1 == LOCK_AUTO_INC);
	ut_ad(mode2 == LOCK_X || mode2 == LOCK_S || mode2 == LOCK_IX
	      || mode2 == LOCK_IS || mode2 == LOCK_AUTO_INC);

	return((LOCK_MODE_STRONGER_OR_EQ) & LK(mode1, mode2));
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
881 882 883
/*********************************************************************//**
Calculates if lock mode 1 is compatible with lock mode 2.
@return	nonzero if mode1 compatible with mode2 */
884 885 886 887
UNIV_INLINE
ulint
lock_mode_compatible(
/*=================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
888 889
	enum lock_mode	mode1,	/*!< in: lock mode */
	enum lock_mode	mode2)	/*!< in: lock mode */
890 891 892 893 894 895 896 897 898
{
	ut_ad(mode1 == LOCK_X || mode1 == LOCK_S || mode1 == LOCK_IX
	      || mode1 == LOCK_IS || mode1 == LOCK_AUTO_INC);
	ut_ad(mode2 == LOCK_X || mode2 == LOCK_S || mode2 == LOCK_IX
	      || mode2 == LOCK_IS || mode2 == LOCK_AUTO_INC);

	return((LOCK_MODE_COMPATIBILITY) & LK(mode1, mode2));
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
899 900 901
/*********************************************************************//**
Checks if a lock request for a new lock has to wait for request lock2.
@return	TRUE if new lock has to wait for lock2 to be removed */
902 903 904 905
UNIV_INLINE
ibool
lock_rec_has_to_wait(
/*=================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
906 907
	const trx_t*	trx,	/*!< in: trx of new lock */
	ulint		type_mode,/*!< in: precise mode of the new lock
908 909 910
				to set: LOCK_S or LOCK_X, possibly
				ORed to LOCK_GAP or LOCK_REC_NOT_GAP,
				LOCK_INSERT_INTENTION */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
911
	const lock_t*	lock2,	/*!< in: another record lock; NOTE that
912 913 914
				it is assumed that this has a lock bit
				set on the same record as in the new
				lock we are setting */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
915
	ibool lock_is_on_supremum)  /*!< in: TRUE if we are setting the
916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980
				lock on the 'supremum' record of an
				index page: we know then that the lock
				request is really for a 'gap' type lock */
{
	ut_ad(trx && lock2);
	ut_ad(lock_get_type_low(lock2) == LOCK_REC);

	if (trx != lock2->trx
	    && !lock_mode_compatible(LOCK_MODE_MASK & type_mode,
				     lock_get_mode(lock2))) {

		/* We have somewhat complex rules when gap type record locks
		cause waits */

		if ((lock_is_on_supremum || (type_mode & LOCK_GAP))
		    && !(type_mode & LOCK_INSERT_INTENTION)) {

			/* Gap type locks without LOCK_INSERT_INTENTION flag
			do not need to wait for anything. This is because
			different users can have conflicting lock types
			on gaps. */

			return(FALSE);
		}

		if (!(type_mode & LOCK_INSERT_INTENTION)
		    && lock_rec_get_gap(lock2)) {

			/* Record lock (LOCK_ORDINARY or LOCK_REC_NOT_GAP
			does not need to wait for a gap type lock */

			return(FALSE);
		}

		if ((type_mode & LOCK_GAP)
		    && lock_rec_get_rec_not_gap(lock2)) {

			/* Lock on gap does not need to wait for
			a LOCK_REC_NOT_GAP type lock */

			return(FALSE);
		}

		if (lock_rec_get_insert_intention(lock2)) {

			/* No lock request needs to wait for an insert
			intention lock to be removed. This is ok since our
			rules allow conflicting locks on gaps. This eliminates
			a spurious deadlock caused by a next-key lock waiting
			for an insert intention lock; when the insert
			intention lock was granted, the insert deadlocked on
			the waiting next-key lock.

			Also, insert intention locks do not disturb each
			other. */

			return(FALSE);
		}

		return(TRUE);
	}

	return(FALSE);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
981 982 983
/*********************************************************************//**
Checks if a lock request lock1 has to wait for request lock2.
@return	TRUE if lock1 has to wait for lock2 to be removed */
984 985 986 987
UNIV_INTERN
ibool
lock_has_to_wait(
/*=============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
988 989
	const lock_t*	lock1,	/*!< in: waiting lock */
	const lock_t*	lock2)	/*!< in: another lock; NOTE that it is
990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018
				assumed that this has a lock bit set
				on the same record as in lock1 if the
				locks are record locks */
{
	ut_ad(lock1 && lock2);

	if (lock1->trx != lock2->trx
	    && !lock_mode_compatible(lock_get_mode(lock1),
				     lock_get_mode(lock2))) {
		if (lock_get_type_low(lock1) == LOCK_REC) {
			ut_ad(lock_get_type_low(lock2) == LOCK_REC);

			/* If this lock request is for a supremum record
			then the second bit on the lock bitmap is set */

			return(lock_rec_has_to_wait(lock1->trx,
						    lock1->type_mode, lock2,
						    lock_rec_get_nth_bit(
							    lock1, 1)));
		}

		return(TRUE);
	}

	return(FALSE);
}

/*============== RECORD LOCK BASIC FUNCTIONS ============================*/

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1019 1020 1021
/*********************************************************************//**
Gets the number of bits in a record lock bitmap.
@return	number of bits */
1022 1023 1024 1025
UNIV_INLINE
ulint
lock_rec_get_n_bits(
/*================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1026
	const lock_t*	lock)	/*!< in: record lock */
1027 1028 1029 1030
{
	return(lock->un_member.rec_lock.n_bits);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1031
/**********************************************************************//**
1032 1033 1034 1035 1036
Sets the nth bit of a record lock to TRUE. */
UNIV_INLINE
void
lock_rec_set_nth_bit(
/*=================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1037 1038
	lock_t*	lock,	/*!< in: record lock */
	ulint	i)	/*!< in: index of the bit */
1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052
{
	ulint	byte_index;
	ulint	bit_index;

	ut_ad(lock);
	ut_ad(lock_get_type_low(lock) == LOCK_REC);
	ut_ad(i < lock->un_member.rec_lock.n_bits);

	byte_index = i / 8;
	bit_index = i % 8;

	((byte*) &lock[1])[byte_index] |= 1 << bit_index;
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1053
/**********************************************************************//**
1054
Looks for a set bit in a record lock bitmap. Returns ULINT_UNDEFINED,
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1055 1056 1057
if none found.
@return bit index == heap number of the record, or ULINT_UNDEFINED if
none found */
1058 1059 1060 1061
UNIV_INTERN
ulint
lock_rec_find_set_bit(
/*==================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1062
	const lock_t*	lock)	/*!< in: record lock with at least one bit set */
1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076
{
	ulint	i;

	for (i = 0; i < lock_rec_get_n_bits(lock); i++) {

		if (lock_rec_get_nth_bit(lock, i)) {

			return(i);
		}
	}

	return(ULINT_UNDEFINED);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1077
/**********************************************************************//**
1078 1079 1080 1081 1082
Resets the nth bit of a record lock. */
UNIV_INLINE
void
lock_rec_reset_nth_bit(
/*===================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1083 1084
	lock_t*	lock,	/*!< in: record lock */
	ulint	i)	/*!< in: index of the bit which must be set to TRUE
1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099
			when this function is called */
{
	ulint	byte_index;
	ulint	bit_index;

	ut_ad(lock);
	ut_ad(lock_get_type_low(lock) == LOCK_REC);
	ut_ad(i < lock->un_member.rec_lock.n_bits);

	byte_index = i / 8;
	bit_index = i % 8;

	((byte*) &lock[1])[byte_index] &= ~(1 << bit_index);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1100 1101 1102
/*********************************************************************//**
Gets the first or next record lock on a page.
@return	next lock, NULL if none exists */
1103 1104 1105 1106
UNIV_INLINE
lock_t*
lock_rec_get_next_on_page(
/*======================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1107
	lock_t*	lock)	/*!< in: a record lock */
1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135
{
	ulint	space;
	ulint	page_no;

	ut_ad(mutex_own(&kernel_mutex));
	ut_ad(lock_get_type_low(lock) == LOCK_REC);

	space = lock->un_member.rec_lock.space;
	page_no = lock->un_member.rec_lock.page_no;

	for (;;) {
		lock = HASH_GET_NEXT(hash, lock);

		if (!lock) {

			break;
		}

		if ((lock->un_member.rec_lock.space == space)
		    && (lock->un_member.rec_lock.page_no == page_no)) {

			break;
		}
	}

	return(lock);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1136
/*********************************************************************//**
1137
Gets the first record lock on a page, where the page is identified by its
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1138 1139
file address.
@return	first lock, NULL if none exists */
1140 1141 1142 1143
UNIV_INLINE
lock_t*
lock_rec_get_first_on_page_addr(
/*============================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1144 1145
	ulint	space,	/*!< in: space */
	ulint	page_no)/*!< in: page number */
1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165
{
	lock_t*	lock;

	ut_ad(mutex_own(&kernel_mutex));

	lock = HASH_GET_FIRST(lock_sys->rec_hash,
			      lock_rec_hash(space, page_no));
	while (lock) {
		if ((lock->un_member.rec_lock.space == space)
		    && (lock->un_member.rec_lock.page_no == page_no)) {

			break;
		}

		lock = HASH_GET_NEXT(hash, lock);
	}

	return(lock);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1166 1167 1168
/*********************************************************************//**
Returns TRUE if there are explicit record locks on a page.
@return	TRUE if there are explicit record locks on the page */
1169 1170 1171 1172
UNIV_INTERN
ibool
lock_rec_expl_exist_on_page(
/*========================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1173 1174
	ulint	space,	/*!< in: space id */
	ulint	page_no)/*!< in: page number */
1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190
{
	ibool	ret;

	mutex_enter(&kernel_mutex);

	if (lock_rec_get_first_on_page_addr(space, page_no)) {
		ret = TRUE;
	} else {
		ret = FALSE;
	}

	mutex_exit(&kernel_mutex);

	return(ret);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1191
/*********************************************************************//**
1192
Gets the first record lock on a page, where the page is identified by a
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1193 1194
pointer to it.
@return	first lock, NULL if none exists */
1195 1196 1197 1198
UNIV_INLINE
lock_t*
lock_rec_get_first_on_page(
/*=======================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1199
	const buf_block_t*	block)	/*!< in: buffer block */
1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224
{
	ulint	hash;
	lock_t*	lock;
	ulint	space	= buf_block_get_space(block);
	ulint	page_no	= buf_block_get_page_no(block);

	ut_ad(mutex_own(&kernel_mutex));

	hash = buf_block_get_lock_hash_val(block);

	lock = HASH_GET_FIRST(lock_sys->rec_hash, hash);

	while (lock) {
		if ((lock->un_member.rec_lock.space == space)
		    && (lock->un_member.rec_lock.page_no == page_no)) {

			break;
		}

		lock = HASH_GET_NEXT(hash, lock);
	}

	return(lock);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1225 1226
/*********************************************************************//**
Gets the next explicit lock request on a record.
1227
@return	next lock, NULL if none exists or if heap_no == ULINT_UNDEFINED */
1228 1229 1230 1231
UNIV_INLINE
lock_t*
lock_rec_get_next(
/*==============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1232 1233
	ulint	heap_no,/*!< in: heap number of the record */
	lock_t*	lock)	/*!< in: lock */
1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244
{
	ut_ad(mutex_own(&kernel_mutex));

	do {
		ut_ad(lock_get_type_low(lock) == LOCK_REC);
		lock = lock_rec_get_next_on_page(lock);
	} while (lock && !lock_rec_get_nth_bit(lock, heap_no));

	return(lock);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1245 1246 1247
/*********************************************************************//**
Gets the first explicit lock request on a record.
@return	first lock, NULL if none exists */
1248 1249 1250 1251
UNIV_INLINE
lock_t*
lock_rec_get_first(
/*===============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1252 1253
	const buf_block_t*	block,	/*!< in: block containing the record */
	ulint			heap_no)/*!< in: heap number of the record */
1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268
{
	lock_t*	lock;

	ut_ad(mutex_own(&kernel_mutex));

	for (lock = lock_rec_get_first_on_page(block); lock;
	     lock = lock_rec_get_next_on_page(lock)) {
		if (lock_rec_get_nth_bit(lock, heap_no)) {
			break;
		}
	}

	return(lock);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1269
/*********************************************************************//**
1270 1271 1272 1273 1274 1275 1276
Resets the record lock bitmap to zero. NOTE: does not touch the wait_lock
pointer in the transaction! This function is used in lock object creation
and resetting. */
static
void
lock_rec_bitmap_reset(
/*==================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1277
	lock_t*	lock)	/*!< in: record lock */
1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292
{
	ulint	n_bytes;

	ut_ad(lock_get_type_low(lock) == LOCK_REC);

	/* Reset to zero the bitmap which resides immediately after the lock
	struct */

	n_bytes = lock_rec_get_n_bits(lock) / 8;

	ut_ad((lock_rec_get_n_bits(lock) % 8) == 0);

	memset(&lock[1], 0, n_bytes);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1293 1294 1295
/*********************************************************************//**
Copies a record lock to heap.
@return	copy of lock */
1296 1297 1298 1299
static
lock_t*
lock_rec_copy(
/*==========*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1300 1301
	const lock_t*	lock,	/*!< in: record lock */
	mem_heap_t*	heap)	/*!< in: memory heap */
1302 1303 1304 1305 1306 1307 1308 1309 1310 1311
{
	ulint	size;

	ut_ad(lock_get_type_low(lock) == LOCK_REC);

	size = sizeof(lock_t) + lock_rec_get_n_bits(lock) / 8;

	return(mem_heap_dup(heap, lock, size));
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1312 1313 1314
/*********************************************************************//**
Gets the previous record lock set on a record.
@return	previous lock on the same record, NULL if none exists */
1315 1316 1317 1318
UNIV_INTERN
const lock_t*
lock_rec_get_prev(
/*==============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1319 1320
	const lock_t*	in_lock,/*!< in: record lock */
	ulint		heap_no)/*!< in: heap number of the record */
1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353
{
	lock_t*	lock;
	ulint	space;
	ulint	page_no;
	lock_t*	found_lock	= NULL;

	ut_ad(mutex_own(&kernel_mutex));
	ut_ad(lock_get_type_low(in_lock) == LOCK_REC);

	space = in_lock->un_member.rec_lock.space;
	page_no = in_lock->un_member.rec_lock.page_no;

	lock = lock_rec_get_first_on_page_addr(space, page_no);

	for (;;) {
		ut_ad(lock);

		if (lock == in_lock) {

			return(found_lock);
		}

		if (lock_rec_get_nth_bit(lock, heap_no)) {

			found_lock = lock;
		}

		lock = lock_rec_get_next_on_page(lock);
	}
}

/*============= FUNCTIONS FOR ANALYZING TABLE LOCK QUEUE ================*/

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1354 1355 1356
/*********************************************************************//**
Checks if a transaction has the specified table lock, or stronger.
@return	lock or NULL */
1357 1358 1359 1360
UNIV_INLINE
lock_t*
lock_table_has(
/*===========*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1361 1362 1363
	trx_t*		trx,	/*!< in: transaction */
	dict_table_t*	table,	/*!< in: table */
	enum lock_mode	mode)	/*!< in: lock mode */
1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393
{
	lock_t*	lock;

	ut_ad(mutex_own(&kernel_mutex));

	/* Look for stronger locks the same trx already has on the table */

	lock = UT_LIST_GET_LAST(table->locks);

	while (lock != NULL) {

		if (lock->trx == trx
		    && lock_mode_stronger_or_eq(lock_get_mode(lock), mode)) {

			/* The same trx already has locked the table in
			a mode stronger or equal to the mode given */

			ut_ad(!lock_get_wait(lock));

			return(lock);
		}

		lock = UT_LIST_GET_PREV(un_member.tab_lock.locks, lock);
	}

	return(NULL);
}

/*============= FUNCTIONS FOR ANALYZING RECORD LOCK QUEUE ================*/

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1394
/*********************************************************************//**
1395
Checks if a transaction has a GRANTED explicit lock on rec stronger or equal
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1396 1397
to precise_mode.
@return	lock or NULL */
1398 1399 1400 1401
UNIV_INLINE
lock_t*
lock_rec_has_expl(
/*==============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1402
	ulint			precise_mode,/*!< in: LOCK_S or LOCK_X
1403 1404 1405 1406
					possibly ORed to LOCK_GAP or
					LOCK_REC_NOT_GAP, for a
					supremum record we regard this
					always a gap type request */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1407
	const buf_block_t*	block,	/*!< in: buffer block containing
1408
					the record */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1409 1410
	ulint			heap_no,/*!< in: heap number of the record */
	trx_t*			trx)	/*!< in: transaction */
1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443
{
	lock_t*	lock;

	ut_ad(mutex_own(&kernel_mutex));
	ut_ad((precise_mode & LOCK_MODE_MASK) == LOCK_S
	      || (precise_mode & LOCK_MODE_MASK) == LOCK_X);
	ut_ad(!(precise_mode & LOCK_INSERT_INTENTION));

	lock = lock_rec_get_first(block, heap_no);

	while (lock) {
		if (lock->trx == trx
		    && lock_mode_stronger_or_eq(lock_get_mode(lock),
						precise_mode & LOCK_MODE_MASK)
		    && !lock_get_wait(lock)
		    && (!lock_rec_get_rec_not_gap(lock)
			|| (precise_mode & LOCK_REC_NOT_GAP)
			|| heap_no == PAGE_HEAP_NO_SUPREMUM)
		    && (!lock_rec_get_gap(lock)
			|| (precise_mode & LOCK_GAP)
			|| heap_no == PAGE_HEAP_NO_SUPREMUM)
		    && (!lock_rec_get_insert_intention(lock))) {

			return(lock);
		}

		lock = lock_rec_get_next(heap_no, lock);
	}

	return(NULL);
}

#ifdef UNIV_DEBUG
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1444 1445 1446
/*********************************************************************//**
Checks if some other transaction has a lock request in the queue.
@return	lock or NULL */
1447 1448 1449 1450
static
lock_t*
lock_rec_other_has_expl_req(
/*========================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1451 1452
	enum lock_mode		mode,	/*!< in: LOCK_S or LOCK_X */
	ulint			gap,	/*!< in: LOCK_GAP if also gap
1453 1454
					locks are taken into account,
					or 0 if not */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1455
	ulint			wait,	/*!< in: LOCK_WAIT if also
1456 1457
					waiting locks are taken into
					account, or 0 if not */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1458
	const buf_block_t*	block,	/*!< in: buffer block containing
1459
					the record */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1460 1461
	ulint			heap_no,/*!< in: heap number of the record */
	const trx_t*		trx)	/*!< in: transaction, or NULL if
1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491
					requests by all transactions
					are taken into account */
{
	lock_t*	lock;

	ut_ad(mutex_own(&kernel_mutex));
	ut_ad(mode == LOCK_X || mode == LOCK_S);
	ut_ad(gap == 0 || gap == LOCK_GAP);
	ut_ad(wait == 0 || wait == LOCK_WAIT);

	lock = lock_rec_get_first(block, heap_no);

	while (lock) {
		if (lock->trx != trx
		    && (gap
			|| !(lock_rec_get_gap(lock)
			     || heap_no == PAGE_HEAP_NO_SUPREMUM))
		    && (wait || !lock_get_wait(lock))
		    && lock_mode_stronger_or_eq(lock_get_mode(lock), mode)) {

			return(lock);
		}

		lock = lock_rec_get_next(heap_no, lock);
	}

	return(NULL);
}
#endif /* UNIV_DEBUG */

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1492
/*********************************************************************//**
1493
Checks if some other transaction has a conflicting explicit lock request
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1494 1495
in the queue, so that we have to wait.
@return	lock or NULL */
1496 1497 1498 1499
static
lock_t*
lock_rec_other_has_conflicting(
/*===========================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1500
	enum lock_mode		mode,	/*!< in: LOCK_S or LOCK_X,
1501 1502 1503
					possibly ORed to LOCK_GAP or
					LOC_REC_NOT_GAP,
					LOCK_INSERT_INTENTION */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1504
	const buf_block_t*	block,	/*!< in: buffer block containing
1505
					the record */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1506 1507
	ulint			heap_no,/*!< in: heap number of the record */
	trx_t*			trx)	/*!< in: our transaction */
1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541
{
	lock_t*	lock;

	ut_ad(mutex_own(&kernel_mutex));

	lock = lock_rec_get_first(block, heap_no);

	if (UNIV_LIKELY_NULL(lock)) {
		if (UNIV_UNLIKELY(heap_no == PAGE_HEAP_NO_SUPREMUM)) {

			do {
				if (lock_rec_has_to_wait(trx, mode, lock,
							 TRUE)) {
					return(lock);
				}

				lock = lock_rec_get_next(heap_no, lock);
			} while (lock);
		} else {

			do {
				if (lock_rec_has_to_wait(trx, mode, lock,
							 FALSE)) {
					return(lock);
				}

				lock = lock_rec_get_next(heap_no, lock);
			} while (lock);
		}
	}

	return(NULL);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1542
/*********************************************************************//**
1543 1544
Looks for a suitable type record lock struct by the same trx on the same page.
This can be used to save space when a new record lock should be set on a page:
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1545 1546
no new struct is needed, if a suitable old is found.
@return	lock or NULL */
1547 1548 1549 1550
UNIV_INLINE
lock_t*
lock_rec_find_similar_on_page(
/*==========================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1551 1552 1553 1554
	ulint		type_mode,	/*!< in: lock type_mode field */
	ulint		heap_no,	/*!< in: heap number of the record */
	lock_t*		lock,		/*!< in: lock_rec_get_first_on_page() */
	const trx_t*	trx)		/*!< in: transaction */
1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571
{
	ut_ad(mutex_own(&kernel_mutex));

	while (lock != NULL) {
		if (lock->trx == trx
		    && lock->type_mode == type_mode
		    && lock_rec_get_n_bits(lock) > heap_no) {

			return(lock);
		}

		lock = lock_rec_get_next_on_page(lock);
	}

	return(NULL);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1572
/*********************************************************************//**
1573
Checks if some transaction has an implicit x-lock on a record in a secondary
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1574 1575
index.
@return	transaction which has the x-lock, or NULL */
1576 1577 1578 1579
static
trx_t*
lock_sec_rec_some_has_impl_off_kernel(
/*==================================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1580 1581 1582
	const rec_t*	rec,	/*!< in: user record */
	dict_index_t*	index,	/*!< in: secondary index */
	const ulint*	offsets)/*!< in: rec_get_offsets(rec, index) */
1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618
{
	const page_t*	page = page_align(rec);

	ut_ad(mutex_own(&kernel_mutex));
	ut_ad(!dict_index_is_clust(index));
	ut_ad(page_rec_is_user_rec(rec));
	ut_ad(rec_offs_validate(rec, index, offsets));

