lock0lock.c 152 KB
Newer Older
osku's avatar
osku committed
1 2 3 4 5 6 7 8
/******************************************************
The transaction lock system

(c) 1996 Innobase Oy

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

9 10
#define LOCK_MODULE_IMPLEMENTATION

osku's avatar
osku committed
11
#include "lock0lock.h"
12
#include "lock0priv.h"
osku's avatar
osku committed
13 14 15

#ifdef UNIV_NONINL
#include "lock0lock.ic"
16
#include "lock0priv.ic"
osku's avatar
osku committed
17 18
#endif

vasil's avatar
vasil committed
19
#include "ha_prototypes.h"
osku's avatar
osku committed
20 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
#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
46
record from read or update.
osku's avatar
osku committed
47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279

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
move the read cursor backward from the the next record for it to pass over
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
280 281 282 283 284
 * IS +	 +  +  -  +
 * IX +	 +  -  -  +
 * S  +	 -  +  -  -
 * X  -	 -  -  -  -
 * AI +	 +  -  -  -
osku's avatar
osku committed
285 286 287 288 289 290 291 292
 *
 * 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().
 */
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
#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)
osku's avatar
osku committed
326 327

#ifdef UNIV_DEBUG
328
UNIV_INTERN ibool	lock_print_waits	= FALSE;
329 330 331 332 333 334 335 336

/*************************************************************************
Validates the lock system. */
static
ibool
lock_validate(void);
/*===============*/
			/* out: TRUE if ok */
337 338 339 340 341 342 343 344 345 346

/*************************************************************************
Validates the record lock queues on a page. */
static
ibool
lock_rec_validate_page(
/*===================*/
			/* out: TRUE if ok */
	ulint	space,	/* in: space id */
	ulint	page_no);/* in: page number */
347 348 349

/* Define the following in order to enable lock_rec_validate_page() checks. */
# undef UNIV_DEBUG_LOCK_VALIDATE
osku's avatar
osku committed
350 351 352
#endif /* UNIV_DEBUG */

/* The lock system */
353
UNIV_INTERN lock_sys_t*	lock_sys	= NULL;
osku's avatar
osku committed
354 355 356

/* We store info on the latest deadlock error to this buffer. InnoDB
Monitor will then fetch it and print */
357 358
UNIV_INTERN ibool	lock_deadlock_found = FALSE;
UNIV_INTERN FILE*	lock_latest_err_file;
osku's avatar
osku committed
359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395

/* Flags for recursive deadlock search */
#define LOCK_VICTIM_IS_START	1
#define LOCK_VICTIM_IS_OTHER	2

/************************************************************************
Checks if a lock request results in a deadlock. */
static
ibool
lock_deadlock_occurs(
/*=================*/
			/* out: 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) */
	lock_t*	lock,	/* in: lock the transaction is requesting */
	trx_t*	trx);	/* in: transaction */
/************************************************************************
Looks recursively for a deadlock. */
static
ulint
lock_deadlock_recursive(
/*====================*/
				/* out: 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
				deadlock! */
	trx_t*	start,		/* in: recursion starting point */
	trx_t*	trx,		/* in: a transaction waiting for a lock */
	lock_t*	wait_lock,	/* in: the lock trx is waiting to be granted */
	ulint*	cost,		/* in/out: number of calculation steps thus
				far: if this exceeds LOCK_MAX_N_STEPS_...
				we return LOCK_VICTIM_IS_START */
396
	ulint	depth);		/* in: recursion depth: if this exceeds
osku's avatar
osku committed
397 398 399 400 401 402 403 404 405
				LOCK_MAX_DEPTH_IN_DEADLOCK_CHECK, we
				return LOCK_VICTIM_IS_START */

/*************************************************************************
Gets the nth bit of a record lock. */
UNIV_INLINE
ibool
lock_rec_get_nth_bit(
/*=================*/
406 407 408
				/* out: TRUE if bit set */
	const lock_t*	lock,	/* in: record lock */
	ulint		i)	/* in: index of the bit */
osku's avatar
osku committed
409 410 411 412 413
{
	ulint	byte_index;
	ulint	bit_index;

	ut_ad(lock);
414
	ut_ad(lock_get_type_low(lock) == LOCK_REC);
osku's avatar
osku committed
415 416 417 418 419 420 421 422 423

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

		return(FALSE);
	}

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

424
	return(1 & ((const byte*) &lock[1])[byte_index] >> bit_index);
425
}
osku's avatar
osku committed
426 427 428 429 430 431 432 433