	/* Some transaction may have an implicit x-lock on the record only
	if the max trx id for the page >= min trx id for the trx list, or
	database recovery is running. We do not write the changes of a page
	max trx id to the log, and therefore during recovery, this value
	for a page may be incorrect. */

	if (!(ut_dulint_cmp(page_get_max_trx_id(page),
			    trx_list_get_min_trx_id()) >= 0)
	    && !recv_recovery_is_on()) {

		return(NULL);
	}

	/* Ok, in this case it is possible that some transaction has an
	implicit x-lock. We have to look in the clustered index. */

	if (!lock_check_trx_id_sanity(page_get_max_trx_id(page),
				      rec, index, offsets, TRUE)) {
		buf_page_print(page, 0);

		/* The page is corrupt: try to avoid a crash by returning
		NULL */
		return(NULL);
	}

	return(row_vers_impl_x_locked_off_kernel(rec, index, offsets));
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1619
/*********************************************************************//**
1620 1621 1622 1623 1624 1625 1626
Return approximate number or record locks (bits set in the bitmap) for
this transaction. Since delete-marked records may be removed, the
record count will not be precise. */
UNIV_INTERN
ulint
lock_number_of_rows_locked(
/*=======================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1627
	trx_t*	trx)	/*!< in: transaction */
1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654
{
	lock_t*	lock;
	ulint   n_records = 0;
	ulint	n_bits;
	ulint	n_bit;

	lock = UT_LIST_GET_FIRST(trx->trx_locks);

	while (lock) {
		if (lock_get_type_low(lock) == LOCK_REC) {
			n_bits = lock_rec_get_n_bits(lock);

			for (n_bit = 0; n_bit < n_bits; n_bit++) {
				if (lock_rec_get_nth_bit(lock, n_bit)) {
					n_records++;
				}
			}
		}

		lock = UT_LIST_GET_NEXT(trx_locks, lock);
	}

	return (n_records);
}

/*============== RECORD LOCK CREATION AND QUEUE MANAGEMENT =============*/

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1655
/*********************************************************************//**
1656
Creates a new record lock and inserts it to the lock queue. Does NOT check
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1657 1658
for deadlocks or lock compatibility!
@return	created lock */
1659 1660 1661 1662
static
lock_t*
lock_rec_create(
/*============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1663
	ulint			type_mode,/*!< in: lock mode and wait
1664 1665
					flag, type is ignored and
					replaced by LOCK_REC */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1666
	const buf_block_t*	block,	/*!< in: buffer block containing
1667
					the record */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1668 1669 1670
	ulint			heap_no,/*!< in: heap number of the record */
	dict_index_t*		index,	/*!< in: index of record */
	trx_t*			trx)	/*!< in: transaction */
1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731
{
	lock_t*		lock;
	ulint		page_no;
	ulint		space;
	ulint		n_bits;
	ulint		n_bytes;
	const page_t*	page;

	ut_ad(mutex_own(&kernel_mutex));

	space = buf_block_get_space(block);
	page_no	= buf_block_get_page_no(block);
	page = block->frame;

	ut_ad(!!page_is_comp(page) == dict_table_is_comp(index->table));

	/* If rec is the supremum record, then we reset the gap and
	LOCK_REC_NOT_GAP bits, as all locks on the supremum are
	automatically of the gap type */

	if (UNIV_UNLIKELY(heap_no == PAGE_HEAP_NO_SUPREMUM)) {
		ut_ad(!(type_mode & LOCK_REC_NOT_GAP));

		type_mode = type_mode & ~(LOCK_GAP | LOCK_REC_NOT_GAP);
	}

	/* Make lock bitmap bigger by a safety margin */
	n_bits = page_dir_get_n_heap(page) + LOCK_PAGE_BITMAP_MARGIN;
	n_bytes = 1 + n_bits / 8;

	lock = mem_heap_alloc(trx->lock_heap, sizeof(lock_t) + n_bytes);

	UT_LIST_ADD_LAST(trx_locks, trx->trx_locks, lock);

	lock->trx = trx;

	lock->type_mode = (type_mode & ~LOCK_TYPE_MASK) | LOCK_REC;
	lock->index = index;

	lock->un_member.rec_lock.space = space;
	lock->un_member.rec_lock.page_no = page_no;
	lock->un_member.rec_lock.n_bits = n_bytes * 8;

	/* Reset to zero the bitmap which resides immediately after the
	lock struct */

	lock_rec_bitmap_reset(lock);

	/* Set the bit corresponding to rec */
	lock_rec_set_nth_bit(lock, heap_no);

	HASH_INSERT(lock_t, hash, lock_sys->rec_hash,
		    lock_rec_fold(space, page_no), lock);
	if (UNIV_UNLIKELY(type_mode & LOCK_WAIT)) {

		lock_set_lock_and_trx_wait(lock, trx);
	}

	return(lock);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1732
/*********************************************************************//**
1733
Enqueues a waiting request for a lock which cannot be granted immediately.
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1734 1735 1736 1737 1738
Checks for deadlocks.
@return DB_LOCK_WAIT, DB_DEADLOCK, or DB_QUE_THR_SUSPENDED, or
DB_SUCCESS; DB_SUCCESS means that there was a deadlock, but another
transaction was chosen as a victim, and we got the lock immediately:
no need to wait then */
1739 1740 1741 1742
static
ulint
lock_rec_enqueue_waiting(
/*=====================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1743
	ulint			type_mode,/*!< in: lock mode this
1744 1745 1746 1747 1748 1749 1750 1751
					transaction is requesting:
					LOCK_S or LOCK_X, possibly
					ORed with LOCK_GAP or
					LOCK_REC_NOT_GAP, ORed with
					LOCK_INSERT_INTENTION if this
					waiting lock request is set
					when performing an insert of
					an index record */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1752
	const buf_block_t*	block,	/*!< in: buffer block containing
1753
					the record */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1754 1755 1756
	ulint			heap_no,/*!< in: heap number of the record */
	dict_index_t*		index,	/*!< in: index of record */
	que_thr_t*		thr)	/*!< in: query thread */
1757 1758 1759
{
	lock_t*	lock;
	trx_t*	trx;
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1760 1761
	ulint   sec;
	ulint   ms;
1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819

	ut_ad(mutex_own(&kernel_mutex));

	/* Test if there already is some other reason to suspend thread:
	we do not enqueue a lock request if the query thread should be
	stopped anyway */

	if (UNIV_UNLIKELY(que_thr_stop(thr))) {

		ut_error;

		return(DB_QUE_THR_SUSPENDED);
	}

	trx = thr_get_trx(thr);

	switch (trx_get_dict_operation(trx)) {
	case TRX_DICT_OP_NONE:
		break;
	case TRX_DICT_OP_TABLE:
	case TRX_DICT_OP_INDEX:
		ut_print_timestamp(stderr);
		fputs("  InnoDB: Error: a record lock wait happens"
		      " in a dictionary operation!\n"
		      "InnoDB: ", stderr);
		dict_index_name_print(stderr, trx, index);
		fputs(".\n"
		      "InnoDB: Submit a detailed bug report"
		      " to http://bugs.mysql.com\n",
		      stderr);
	}

	/* Enqueue the lock request that will wait to be granted */
	lock = lock_rec_create(type_mode | LOCK_WAIT,
			       block, heap_no, index, trx);

	/* Check if a deadlock occurs: if yes, remove the lock request and
	return an error code */

	if (UNIV_UNLIKELY(lock_deadlock_occurs(lock, trx))) {

		lock_reset_lock_and_trx_wait(lock);
		lock_rec_reset_nth_bit(lock, heap_no);

		return(DB_DEADLOCK);
	}

	/* If there was a deadlock but we chose another transaction as a
	victim, it is possible that we already have the lock now granted! */

	if (trx->wait_lock == NULL) {

		return(DB_SUCCESS);
	}

	trx->que_state = TRX_QUE_LOCK_WAIT;
	trx->was_chosen_as_deadlock_victim = FALSE;
	trx->wait_started = time(NULL);
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1820 1821 1822 1823
	if (innobase_get_slow_log() && trx->take_stats) {
		ut_usectime(&sec, &ms);
		trx->lock_que_wait_ustarted = (ib_uint64_t)sec * 1000000 + ms;
	}
1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837

	ut_a(que_thr_stop(thr));

#ifdef UNIV_DEBUG
	if (lock_print_waits) {
		fprintf(stderr, "Lock wait for trx %lu in index ",
			(ulong) ut_dulint_get_low(trx->id));
		ut_print_name(stderr, trx, FALSE, index->name);
	}
#endif /* UNIV_DEBUG */

	return(DB_LOCK_WAIT);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1838
/*********************************************************************//**
1839 1840 1841 1842 1843
Adds a record lock request in the record queue. The request is normally
added as the last in the queue, but if there are no waiting lock requests
on the record, and the request to be added is not a waiting request, we
can reuse a suitable record lock object already existing on the same page,
just setting the appropriate bit in its bitmap. This is a low-level function
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1844 1845
which does NOT check for deadlocks or lock compatibility!
@return	lock where the bit was set */
1846 1847 1848 1849
static
lock_t*
lock_rec_add_to_queue(
/*==================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1850
	ulint			type_mode,/*!< in: lock mode, wait, gap
1851 1852
					etc. flags; type is ignored
					and replaced by LOCK_REC */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1853
	const buf_block_t*	block,	/*!< in: buffer block containing
1854
					the record */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1855 1856 1857
	ulint			heap_no,/*!< in: heap number of the record */
	dict_index_t*		index,	/*!< in: index of record */
	trx_t*			trx)	/*!< in: transaction */
1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933
{
	lock_t*	lock;

	ut_ad(mutex_own(&kernel_mutex));
#ifdef UNIV_DEBUG
	switch (type_mode & LOCK_MODE_MASK) {
	case LOCK_X:
	case LOCK_S:
		break;
	default:
		ut_error;
	}

	if (!(type_mode & (LOCK_WAIT | LOCK_GAP))) {
		enum lock_mode	mode = (type_mode & LOCK_MODE_MASK) == LOCK_S
			? LOCK_X
			: LOCK_S;
		lock_t*		other_lock
			= lock_rec_other_has_expl_req(mode, 0, LOCK_WAIT,
						      block, heap_no, trx);
		ut_a(!other_lock);
	}
#endif /* UNIV_DEBUG */

	type_mode |= LOCK_REC;

	/* If rec is the supremum record, then we can reset the gap bit, as
	all locks on the supremum are automatically of the gap type, and we
	try to avoid unnecessary memory consumption of a new record lock
	struct for a gap type lock */

	if (UNIV_UNLIKELY(heap_no == PAGE_HEAP_NO_SUPREMUM)) {
		ut_ad(!(type_mode & LOCK_REC_NOT_GAP));

		/* There should never be LOCK_REC_NOT_GAP on a supremum
		record, but let us play safe */

		type_mode = type_mode & ~(LOCK_GAP | LOCK_REC_NOT_GAP);
	}

	/* Look for a waiting lock request on the same record or on a gap */

	lock = lock_rec_get_first_on_page(block);

	while (lock != NULL) {
		if (lock_get_wait(lock)
		    && (lock_rec_get_nth_bit(lock, heap_no))) {

			goto somebody_waits;
		}

		lock = lock_rec_get_next_on_page(lock);
	}

	if (UNIV_LIKELY(!(type_mode & LOCK_WAIT))) {

		/* Look for a similar record lock on the same page:
		if one is found and there are no waiting lock requests,
		we can just set the bit */

		lock = lock_rec_find_similar_on_page(
			type_mode, heap_no,
			lock_rec_get_first_on_page(block), trx);

		if (lock) {

			lock_rec_set_nth_bit(lock, heap_no);

			return(lock);
		}
	}

somebody_waits:
	return(lock_rec_create(type_mode, block, heap_no, index, trx));
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1934
/*********************************************************************//**
1935 1936 1937 1938 1939
This is a fast routine for locking a record in the most common cases:
there are no explicit locks on the page, or there is just one lock, owned
by this transaction, and of the right type_mode. This is a low-level function
which does NOT look at implicit locks! Checks lock compatibility within
explicit locks. This function sets a normal next-key lock, or in the case of
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1940 1941
a page supremum record, a gap type lock.
@return	TRUE if locking succeeded */
1942 1943 1944 1945
UNIV_INLINE
ibool
lock_rec_lock_fast(
/*===============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1946
	ibool			impl,	/*!< in: if TRUE, no lock is set
1947 1948 1949
					if no wait is necessary: we
					assume that the caller will
					set an implicit lock */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1950
	ulint			mode,	/*!< in: lock mode: LOCK_X or
1951 1952
					LOCK_S possibly ORed to either
					LOCK_GAP or LOCK_REC_NOT_GAP */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1953
	const buf_block_t*	block,	/*!< in: buffer block containing
1954
					the record */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
1955 1956 1957
	ulint			heap_no,/*!< in: heap number of record */
	dict_index_t*		index,	/*!< in: index of record */
	que_thr_t*		thr)	/*!< in: query thread */
1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008
{
	lock_t*	lock;
	trx_t*	trx;

	ut_ad(mutex_own(&kernel_mutex));
	ut_ad((LOCK_MODE_MASK & mode) != LOCK_S
	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IS));
	ut_ad((LOCK_MODE_MASK & mode) != LOCK_X
	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IX));
	ut_ad((LOCK_MODE_MASK & mode) == LOCK_S
	      || (LOCK_MODE_MASK & mode) == LOCK_X);
	ut_ad(mode - (LOCK_MODE_MASK & mode) == LOCK_GAP
	      || mode - (LOCK_MODE_MASK & mode) == 0
	      || mode - (LOCK_MODE_MASK & mode) == LOCK_REC_NOT_GAP);

	lock = lock_rec_get_first_on_page(block);

	trx = thr_get_trx(thr);

	if (lock == NULL) {
		if (!impl) {
			lock_rec_create(mode, block, heap_no, index, trx);
		}

		return(TRUE);
	}

	if (lock_rec_get_next_on_page(lock)) {

		return(FALSE);
	}

	if (lock->trx != trx
	    || lock->type_mode != (mode | LOCK_REC)
	    || lock_rec_get_n_bits(lock) <= heap_no) {

		return(FALSE);
	}

	if (!impl) {
		/* If the nth bit of the record lock is already set then we
		do not set a new lock bit, otherwise we do set */

		if (!lock_rec_get_nth_bit(lock, heap_no)) {
			lock_rec_set_nth_bit(lock, heap_no);
		}
	}

	return(TRUE);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2009
/*********************************************************************//**
2010 2011 2012
This is the general, and slower, routine for locking a record. This is a
low-level function which does NOT look at implicit locks! Checks lock
compatibility within explicit locks. This function sets a normal next-key
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2013 2014
lock, or in the case of a page supremum record, a gap type lock.
@return	DB_SUCCESS, DB_LOCK_WAIT, or error code */
2015 2016 2017 2018
static
ulint
lock_rec_lock_slow(
/*===============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2019
	ibool			impl,	/*!< in: if TRUE, no lock is set
2020 2021 2022
					if no wait is necessary: we
					assume that the caller will
					set an implicit lock */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2023
	ulint			mode,	/*!< in: lock mode: LOCK_X or
2024 2025
					LOCK_S possibly ORed to either
					LOCK_GAP or LOCK_REC_NOT_GAP */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2026
	const buf_block_t*	block,	/*!< in: buffer block containing
2027
					the record */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2028 2029 2030
	ulint			heap_no,/*!< in: heap number of record */
	dict_index_t*		index,	/*!< in: index of record */
	que_thr_t*		thr)	/*!< in: query thread */
2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074
{
	trx_t*	trx;
	ulint	err;

	ut_ad(mutex_own(&kernel_mutex));
	ut_ad((LOCK_MODE_MASK & mode) != LOCK_S
	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IS));
	ut_ad((LOCK_MODE_MASK & mode) != LOCK_X
	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IX));
	ut_ad((LOCK_MODE_MASK & mode) == LOCK_S
	      || (LOCK_MODE_MASK & mode) == LOCK_X);
	ut_ad(mode - (LOCK_MODE_MASK & mode) == LOCK_GAP
	      || mode - (LOCK_MODE_MASK & mode) == 0
	      || mode - (LOCK_MODE_MASK & mode) == LOCK_REC_NOT_GAP);

	trx = thr_get_trx(thr);

	if (lock_rec_has_expl(mode, block, heap_no, trx)) {
		/* The trx already has a strong enough lock on rec: do
		nothing */

		err = DB_SUCCESS;
	} else if (lock_rec_other_has_conflicting(mode, block, heap_no, trx)) {

		/* If another transaction has a non-gap conflicting request in
		the queue, as this transaction does not have a lock strong
		enough already granted on the record, we have to wait. */

		err = lock_rec_enqueue_waiting(mode, block, heap_no,
					       index, thr);
	} else {
		if (!impl) {
			/* Set the requested lock on the record */

			lock_rec_add_to_queue(LOCK_REC | mode, block,
					      heap_no, index, trx);
		}

		err = DB_SUCCESS;
	}

	return(err);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2075
/*********************************************************************//**
2076 2077 2078 2079
Tries to lock the specified record in the mode requested. If not immediately
possible, enqueues a waiting lock request. This is a low-level function
which does NOT look at implicit locks! Checks lock compatibility within
explicit locks. This function sets a normal next-key lock, or in the case
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2080 2081
of a page supremum record, a gap type lock.
@return	DB_SUCCESS, DB_LOCK_WAIT, or error code */
2082 2083 2084 2085
static
ulint
lock_rec_lock(
/*==========*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2086
	ibool			impl,	/*!< in: if TRUE, no lock is set
2087 2088 2089
					if no wait is necessary: we
					assume that the caller will
					set an implicit lock */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2090
	ulint			mode,	/*!< in: lock mode: LOCK_X or
2091 2092
					LOCK_S possibly ORed to either
					LOCK_GAP or LOCK_REC_NOT_GAP */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2093
	const buf_block_t*	block,	/*!< in: buffer block containing
2094
					the record */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2095 2096 2097
	ulint			heap_no,/*!< in: heap number of record */
	dict_index_t*		index,	/*!< in: index of record */
	que_thr_t*		thr)	/*!< in: query thread */
2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125
{
	ulint	err;

	ut_ad(mutex_own(&kernel_mutex));
	ut_ad((LOCK_MODE_MASK & mode) != LOCK_S
	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IS));
	ut_ad((LOCK_MODE_MASK & mode) != LOCK_X
	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IX));
	ut_ad((LOCK_MODE_MASK & mode) == LOCK_S
	      || (LOCK_MODE_MASK & mode) == LOCK_X);
	ut_ad(mode - (LOCK_MODE_MASK & mode) == LOCK_GAP
	      || mode - (LOCK_MODE_MASK & mode) == LOCK_REC_NOT_GAP
	      || mode - (LOCK_MODE_MASK & mode) == 0);

	if (lock_rec_lock_fast(impl, mode, block, heap_no, index, thr)) {

		/* We try a simplified and faster subroutine for the most
		common cases */

		err = DB_SUCCESS;
	} else {
		err = lock_rec_lock_slow(impl, mode, block,
					 heap_no, index, thr);
	}

	return(err);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2126 2127 2128
/*********************************************************************//**
Checks if a waiting record lock request still has to wait in a queue.
@return	TRUE if still has to wait */
2129 2130 2131 2132
static
ibool
lock_rec_has_to_wait_in_queue(
/*==========================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2133
	lock_t*	wait_lock)	/*!< in: waiting record lock */
2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163
{
	lock_t*	lock;
	ulint	space;
	ulint	page_no;
	ulint	heap_no;

	ut_ad(mutex_own(&kernel_mutex));
	ut_ad(lock_get_wait(wait_lock));
	ut_ad(lock_get_type_low(wait_lock) == LOCK_REC);

	space = wait_lock->un_member.rec_lock.space;
	page_no = wait_lock->un_member.rec_lock.page_no;
	heap_no = lock_rec_find_set_bit(wait_lock);

	lock = lock_rec_get_first_on_page_addr(space, page_no);

	while (lock != wait_lock) {

		if (lock_rec_get_nth_bit(lock, heap_no)
		    && lock_has_to_wait(wait_lock, lock)) {

			return(TRUE);
		}

		lock = lock_rec_get_next_on_page(lock);
	}

	return(FALSE);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2164
/*************************************************************//**
2165 2166 2167 2168 2169 2170
Grants a lock to a waiting lock request and releases the waiting
transaction. */
static
void
lock_grant(
/*=======*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2171
	lock_t*	lock)	/*!< in/out: waiting lock request */
2172 2173 2174 2175 2176 2177
{
	ut_ad(mutex_own(&kernel_mutex));

	lock_reset_lock_and_trx_wait(lock);

	if (lock_get_mode(lock) == LOCK_AUTO_INC) {
Vadim Tkachenko's avatar
Vadim Tkachenko committed
2178 2179
		trx_t*		trx = lock->trx;
		dict_table_t*	table = lock->un_member.tab_lock.table;
2180