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

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

/*************************************************************************
Checks that a transaction id is sensible, i.e., not in the future. */
434
UNIV_INTERN
osku's avatar
osku committed
435 436 437 438 439
ibool
lock_check_trx_id_sanity(
/*=====================*/
					/* out: TRUE if ok */
	dulint		trx_id,		/* in: trx id */
440
	const rec_t*	rec,		/* in: user record */
osku's avatar
osku committed
441 442 443 444 445 446
	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
					kernel mutex */
{
	ibool	is_ok		= TRUE;
447

osku's avatar
osku committed
448 449 450 451 452 453 454 455 456 457 458
	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);
459
		fputs("  InnoDB: Error: transaction id associated"
460 461
		      " with record\n",
		      stderr);
osku's avatar
osku committed
462 463 464 465
		rec_print_new(stderr, rec, offsets);
		fputs("InnoDB: in ", stderr);
		dict_index_name_print(stderr, NULL, index);
		fprintf(stderr, "\n"
vasil's avatar
vasil committed
466 467
			"InnoDB: is " TRX_ID_FMT " which is higher than the"
			" global trx id counter " TRX_ID_FMT "!\n"
468 469
			"InnoDB: The table is corrupt. You have to do"
			" dump + drop + reimport.\n",
vasil's avatar
vasil committed
470 471
			TRX_ID_PREP_PRINTF(trx_id),
			TRX_ID_PREP_PRINTF(trx_sys->max_trx_id));
osku's avatar
osku committed
472 473 474

		is_ok = FALSE;
	}
475

osku's avatar
osku committed
476 477 478 479 480 481 482 483 484
	if (!has_kernel_mutex) {
		mutex_exit(&kernel_mutex);
	}

	return(is_ok);
}

/*************************************************************************
Checks that a record is seen in a consistent read. */
485
UNIV_INTERN
osku's avatar
osku committed
486 487 488 489 490
ibool
lock_clust_rec_cons_read_sees(
/*==========================*/
				/* out: TRUE if sees, or FALSE if an earlier
				version of the record should be retrieved */
491
	const rec_t*	rec,	/* in: user record which should be read or
osku's avatar
osku committed
492 493 494 495 496 497 498
				passed over by a read cursor */
	dict_index_t*	index,	/* in: clustered index */
	const ulint*	offsets,/* in: rec_get_offsets(rec, index) */
	read_view_t*	view)	/* in: consistent read view */
{
	dulint	trx_id;

499
	ut_ad(dict_index_is_clust(index));
osku's avatar
osku committed
500 501 502 503 504 505 506 507
	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);
508

osku's avatar
osku committed
509 510 511 512 513
	return(read_view_sees_trx_id(view, trx_id));
}

/*************************************************************************
Checks that a non-clustered index record is seen in a consistent read. */
514
UNIV_INTERN
osku's avatar
osku committed
515 516 517
ulint
lock_sec_rec_cons_read_sees(
/*========================*/
518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534
					/* out: TRUE if certainly
					sees, or FALSE if an earlier
					version of the clustered index
					record might be needed: 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 */
	const rec_t*		rec,	/* in: user record which
					should be read or passed over
					by a read cursor */
	const read_view_t*	view)	/* in: consistent read view */
osku's avatar
osku committed
535 536
{
	dulint	max_trx_id;
537

osku's avatar
osku committed
538 539 540 541 542 543 544 545 546 547 548
	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);
	}

549
	max_trx_id = page_get_max_trx_id(page_align(rec));
osku's avatar
osku committed
550

551
	return(ut_dulint_cmp(max_trx_id, view->up_limit_id) < 0);
osku's avatar
osku committed
552 553 554 555
}

/*************************************************************************
Creates the lock system at database start. */
556
UNIV_INTERN
osku's avatar
osku committed
557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573
void
lock_sys_create(
/*============*/
	ulint	n_cells)	/* in: number of slots in lock hash table */
{
	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);
}

/*************************************************************************
Gets the size of a lock struct. */
574
UNIV_INTERN
osku's avatar
osku committed
575 576 577 578 579 580 581 582 583 584 585
ulint
lock_get_size(void)
/*===============*/
			/* out: size in bytes */
{
	return((ulint)sizeof(lock_t));
}