Vadim Tkachenko's avatar
Vadim Tkachenko committed
2181
		if (table->autoinc_trx == trx) {
2182 2183 2184
			fprintf(stderr,
				"InnoDB: Error: trx already had"
				" an AUTO-INC lock!\n");
Vadim Tkachenko's avatar
Vadim Tkachenko committed
2185 2186
		} else {
			table->autoinc_trx = trx;
2187

Vadim Tkachenko's avatar
Vadim Tkachenko committed
2188 2189
			ib_vector_push(trx->autoinc_locks, lock);
		}
2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208
	}

#ifdef UNIV_DEBUG
	if (lock_print_waits) {
		fprintf(stderr, "Lock wait for trx %lu ends\n",
			(ulong) ut_dulint_get_low(lock->trx->id));
	}
#endif /* UNIV_DEBUG */

	/* If we are resolving a deadlock by choosing another transaction
	as a victim, then our original transaction may not be in the
	TRX_QUE_LOCK_WAIT state, and there is no need to end the lock wait
	for it */

	if (lock->trx->que_state == TRX_QUE_LOCK_WAIT) {
		trx_end_lock_wait(lock->trx);
	}
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2209
/*************************************************************//**
2210 2211 2212 2213 2214 2215 2216
Cancels a waiting record lock request and releases the waiting transaction
that requested it. NOTE: does NOT check if waiting lock requests behind this
one can now be granted! */
static
void
lock_rec_cancel(
/*============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2217
	lock_t*	lock)	/*!< in: waiting record lock request */
2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233
{
	ut_ad(mutex_own(&kernel_mutex));
	ut_ad(lock_get_type_low(lock) == LOCK_REC);

	/* Reset the bit (there can be only one set bit) in the lock bitmap */
	lock_rec_reset_nth_bit(lock, lock_rec_find_set_bit(lock));

	/* Reset the wait flag and the back pointer to lock in trx */

	lock_reset_lock_and_trx_wait(lock);

	/* The following function releases the trx from lock wait */

	trx_end_lock_wait(lock->trx);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2234
/*************************************************************//**
2235 2236 2237 2238 2239 2240 2241
Removes a record lock request, waiting or granted, from the queue and
grants locks to other transactions in the queue if they now are entitled
to a lock. NOTE: all record locks contained in in_lock are removed. */
static
void
lock_rec_dequeue_from_page(
/*=======================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2242
	lock_t*	in_lock)/*!< in: record lock object: all record locks which
2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281
			are contained in this lock object are removed;
			transactions waiting behind will get their lock
			requests granted, if they are now qualified to it */
{
	ulint	space;
	ulint	page_no;
	lock_t*	lock;
	trx_t*	trx;

	ut_ad(mutex_own(&kernel_mutex));
	ut_ad(lock_get_type_low(in_lock) == LOCK_REC);

	trx = in_lock->trx;

	space = in_lock->un_member.rec_lock.space;
	page_no = in_lock->un_member.rec_lock.page_no;

	HASH_DELETE(lock_t, hash, lock_sys->rec_hash,
		    lock_rec_fold(space, page_no), in_lock);

	UT_LIST_REMOVE(trx_locks, trx->trx_locks, in_lock);

	/* Check if waiting locks in the queue can now be granted: grant
	locks if there are no conflicting locks ahead. */

	lock = lock_rec_get_first_on_page_addr(space, page_no);

	while (lock != NULL) {
		if (lock_get_wait(lock)
		    && !lock_rec_has_to_wait_in_queue(lock)) {

			/* Grant the lock */
			lock_grant(lock);
		}

		lock = lock_rec_get_next_on_page(lock);
	}
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2282
/*************************************************************//**
2283 2284 2285 2286 2287
Removes a record lock request, waiting or granted, from the queue. */
static
void
lock_rec_discard(
/*=============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2288
	lock_t*	in_lock)/*!< in: record lock object: all record locks which
2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308
			are contained in this lock object are removed */
{
	ulint	space;
	ulint	page_no;
	trx_t*	trx;

	ut_ad(mutex_own(&kernel_mutex));
	ut_ad(lock_get_type_low(in_lock) == LOCK_REC);

	trx = in_lock->trx;

	space = in_lock->un_member.rec_lock.space;
	page_no = in_lock->un_member.rec_lock.page_no;

	HASH_DELETE(lock_t, hash, lock_sys->rec_hash,
		    lock_rec_fold(space, page_no), in_lock);

	UT_LIST_REMOVE(trx_locks, trx->trx_locks, in_lock);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2309
/*************************************************************//**
2310 2311 2312 2313 2314 2315 2316
Removes record lock objects set on an index page which is discarded. This
function does not move locks, or check for waiting locks, therefore the
lock bitmaps must already be reset when this function is called. */
static
void
lock_rec_free_all_from_discard_page(
/*================================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2317
	const buf_block_t*	block)	/*!< in: page to be discarded */
2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344
{
	ulint	space;
	ulint	page_no;
	lock_t*	lock;
	lock_t*	next_lock;

	ut_ad(mutex_own(&kernel_mutex));

	space = buf_block_get_space(block);
	page_no = buf_block_get_page_no(block);

	lock = lock_rec_get_first_on_page_addr(space, page_no);

	while (lock != NULL) {
		ut_ad(lock_rec_find_set_bit(lock) == ULINT_UNDEFINED);
		ut_ad(!lock_get_wait(lock));

		next_lock = lock_rec_get_next_on_page(lock);

		lock_rec_discard(lock);

		lock = next_lock;
	}
}

/*============= RECORD LOCK MOVING AND INHERITING ===================*/

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2345
/*************************************************************//**
2346 2347 2348 2349 2350 2351
Resets the lock bits for a single record. Releases transactions waiting for
lock requests here. */
static
void
lock_rec_reset_and_release_wait(
/*============================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2352
	const buf_block_t*	block,	/*!< in: buffer block containing
2353
					the record */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2354
	ulint			heap_no)/*!< in: heap number of record */
2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372
{
	lock_t*	lock;

	ut_ad(mutex_own(&kernel_mutex));

	lock = lock_rec_get_first(block, heap_no);

	while (lock != NULL) {
		if (lock_get_wait(lock)) {
			lock_rec_cancel(lock);
		} else {
			lock_rec_reset_nth_bit(lock, heap_no);
		}

		lock = lock_rec_get_next(heap_no, lock);
	}
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2373
/*************************************************************//**
2374 2375 2376 2377 2378 2379 2380 2381
Makes a record to inherit the locks (except LOCK_INSERT_INTENTION type)
of another record as gap type locks, but does not reset the lock bits of
the other record. Also waiting lock requests on rec are inherited as
GRANTED gap locks. */
static
void
lock_rec_inherit_to_gap(
/*====================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2382
	const buf_block_t*	heir_block,	/*!< in: block containing the
2383
						record which inherits */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2384
	const buf_block_t*	block,		/*!< in: block containing the
2385 2386 2387
						record from which inherited;
						does NOT reset the locks on
						this record */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2388
	ulint			heir_heap_no,	/*!< in: heap_no of the
2389
						inheriting record */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2390
	ulint			heap_no)	/*!< in: heap_no of the
2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408
						donating record */
{
	lock_t*	lock;

	ut_ad(mutex_own(&kernel_mutex));

	lock = lock_rec_get_first(block, heap_no);

	/* If srv_locks_unsafe_for_binlog is TRUE or session is using
	READ COMMITTED isolation level, we do not want locks set
	by an UPDATE or a DELETE to be inherited as gap type locks. But we
	DO want S-locks set by a consistency constraint to be inherited also
	then. */

	while (lock != NULL) {
		if (!lock_rec_get_insert_intention(lock)
		    && !((srv_locks_unsafe_for_binlog
			  || lock->trx->isolation_level
2409
			  <= TRX_ISO_READ_COMMITTED)
2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421
			 && lock_get_mode(lock) == LOCK_X)) {

			lock_rec_add_to_queue(LOCK_REC | LOCK_GAP
					      | lock_get_mode(lock),
					      heir_block, heir_heap_no,
					      lock->index, lock->trx);
		}

		lock = lock_rec_get_next(heap_no, lock);
	}
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2422
/*************************************************************//**
2423 2424 2425 2426 2427 2428 2429
Makes a record to inherit the gap locks (except LOCK_INSERT_INTENTION type)
of another record as gap type locks, but does not reset the lock bits of the
other record. Also waiting lock requests are inherited as GRANTED gap locks. */
static
void
lock_rec_inherit_to_gap_if_gap_lock(
/*================================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2430 2431
	const buf_block_t*	block,		/*!< in: buffer block */
	ulint			heir_heap_no,	/*!< in: heap_no of
2432
						record which inherits */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2433
	ulint			heap_no)	/*!< in: heap_no of record
2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458
						from which inherited;
						does NOT reset the locks
						on this record */
{
	lock_t*	lock;

	ut_ad(mutex_own(&kernel_mutex));

	lock = lock_rec_get_first(block, heap_no);

	while (lock != NULL) {
		if (!lock_rec_get_insert_intention(lock)
		    && (heap_no == PAGE_HEAP_NO_SUPREMUM
			|| !lock_rec_get_rec_not_gap(lock))) {

			lock_rec_add_to_queue(LOCK_REC | LOCK_GAP
					      | lock_get_mode(lock),
					      block, heir_heap_no,
					      lock->index, lock->trx);
		}

		lock = lock_rec_get_next(heap_no, lock);
	}
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2459
/*************************************************************//**
2460 2461 2462 2463 2464 2465
Moves the locks of a record to another record and resets the lock bits of
the donating record. */
static
void
lock_rec_move(
/*==========*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2466
	const buf_block_t*	receiver,	/*!< in: buffer block containing
2467
						the receiving record */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2468
	const buf_block_t*	donator,	/*!< in: buffer block containing
2469
						the donating record */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2470
	ulint			receiver_heap_no,/*!< in: heap_no of the record
2471 2472 2473
						which gets the locks; there
						must be no lock requests
						on it! */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2474
	ulint			donator_heap_no)/*!< in: heap_no of the record
2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504
						which gives the locks */
{
	lock_t*	lock;

	ut_ad(mutex_own(&kernel_mutex));

	lock = lock_rec_get_first(donator, donator_heap_no);

	ut_ad(lock_rec_get_first(receiver, receiver_heap_no) == NULL);

	while (lock != NULL) {
		const ulint	type_mode = lock->type_mode;

		lock_rec_reset_nth_bit(lock, donator_heap_no);

		if (UNIV_UNLIKELY(type_mode & LOCK_WAIT)) {
			lock_reset_lock_and_trx_wait(lock);
		}

		/* Note that we FIRST reset the bit, and then set the lock:
		the function works also if donator == receiver */

		lock_rec_add_to_queue(type_mode, receiver, receiver_heap_no,
				      lock->index, lock->trx);
		lock = lock_rec_get_next(donator_heap_no, lock);
	}

	ut_ad(lock_rec_get_first(donator, donator_heap_no) == NULL);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2505
/*************************************************************//**
2506 2507 2508 2509 2510 2511 2512 2513
Updates the lock table when we have reorganized a page. NOTE: we copy
also the locks set on the infimum of the page; the infimum may carry
locks if an update of a record is occurring on the page, and its locks
were temporarily stored on the infimum. */
UNIV_INTERN
void
lock_move_reorganize_page(
/*======================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2514
	const buf_block_t*	block,	/*!< in: old index page, now
2515
					reorganized */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2516
	const buf_block_t*	oblock)	/*!< in: copy of the old, not
2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648
					reorganized page */
{
	lock_t*		lock;
	UT_LIST_BASE_NODE_T(lock_t)	old_locks;
	mem_heap_t*	heap		= NULL;
	ulint		comp;

	lock_mutex_enter_kernel();

	lock = lock_rec_get_first_on_page(block);

	if (lock == NULL) {
		lock_mutex_exit_kernel();

		return;
	}

	heap = mem_heap_create(256);

	/* Copy first all the locks on the page to heap and reset the
	bitmaps in the original locks; chain the copies of the locks
	using the trx_locks field in them. */

	UT_LIST_INIT(old_locks);

	do {
		/* Make a copy of the lock */
		lock_t*	old_lock = lock_rec_copy(lock, heap);

		UT_LIST_ADD_LAST(trx_locks, old_locks, old_lock);

		/* Reset bitmap of lock */
		lock_rec_bitmap_reset(lock);

		if (lock_get_wait(lock)) {
			lock_reset_lock_and_trx_wait(lock);
		}

		lock = lock_rec_get_next_on_page(lock);
	} while (lock != NULL);

	comp = page_is_comp(block->frame);
	ut_ad(comp == page_is_comp(oblock->frame));

	for (lock = UT_LIST_GET_FIRST(old_locks); lock;
	     lock = UT_LIST_GET_NEXT(trx_locks, lock)) {
		/* NOTE: we copy also the locks set on the infimum and
		supremum of the page; the infimum may carry locks if an
		update of a record is occurring on the page, and its locks
		were temporarily stored on the infimum */
		page_cur_t	cur1;
		page_cur_t	cur2;

		page_cur_set_before_first(block, &cur1);
		page_cur_set_before_first(oblock, &cur2);

		/* Set locks according to old locks */
		for (;;) {
			ulint	old_heap_no;
			ulint	new_heap_no;

			ut_ad(comp || !memcmp(page_cur_get_rec(&cur1),
					      page_cur_get_rec(&cur2),
					      rec_get_data_size_old(
						      page_cur_get_rec(
							      &cur2))));
			if (UNIV_LIKELY(comp)) {
				old_heap_no = rec_get_heap_no_new(
					page_cur_get_rec(&cur2));
				new_heap_no = rec_get_heap_no_new(
					page_cur_get_rec(&cur1));
			} else {
				old_heap_no = rec_get_heap_no_old(
					page_cur_get_rec(&cur2));
				new_heap_no = rec_get_heap_no_old(
					page_cur_get_rec(&cur1));
			}

			if (lock_rec_get_nth_bit(lock, old_heap_no)) {

				/* Clear the bit in old_lock. */
				ut_d(lock_rec_reset_nth_bit(lock,
							    old_heap_no));

				/* NOTE that the old lock bitmap could be too
				small for the new heap number! */

				lock_rec_add_to_queue(lock->type_mode, block,
						      new_heap_no,
						      lock->index, lock->trx);

				/* if (new_heap_no == PAGE_HEAP_NO_SUPREMUM
				&& lock_get_wait(lock)) {
				fprintf(stderr,
				"---\n--\n!!!Lock reorg: supr type %lu\n",
				lock->type_mode);
				} */
			}

			if (UNIV_UNLIKELY
			    (new_heap_no == PAGE_HEAP_NO_SUPREMUM)) {

				ut_ad(old_heap_no == PAGE_HEAP_NO_SUPREMUM);
				break;
			}

			page_cur_move_to_next(&cur1);
			page_cur_move_to_next(&cur2);
		}

#ifdef UNIV_DEBUG
		{
			ulint	i = lock_rec_find_set_bit(lock);

			/* Check that all locks were moved. */
			if (UNIV_UNLIKELY(i != ULINT_UNDEFINED)) {
				fprintf(stderr,
					"lock_move_reorganize_page():"
					" %lu not moved in %p\n",
					(ulong) i, (void*) lock);
				ut_error;
			}
		}
#endif /* UNIV_DEBUG */
	}

	lock_mutex_exit_kernel();

	mem_heap_free(heap);

#ifdef UNIV_DEBUG_LOCK_VALIDATE
	ut_ad(lock_rec_validate_page(buf_block_get_space(block),
2649
				     buf_block_get_zip_size(block),
2650 2651 2652 2653
				     buf_block_get_page_no(block)));
#endif
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2654
/*************************************************************//**
2655 2656 2657 2658 2659 2660
Moves the explicit locks on user records to another page if a record
list end is moved to another page. */
UNIV_INTERN
void
lock_move_rec_list_end(
/*===================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2661 2662 2663
	const buf_block_t*	new_block,	/*!< in: index page to move to */
	const buf_block_t*	block,		/*!< in: index page */
	const rec_t*		rec)		/*!< in: record on page: this
2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738
						is the first record moved */
{
	lock_t*		lock;
	const ulint	comp	= page_rec_is_comp(rec);

	lock_mutex_enter_kernel();

	/* Note: when we move locks from record to record, waiting locks
	and possible granted gap type locks behind them are enqueued in
	the original order, because new elements are inserted to a hash
	table to the end of the hash chain, and lock_rec_add_to_queue
	does not reuse locks if there are waiters in the queue. */

	for (lock = lock_rec_get_first_on_page(block); lock;
	     lock = lock_rec_get_next_on_page(lock)) {
		page_cur_t	cur1;
		page_cur_t	cur2;
		const ulint	type_mode = lock->type_mode;

		page_cur_position(rec, block, &cur1);

		if (page_cur_is_before_first(&cur1)) {
			page_cur_move_to_next(&cur1);
		}

		page_cur_set_before_first(new_block, &cur2);
		page_cur_move_to_next(&cur2);

		/* Copy lock requests on user records to new page and
		reset the lock bits on the old */

		while (!page_cur_is_after_last(&cur1)) {
			ulint	heap_no;

			if (comp) {
				heap_no = rec_get_heap_no_new(
					page_cur_get_rec(&cur1));
			} else {
				heap_no = rec_get_heap_no_old(
					page_cur_get_rec(&cur1));
				ut_ad(!memcmp(page_cur_get_rec(&cur1),
					 page_cur_get_rec(&cur2),
					 rec_get_data_size_old(
						 page_cur_get_rec(&cur2))));
			}

			if (lock_rec_get_nth_bit(lock, heap_no)) {
				lock_rec_reset_nth_bit(lock, heap_no);

				if (UNIV_UNLIKELY(type_mode & LOCK_WAIT)) {
					lock_reset_lock_and_trx_wait(lock);
				}

				if (comp) {
					heap_no = rec_get_heap_no_new(
						page_cur_get_rec(&cur2));
				} else {
					heap_no = rec_get_heap_no_old(
						page_cur_get_rec(&cur2));
				}

				lock_rec_add_to_queue(type_mode,
						      new_block, heap_no,
						      lock->index, lock->trx);
			}

			page_cur_move_to_next(&cur1);
			page_cur_move_to_next(&cur2);
		}
	}

	lock_mutex_exit_kernel();

#ifdef UNIV_DEBUG_LOCK_VALIDATE
	ut_ad(lock_rec_validate_page(buf_block_get_space(block),
2739
				     buf_block_get_zip_size(block),
2740 2741
				     buf_block_get_page_no(block)));
	ut_ad(lock_rec_validate_page(buf_block_get_space(new_block),
2742
				     buf_block_get_zip_size(block),
2743 2744 2745 2746
				     buf_block_get_page_no(new_block)));
#endif
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2747
/*************************************************************//**
2748 2749 2750 2751 2752 2753
Moves the explicit locks on user records to another page if a record
list start is moved to another page. */
UNIV_INTERN
void
lock_move_rec_list_start(
/*=====================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2754 2755 2756
	const buf_block_t*	new_block,	/*!< in: index page to move to */
	const buf_block_t*	block,		/*!< in: index page */
	const rec_t*		rec,		/*!< in: record on page:
2757 2758
						this is the first
						record NOT copied */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2759
	const rec_t*		old_end)	/*!< in: old
2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851
						previous-to-last
						record on new_page
						before the records
						were copied */
{
	lock_t*		lock;
	const ulint	comp	= page_rec_is_comp(rec);

	ut_ad(block->frame == page_align(rec));
	ut_ad(new_block->frame == page_align(old_end));

	lock_mutex_enter_kernel();

	for (lock = lock_rec_get_first_on_page(block); lock;
	     lock = lock_rec_get_next_on_page(lock)) {
		page_cur_t	cur1;
		page_cur_t	cur2;
		const ulint	type_mode = lock->type_mode;

		page_cur_set_before_first(block, &cur1);
		page_cur_move_to_next(&cur1);

		page_cur_position(old_end, new_block, &cur2);
		page_cur_move_to_next(&cur2);

		/* Copy lock requests on user records to new page and
		reset the lock bits on the old */

		while (page_cur_get_rec(&cur1) != rec) {
			ulint	heap_no;

			if (comp) {
				heap_no = rec_get_heap_no_new(
					page_cur_get_rec(&cur1));
			} else {
				heap_no = rec_get_heap_no_old(
					page_cur_get_rec(&cur1));
				ut_ad(!memcmp(page_cur_get_rec(&cur1),
					      page_cur_get_rec(&cur2),
					      rec_get_data_size_old(
						      page_cur_get_rec(
							      &cur2))));
			}

			if (lock_rec_get_nth_bit(lock, heap_no)) {
				lock_rec_reset_nth_bit(lock, heap_no);

				if (UNIV_UNLIKELY(type_mode & LOCK_WAIT)) {
					lock_reset_lock_and_trx_wait(lock);
				}

				if (comp) {
					heap_no = rec_get_heap_no_new(
						page_cur_get_rec(&cur2));
				} else {
					heap_no = rec_get_heap_no_old(
						page_cur_get_rec(&cur2));
				}

				lock_rec_add_to_queue(type_mode,
						      new_block, heap_no,
						      lock->index, lock->trx);
			}

			page_cur_move_to_next(&cur1);
			page_cur_move_to_next(&cur2);
		}

#ifdef UNIV_DEBUG
		if (page_rec_is_supremum(rec)) {
			ulint	i;

			for (i = PAGE_HEAP_NO_USER_LOW;
			     i < lock_rec_get_n_bits(lock); i++) {
				if (UNIV_UNLIKELY
				    (lock_rec_get_nth_bit(lock, i))) {

					fprintf(stderr,
						"lock_move_rec_list_start():"
						" %lu not moved in %p\n",
						(ulong) i, (void*) lock);
					ut_error;
				}
			}
		}
#endif /* UNIV_DEBUG */
	}

	lock_mutex_exit_kernel();

#ifdef UNIV_DEBUG_LOCK_VALIDATE
	ut_ad(lock_rec_validate_page(buf_block_get_space(block),
2852
				     buf_block_get_zip_size(block),
2853 2854 2855 2856
				     buf_block_get_page_no(block)));
#endif
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2857
/*************************************************************//**
2858 2859 2860 2861 2862
Updates the lock table when a page is split to the right. */
UNIV_INTERN
void
lock_update_split_right(
/*====================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2863 2864
	const buf_block_t*	right_block,	/*!< in: right page */
	const buf_block_t*	left_block)	/*!< in: left page */
2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884
{
	ulint	heap_no = lock_get_min_heap_no(right_block);

	lock_mutex_enter_kernel();

	/* Move the locks on the supremum of the left page to the supremum
	of the right page */

	lock_rec_move(right_block, left_block,
		      PAGE_HEAP_NO_SUPREMUM, PAGE_HEAP_NO_SUPREMUM);

	/* Inherit the locks to the supremum of left page from the successor
	of the infimum on right page */

	lock_rec_inherit_to_gap(left_block, right_block,
				PAGE_HEAP_NO_SUPREMUM, heap_no);

	lock_mutex_exit_kernel();
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2885
/*************************************************************//**
2886 2887 2888 2889 2890
Updates the lock table when a page is merged to the right. */
UNIV_INTERN
void
lock_update_merge_right(
/*====================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2891
	const buf_block_t*	right_block,	/*!< in: right page to
2892
						which merged */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2893
	const rec_t*		orig_succ,	/*!< in: original
2894 2895 2896
						successor of infimum
						on the right page
						before merge */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2897
	const buf_block_t*	left_block)	/*!< in: merged index
2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921
						page which will be
						discarded */
{
	lock_mutex_enter_kernel();

	/* Inherit the locks from the supremum of the left page to the
	original successor of infimum on the right page, to which the left
	page was merged */

	lock_rec_inherit_to_gap(right_block, left_block,
				page_rec_get_heap_no(orig_succ),
				PAGE_HEAP_NO_SUPREMUM);

	/* Reset the locks on the supremum of the left page, releasing
	waiting transactions */

	lock_rec_reset_and_release_wait(left_block,
					PAGE_HEAP_NO_SUPREMUM);

	lock_rec_free_all_from_discard_page(left_block);

	lock_mutex_exit_kernel();
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2922
/*************************************************************//**
2923 2924 2925 2926 2927 2928 2929 2930 2931 2932
Updates the lock table when the root page is copied to another in
btr_root_raise_and_insert. Note that we leave lock structs on the
root page, even though they do not make sense on other than leaf
pages: the reason is that in a pessimistic update the infimum record
of the root page will act as a dummy carrier of the locks of the record
to be updated. */
UNIV_INTERN
void
lock_update_root_raise(
/*===================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2933 2934
	const buf_block_t*	block,	/*!< in: index page to which copied */
	const buf_block_t*	root)	/*!< in: root page */
2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945
{
	lock_mutex_enter_kernel();

	/* Move the locks on the supremum of the root to the supremum
	of block */

	lock_rec_move(block, root,
		      PAGE_HEAP_NO_SUPREMUM, PAGE_HEAP_NO_SUPREMUM);
	lock_mutex_exit_kernel();
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2946
/*************************************************************//**
2947 2948 2949 2950 2951 2952
Updates the lock table when a page is copied to another and the original page
is removed from the chain of leaf pages, except if page is the root! */
UNIV_INTERN
void
lock_update_copy_and_discard(
/*=========================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2953
	const buf_block_t*	new_block,	/*!< in: index page to
2954
						which copied */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2955
	const buf_block_t*	block)		/*!< in: index page;
2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969
						NOT the root! */
{
	lock_mutex_enter_kernel();

	/* Move the locks on the supremum of the old page to the supremum
	of new_page */

	lock_rec_move(new_block, block,
		      PAGE_HEAP_NO_SUPREMUM, PAGE_HEAP_NO_SUPREMUM);
	lock_rec_free_all_from_discard_page(block);

	lock_mutex_exit_kernel();
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2970
/*************************************************************//**
2971 2972 2973 2974 2975
Updates the lock table when a page is split to the left. */
UNIV_INTERN
void
lock_update_split_left(
/*===================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2976 2977
	const buf_block_t*	right_block,	/*!< in: right page */
	const buf_block_t*	left_block)	/*!< in: left page */
2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991
{
	ulint	heap_no = lock_get_min_heap_no(right_block);

	lock_mutex_enter_kernel();

	/* Inherit the locks to the supremum of the left page from the
	successor of the infimum on the right page */

	lock_rec_inherit_to_gap(left_block, right_block,
				PAGE_HEAP_NO_SUPREMUM, heap_no);

	lock_mutex_exit_kernel();
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2992
/*************************************************************//**
2993 2994 2995 2996 2997
Updates the lock table when a page is merged to the left. */
UNIV_INTERN
void
lock_update_merge_left(
/*===================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
2998
	const buf_block_t*	left_block,	/*!< in: left page to
2999
						which merged */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3000
	const rec_t*		orig_pred,	/*!< in: original predecessor
3001 3002
						of supremum on the left page
						before merge */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3003
	const buf_block_t*	right_block)	/*!< in: merged index page
3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040
						which will be discarded */
{
	const rec_t*	left_next_rec;

	ut_ad(left_block->frame == page_align(orig_pred));

	lock_mutex_enter_kernel();

	left_next_rec = page_rec_get_next_const(orig_pred);

	if (!page_rec_is_supremum(left_next_rec)) {

		/* Inherit the locks on the supremum of the left page to the
		first record which was moved from the right page */

		lock_rec_inherit_to_gap(left_block, left_block,
					page_rec_get_heap_no(left_next_rec),
					PAGE_HEAP_NO_SUPREMUM);

		/* Reset the locks on the supremum of the left page,
		releasing waiting transactions */

		lock_rec_reset_and_release_wait(left_block,
						PAGE_HEAP_NO_SUPREMUM);
	}

	/* Move the locks from the supremum of right page to the supremum
	of the left page */

	lock_rec_move(left_block, right_block,
		      PAGE_HEAP_NO_SUPREMUM, PAGE_HEAP_NO_SUPREMUM);

	lock_rec_free_all_from_discard_page(right_block);

	lock_mutex_exit_kernel();
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3041
/*************************************************************//**
3042 3043 3044 3045 3046 3047
Resets the original locks on heir and replaces them with gap type locks
inherited from rec. */
UNIV_INTERN
void
lock_rec_reset_and_inherit_gap_locks(
/*=================================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3048
	const buf_block_t*	heir_block,	/*!< in: block containing the
3049
						record which inherits */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3050
	const buf_block_t*	block,		/*!< in: block containing the
3051 3052 3053
						record from which inherited;
						does NOT reset the locks on
						this record */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3054
	ulint			heir_heap_no,	/*!< in: heap_no of the
3055
						inheriting record */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3056
	ulint			heap_no)	/*!< in: heap_no of the
3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067
						donating record */
{
	mutex_enter(&kernel_mutex);

	lock_rec_reset_and_release_wait(heir_block, heir_heap_no);

	lock_rec_inherit_to_gap(heir_block, block, heir_heap_no, heap_no);

	mutex_exit(&kernel_mutex);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3068
/*************************************************************//**
3069 3070 3071 3072 3073
Updates the lock table when a page is discarded. */
UNIV_INTERN
void
lock_update_discard(
/*================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3074
	const buf_block_t*	heir_block,	/*!< in: index page
3075
						which will inherit the locks */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3076
	ulint			heir_heap_no,	/*!< in: heap_no of the record
3077
						which will inherit the locks */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3078
	const buf_block_t*	block)		/*!< in: index page
3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130
						which will be discarded */
{
	const page_t*	page = block->frame;
	const rec_t*	rec;
	ulint		heap_no;

	lock_mutex_enter_kernel();

	if (!lock_rec_get_first_on_page(block)) {
		/* No locks exist on page, nothing to do */

		lock_mutex_exit_kernel();

		return;
	}

	/* Inherit all the locks on the page to the record and reset all
	the locks on the page */

	if (page_is_comp(page)) {
		rec = page + PAGE_NEW_INFIMUM;

		do {
			heap_no = rec_get_heap_no_new(rec);

			lock_rec_inherit_to_gap(heir_block, block,
						heir_heap_no, heap_no);

			lock_rec_reset_and_release_wait(block, heap_no);

			rec = page + rec_get_next_offs(rec, TRUE);
		} while (heap_no != PAGE_HEAP_NO_SUPREMUM);
	} else {
		rec = page + PAGE_OLD_INFIMUM;

		do {
			heap_no = rec_get_heap_no_old(rec);

			lock_rec_inherit_to_gap(heir_block, block,
						heir_heap_no, heap_no);

			lock_rec_reset_and_release_wait(block, heap_no);

			rec = page + rec_get_next_offs(rec, FALSE);
		} while (heap_no != PAGE_HEAP_NO_SUPREMUM);
	}

	lock_rec_free_all_from_discard_page(block);

	lock_mutex_exit_kernel();
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3131
/*************************************************************//**
3132 3133 3134 3135 3136
Updates the lock table when a new user record is inserted. */
UNIV_INTERN
void
lock_update_insert(
/*===============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3137 3138
	const buf_block_t*	block,	/*!< in: buffer block containing rec */
	const rec_t*		rec)	/*!< in: the inserted record */
3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163
{
	ulint	receiver_heap_no;
	ulint	donator_heap_no;

	ut_ad(block->frame == page_align(rec));

	/* Inherit the gap-locking locks for rec, in gap mode, from the next
	record */

	if (page_rec_is_comp(rec)) {
		receiver_heap_no = rec_get_heap_no_new(rec);
		donator_heap_no = rec_get_heap_no_new(
			page_rec_get_next_low(rec, TRUE));
	} else {
		receiver_heap_no = rec_get_heap_no_old(rec);
		donator_heap_no = rec_get_heap_no_old(
			page_rec_get_next_low(rec, FALSE));
	}

	lock_mutex_enter_kernel();
	lock_rec_inherit_to_gap_if_gap_lock(block,
					    receiver_heap_no, donator_heap_no);
	lock_mutex_exit_kernel();
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3164
/*************************************************************//**
3165 3166 3167 3168 3169
Updates the lock table when a record is removed. */
UNIV_INTERN
void
lock_update_delete(
/*===============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3170 3171
	const buf_block_t*	block,	/*!< in: buffer block containing rec */
	const rec_t*		rec)	/*!< in: the record to be removed */
3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203
{
	const page_t*	page = block->frame;
	ulint		heap_no;
	ulint		next_heap_no;

	ut_ad(page == page_align(rec));

	if (page_is_comp(page)) {
		heap_no = rec_get_heap_no_new(rec);
		next_heap_no = rec_get_heap_no_new(page
						   + rec_get_next_offs(rec,
								       TRUE));
	} else {
		heap_no = rec_get_heap_no_old(rec);
		next_heap_no = rec_get_heap_no_old(page
						   + rec_get_next_offs(rec,
								       FALSE));
	}

	lock_mutex_enter_kernel();

	/* Let the next record inherit the locks from rec, in gap mode */

	lock_rec_inherit_to_gap(block, block, next_heap_no, heap_no);

	/* Reset the lock bits on rec and release waiting transactions */

	lock_rec_reset_and_release_wait(block, heap_no);

	lock_mutex_exit_kernel();
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3204
/*********************************************************************//**
3205 3206 3207 3208 3209 3210 3211 3212 3213 3214
Stores on the page infimum record the explicit locks of another record.
This function is used to store the lock state of a record when it is
updated and the size of the record changes in the update. The record
is moved in such an update, perhaps to another page. The infimum record
acts as a dummy carrier record, taking care of lock releases while the
actual record is being moved. */
UNIV_INTERN
void
lock_rec_store_on_page_infimum(
/*===========================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3215 3216
	const buf_block_t*	block,	/*!< in: buffer block containing rec */
	const rec_t*		rec)	/*!< in: record whose lock state
3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232
					is stored on the infimum
					record of the same page; lock
					bits are reset on the
					record */
{
	ulint	heap_no = page_rec_get_heap_no(rec);

	ut_ad(block->frame == page_align(rec));

	lock_mutex_enter_kernel();

	lock_rec_move(block, block, PAGE_HEAP_NO_INFIMUM, heap_no);

	lock_mutex_exit_kernel();
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3233
/*********************************************************************//**
3234 3235 3236 3237 3238 3239
Restores the state of explicit lock requests on a single record, where the
state was stored on the infimum of the page. */
UNIV_INTERN
void
lock_rec_restore_from_page_infimum(
/*===============================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3240 3241
	const buf_block_t*	block,	/*!< in: buffer block containing rec */
	const rec_t*		rec,	/*!< in: record whose lock state
3242
					is restored */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3243
	const buf_block_t*	donator)/*!< in: page (rec is not
3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259
					necessarily on this page)
					whose infimum stored the lock
					state; lock bits are reset on
					the infimum */
{
	ulint	heap_no = page_rec_get_heap_no(rec);

	lock_mutex_enter_kernel();

	lock_rec_move(block, donator, heap_no, PAGE_HEAP_NO_INFIMUM);

	lock_mutex_exit_kernel();
}

/*=========== DEADLOCK CHECKING ======================================*/

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3260 3261 3262 3263 3264
/********************************************************************//**
Checks if a lock request results in a deadlock.
@return TRUE if a deadlock was detected and we chose trx as a victim;
FALSE if no deadlock, or there was a deadlock, but we chose other
transaction(s) as victim(s) */
3265 3266 3267 3268
static
ibool
lock_deadlock_occurs(
/*=================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3269 3270
	lock_t*	lock,	/*!< in: lock the transaction is requesting */
	trx_t*	trx)	/*!< in: transaction */
3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292
{
	trx_t*		mark_trx;
	ulint		ret;
	ulint		cost	= 0;

	ut_ad(trx);
	ut_ad(lock);
	ut_ad(mutex_own(&kernel_mutex));
retry:
	/* We check that adding this trx to the waits-for graph
	does not produce a cycle. First mark all active transactions
	with 0: */

	mark_trx = UT_LIST_GET_FIRST(trx_sys->trx_list);

	while (mark_trx) {
		mark_trx->deadlock_mark = 0;
		mark_trx = UT_LIST_GET_NEXT(trx_list, mark_trx);
	}

	ret = lock_deadlock_recursive(trx, trx, lock, &cost, 0);

3293 3294
	switch (ret) {
	case LOCK_VICTIM_IS_OTHER:
3295 3296 3297 3298
		/* We chose some other trx as a victim: retry if there still
		is a deadlock */
		goto retry;

3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317 3318
	case LOCK_EXCEED_MAX_DEPTH:
		/* If the lock search exceeds the max step
		or the max depth, the current trx will be
		the victim. Print its information. */
		rewind(lock_latest_err_file);
		ut_print_timestamp(lock_latest_err_file);

		fputs("TOO DEEP OR LONG SEARCH IN THE LOCK TABLE"
		      " WAITS-FOR GRAPH, WE WILL ROLL BACK"
		      " FOLLOWING TRANSACTION \n",
		      lock_latest_err_file);

		fputs("\n*** TRANSACTION:\n", lock_latest_err_file);
		      trx_print(lock_latest_err_file, trx, 3000);

		fputs("*** WAITING FOR THIS LOCK TO BE GRANTED:\n",
		      lock_latest_err_file);

		if (lock_get_type(lock) == LOCK_REC) {
			lock_rec_print(lock_latest_err_file, lock);
3319
		} else {
3320
			lock_table_print(lock_latest_err_file, lock);
3321
		}
3322
		break;
3323

3324 3325
	case LOCK_VICTIM_IS_START:
		srv_n_lock_deadlock_count++;
3326 3327
		fputs("*** WE ROLL BACK TRANSACTION (2)\n",
		      lock_latest_err_file);
3328
		break;
3329

3330 3331 3332
	default:
		/* No deadlock detected*/
		return(FALSE);
3333 3334
	}

3335 3336 3337
	lock_deadlock_found = TRUE;

	return(TRUE);
3338 3339
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3340 3341 3342 3343 3344 3345
/********************************************************************//**
Looks recursively for a deadlock.
@return 0 if no deadlock found, LOCK_VICTIM_IS_START if there was a
deadlock and we chose 'start' as the victim, LOCK_VICTIM_IS_OTHER if a
deadlock was found and we chose some other trx as a victim: we must do
the search again in this last case because there may be another
3346 3347
deadlock!
LOCK_EXCEED_MAX_DEPTH if the lock search exceeds max steps or max depth. */
3348 3349 3350 3351
static
ulint
lock_deadlock_recursive(
/*====================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3352 3353
	trx_t*	start,		/*!< in: recursion starting point */
	trx_t*	trx,		/*!< in: a transaction waiting for a lock */
3354
	lock_t*	wait_lock,	/*!< in: lock that is waiting to be granted */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3355
	ulint*	cost,		/*!< in/out: number of calculation steps thus
3356
				far: if this exceeds LOCK_MAX_N_STEPS_...
3357
				we return LOCK_EXCEED_MAX_DEPTH */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3358
	ulint	depth)		/*!< in: recursion depth: if this exceeds
3359
				LOCK_MAX_DEPTH_IN_DEADLOCK_CHECK, we
3360
				return LOCK_EXCEED_MAX_DEPTH */
3361
{
3362
	ulint	ret;
3363 3364
	lock_t*	lock;
	trx_t*	lock_trx;
3365
	ulint	heap_no		= ULINT_UNDEFINED;
3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381

	ut_a(trx);
	ut_a(start);
	ut_a(wait_lock);
	ut_ad(mutex_own(&kernel_mutex));

	if (trx->deadlock_mark == 1) {
		/* We have already exhaustively searched the subtree starting
		from this trx */

		return(0);
	}

	*cost = *cost + 1;

	if (lock_get_type_low(wait_lock) == LOCK_REC) {
3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403
		ulint		space;
		ulint		page_no;

		heap_no = lock_rec_find_set_bit(wait_lock);
		ut_a(heap_no != ULINT_UNDEFINED);

		space = wait_lock->un_member.rec_lock.space;
		page_no = wait_lock->un_member.rec_lock.page_no;

		lock = lock_rec_get_first_on_page_addr(space, page_no);

		/* Position the iterator on the first matching record lock. */
		while (lock != NULL
		       && lock != wait_lock
		       && !lock_rec_get_nth_bit(lock, heap_no)) {

			lock = lock_rec_get_next_on_page(lock);
		}

		if (lock == wait_lock) {
			lock = NULL;
		}
3404

3405
		ut_ad(lock == NULL || lock_rec_get_nth_bit(lock, heap_no));
3406

3407 3408
	} else {
		lock = wait_lock;
3409
	}
3410 3411 3412 3413

	/* Look at the locks ahead of wait_lock in the lock queue */

	for (;;) {
3414 3415
		/* Get previous table lock. */
		if (heap_no == ULINT_UNDEFINED) {
3416

3417 3418
			lock = UT_LIST_GET_PREV(
				un_member.tab_lock.locks, lock);
3419
		}
3420

3421
		if (lock == NULL) {
3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435
			/* We can mark this subtree as searched */
			trx->deadlock_mark = 1;

			return(FALSE);
		}

		if (lock_has_to_wait(wait_lock, lock)) {

			ibool	too_far
				= depth > LOCK_MAX_DEPTH_IN_DEADLOCK_CHECK
				|| *cost > LOCK_MAX_N_STEPS_IN_DEADLOCK_CHECK;

			lock_trx = lock->trx;

3436
			if (lock_trx == start) {
3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482

				/* We came back to the recursion starting
				point: a deadlock detected; or we have
				searched the waits-for graph too long */