/*************************************************************************
Gets the mode of a lock. */
UNIV_INLINE
586
enum lock_mode
osku's avatar
osku committed
587 588
lock_get_mode(
/*==========*/
589 590
				/* out: mode */
	const lock_t*	lock)	/* in: lock */
osku's avatar
osku committed
591 592 593 594 595 596 597 598 599 600 601 602
{
	ut_ad(lock);

	return(lock->type_mode & LOCK_MODE_MASK);
}

/*************************************************************************
Gets the wait flag of a lock. */
UNIV_INLINE
ibool
lock_get_wait(
/*==========*/
603 604
				/* out: TRUE if waiting */
	const lock_t*	lock)	/* in: lock */
osku's avatar
osku committed
605 606 607
{
	ut_ad(lock);

608
	if (UNIV_UNLIKELY(lock->type_mode & LOCK_WAIT)) {
osku's avatar
osku committed
609 610 611 612 613 614 615 616 617 618

		return(TRUE);
	}

	return(FALSE);
}

/*************************************************************************
Gets the source table of an ALTER TABLE transaction.  The table must be
covered by an IX or IS table lock. */
619
UNIV_INTERN
osku's avatar
osku committed
620 621 622 623 624 625 626 627 628 629
dict_table_t*
lock_get_src_table(
/*===============*/
				/* out: 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 */
	trx_t*		trx,	/* in: transaction */
	dict_table_t*	dest,	/* in: destination of ALTER TABLE */
630
	enum lock_mode*	mode)	/* out: lock mode of the source table */
osku's avatar
osku committed
631 632 633 634 635 636 637 638 639 640 641
{
	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;
642
		enum lock_mode	lock_mode;
643
		if (!(lock_get_type_low(lock) & LOCK_TABLE)) {
osku's avatar
osku committed
644 645 646 647 648 649 650 651 652 653
			/* 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;
654 655
			if (UT_LIST_GET_LEN(src->locks) != 1
			    || UT_LIST_GET_FIRST(src->locks) != lock) {
osku's avatar
osku committed
656 657 658 659 660 661 662 663 664 665 666 667 668
				/* 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);
669
		if (lock_mode == LOCK_IX || lock_mode == LOCK_IS) {
osku's avatar
osku committed
670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689
			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);
}

/*************************************************************************
Determine if the given table is exclusively "owned" by the given
transaction, i.e., transaction holds LOCK_IX and possibly LOCK_AUTO_INC
on the table. */
690
UNIV_INTERN
osku's avatar
osku committed
691 692 693 694 695 696 697 698
ibool
lock_is_table_exclusive(
/*====================*/
				/* out: TRUE if table is only locked by trx,
				with LOCK_IX, and possibly LOCK_AUTO_INC */
	dict_table_t*	table,	/* in: table */
	trx_t*		trx)	/* in: transaction */
{
699 700
	const lock_t*	lock;
	ibool		ok	= FALSE;
osku's avatar
osku committed
701 702 703 704 705 706 707 708 709 710 711 712

	ut_ad(table && trx);

	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. */
			return(FALSE);
		}

713
		if (!(lock_get_type_low(lock) & LOCK_TABLE)) {
osku's avatar
osku committed
714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745
			/* 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:
			/* Other table locks than LOCK_IX are not allowed. */
			return(FALSE);
		}
	}

	return(ok);
}

/*************************************************************************
Sets the wait flag of a lock and the back pointer in trx to lock. */
UNIV_INLINE
void
lock_set_lock_and_trx_wait(
/*=======================*/
	lock_t*	lock,	/* in: lock */
	trx_t*	trx)	/* in: trx */
{
	ut_ad(lock);
	ut_ad(trx->wait_lock == NULL);
746

osku's avatar
osku committed
747
	trx->wait_lock = lock;
748
	lock->type_mode |= LOCK_WAIT;
osku's avatar
osku committed
749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765
}

/**************************************************************************
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(
/*=========================*/
	lock_t*	lock)	/* in: record lock */
{
	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;
766
	lock->type_mode &= ~LOCK_WAIT;
osku's avatar
osku committed
767 768 769 770 771 772 773 774
}