				FILE*	ef = lock_latest_err_file;

				rewind(ef);
				ut_print_timestamp(ef);

				fputs("\n*** (1) TRANSACTION:\n", ef);

				trx_print(ef, wait_lock->trx, 3000);

				fputs("*** (1) WAITING FOR THIS LOCK"
				      " TO BE GRANTED:\n", ef);

				if (lock_get_type_low(wait_lock) == LOCK_REC) {
					lock_rec_print(ef, wait_lock);
				} else {
					lock_table_print(ef, wait_lock);
				}

				fputs("*** (2) TRANSACTION:\n", ef);

				trx_print(ef, lock->trx, 3000);

				fputs("*** (2) HOLDS THE LOCK(S):\n", ef);

				if (lock_get_type_low(lock) == LOCK_REC) {
					lock_rec_print(ef, lock);
				} else {
					lock_table_print(ef, lock);
				}

				fputs("*** (2) WAITING FOR THIS LOCK"
				      " TO BE GRANTED:\n", ef);

				if (lock_get_type_low(start->wait_lock)
				    == LOCK_REC) {
					lock_rec_print(ef, start->wait_lock);
				} else {
					lock_table_print(ef, start->wait_lock);
				}
#ifdef UNIV_DEBUG
				if (lock_print_waits) {
3483
					fputs("Deadlock detected\n",
3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521
					      stderr);
				}
#endif /* UNIV_DEBUG */

				if (trx_weight_cmp(wait_lock->trx,
						   start) >= 0) {
					/* Our recursion starting point
					transaction is 'smaller', let us
					choose 'start' as the victim and roll
					back it */

					return(LOCK_VICTIM_IS_START);
				}

				lock_deadlock_found = TRUE;

				/* Let us choose the transaction of wait_lock
				as a victim to try to avoid deadlocking our
				recursion starting point transaction */

				fputs("*** WE ROLL BACK TRANSACTION (1)\n",
				      ef);

				wait_lock->trx->was_chosen_as_deadlock_victim
					= TRUE;

				lock_cancel_waiting_and_release(wait_lock);

				/* Since trx and wait_lock are no longer
				in the waits-for graph, we can return FALSE;
				note that our selective algorithm can choose
				several transactions as victims, but still
				we may end up rolling back also the recursion
				starting point transaction! */

				return(LOCK_VICTIM_IS_OTHER);
			}

3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536
			if (too_far) {

#ifdef UNIV_DEBUG
				if (lock_print_waits) {
					fputs("Deadlock search exceeds"
					      " max steps or depth.\n",
					      stderr);
				}
#endif /* UNIV_DEBUG */
				/* The information about transaction/lock
				to be rolled back is available in the top
				level. Do not print anything here. */
				return(LOCK_EXCEED_MAX_DEPTH);
			}

3537 3538 3539 3540 3541 3542 3543 3544 3545
			if (lock_trx->que_state == TRX_QUE_LOCK_WAIT) {

				/* Another trx ahead has requested lock	in an
				incompatible mode, and is itself waiting for
				a lock */

				ret = lock_deadlock_recursive(
					start, lock_trx,
					lock_trx->wait_lock, cost, depth + 1);
3546

3547 3548 3549 3550 3551 3552
				if (ret != 0) {

					return(ret);
				}
			}
		}
3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567
		/* Get the next record lock to check. */
		if (heap_no != ULINT_UNDEFINED) {

			ut_a(lock != NULL);

			do {
				lock = lock_rec_get_next_on_page(lock);
			} while (lock != NULL
				&& lock != wait_lock
				&& !lock_rec_get_nth_bit(lock, heap_no));

			if (lock == wait_lock) {
				lock = NULL;
			}
		}
3568 3569 3570 3571 3572
	}/* end of the 'for (;;)'-loop */
}

/*========================= TABLE LOCKS ==============================*/

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3573
/*********************************************************************//**
3574
Creates a table lock object and adds it as the last in the lock queue
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3575 3576
of the table. Does NOT check for deadlocks or lock compatibility.
@return	own: new lock object */
3577 3578 3579 3580
UNIV_INLINE
lock_t*
lock_table_create(
/*==============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3581 3582
	dict_table_t*	table,	/*!< in: database table in dictionary cache */
	ulint		type_mode,/*!< in: lock mode possibly ORed with
3583
				LOCK_WAIT */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3584
	trx_t*		trx)	/*!< in: trx */
3585 3586 3587 3588 3589 3590 3591 3592 3593 3594
{
	lock_t*	lock;

	ut_ad(table && trx);
	ut_ad(mutex_own(&kernel_mutex));

	if ((type_mode & LOCK_MODE_MASK) == LOCK_AUTO_INC) {
		++table->n_waiting_or_granted_auto_inc_locks;
	}

Vadim Tkachenko's avatar
Vadim Tkachenko committed
3595 3596 3597
	/* For AUTOINC locking we reuse the lock instance only if
	there is no wait involved else we allocate the waiting lock
	from the transaction lock heap. */
3598 3599
	if (type_mode == LOCK_AUTO_INC) {

Vadim Tkachenko's avatar
Vadim Tkachenko committed
3600 3601 3602
		lock = table->autoinc_lock;

		table->autoinc_trx = trx;
3603

Vadim Tkachenko's avatar
Vadim Tkachenko committed
3604
		ib_vector_push(trx->autoinc_locks, lock);
3605 3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623 3624 3625
	} else {
		lock = mem_heap_alloc(trx->lock_heap, sizeof(lock_t));
	}

	UT_LIST_ADD_LAST(trx_locks, trx->trx_locks, lock);

	lock->type_mode = type_mode | LOCK_TABLE;
	lock->trx = trx;

	lock->un_member.tab_lock.table = table;

	UT_LIST_ADD_LAST(un_member.tab_lock.locks, table->locks, lock);

	if (UNIV_UNLIKELY(type_mode & LOCK_WAIT)) {

		lock_set_lock_and_trx_wait(lock, trx);
	}

	return(lock);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3626
/*************************************************************//**
3627 3628 3629 3630 3631 3632 3633
Removes a table lock request from the queue and the trx list of locks;
this is a low-level function which does NOT check if waiting requests
can now be granted. */
UNIV_INLINE
void
lock_table_remove_low(
/*==================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3634
	lock_t*	lock)	/*!< in: table lock */
3635 3636
{
	trx_t*		trx;
Vadim Tkachenko's avatar
Vadim Tkachenko committed
3637
	dict_table_t*	table;
3638 3639 3640 3641

	ut_ad(mutex_own(&kernel_mutex));

	trx = lock->trx;
Vadim Tkachenko's avatar
Vadim Tkachenko committed
3642 3643 3644 3645 3646 3647 3648 3649 3650 3651 3652
	table = lock->un_member.tab_lock.table;

	/* Remove the table from the transaction's AUTOINC vector, if
	the lock that is being release is an AUTOINC lock. */
	if (lock_get_mode(lock) == LOCK_AUTO_INC) {

		/* The table's AUTOINC lock can get transferred to
		another transaction before we get here. */
		if (table->autoinc_trx == trx) {
			table->autoinc_trx = NULL;
		}
3653

Vadim Tkachenko's avatar
Vadim Tkachenko committed
3654 3655 3656 3657 3658 3659 3660 3661 3662
		/* The locks must be freed in the reverse order from
		the one in which they were acquired. This is to avoid
		traversing the AUTOINC lock vector unnecessarily. 

		We only store locks that were granted in the
		trx->autoinc_locks vector (see lock_table_create()
		and lock_grant()). Therefore it can be empty and we
		need to check for that. */

3663 3664
		if (!lock_get_wait(lock)
		    && !ib_vector_is_empty(trx->autoinc_locks)) {
Vadim Tkachenko's avatar
Vadim Tkachenko committed
3665 3666 3667 3668 3669
			lock_t*	autoinc_lock;

			autoinc_lock = ib_vector_pop(trx->autoinc_locks);
			ut_a(autoinc_lock == lock);
		}
3670 3671 3672 3673 3674 3675 3676 3677 3678

		ut_a(table->n_waiting_or_granted_auto_inc_locks > 0);
		--table->n_waiting_or_granted_auto_inc_locks;
	}

	UT_LIST_REMOVE(trx_locks, trx->trx_locks, lock);
	UT_LIST_REMOVE(un_member.tab_lock.locks, table->locks, lock);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3679
/*********************************************************************//**
3680
Enqueues a waiting request for a table lock which cannot be granted
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3681 3682 3683 3684 3685
immediately. Checks for deadlocks.
@return DB_LOCK_WAIT, DB_DEADLOCK, or DB_QUE_THR_SUSPENDED, or
DB_SUCCESS; DB_SUCCESS means that there was a deadlock, but another
transaction was chosen as a victim, and we got the lock immediately:
no need to wait then */
3686 3687 3688 3689
static
ulint
lock_table_enqueue_waiting(
/*=======================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3690
	ulint		mode,	/*!< in: lock mode this transaction is
3691
				requesting */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3692 3693
	dict_table_t*	table,	/*!< in: table */
	que_thr_t*	thr)	/*!< in: query thread */
3694 3695 3696
{
	lock_t*	lock;
	trx_t*	trx;
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3697 3698
	ulint   sec;
	ulint   ms;
3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738

	ut_ad(mutex_own(&kernel_mutex));

	/* Test if there already is some other reason to suspend thread:
	we do not enqueue a lock request if the query thread should be
	stopped anyway */

	if (que_thr_stop(thr)) {
		ut_error;

		return(DB_QUE_THR_SUSPENDED);
	}

	trx = thr_get_trx(thr);

	switch (trx_get_dict_operation(trx)) {
	case TRX_DICT_OP_NONE:
		break;
	case TRX_DICT_OP_TABLE:
	case TRX_DICT_OP_INDEX:
		ut_print_timestamp(stderr);
		fputs("  InnoDB: Error: a table lock wait happens"
		      " in a dictionary operation!\n"
		      "InnoDB: Table name ", stderr);
		ut_print_name(stderr, trx, TRUE, table->name);
		fputs(".\n"
		      "InnoDB: Submit a detailed bug report"
		      " to http://bugs.mysql.com\n",
		      stderr);
	}

	/* Enqueue the lock request that will wait to be granted */

	lock = lock_table_create(table, mode | LOCK_WAIT, trx);

	/* Check if a deadlock occurs: if yes, remove the lock request and
	return an error code */

	if (lock_deadlock_occurs(lock, trx)) {

3739 3740
		/* The order here is important, we don't want to
		lose the state of the lock before calling remove. */
3741
		lock_table_remove_low(lock);
3742
		lock_reset_lock_and_trx_wait(lock);
3743 3744 3745 3746 3747 3748 3749 3750 3751 3752 3753

		return(DB_DEADLOCK);
	}

	if (trx->wait_lock == NULL) {
		/* Deadlock resolution chose another transaction as a victim,
		and we accidentally got our lock granted! */

		return(DB_SUCCESS);
	}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3754 3755 3756 3757
	if (innobase_get_slow_log() && trx->take_stats) {
		ut_usectime(&sec, &ms);
		trx->lock_que_wait_ustarted = (ib_uint64_t)sec * 1000000 + ms;
	}
3758 3759 3760 3761 3762 3763 3764 3765 3766
	trx->que_state = TRX_QUE_LOCK_WAIT;
	trx->was_chosen_as_deadlock_victim = FALSE;
	trx->wait_started = time(NULL);

	ut_a(que_thr_stop(thr));

	return(DB_LOCK_WAIT);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3767
/*********************************************************************//**
3768
Checks if other transactions have an incompatible mode lock request in
3769 3770
the lock queue.
@return	lock or NULL */
3771
UNIV_INLINE
3772
lock_t*
3773 3774
lock_table_other_has_incompatible(
/*==============================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3775
	trx_t*		trx,	/*!< in: transaction, or NULL if all
3776
				transactions should be included */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3777
	ulint		wait,	/*!< in: LOCK_WAIT if also waiting locks are
3778
				taken into account, or 0 if not */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3779 3780
	dict_table_t*	table,	/*!< in: table */
	enum lock_mode	mode)	/*!< in: lock mode */
3781 3782 3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793
{
	lock_t*	lock;

	ut_ad(mutex_own(&kernel_mutex));

	lock = UT_LIST_GET_LAST(table->locks);

	while (lock != NULL) {

		if ((lock->trx != trx)
		    && (!lock_mode_compatible(lock_get_mode(lock), mode))
		    && (wait || !(lock_get_wait(lock)))) {

3794
			return(lock);
3795 3796 3797 3798 3799
		}

		lock = UT_LIST_GET_PREV(un_member.tab_lock.locks, lock);
	}

3800
	return(NULL);
3801 3802
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3803
/*********************************************************************//**
3804
Locks the specified database table in the mode given. If the lock cannot
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3805 3806
be granted immediately, the query thread is put to wait.
@return	DB_SUCCESS, DB_LOCK_WAIT, DB_DEADLOCK, or DB_QUE_THR_SUSPENDED */
3807 3808 3809 3810
UNIV_INTERN
ulint
lock_table(
/*=======*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3811
	ulint		flags,	/*!< in: if BTR_NO_LOCKING_FLAG bit is set,
3812
				does nothing */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3813 3814 3815
	dict_table_t*	table,	/*!< in: database table in dictionary cache */
	enum lock_mode	mode,	/*!< in: lock mode */
	que_thr_t*	thr)	/*!< in: query thread */
3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848 3849 3850 3851 3852 3853 3854 3855 3856 3857 3858 3859 3860 3861 3862 3863 3864 3865
{
	trx_t*	trx;
	ulint	err;

	ut_ad(table && thr);

	if (flags & BTR_NO_LOCKING_FLAG) {

		return(DB_SUCCESS);
	}

	ut_a(flags == 0);

	trx = thr_get_trx(thr);

	lock_mutex_enter_kernel();

	/* Look for stronger locks the same trx already has on the table */

	if (lock_table_has(trx, table, mode)) {

		lock_mutex_exit_kernel();

		return(DB_SUCCESS);
	}

	/* We have to check if the new lock is compatible with any locks
	other transactions have in the table lock queue. */

	if (lock_table_other_has_incompatible(trx, LOCK_WAIT, table, mode)) {

		/* Another trx has a request on the table in an incompatible
		mode: this trx may have to wait */

		err = lock_table_enqueue_waiting(mode | flags, table, thr);

		lock_mutex_exit_kernel();

		return(err);
	}

	lock_table_create(table, mode | flags, trx);

	ut_a(!flags || mode == LOCK_S || mode == LOCK_X);

	lock_mutex_exit_kernel();

	return(DB_SUCCESS);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3866 3867 3868
/*********************************************************************//**
Checks if a waiting table lock request still has to wait in a queue.
@return	TRUE if still has to wait */
3869 3870 3871 3872
static
ibool
lock_table_has_to_wait_in_queue(
/*============================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3873
	lock_t*	wait_lock)	/*!< in: waiting table lock */
3874 3875 3876 3877
{
	dict_table_t*	table;
	lock_t*		lock;

Vadim Tkachenko's avatar
Vadim Tkachenko committed
3878
	ut_ad(mutex_own(&kernel_mutex));
3879 3880 3881 3882 3883 3884 3885 3886 3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897
	ut_ad(lock_get_wait(wait_lock));

	table = wait_lock->un_member.tab_lock.table;

	lock = UT_LIST_GET_FIRST(table->locks);

	while (lock != wait_lock) {

		if (lock_has_to_wait(wait_lock, lock)) {

			return(TRUE);
		}

		lock = UT_LIST_GET_NEXT(un_member.tab_lock.locks, lock);
	}

	return(FALSE);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3898
/*************************************************************//**
3899 3900 3901 3902 3903 3904 3905
Removes a table lock request, waiting or granted, from the queue and grants
locks to other transactions in the queue, if they now are entitled to a
lock. */
static
void
lock_table_dequeue(
/*===============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3906
	lock_t*	in_lock)/*!< in: table lock object; transactions waiting
3907 3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927 3928 3929 3930 3931 3932 3933 3934 3935 3936
			behind will get their lock requests granted, if
			they are now qualified to it */
{
	lock_t*	lock;

	ut_ad(mutex_own(&kernel_mutex));
	ut_a(lock_get_type_low(in_lock) == LOCK_TABLE);

	lock = UT_LIST_GET_NEXT(un_member.tab_lock.locks, in_lock);

	lock_table_remove_low(in_lock);

	/* Check if waiting locks in the queue can now be granted: grant
	locks if there are no conflicting locks ahead. */

	while (lock != NULL) {

		if (lock_get_wait(lock)
		    && !lock_table_has_to_wait_in_queue(lock)) {

			/* Grant the lock */
			lock_grant(lock);
		}

		lock = UT_LIST_GET_NEXT(un_member.tab_lock.locks, lock);
	}
}

/*=========================== LOCK RELEASE ==============================*/

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3937
/*************************************************************//**
3938 3939 3940 3941 3942 3943 3944
Removes a granted record lock of a transaction from the queue and grants
locks to other transactions waiting in the queue if they now are entitled
to a lock. */
UNIV_INTERN
void
lock_rec_unlock(
/*============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3945
	trx_t*			trx,	/*!< in: transaction that has
3946
					set a record lock */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
3947 3948 3949
	const buf_block_t*	block,	/*!< in: buffer block containing rec */
	const rec_t*		rec,	/*!< in: record */
	enum lock_mode		lock_mode)/*!< in: LOCK_S or LOCK_X */
3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967 3968 3969 3970 3971 3972 3973 3974 3975 3976 3977 3978 3979 3980 3981 3982 3983 3984 3985 3986 3987 3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004 4005 4006 4007 4008
{
	lock_t*	lock;
	lock_t*	release_lock	= NULL;
	ulint	heap_no;

	ut_ad(trx && rec);
	ut_ad(block->frame == page_align(rec));

	heap_no = page_rec_get_heap_no(rec);

	mutex_enter(&kernel_mutex);

	lock = lock_rec_get_first(block, heap_no);

	/* Find the last lock with the same lock_mode and transaction
	from the record. */

	while (lock != NULL) {
		if (lock->trx == trx && lock_get_mode(lock) == lock_mode) {
			release_lock = lock;
			ut_a(!lock_get_wait(lock));
		}

		lock = lock_rec_get_next(heap_no, lock);
	}

	/* If a record lock is found, release the record lock */

	if (UNIV_LIKELY(release_lock != NULL)) {
		lock_rec_reset_nth_bit(release_lock, heap_no);
	} else {
		mutex_exit(&kernel_mutex);
		ut_print_timestamp(stderr);
		fprintf(stderr,
			"  InnoDB: Error: unlock row could not"
			" find a %lu mode lock on the record\n",
			(ulong) lock_mode);

		return;
	}

	/* Check if we can now grant waiting lock requests */

	lock = lock_rec_get_first(block, heap_no);

	while (lock != NULL) {
		if (lock_get_wait(lock)
		    && !lock_rec_has_to_wait_in_queue(lock)) {

			/* Grant the lock */
			lock_grant(lock);
		}

		lock = lock_rec_get_next(heap_no, lock);
	}

	mutex_exit(&kernel_mutex);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4009
/*********************************************************************//**
4010 4011 4012 4013 4014 4015
Releases transaction locks, and releases possible other transactions waiting
because of these locks. */
UNIV_INTERN
void
lock_release_off_kernel(
/*====================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4016
	trx_t*	trx)	/*!< in: transaction */
4017 4018 4019 4020 4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036 4037 4038 4039 4040 4041 4042 4043 4044 4045 4046 4047 4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059 4060 4061 4062 4063 4064 4065 4066 4067
{
	dict_table_t*	table;
	ulint		count;
	lock_t*		lock;

	ut_ad(mutex_own(&kernel_mutex));

	lock = UT_LIST_GET_LAST(trx->trx_locks);

	count = 0;

	while (lock != NULL) {

		count++;

		if (lock_get_type_low(lock) == LOCK_REC) {

			lock_rec_dequeue_from_page(lock);
		} else {
			ut_ad(lock_get_type_low(lock) & LOCK_TABLE);

			if (lock_get_mode(lock) != LOCK_IS
			    && !ut_dulint_is_zero(trx->undo_no)) {

				/* The trx may have modified the table. We
				block the use of the MySQL query cache for
				all currently active transactions. */

				table = lock->un_member.tab_lock.table;

				table->query_cache_inv_trx_id
					= trx_sys->max_trx_id;
			}

			lock_table_dequeue(lock);
		}

		if (count == LOCK_RELEASE_KERNEL_INTERVAL) {
			/* Release the kernel mutex for a while, so that we
			do not monopolize it */

			lock_mutex_exit_kernel();

			lock_mutex_enter_kernel();

			count = 0;
		}

		lock = UT_LIST_GET_LAST(trx->trx_locks);
	}

Vadim Tkachenko's avatar
Vadim Tkachenko committed
4068
	ut_a(ib_vector_size(trx->autoinc_locks) == 0);
4069

Vadim Tkachenko's avatar
Vadim Tkachenko committed
4070
	mem_heap_empty(trx->lock_heap);
4071 4072
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4073
/*********************************************************************//**
4074 4075 4076 4077 4078 4079
Cancels a waiting lock request and releases possible other transactions
waiting behind it. */
UNIV_INTERN
void
lock_cancel_waiting_and_release(
/*============================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4080
	lock_t*	lock)	/*!< in: waiting lock request */
4081 4082 4083 4084 4085 4086 4087 4088 4089
{
	ut_ad(mutex_own(&kernel_mutex));

	if (lock_get_type_low(lock) == LOCK_REC) {

		lock_rec_dequeue_from_page(lock);
	} else {
		ut_ad(lock_get_type_low(lock) & LOCK_TABLE);

Vadim Tkachenko's avatar
Vadim Tkachenko committed
4090 4091 4092 4093 4094
		if (lock->trx->autoinc_locks != NULL) {
			/* Release the transaction's AUTOINC locks/ */
			lock_release_autoinc_locks(lock->trx);
		}

4095 4096 4097 4098 4099 4100 4101 4102 4103 4104 4105 4106
		lock_table_dequeue(lock);
	}

	/* Reset the wait flag and the back pointer to lock in trx */

	lock_reset_lock_and_trx_wait(lock);

	/* The following function releases the trx from lock wait */

	trx_end_lock_wait(lock->trx);
}

Vadim Tkachenko's avatar
Vadim Tkachenko committed
4107 4108 4109 4110 4111 4112
/* True if a lock mode is S or X */
#define IS_LOCK_S_OR_X(lock) \
	(lock_get_mode(lock) == LOCK_S \
	 || lock_get_mode(lock) == LOCK_X)


Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4113
/*********************************************************************//**
Vadim Tkachenko's avatar
Vadim Tkachenko committed
4114 4115 4116 4117
Removes locks of a transaction on a table to be dropped.
If remove_also_table_sx_locks is TRUE then table-level S and X locks are
also removed in addition to other table-level and record-level locks.
No lock, that is going to be removed, is allowed to be a wait lock. */
4118 4119
static
void
Vadim Tkachenko's avatar
Vadim Tkachenko committed
4120 4121
lock_remove_all_on_table_for_trx(
/*=============================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4122 4123 4124
	dict_table_t*	table,			/*!< in: table to be dropped */
	trx_t*		trx,			/*!< in: a transaction */
	ibool		remove_also_table_sx_locks)/*!< in: also removes
Vadim Tkachenko's avatar
Vadim Tkachenko committed
4125
						table S and X locks */
4126 4127 4128 4129 4130 4131 4132 4133 4134 4135 4136 4137 4138 4139 4140 4141 4142
{
	lock_t*	lock;
	lock_t*	prev_lock;

	ut_ad(mutex_own(&kernel_mutex));

	lock = UT_LIST_GET_LAST(trx->trx_locks);

	while (lock != NULL) {
		prev_lock = UT_LIST_GET_PREV(trx_locks, lock);

		if (lock_get_type_low(lock) == LOCK_REC
		    && lock->index->table == table) {
			ut_a(!lock_get_wait(lock));

			lock_rec_discard(lock);
		} else if (lock_get_type_low(lock) & LOCK_TABLE
Vadim Tkachenko's avatar
Vadim Tkachenko committed
4143 4144 4145
			   && lock->un_member.tab_lock.table == table
			   && (remove_also_table_sx_locks
			       || !IS_LOCK_S_OR_X(lock))) {
4146 4147 4148 4149 4150 4151 4152 4153 4154 4155

			ut_a(!lock_get_wait(lock));

			lock_table_remove_low(lock);
		}

		lock = prev_lock;
	}
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4156
/*********************************************************************//**
Vadim Tkachenko's avatar
Vadim Tkachenko committed
4157 4158 4159 4160
Removes locks on a table to be dropped or truncated.
If remove_also_table_sx_locks is TRUE then table-level S and X locks are
also removed in addition to other table-level and record-level locks.
No lock, that is going to be removed, is allowed to be a wait lock. */
4161 4162
UNIV_INTERN
void
Vadim Tkachenko's avatar
Vadim Tkachenko committed
4163 4164
lock_remove_all_on_table(
/*=====================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4165
	dict_table_t*	table,			/*!< in: table to be dropped
Vadim Tkachenko's avatar
Vadim Tkachenko committed
4166
						or truncated */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4167
	ibool		remove_also_table_sx_locks)/*!< in: also removes
Vadim Tkachenko's avatar
Vadim Tkachenko committed
4168
						table S and X locks */
4169 4170
{
	lock_t*	lock;
Vadim Tkachenko's avatar
Vadim Tkachenko committed
4171
	lock_t*	prev_lock;
4172 4173 4174 4175 4176

	mutex_enter(&kernel_mutex);

	lock = UT_LIST_GET_FIRST(table->locks);

Vadim Tkachenko's avatar
Vadim Tkachenko committed
4177 4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189 4190
	while (lock != NULL) {

		prev_lock = UT_LIST_GET_PREV(un_member.tab_lock.locks,
					     lock);

		/* If we should remove all locks (remove_also_table_sx_locks
		is TRUE), or if the lock is not table-level S or X lock,
		then check we are not going to remove a wait lock. */
		if (remove_also_table_sx_locks
		    || !(lock_get_type(lock) == LOCK_TABLE
			 && IS_LOCK_S_OR_X(lock))) {

			ut_a(!lock_get_wait(lock));
		}
4191

Vadim Tkachenko's avatar
Vadim Tkachenko committed
4192 4193
		lock_remove_all_on_table_for_trx(table, lock->trx,
						 remove_also_table_sx_locks);
4194

Vadim Tkachenko's avatar
Vadim Tkachenko committed
4195 4196 4197 4198 4199 4200 4201 4202 4203 4204 4205 4206 4207 4208 4209 4210 4211 4212 4213 4214 4215
		if (prev_lock == NULL) {
			if (lock == UT_LIST_GET_FIRST(table->locks)) {
				/* lock was not removed, pick its successor */
				lock = UT_LIST_GET_NEXT(
					un_member.tab_lock.locks, lock);
			} else {
				/* lock was removed, pick the first one */
				lock = UT_LIST_GET_FIRST(table->locks);
			}
		} else if (UT_LIST_GET_NEXT(un_member.tab_lock.locks,
					    prev_lock) != lock) {
			/* If lock was removed by
			lock_remove_all_on_table_for_trx() then pick the
			successor of prev_lock ... */
			lock = UT_LIST_GET_NEXT(
				un_member.tab_lock.locks, prev_lock);
		} else {
			/* ... otherwise pick the successor of lock. */
			lock = UT_LIST_GET_NEXT(
				un_member.tab_lock.locks, lock);
		}
4216 4217 4218 4219 4220 4221 4222
	}

	mutex_exit(&kernel_mutex);
}

/*===================== VALIDATION AND DEBUGGING  ====================*/

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4223
/*********************************************************************//**
4224 4225 4226 4227 4228
Prints info of a table lock. */
UNIV_INTERN
void
lock_table_print(
/*=============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4229 4230
	FILE*		file,	/*!< in: file where to print */
	const lock_t*	lock)	/*!< in: table type lock */
4231 4232 4233 4234 4235 4236 4237 4238 4239 4240 4241 4242 4243 4244 4245 4246 4247 4248 4249 4250 4251 4252 4253 4254 4255 4256 4257 4258 4259 4260 4261 4262
{
	ut_ad(mutex_own(&kernel_mutex));
	ut_a(lock_get_type_low(lock) == LOCK_TABLE);

	fputs("TABLE LOCK table ", file);
	ut_print_name(file, lock->trx, TRUE,
		      lock->un_member.tab_lock.table->name);
	fprintf(file, " trx id " TRX_ID_FMT,
		TRX_ID_PREP_PRINTF(lock->trx->id));

	if (lock_get_mode(lock) == LOCK_S) {
		fputs(" lock mode S", file);
	} else if (lock_get_mode(lock) == LOCK_X) {
		fputs(" lock mode X", file);
	} else if (lock_get_mode(lock) == LOCK_IS) {
		fputs(" lock mode IS", file);
	} else if (lock_get_mode(lock) == LOCK_IX) {
		fputs(" lock mode IX", file);
	} else if (lock_get_mode(lock) == LOCK_AUTO_INC) {
		fputs(" lock mode AUTO-INC", file);
	} else {
		fprintf(file, " unknown lock mode %lu",
			(ulong) lock_get_mode(lock));
	}

	if (lock_get_wait(lock)) {
		fputs(" waiting", file);
	}

	putc('\n', file);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4263
/*********************************************************************//**
4264 4265 4266 4267 4268
Prints info of a record lock. */
UNIV_INTERN
void
lock_rec_print(
/*===========*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4269 4270
	FILE*		file,	/*!< in: file where to print */
	const lock_t*	lock)	/*!< in: record type lock */
4271 4272 4273 4274 4275 4276 4277 4278 4279 4280 4281 4282 4283 4284 4285 4286 4287 4288 4289 4290 4291 4292 4293 4294 4295 4296 4297 4298 4299 4300 4301 4302 4303 4304 4305 4306 4307 4308 4309 4310 4311 4312 4313 4314 4315 4316 4317 4318 4319 4320 4321 4322
{
	const buf_block_t*	block;
	ulint			space;
	ulint			page_no;
	ulint			i;
	mtr_t			mtr;
	mem_heap_t*		heap		= NULL;
	ulint			offsets_[REC_OFFS_NORMAL_SIZE];
	ulint*			offsets		= offsets_;
	rec_offs_init(offsets_);

	ut_ad(mutex_own(&kernel_mutex));
	ut_a(lock_get_type_low(lock) == LOCK_REC);

	space = lock->un_member.rec_lock.space;
	page_no = lock->un_member.rec_lock.page_no;

	fprintf(file, "RECORD LOCKS space id %lu page no %lu n bits %lu ",
		(ulong) space, (ulong) page_no,
		(ulong) lock_rec_get_n_bits(lock));
	dict_index_name_print(file, lock->trx, lock->index);
	fprintf(file, " trx id " TRX_ID_FMT,
		TRX_ID_PREP_PRINTF(lock->trx->id));

	if (lock_get_mode(lock) == LOCK_S) {
		fputs(" lock mode S", file);
	} else if (lock_get_mode(lock) == LOCK_X) {
		fputs(" lock_mode X", file);
	} else {
		ut_error;
	}

	if (lock_rec_get_gap(lock)) {
		fputs(" locks gap before rec", file);
	}

	if (lock_rec_get_rec_not_gap(lock)) {
		fputs(" locks rec but not gap", file);
	}

	if (lock_rec_get_insert_intention(lock)) {
		fputs(" insert intention", file);
	}

	if (lock_get_wait(lock)) {
		fputs(" waiting", file);
	}

	mtr_start(&mtr);

	putc('\n', file);

4323
	if ( srv_show_verbose_locks ) {
4324 4325 4326 4327 4328 4329 4330 4331 4332 4333
	block = buf_page_try_get(space, page_no, &mtr);

	for (i = 0; i < lock_rec_get_n_bits(lock); ++i) {

		if (!lock_rec_get_nth_bit(lock, i)) {
			continue;
		}

		fprintf(file, "Record lock, heap no %lu", (ulong) i);

4334
		if (block) {
4335
			const rec_t*	rec;
4336

4337 4338
			rec = page_find_rec_with_heap_no(
				buf_block_get_frame(block), i);
4339

4340 4341 4342
			offsets = rec_get_offsets(
				rec, lock->index, offsets,
				ULINT_UNDEFINED, &heap);
4343

4344 4345
			putc(' ', file);
			rec_print_new(file, rec, offsets);
Vadim Tkachenko's avatar
Vadim Tkachenko committed
4346
		}
4347 4348 4349

		putc('\n', file);
	}
4350
	}
4351

4352 4353 4354 4355 4356 4357
	mtr_commit(&mtr);
	if (UNIV_LIKELY_NULL(heap)) {
		mem_heap_free(heap);
	}
}

Vadim Tkachenko's avatar
Vadim Tkachenko committed
4358 4359 4360 4361 4362 4363 4364 4365
#ifdef UNIV_DEBUG
/* Print the number of lock structs from lock_print_info_summary() only
in non-production builds for performance reasons, see
http://bugs.mysql.com/36942 */
#define PRINT_NUM_OF_LOCK_STRUCTS
#endif /* UNIV_DEBUG */

#ifdef PRINT_NUM_OF_LOCK_STRUCTS
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4366 4367 4368
/*********************************************************************//**
Calculates the number of record lock structs in the record lock hash table.
@return	number of record locks */
4369 4370 4371 4372 4373 4374 4375 4376 4377 4378 4379 4380 4381 4382 4383 4384 4385 4386 4387 4388 4389 4390 4391 4392
static
ulint
lock_get_n_rec_locks(void)
/*======================*/
{
	lock_t*	lock;
	ulint	n_locks	= 0;
	ulint	i;

	ut_ad(mutex_own(&kernel_mutex));

	for (i = 0; i < hash_get_n_cells(lock_sys->rec_hash); i++) {

		lock = HASH_GET_FIRST(lock_sys->rec_hash, i);

		while (lock) {
			n_locks++;

			lock = HASH_GET_NEXT(hash, lock);
		}
	}

	return(n_locks);
}
Vadim Tkachenko's avatar
Vadim Tkachenko committed
4393
#endif /* PRINT_NUM_OF_LOCK_STRUCTS */
4394

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4395
/*********************************************************************//**
4396 4397 4398
Prints info of locks for all transactions.
@return FALSE if not able to obtain kernel mutex
and exits without printing info */
4399
UNIV_INTERN
4400
ibool
4401 4402
lock_print_info_summary(
/*====================*/
4403 4404
	FILE*	file,	/*!< in: file where to print */
	ibool   nowait)	/*!< in: whether to wait for the kernel mutex */
4405
{
4406 4407 4408 4409 4410 4411 4412 4413 4414 4415
	/* if nowait is FALSE, wait on the kernel mutex,
	otherwise return immediately if fail to obtain the
	mutex. */
	if (!nowait) {
		lock_mutex_enter_kernel();
	} else if (mutex_enter_nowait(&kernel_mutex)) {
		fputs("FAIL TO OBTAIN KERNEL MUTEX, "
		      "SKIP LOCK INFO PRINTING\n", file);
		return(FALSE);
	}
4416 4417 4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430 4431 4432 4433 4434 4435 4436 4437 4438 4439 4440 4441

	if (lock_deadlock_found) {
		fputs("------------------------\n"
		      "LATEST DETECTED DEADLOCK\n"
		      "------------------------\n", file);

		ut_copy_file(file, lock_latest_err_file);
	}

	fputs("------------\n"
	      "TRANSACTIONS\n"
	      "------------\n", file);

	fprintf(file, "Trx id counter " TRX_ID_FMT "\n",
		TRX_ID_PREP_PRINTF(trx_sys->max_trx_id));

	fprintf(file,
		"Purge done for trx's n:o < " TRX_ID_FMT
		" undo n:o < " TRX_ID_FMT "\n",
		TRX_ID_PREP_PRINTF(purge_sys->purge_trx_no),
		TRX_ID_PREP_PRINTF(purge_sys->purge_undo_no));

	fprintf(file,
		"History list length %lu\n",
		(ulong) trx_sys->rseg_history_len);

Vadim Tkachenko's avatar
Vadim Tkachenko committed
4442
#ifdef PRINT_NUM_OF_LOCK_STRUCTS
4443 4444 4445
	fprintf(file,
		"Total number of lock structs in row lock hash table %lu\n",
		(ulong) lock_get_n_rec_locks());
Vadim Tkachenko's avatar
Vadim Tkachenko committed
4446
#endif /* PRINT_NUM_OF_LOCK_STRUCTS */
4447
	return(TRUE);
4448 4449
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4450
/*********************************************************************//**
4451 4452 4453 4454 4455
Prints info of locks for each transaction. */
UNIV_INTERN
void
lock_print_info_all_transactions(
/*=============================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4456
	FILE*	file)	/*!< in: file where to print */
4457 4458 4459 4460 4461 4462 4463 4464 4465 4466 4467 4468 4469 4470 4471 4472 4473 4474 4475 4476 4477 4478 4479 4480 4481 4482 4483 4484 4485 4486 4487 4488 4489 4490 4491 4492 4493 4494 4495 4496 4497 4498 4499 4500 4501 4502 4503 4504 4505 4506 4507 4508 4509 4510 4511 4512 4513 4514 4515 4516 4517 4518 4519 4520 4521 4522 4523 4524 4525 4526 4527 4528 4529 4530 4531 4532 4533 4534 4535
{
	lock_t*	lock;
	ibool	load_page_first = TRUE;
	ulint	nth_trx		= 0;
	ulint	nth_lock	= 0;
	ulint	i;
	mtr_t	mtr;
	trx_t*	trx;

	fprintf(file, "LIST OF TRANSACTIONS FOR EACH SESSION:\n");

	/* First print info on non-active transactions */

	trx = UT_LIST_GET_FIRST(trx_sys->mysql_trx_list);

	while (trx) {
		if (trx->conc_state == TRX_NOT_STARTED) {
			fputs("---", file);
			trx_print(file, trx, 600);
		}

		trx = UT_LIST_GET_NEXT(mysql_trx_list, trx);
	}

loop:
	trx = UT_LIST_GET_FIRST(trx_sys->trx_list);

	i = 0;

	/* Since we temporarily release the kernel mutex when
	reading a database page in below, variable trx may be
	obsolete now and we must loop through the trx list to
	get probably the same trx, or some other trx. */

	while (trx && (i < nth_trx)) {
		trx = UT_LIST_GET_NEXT(trx_list, trx);
		i++;
	}

	if (trx == NULL) {
		lock_mutex_exit_kernel();

		ut_ad(lock_validate());

		return;
	}

	if (nth_lock == 0) {
		fputs("---", file);
		trx_print(file, trx, 600);

		if (trx->read_view) {
			fprintf(file,
				"Trx read view will not see trx with"
				" id >= " TRX_ID_FMT
				", sees < " TRX_ID_FMT "\n",
				TRX_ID_PREP_PRINTF(
					trx->read_view->low_limit_id),
				TRX_ID_PREP_PRINTF(
					trx->read_view->up_limit_id));
		}

		if (trx->que_state == TRX_QUE_LOCK_WAIT) {
			fprintf(file,
				"------- TRX HAS BEEN WAITING %lu SEC"
				" FOR THIS LOCK TO BE GRANTED:\n",
				(ulong) difftime(time(NULL),
						 trx->wait_started));

			if (lock_get_type_low(trx->wait_lock) == LOCK_REC) {
				lock_rec_print(file, trx->wait_lock);
			} else {
				lock_table_print(file, trx->wait_lock);
			}

			fputs("------------------\n", file);
		}
	}

4536
        if (!srv_print_innodb_lock_monitor && !srv_show_locks_held) {
4537 4538 4539 4540 4541 4542 4543 4544 4545 4546 4547 4548 4549 4550 4551 4552 4553 4554 4555 4556 4557 4558 4559 4560 4561 4562 4563 4564 4565
		nth_trx++;
		goto loop;
	}

	i = 0;

	/* Look at the note about the trx loop above why we loop here:
	lock may be an obsolete pointer now. */

	lock = UT_LIST_GET_FIRST(trx->trx_locks);

	while (lock && (i < nth_lock)) {
		lock = UT_LIST_GET_NEXT(trx_locks, lock);
		i++;
	}

	if (lock == NULL) {
		nth_trx++;
		nth_lock = 0;

		goto loop;
	}

	if (lock_get_type_low(lock) == LOCK_REC) {
		if (load_page_first) {
			ulint	space	= lock->un_member.rec_lock.space;
			ulint	zip_size= fil_space_get_zip_size(space);
			ulint	page_no = lock->un_member.rec_lock.page_no;

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4566 4567 4568 4569 4570 4571 4572 4573 4574 4575 4576 4577 4578 4579
			if (UNIV_UNLIKELY(zip_size == ULINT_UNDEFINED)) {

				/* It is a single table tablespace and
				the .ibd file is missing (TRUNCATE
				TABLE probably stole the locks): just
				print the lock without attempting to
				load the page in the buffer pool. */

				fprintf(file, "RECORD LOCKS on"
					" non-existing space %lu\n",
					(ulong) space);
				goto print_rec;
			}

4580 4581 4582 4583 4584 4585 4586 4587 4588 4589 4590 4591 4592 4593 4594 4595
			lock_mutex_exit_kernel();

			mtr_start(&mtr);

			buf_page_get_with_no_latch(space, zip_size,
						   page_no, &mtr);

			mtr_commit(&mtr);

			load_page_first = FALSE;

			lock_mutex_enter_kernel();

			goto loop;
		}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4596
print_rec:
4597 4598 4599 4600 4601 4602 4603 4604 4605 4606 4607
		lock_rec_print(file, lock);
	} else {
		ut_ad(lock_get_type_low(lock) & LOCK_TABLE);

		lock_table_print(file, lock);
	}

	load_page_first = TRUE;

	nth_lock++;

4608 4609
	if (nth_lock >= srv_show_locks_held) {
		fputs("TOO LOCKS PRINTED FOR THIS TRX:"
4610 4611 4612 4613 4614 4615 4616 4617 4618 4619 4620 4621
		      " SUPPRESSING FURTHER PRINTS\n",
		      file);

		nth_trx++;
		nth_lock = 0;

		goto loop;
	}

	goto loop;
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4622 4623 4624 4625
#ifdef UNIV_DEBUG
/*********************************************************************//**
Validates the lock queue on a table.
@return	TRUE if ok */
4626 4627 4628 4629
static
ibool
lock_table_queue_validate(
/*======================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4630
	dict_table_t*	table)	/*!< in: table */
4631 4632 4633 4634 4635 4636 4637 4638 4639 4640 4641 4642 4643 4644 4645 4646 4647 4648 4649 4650 4651 4652 4653 4654 4655 4656 4657 4658
{
	lock_t*	lock;

	ut_ad(mutex_own(&kernel_mutex));

	lock = UT_LIST_GET_FIRST(table->locks);

	while (lock) {
		ut_a(((lock->trx)->conc_state == TRX_ACTIVE)
		     || ((lock->trx)->conc_state == TRX_PREPARED)
		     || ((lock->trx)->conc_state == TRX_COMMITTED_IN_MEMORY));

		if (!lock_get_wait(lock)) {

			ut_a(!lock_table_other_has_incompatible(
				     lock->trx, 0, table,
				     lock_get_mode(lock)));
		} else {

			ut_a(lock_table_has_to_wait_in_queue(lock));
		}

		lock = UT_LIST_GET_NEXT(un_member.tab_lock.locks, lock);
	}

	return(TRUE);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4659 4660 4661
/*********************************************************************//**
Validates the lock queue on a single record.
@return	TRUE if ok */
4662 4663 4664 4665
static
ibool
lock_rec_queue_validate(
/*====================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4666 4667 4668 4669
	const buf_block_t*	block,	/*!< in: buffer block containing rec */
	const rec_t*		rec,	/*!< in: record to look at */
	dict_index_t*		index,	/*!< in: index, or NULL if not known */
	const ulint*		offsets)/*!< in: rec_get_offsets(rec, index) */
4670 4671 4672 4673 4674 4675 4676 4677 4678 4679 4680 4681 4682 4683 4684 4685 4686 4687 4688 4689 4690 4691 4692 4693 4694 4695 4696 4697 4698 4699 4700 4701 4702 4703 4704 4705 4706 4707 4708 4709 4710 4711 4712 4713 4714 4715 4716 4717 4718 4719 4720 4721 4722 4723 4724 4725 4726 4727
{
	trx_t*	impl_trx;
	lock_t*	lock;
	ulint	heap_no;

	ut_a(rec);
	ut_a(block->frame == page_align(rec));
	ut_ad(rec_offs_validate(rec, index, offsets));
	ut_ad(!page_rec_is_comp(rec) == !rec_offs_comp(offsets));

	heap_no = page_rec_get_heap_no(rec);

	lock_mutex_enter_kernel();

	if (!page_rec_is_user_rec(rec)) {

		lock = lock_rec_get_first(block, heap_no);

		while (lock) {
			switch(lock->trx->conc_state) {
			case TRX_ACTIVE:
			case TRX_PREPARED:
			case TRX_COMMITTED_IN_MEMORY:
				break;
			default:
				ut_error;
			}

			ut_a(trx_in_trx_list(lock->trx));

			if (lock_get_wait(lock)) {
				ut_a(lock_rec_has_to_wait_in_queue(lock));
			}

			if (index) {
				ut_a(lock->index == index);
			}

			lock = lock_rec_get_next(heap_no, lock);
		}

		lock_mutex_exit_kernel();

		return(TRUE);
	}

	if (!index);
	else if (dict_index_is_clust(index)) {

		impl_trx = lock_clust_rec_some_has_impl(rec, index, offsets);

		if (impl_trx
		    && lock_rec_other_has_expl_req(LOCK_S, 0, LOCK_WAIT,
						   block, heap_no, impl_trx)) {

			ut_a(lock_rec_has_expl(LOCK_X | LOCK_REC_NOT_GAP,
					       block, heap_no, impl_trx));
		}
4728
#if 0
4729 4730 4731 4732 4733 4734
	} else {

		/* The kernel mutex may get released temporarily in the
		next function call: we have to release lock table mutex
		to obey the latching order */

4735 4736 4737 4738
		/* If this thread is holding the file space latch
		(fil_space_t::latch), the following check WILL break
		latching order and may cause a deadlock of threads. */

4739 4740 4741 4742 4743 4744 4745 4746 4747 4748 4749 4750 4751 4752 4753 4754 4755 4756 4757 4758 4759
		/* NOTE: This is a bogus check that would fail in the
		following case: Our transaction is updating a
		row. After it has updated the clustered index record,
		it goes to a secondary index record and finds someone
		else holding an explicit S- or X-lock on that
		secondary index record, presumably from a locking
		read. Our transaction cannot update the secondary
		index immediately, but places a waiting X-lock request
		on the secondary index record. There is nothing
		illegal in this. The assertion is simply too strong. */

		/* From the locking point of view, each secondary
		index is a separate table. A lock that is held on
		secondary index rec does not give any rights to modify
		or read the clustered index rec. Therefore, we can
		think of the sec index as a separate 'table' from the
		clust index 'table'. Conversely, a transaction that
		has acquired a lock on and modified a clustered index
		record may need to wait for a lock on the
		corresponding record in a secondary index. */

4760 4761 4762 4763 4764 4765 4766 4767 4768 4769
		impl_trx = lock_sec_rec_some_has_impl_off_kernel(
			rec, index, offsets);

		if (impl_trx
		    && lock_rec_other_has_expl_req(LOCK_S, 0, LOCK_WAIT,
						   block, heap_no, impl_trx)) {

			ut_a(lock_rec_has_expl(LOCK_X | LOCK_REC_NOT_GAP,
					       block, heap_no, impl_trx));
		}
4770
#endif
4771 4772 4773 4774 4775 4776 4777 4778 4779 4780 4781 4782 4783 4784 4785 4786 4787 4788 4789 4790 4791 4792 4793 4794 4795 4796 4797 4798 4799 4800 4801 4802 4803 4804 4805 4806 4807 4808 4809
	}

	lock = lock_rec_get_first(block, heap_no);

	while (lock) {
		ut_a(lock->trx->conc_state == TRX_ACTIVE
		     || lock->trx->conc_state == TRX_PREPARED
		     || lock->trx->conc_state == TRX_COMMITTED_IN_MEMORY);
		ut_a(trx_in_trx_list(lock->trx));

		if (index) {
			ut_a(lock->index == index);
		}

		if (!lock_rec_get_gap(lock) && !lock_get_wait(lock)) {

			enum lock_mode	mode;

			if (lock_get_mode(lock) == LOCK_S) {
				mode = LOCK_X;
			} else {
				mode = LOCK_S;
			}
			ut_a(!lock_rec_other_has_expl_req(
				     mode, 0, 0, block, heap_no, lock->trx));

		} else if (lock_get_wait(lock) && !lock_rec_get_gap(lock)) {

			ut_a(lock_rec_has_to_wait_in_queue(lock));
		}

		lock = lock_rec_get_next(heap_no, lock);
	}

	lock_mutex_exit_kernel();

	return(TRUE);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4810 4811 4812
/*********************************************************************//**
Validates the record lock queues on a page.
@return	TRUE if ok */
4813 4814 4815 4816
static
ibool
lock_rec_validate_page(
/*===================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4817
	ulint	space,	/*!< in: space id */
4818 4819
	ulint	zip_size,/*!< in: compressed page size in bytes
			or 0 for uncompressed pages */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4820
	ulint	page_no)/*!< in: page number */
4821 4822 4823 4824 4825 4826 4827 4828 4829 4830 4831 4832 4833 4834 4835 4836 4837 4838 4839
{
	dict_index_t*	index;
	buf_block_t*	block;
	const page_t*	page;
	lock_t*		lock;
	const rec_t*	rec;
	ulint		nth_lock	= 0;
	ulint		nth_bit		= 0;
	ulint		i;
	mtr_t		mtr;
	mem_heap_t*	heap		= NULL;
	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
	ulint*		offsets		= offsets_;
	rec_offs_init(offsets_);

	ut_ad(!mutex_own(&kernel_mutex));

	mtr_start(&mtr);

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4840 4841
	ut_ad(zip_size != ULINT_UNDEFINED);
	block = buf_page_get(space, zip_size, page_no, RW_X_LATCH, &mtr);
4842
	buf_block_dbg_add_level(block, SYNC_NO_ORDER_CHECK);
Vadim Tkachenko's avatar
Vadim Tkachenko committed
4843

4844 4845 4846 4847 4848 4849 4850 4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861 4862 4863 4864 4865 4866 4867
	page = block->frame;

	lock_mutex_enter_kernel();
loop:
	lock = lock_rec_get_first_on_page_addr(space, page_no);

	if (!lock) {
		goto function_exit;
	}

	for (i = 0; i < nth_lock; i++) {

		lock = lock_rec_get_next_on_page(lock);

		if (!lock) {
			goto function_exit;
		}
	}

	ut_a(trx_in_trx_list(lock->trx));
	ut_a(lock->trx->conc_state == TRX_ACTIVE
	     || lock->trx->conc_state == TRX_PREPARED
	     || lock->trx->conc_state == TRX_COMMITTED_IN_MEMORY);

4868 4869 4870 4871 4872 4873 4874
# ifdef UNIV_SYNC_DEBUG
	/* Only validate the record queues when this thread is not
	holding a space->latch.  Deadlocks are possible due to
	latching order violation when UNIV_DEBUG is defined while
	UNIV_SYNC_DEBUG is not. */
	if (!sync_thread_levels_contains(SYNC_FSP))
# endif /* UNIV_SYNC_DEBUG */
4875 4876 4877 4878 4879 4880 4881 4882 4883 4884 4885 4886 4887 4888 4889 4890
	for (i = nth_bit; i < lock_rec_get_n_bits(lock); i++) {

		if (i == 1 || lock_rec_get_nth_bit(lock, i)) {

			index = lock->index;
			rec = page_find_rec_with_heap_no(page, i);
			ut_a(rec);
			offsets = rec_get_offsets(rec, index, offsets,
						  ULINT_UNDEFINED, &heap);

			fprintf(stderr,
				"Validating %lu %lu\n",
				(ulong) space, (ulong) page_no);

			lock_mutex_exit_kernel();

4891 4892 4893 4894 4895
			/* If this thread is holding the file space
			latch (fil_space_t::latch), the following
			check WILL break the latching order and may
			cause a deadlock of threads. */

4896 4897 4898 4899 4900 4901 4902 4903 4904 4905 4906 4907 4908 4909 4910 4911 4912 4913 4914 4915 4916 4917 4918 4919 4920 4921
			lock_rec_queue_validate(block, rec, index, offsets);

			lock_mutex_enter_kernel();

			nth_bit = i + 1;

			goto loop;
		}
	}

	nth_bit = 0;
	nth_lock++;

	goto loop;

function_exit:
	lock_mutex_exit_kernel();

	mtr_commit(&mtr);

	if (UNIV_LIKELY_NULL(heap)) {
		mem_heap_free(heap);
	}
	return(TRUE);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
4922 4923 4924
/*********************************************************************//**
Validates the lock system.
@return	TRUE if ok */
4925 4926 4927 4928 4929 4930 4931 4932 4933 4934 4935 4936 4937 4938 4939 4940 4941 4942 4943 4944 4945 4946 4947 4948 4949 4950 4951 4952 4953 4954 4955 4956 4957 4958 4959 4960 4961 4962 4963 4964 4965 4966 4967 4968 4969 4970 4971 4972 4973 4974 4975 4976 4977 4978 4979 4980 4981 4982 4983 4984 4985
static
ibool
lock_validate(void)
/*===============*/
{
	lock_t*	lock;
	trx_t*	trx;
	dulint	limit;
	ulint	space;
	ulint	page_no;
	ulint	i;

	lock_mutex_enter_kernel();

	trx = UT_LIST_GET_FIRST(trx_sys->trx_list);

	while (trx) {
		lock = UT_LIST_GET_FIRST(trx->trx_locks);

		while (lock) {
			if (lock_get_type_low(lock) & LOCK_TABLE) {

				lock_table_queue_validate(
					lock->un_member.tab_lock.table);
			}

			lock = UT_LIST_GET_NEXT(trx_locks, lock);
		}

		trx = UT_LIST_GET_NEXT(trx_list, trx);
	}

	for (i = 0; i < hash_get_n_cells(lock_sys->rec_hash); i++) {

		limit = ut_dulint_zero;

		for (;;) {
			lock = HASH_GET_FIRST(lock_sys->rec_hash, i);

			while (lock) {
				ut_a(trx_in_trx_list(lock->trx));

				space = lock->un_member.rec_lock.space;
				page_no = lock->un_member.rec_lock.page_no;

				if (ut_dulint_cmp(
					    ut_dulint_create(space, page_no),
					    limit) >= 0) {
					break;
				}

				lock = HASH_GET_NEXT(hash, lock);
			}

			if (!lock) {

				break;
			}

			lock_mutex_exit_kernel();

4986 4987 4988
			lock_rec_validate_page(space,
					       fil_space_get_zip_size(space),
					       page_no);
4989 4990 4991 4992 4993 4994 4995 4996 4997 4998 4999

			lock_mutex_enter_kernel();

			limit = ut_dulint_create(space, page_no + 1);
		}
	}

	lock_mutex_exit_kernel();

	return(TRUE);
}
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5000
#endif /* UNIV_DEBUG */
5001 5002
/*============ RECORD LOCK CHECKS FOR ROW OPERATIONS ====================*/

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5003
/*********************************************************************//**
5004 5005 5006 5007
Checks if locks of other transactions prevent an immediate insert of
a record. If they do, first tests if the query thread should anyway
be suspended for some reason; if not, then puts the transaction and
the query thread to the lock wait state and inserts a waiting request
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5008 5009
for a gap x-lock to the lock queue.
@return	DB_SUCCESS, DB_LOCK_WAIT, DB_DEADLOCK, or DB_QUE_THR_SUSPENDED */
5010 5011 5012 5013
UNIV_INTERN
ulint
lock_rec_insert_check_and_lock(
/*===========================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5014
	ulint		flags,	/*!< in: if BTR_NO_LOCKING_FLAG bit is
5015
				set, does nothing */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5016 5017 5018 5019 5020 5021
	const rec_t*	rec,	/*!< in: record after which to insert */
	buf_block_t*	block,	/*!< in/out: buffer block of rec */
	dict_index_t*	index,	/*!< in: index */
	que_thr_t*	thr,	/*!< in: query thread */
	mtr_t*		mtr,	/*!< in/out: mini-transaction */
	ibool*		inherit)/*!< out: set to TRUE if the new
5022 5023 5024 5025 5026 5027 5028 5029 5030 5031 5032 5033 5034 5035 5036 5037 5038 5039
				inserted record maybe should inherit
				LOCK_GAP type locks from the successor
				record */
{
	const rec_t*	next_rec;
	trx_t*		trx;
	lock_t*		lock;
	ulint		err;
	ulint		next_rec_heap_no;

	ut_ad(block->frame == page_align(rec));

	if (flags & BTR_NO_LOCKING_FLAG) {

		return(DB_SUCCESS);
	}

	trx = thr_get_trx(thr);
5040
	next_rec = page_rec_get_next_const(rec);
5041 5042 5043 5044 5045 5046 5047 5048 5049 5050 5051 5052 5053 5054 5055 5056 5057 5058 5059 5060 5061 5062
	next_rec_heap_no = page_rec_get_heap_no(next_rec);

	lock_mutex_enter_kernel();

	/* When inserting a record into an index, the table must be at
	least IX-locked or we must be building an index, in which case
	the table must be at least S-locked. */
	ut_ad(lock_table_has(trx, index->table, LOCK_IX)
	      || (*index->name == TEMP_INDEX_PREFIX
		  && lock_table_has(trx, index->table, LOCK_S)));

	lock = lock_rec_get_first(block, next_rec_heap_no);

	if (UNIV_LIKELY(lock == NULL)) {
		/* We optimize CPU time usage in the simplest case */

		lock_mutex_exit_kernel();

		if (!dict_index_is_clust(index)) {
			/* Update the page max trx id field */
			page_update_max_trx_id(block,
					       buf_block_get_page_zip(block),
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5063
					       trx->id, mtr);
5064 5065 5066 5067 5068 5069 5070 5071 5072 5073 5074 5075 5076 5077 5078 5079 5080 5081 5082 5083 5084 5085 5086 5087 5088 5089 5090 5091 5092 5093 5094 5095 5096 5097 5098 5099 5100 5101
		}

		*inherit = FALSE;

		return(DB_SUCCESS);
	}

	*inherit = TRUE;

	/* If another transaction has an explicit lock request which locks
	the gap, waiting or granted, on the successor, the insert has to wait.

	An exception is the case where the lock by the another transaction
	is a gap type lock which it placed to wait for its turn to insert. We
	do not consider that kind of a lock conflicting with our insert. This
	eliminates an unnecessary deadlock which resulted when 2 transactions
	had to wait for their insert. Both had waiting gap type lock requests
	on the successor, which produced an unnecessary deadlock. */

	if (lock_rec_other_has_conflicting(
		    LOCK_X | LOCK_GAP | LOCK_INSERT_INTENTION,
		    block, next_rec_heap_no, trx)) {

		/* Note that we may get DB_SUCCESS also here! */
		err = lock_rec_enqueue_waiting(LOCK_X | LOCK_GAP
					       | LOCK_INSERT_INTENTION,
					       block, next_rec_heap_no,
					       index, thr);
	} else {
		err = DB_SUCCESS;
	}

	lock_mutex_exit_kernel();

	if ((err == DB_SUCCESS) && !dict_index_is_clust(index)) {
		/* Update the page max trx id field */
		page_update_max_trx_id(block,
				       buf_block_get_page_zip(block),
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5102
				       trx->id, mtr);
5103 5104 5105 5106 5107 5108 5109 5110 5111 5112 5113 5114 5115 5116 5117 5118 5119 5120 5121 5122 5123 5124
	}

#ifdef UNIV_DEBUG
	{
		mem_heap_t*	heap		= NULL;
		ulint		offsets_[REC_OFFS_NORMAL_SIZE];
		const ulint*	offsets;
		rec_offs_init(offsets_);

		offsets = rec_get_offsets(next_rec, index, offsets_,
					  ULINT_UNDEFINED, &heap);
		ut_ad(lock_rec_queue_validate(block,
					      next_rec, index, offsets));
		if (UNIV_LIKELY_NULL(heap)) {
			mem_heap_free(heap);
		}
	}
#endif /* UNIV_DEBUG */

	return(err);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5125
/*********************************************************************//**
5126 5127 5128 5129 5130 5131 5132
If a transaction has an implicit x-lock on a record, but no explicit x-lock
set on the record, sets one for it. NOTE that in the case of a secondary
index, the kernel mutex may get temporarily released. */
static
void
lock_rec_convert_impl_to_expl(
/*==========================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5133 5134 5135 5136
	const buf_block_t*	block,	/*!< in: buffer block of rec */
	const rec_t*		rec,	/*!< in: user record on page */
	dict_index_t*		index,	/*!< in: index of record */
	const ulint*		offsets)/*!< in: rec_get_offsets(rec, index) */
5137 5138 5139 5140 5141 5142 5143 5144 5145 5146 5147 5148 5149 5150 5151 5152 5153 5154 5155 5156 5157 5158 5159 5160 5161 5162 5163 5164 5165 5166 5167
{
	trx_t*	impl_trx;

	ut_ad(mutex_own(&kernel_mutex));
	ut_ad(page_rec_is_user_rec(rec));
	ut_ad(rec_offs_validate(rec, index, offsets));
	ut_ad(!page_rec_is_comp(rec) == !rec_offs_comp(offsets));

	if (dict_index_is_clust(index)) {
		impl_trx = lock_clust_rec_some_has_impl(rec, index, offsets);
	} else {
		impl_trx = lock_sec_rec_some_has_impl_off_kernel(
			rec, index, offsets);
	}

	if (impl_trx) {
		ulint	heap_no = page_rec_get_heap_no(rec);

		/* If the transaction has no explicit x-lock set on the
		record, set one for it */

		if (!lock_rec_has_expl(LOCK_X | LOCK_REC_NOT_GAP, block,
				       heap_no, impl_trx)) {

			lock_rec_add_to_queue(
				LOCK_REC | LOCK_X | LOCK_REC_NOT_GAP,
				block, heap_no, index, impl_trx);
		}
	}
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5168
/*********************************************************************//**
5169 5170 5171 5172 5173
Checks if locks of other transactions prevent an immediate modify (update,
delete mark, or delete unmark) of a clustered index record. If they do,
first tests if the query thread should anyway be suspended for some
reason; if not, then puts the transaction and the query thread to the
lock wait state and inserts a waiting request for a record x-lock to the
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5174 5175
lock queue.
@return	DB_SUCCESS, DB_LOCK_WAIT, DB_DEADLOCK, or DB_QUE_THR_SUSPENDED */
5176 5177 5178 5179
UNIV_INTERN
ulint
lock_clust_rec_modify_check_and_lock(
/*=================================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5180
	ulint			flags,	/*!< in: if BTR_NO_LOCKING_FLAG
5181
					bit is set, does nothing */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5182 5183
	const buf_block_t*	block,	/*!< in: buffer block of rec */
	const rec_t*		rec,	/*!< in: record which should be
5184
					modified */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5185 5186 5187
	dict_index_t*		index,	/*!< in: clustered index */
	const ulint*		offsets,/*!< in: rec_get_offsets(rec, index) */
	que_thr_t*		thr)	/*!< in: query thread */
5188 5189 5190 5191 5192 5193 5194 5195 5196 5197 5198 5199 5200 5201 5202 5203 5204 5205 5206 5207 5208 5209 5210 5211 5212 5213 5214 5215 5216 5217 5218 5219 5220 5221 5222 5223
{
	ulint	err;
	ulint	heap_no;

	ut_ad(rec_offs_validate(rec, index, offsets));
	ut_ad(dict_index_is_clust(index));
	ut_ad(block->frame == page_align(rec));

	if (flags & BTR_NO_LOCKING_FLAG) {

		return(DB_SUCCESS);
	}

	heap_no = rec_offs_comp(offsets)
		? rec_get_heap_no_new(rec)
		: rec_get_heap_no_old(rec);

	lock_mutex_enter_kernel();

	ut_ad(lock_table_has(thr_get_trx(thr), index->table, LOCK_IX));

	/* If a transaction has no explicit x-lock set on the record, set one
	for it */

	lock_rec_convert_impl_to_expl(block, rec, index, offsets);

	err = lock_rec_lock(TRUE, LOCK_X | LOCK_REC_NOT_GAP,
			    block, heap_no, index, thr);

	lock_mutex_exit_kernel();

	ut_ad(lock_rec_queue_validate(block, rec, index, offsets));

	return(err);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5224
/*********************************************************************//**
5225
Checks if locks of other transactions prevent an immediate modify (delete
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5226 5227
mark or delete unmark) of a secondary index record.
@return	DB_SUCCESS, DB_LOCK_WAIT, DB_DEADLOCK, or DB_QUE_THR_SUSPENDED */
5228 5229 5230 5231
UNIV_INTERN
ulint
lock_sec_rec_modify_check_and_lock(
/*===============================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5232
	ulint		flags,	/*!< in: if BTR_NO_LOCKING_FLAG
5233
				bit is set, does nothing */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5234 5235
	buf_block_t*	block,	/*!< in/out: buffer block of rec */
	const rec_t*	rec,	/*!< in: record which should be
5236 5237 5238 5239
				modified; NOTE: as this is a secondary
				index, we always have to modify the
				clustered index record first: see the
				comment below */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5240 5241 5242
	dict_index_t*	index,	/*!< in: secondary index */
	que_thr_t*	thr,	/*!< in: query thread */
	mtr_t*		mtr)	/*!< in/out: mini-transaction */
5243 5244 5245 5246 5247 5248 5249 5250 5251 5252 5253 5254 5255 5256 5257 5258 5259 5260 5261 5262 5263 5264 5265 5266 5267 5268 5269 5270 5271 5272 5273 5274 5275 5276 5277 5278 5279 5280 5281 5282 5283 5284 5285 5286 5287 5288 5289 5290
{
	ulint	err;
	ulint	heap_no;

	ut_ad(!dict_index_is_clust(index));
	ut_ad(block->frame == page_align(rec));

	if (flags & BTR_NO_LOCKING_FLAG) {

		return(DB_SUCCESS);
	}

	heap_no = page_rec_get_heap_no(rec);

	/* Another transaction cannot have an implicit lock on the record,
	because when we come here, we already have modified the clustered
	index record, and this would not have been possible if another active
	transaction had modified this secondary index record. */

	lock_mutex_enter_kernel();

	ut_ad(lock_table_has(thr_get_trx(thr), index->table, LOCK_IX));

	err = lock_rec_lock(TRUE, LOCK_X | LOCK_REC_NOT_GAP,
			    block, heap_no, index, thr);

	lock_mutex_exit_kernel();

#ifdef UNIV_DEBUG
	{
		mem_heap_t*	heap		= NULL;
		ulint		offsets_[REC_OFFS_NORMAL_SIZE];
		const ulint*	offsets;
		rec_offs_init(offsets_);

		offsets = rec_get_offsets(rec, index, offsets_,
					  ULINT_UNDEFINED, &heap);
		ut_ad(lock_rec_queue_validate(block, rec, index, offsets));
		if (UNIV_LIKELY_NULL(heap)) {
			mem_heap_free(heap);
		}
	}
#endif /* UNIV_DEBUG */

	if (err == DB_SUCCESS) {
		/* Update the page max trx id field */
		page_update_max_trx_id(block,
				       buf_block_get_page_zip(block),
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5291
				       thr_get_trx(thr)->id, mtr);
5292 5293 5294 5295 5296
	}

	return(err);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5297
/*********************************************************************//**
5298
Like the counterpart for a clustered index below, but now we read a
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5299 5300
secondary index record.
@return	DB_SUCCESS, DB_LOCK_WAIT, DB_DEADLOCK, or DB_QUE_THR_SUSPENDED */
5301 5302 5303 5304
UNIV_INTERN
ulint
lock_sec_rec_read_check_and_lock(
/*=============================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5305
	ulint			flags,	/*!< in: if BTR_NO_LOCKING_FLAG
5306
					bit is set, does nothing */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5307 5308
	const buf_block_t*	block,	/*!< in: buffer block of rec */
	const rec_t*		rec,	/*!< in: user record or page
5309 5310 5311
					supremum record which should
					be read or passed over by a
					read cursor */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5312 5313 5314
	dict_index_t*		index,	/*!< in: secondary index */
	const ulint*		offsets,/*!< in: rec_get_offsets(rec, index) */
	enum lock_mode		mode,	/*!< in: mode of the lock which
5315 5316 5317 5318
					the read cursor should set on
					records: LOCK_S or LOCK_X; the
					latter is possible in
					SELECT FOR UPDATE */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5319
	ulint			gap_mode,/*!< in: LOCK_ORDINARY, LOCK_GAP, or
5320
					LOCK_REC_NOT_GAP */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5321
	que_thr_t*		thr)	/*!< in: query thread */
5322 5323 5324 5325 5326 5327 5328 5329 5330 5331 5332 5333 5334 5335 5336 5337 5338 5339 5340 5341 5342 5343 5344 5345 5346 5347 5348 5349 5350 5351 5352 5353 5354 5355 5356 5357 5358 5359 5360 5361 5362 5363 5364 5365 5366 5367
{
	ulint	err;
	ulint	heap_no;

	ut_ad(!dict_index_is_clust(index));
	ut_ad(block->frame == page_align(rec));
	ut_ad(page_rec_is_user_rec(rec) || page_rec_is_supremum(rec));
	ut_ad(rec_offs_validate(rec, index, offsets));
	ut_ad(mode == LOCK_X || mode == LOCK_S);

	if (flags & BTR_NO_LOCKING_FLAG) {

		return(DB_SUCCESS);
	}

	heap_no = page_rec_get_heap_no(rec);

	lock_mutex_enter_kernel();

	ut_ad(mode != LOCK_X
	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IX));
	ut_ad(mode != LOCK_S
	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IS));

	/* Some transaction may have an implicit x-lock on the record only
	if the max trx id for the page >= min trx id for the trx list or a
	database recovery is running. */

	if (((ut_dulint_cmp(page_get_max_trx_id(block->frame),
			    trx_list_get_min_trx_id()) >= 0)
	     || recv_recovery_is_on())
	    && !page_rec_is_supremum(rec)) {

		lock_rec_convert_impl_to_expl(block, rec, index, offsets);
	}

	err = lock_rec_lock(FALSE, mode | gap_mode,
			    block, heap_no, index, thr);

	lock_mutex_exit_kernel();

	ut_ad(lock_rec_queue_validate(block, rec, index, offsets));

	return(err);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5368
/*********************************************************************//**
5369 5370 5371 5372 5373
Checks if locks of other transactions prevent an immediate read, or passing
over by a read cursor, of a clustered index record. If they do, first tests
if the query thread should anyway be suspended for some reason; if not, then
puts the transaction and the query thread to the lock wait state and inserts a
waiting request for a record lock to the lock queue. Sets the requested mode
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5374 5375
lock on the record.
@return	DB_SUCCESS, DB_LOCK_WAIT, DB_DEADLOCK, or DB_QUE_THR_SUSPENDED */
5376 5377 5378 5379
UNIV_INTERN
ulint
lock_clust_rec_read_check_and_lock(
/*===============================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5380
	ulint			flags,	/*!< in: if BTR_NO_LOCKING_FLAG
5381
					bit is set, does nothing */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5382 5383
	const buf_block_t*	block,	/*!< in: buffer block of rec */
	const rec_t*		rec,	/*!< in: user record or page
5384 5385 5386
					supremum record which should
					be read or passed over by a
					read cursor */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5387 5388 5389
	dict_index_t*		index,	/*!< in: clustered index */
	const ulint*		offsets,/*!< in: rec_get_offsets(rec, index) */
	enum lock_mode		mode,	/*!< in: mode of the lock which
5390 5391 5392 5393
					the read cursor should set on
					records: LOCK_S or LOCK_X; the
					latter is possible in
					SELECT FOR UPDATE */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5394
	ulint			gap_mode,/*!< in: LOCK_ORDINARY, LOCK_GAP, or
5395
					LOCK_REC_NOT_GAP */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5396
	que_thr_t*		thr)	/*!< in: query thread */
5397 5398 5399 5400 5401 5402 5403 5404 5405 5406 5407 5408 5409 5410 5411 5412 5413 5414 5415 5416 5417 5418 5419 5420 5421 5422 5423 5424 5425 5426 5427 5428 5429 5430 5431 5432 5433 5434 5435
{
	ulint	err;
	ulint	heap_no;

	ut_ad(dict_index_is_clust(index));
	ut_ad(block->frame == page_align(rec));
	ut_ad(page_rec_is_user_rec(rec) || page_rec_is_supremum(rec));
	ut_ad(gap_mode == LOCK_ORDINARY || gap_mode == LOCK_GAP
	      || gap_mode == LOCK_REC_NOT_GAP);
	ut_ad(rec_offs_validate(rec, index, offsets));

	if (flags & BTR_NO_LOCKING_FLAG) {

		return(DB_SUCCESS);
	}

	heap_no = page_rec_get_heap_no(rec);

	lock_mutex_enter_kernel();

	ut_ad(mode != LOCK_X
	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IX));
	ut_ad(mode != LOCK_S
	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IS));

	if (UNIV_LIKELY(heap_no != PAGE_HEAP_NO_SUPREMUM)) {

		lock_rec_convert_impl_to_expl(block, rec, index, offsets);
	}

	err = lock_rec_lock(FALSE, mode | gap_mode,
			    block, heap_no, index, thr);

	lock_mutex_exit_kernel();

	ut_ad(lock_rec_queue_validate(block, rec, index, offsets));

	return(err);
}
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5436
/*********************************************************************//**
5437 5438 5439 5440 5441 5442 5443
Checks if locks of other transactions prevent an immediate read, or passing
over by a read cursor, of a clustered index record. If they do, first tests
if the query thread should anyway be suspended for some reason; if not, then
puts the transaction and the query thread to the lock wait state and inserts a
waiting request for a record lock to the lock queue. Sets the requested mode
lock on the record. This is an alternative version of
lock_clust_rec_read_check_and_lock() that does not require the parameter
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5444 5445
"offsets".
@return	DB_SUCCESS, DB_LOCK_WAIT, DB_DEADLOCK, or DB_QUE_THR_SUSPENDED */
5446 5447 5448 5449
UNIV_INTERN
ulint
lock_clust_rec_read_check_and_lock_alt(
/*===================================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5450
	ulint			flags,	/*!< in: if BTR_NO_LOCKING_FLAG
5451
					bit is set, does nothing */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5452 5453
	const buf_block_t*	block,	/*!< in: buffer block of rec */
	const rec_t*		rec,	/*!< in: user record or page
5454 5455 5456
					supremum record which should
					be read or passed over by a
					read cursor */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5457 5458
	dict_index_t*		index,	/*!< in: clustered index */
	enum lock_mode		mode,	/*!< in: mode of the lock which
5459 5460 5461 5462
					the read cursor should set on
					records: LOCK_S or LOCK_X; the
					latter is possible in
					SELECT FOR UPDATE */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5463
	ulint			gap_mode,/*!< in: LOCK_ORDINARY, LOCK_GAP, or
5464
					LOCK_REC_NOT_GAP */
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5465
	que_thr_t*		thr)	/*!< in: query thread */
5466 5467 5468 5469 5470 5471 5472 5473 5474 5475 5476 5477 5478 5479 5480 5481 5482
{
	mem_heap_t*	tmp_heap	= NULL;
	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
	ulint*		offsets		= offsets_;
	ulint		ret;
	rec_offs_init(offsets_);

	offsets = rec_get_offsets(rec, index, offsets,
				  ULINT_UNDEFINED, &tmp_heap);
	ret = lock_clust_rec_read_check_and_lock(flags, block, rec, index,
						 offsets, mode, gap_mode, thr);
	if (tmp_heap) {
		mem_heap_free(tmp_heap);
	}
	return(ret);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5483
/*******************************************************************//**
Vadim Tkachenko's avatar
Vadim Tkachenko committed
5484 5485 5486 5487 5488
Release the last lock from the transaction's autoinc locks. */
UNIV_INLINE
void
lock_release_autoinc_last_lock(
/*===========================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5489
	ib_vector_t*	autoinc_locks)	/*!< in/out: vector of AUTOINC locks */
Vadim Tkachenko's avatar
Vadim Tkachenko committed
5490 5491 5492 5493 5494 5495 5496 5497 5498 5499 5500 5501 5502 5503 5504 5505 5506 5507 5508 5509 5510
{
	ulint		last;
	lock_t*		lock;

	ut_ad(mutex_own(&kernel_mutex));
	ut_a(!ib_vector_is_empty(autoinc_locks));

	/* The lock to be release must be the last lock acquired. */
	last = ib_vector_size(autoinc_locks) - 1;
	lock = ib_vector_get(autoinc_locks, last);

	/* Should have only AUTOINC locks in the vector. */
	ut_a(lock_get_mode(lock) == LOCK_AUTO_INC);
	ut_a(lock_get_type(lock) == LOCK_TABLE);

	ut_a(lock->un_member.tab_lock.table != NULL);

	/* This will remove the lock from the trx autoinc_locks too. */
	lock_table_dequeue(lock);
}

5511 5512 5513 5514 5515 5516 5517 5518 5519 5520 5521 5522 5523 5524
/*******************************************************************//**
Check if a transaction holds any autoinc locks. 
@return TRUE if the transaction holds any AUTOINC locks. */
UNIV_INTERN
ibool
lock_trx_holds_autoinc_locks(
/*=========================*/
	const trx_t*	trx)		/*!< in: transaction */
{
	ut_a(trx->autoinc_locks != NULL);

	return(!ib_vector_is_empty(trx->autoinc_locks));
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5525
/*******************************************************************//**
Vadim Tkachenko's avatar
Vadim Tkachenko committed
5526 5527 5528 5529 5530
Release all the transaction's autoinc locks. */
UNIV_INTERN
void
lock_release_autoinc_locks(
/*=======================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5531
	trx_t*		trx)		/*!< in/out: transaction */
Vadim Tkachenko's avatar
Vadim Tkachenko committed
5532 5533 5534 5535 5536 5537 5538 5539 5540 5541 5542 5543 5544 5545 5546 5547 5548 5549 5550
{
	ut_ad(mutex_own(&kernel_mutex));

	ut_a(trx->autoinc_locks != NULL);

	/* We release the locks in the reverse order. This is to
	avoid searching the vector for the element to delete at
	the lower level. See (lock_table_remove_low()) for details. */
	while (!ib_vector_is_empty(trx->autoinc_locks)) {

		/* lock_table_remove_low() will also remove the lock from
		the transaction's autoinc_locks vector. */
		lock_release_autoinc_last_lock(trx->autoinc_locks);
	}

	/* Should release all locks. */
	ut_a(ib_vector_is_empty(trx->autoinc_locks));
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5551
/*******************************************************************//**
5552
Gets the type of a lock. Non-inline version for using outside of the
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5553 5554
lock module.
@return	LOCK_TABLE or LOCK_REC */
5555 5556 5557 5558
UNIV_INTERN
ulint
lock_get_type(
/*==========*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5559
	const lock_t*	lock)	/*!< in: lock */
5560 5561 5562 5563
{
	return(lock_get_type_low(lock));
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5564 5565 5566
/*******************************************************************//**
Gets the id of the transaction owning a lock.
@return	transaction id */
5567 5568 5569 5570
UNIV_INTERN
ullint
lock_get_trx_id(
/*============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5571
	const lock_t*	lock)	/*!< in: lock */
5572 5573 5574 5575
{
	return(trx_get_id(lock->trx));
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5576
/*******************************************************************//**
5577
Gets the mode of a lock in a human readable string.
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5578 5579
The string should not be free()'d or modified.
@return	lock mode */
5580 5581 5582 5583
UNIV_INTERN
const char*
lock_get_mode_str(
/*==============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5584
	const lock_t*	lock)	/*!< in: lock */
5585 5586 5587 5588 5589 5590 5591 5592 5593 5594 5595 5596 5597 5598 5599 5600 5601 5602 5603 5604 5605 5606 5607 5608 5609 5610 5611 5612 5613 5614 5615 5616 5617 5618 5619 5620 5621 5622
{
	ibool	is_gap_lock;

	is_gap_lock = lock_get_type_low(lock) == LOCK_REC
		&& lock_rec_get_gap(lock);

	switch (lock_get_mode(lock)) {
	case LOCK_S:
		if (is_gap_lock) {
			return("S,GAP");
		} else {
			return("S");
		}
	case LOCK_X:
		if (is_gap_lock) {
			return("X,GAP");
		} else {
			return("X");
		}
	case LOCK_IS:
		if (is_gap_lock) {
			return("IS,GAP");
		} else {
			return("IS");
		}
	case LOCK_IX:
		if (is_gap_lock) {
			return("IX,GAP");
		} else {
			return("IX");
		}
	case LOCK_AUTO_INC:
		return("AUTO_INC");
	default:
		return("UNKNOWN");
	}
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5623
/*******************************************************************//**
5624
Gets the type of a lock in a human readable string.
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5625 5626
The string should not be free()'d or modified.
@return	lock type */
5627 5628 5629 5630
UNIV_INTERN
const char*
lock_get_type_str(
/*==============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5631
	const lock_t*	lock)	/*!< in: lock */
5632 5633 5634 5635 5636 5637 5638 5639 5640 5641 5642
{
	switch (lock_get_type_low(lock)) {
	case LOCK_REC:
		return("RECORD");
	case LOCK_TABLE:
		return("TABLE");
	default:
		return("UNKNOWN");
	}
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5643 5644 5645
/*******************************************************************//**
Gets the table on which the lock is.
@return	table */
5646 5647 5648 5649
UNIV_INLINE
dict_table_t*
lock_get_table(
/*===========*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5650
	const lock_t*	lock)	/*!< in: lock */
5651 5652 5653 5654 5655 5656 5657 5658 5659 5660 5661 5662
{
	switch (lock_get_type_low(lock)) {
	case LOCK_REC:
		return(lock->index->table);
	case LOCK_TABLE:
		return(lock->un_member.tab_lock.table);
	default:
		ut_error;
		return(NULL);
	}
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5663 5664 5665
/*******************************************************************//**
Gets the id of the table on which the lock is.
@return	id of the table */
5666 5667 5668 5669
UNIV_INTERN
ullint
lock_get_table_id(
/*==============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5670
	const lock_t*	lock)	/*!< in: lock */
5671 5672 5673 5674 5675 5676 5677 5678
{
	dict_table_t*	table;

	table = lock_get_table(lock);

	return((ullint)ut_conv_dulint_to_longlong(table->id));
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5679
/*******************************************************************//**
5680
Gets the name of the table on which the lock is.
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5681 5682
The string should not be free()'d or modified.
@return	name of the table */
5683 5684 5685 5686
UNIV_INTERN
const char*
lock_get_table_name(
/*================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5687
	const lock_t*	lock)	/*!< in: lock */
5688 5689 5690 5691 5692 5693 5694 5695
{
	dict_table_t*	table;

	table = lock_get_table(lock);

	return(table->name);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5696 5697 5698
/*******************************************************************//**
For a record lock, gets the index on which the lock is.
@return	index */
5699 5700 5701 5702
UNIV_INTERN
const dict_index_t*
lock_rec_get_index(
/*===============*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5703
	const lock_t*	lock)	/*!< in: lock */
5704 5705 5706 5707 5708 5709
{
	ut_a(lock_get_type_low(lock) == LOCK_REC);

	return(lock->index);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5710
/*******************************************************************//**
5711
For a record lock, gets the name of the index on which the lock is.
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5712 5713
The string should not be free()'d or modified.
@return	name of the index */
5714 5715 5716 5717
UNIV_INTERN
const char*
lock_rec_get_index_name(
/*====================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5718
	const lock_t*	lock)	/*!< in: lock */
5719 5720 5721 5722 5723 5724
{
	ut_a(lock_get_type_low(lock) == LOCK_REC);

	return(lock->index->name);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5725 5726 5727
/*******************************************************************//**
For a record lock, gets the tablespace number on which the lock is.
@return	tablespace number */
5728 5729 5730 5731
UNIV_INTERN
ulint
lock_rec_get_space_id(
/*==================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5732
	const lock_t*	lock)	/*!< in: lock */
5733 5734 5735 5736 5737 5738
{
	ut_a(lock_get_type_low(lock) == LOCK_REC);

	return(lock->un_member.rec_lock.space);
}

Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5739 5740 5741
/*******************************************************************//**
For a record lock, gets the page number on which the lock is.
@return	page number */
5742 5743 5744 5745
UNIV_INTERN
ulint
lock_rec_get_page_no(
/*=================*/
Aleksandr Kuzminsky's avatar
Aleksandr Kuzminsky committed
5746
	const lock_t*	lock)	/*!< in: lock */
5747 5748 5749 5750 5751
{
	ut_a(lock_get_type_low(lock) == LOCK_REC);

	return(lock->un_member.rec_lock.page_no);
}