/*************************************************************************
Gets the gap flag of a record lock. */
UNIV_INLINE
ibool
lock_rec_get_gap(
/*=============*/
775 776
				/* out: TRUE if gap flag set */
	const lock_t*	lock)	/* in: record lock */
osku's avatar
osku committed
777 778
{
	ut_ad(lock);
779
	ut_ad(lock_get_type_low(lock) == LOCK_REC);
osku's avatar
osku committed
780 781 782 783 784 785 786 787 788 789 790 791 792 793 794

	if (lock->type_mode & LOCK_GAP) {

		return(TRUE);
	}

	return(FALSE);
}

/*************************************************************************
Gets the LOCK_REC_NOT_GAP flag of a record lock. */
UNIV_INLINE
ibool
lock_rec_get_rec_not_gap(
/*=====================*/
795 796
				/* out: TRUE if LOCK_REC_NOT_GAP flag set */
	const lock_t*	lock)	/* in: record lock */
osku's avatar
osku committed
797 798
{
	ut_ad(lock);
799
	ut_ad(lock_get_type_low(lock) == LOCK_REC);
osku's avatar
osku committed
800 801 802 803 804 805 806 807 808 809 810 811 812 813 814

	if (lock->type_mode & LOCK_REC_NOT_GAP) {

		return(TRUE);
	}

	return(FALSE);
}

/*************************************************************************
Gets the waiting insert flag of a record lock. */
UNIV_INLINE
ibool
lock_rec_get_insert_intention(
/*==========================*/
815 816
				/* out: TRUE if gap flag set */
	const lock_t*	lock)	/* in: record lock */
osku's avatar
osku committed
817 818
{
	ut_ad(lock);
819
	ut_ad(lock_get_type_low(lock) == LOCK_REC);
osku's avatar
osku committed
820 821 822 823 824 825 826 827 828 829 830 831

	if (lock->type_mode & LOCK_INSERT_INTENTION) {

		return(TRUE);
	}

	return(FALSE);
}

/*************************************************************************
Calculates if lock mode 1 is stronger or equal to lock mode 2. */
UNIV_INLINE
832
ulint
osku's avatar
osku committed
833 834
lock_mode_stronger_or_eq(
/*=====================*/
835 836 837 838
				/* out: nonzero
				if mode1 stronger or equal to mode2 */
	enum lock_mode	mode1,	/* in: lock mode */
	enum lock_mode	mode2)	/* in: lock mode */
osku's avatar
osku committed
839 840
{
	ut_ad(mode1 == LOCK_X || mode1 == LOCK_S || mode1 == LOCK_IX
841
	      || mode1 == LOCK_IS || mode1 == LOCK_AUTO_INC);
osku's avatar
osku committed
842
	ut_ad(mode2 == LOCK_X || mode2 == LOCK_S || mode2 == LOCK_IX
843
	      || mode2 == LOCK_IS || mode2 == LOCK_AUTO_INC);
osku's avatar
osku committed
844

845
	return((LOCK_MODE_STRONGER_OR_EQ) & LK(mode1, mode2));
osku's avatar
osku committed
846 847 848 849 850
}

/*************************************************************************
Calculates if lock mode 1 is compatible with lock mode 2. */
UNIV_INLINE
851
ulint
osku's avatar
osku committed
852 853
lock_mode_compatible(
/*=================*/
854 855 856
			/* out: nonzero if mode1 compatible with mode2 */
	enum lock_mode	mode1,	/* in: lock mode */
	enum lock_mode	mode2)	/* in: lock mode */
osku's avatar
osku committed
857 858
{
	ut_ad(mode1 == LOCK_X || mode1 == LOCK_S || mode1 == LOCK_IX
859
	      || mode1 == LOCK_IS || mode1 == LOCK_AUTO_INC);
osku's avatar
osku committed
860
	ut_ad(mode2 == LOCK_X || mode2 == LOCK_S || mode2 == LOCK_IX
861
	      || mode2 == LOCK_IS || mode2 == LOCK_AUTO_INC);
osku's avatar
osku committed
862

863
	return((LOCK_MODE_COMPATIBILITY) & LK(mode1, mode2));
osku's avatar
osku committed
864 865 866 867 868 869 870 871
}

/*************************************************************************
Checks if a lock request for a new lock has to wait for request lock2. */
UNIV_INLINE
ibool
lock_rec_has_to_wait(
/*=================*/
872 873 874 875 876 877 878 879 880 881 882 883 884 885 886
				/* out: TRUE if new lock has to wait
				for lock2 to be removed */
	const trx_t*	trx,	/* in: trx of new lock */
	ulint		type_mode,/* in: precise mode of the new lock
				to set: LOCK_S or LOCK_X, possibly
				ORed to LOCK_GAP or LOCK_REC_NOT_GAP,
				LOCK_INSERT_INTENTION */
	const lock_t*	lock2,	/* in: another record lock; NOTE that
				it is assumed that this has a lock bit
				set on the same record as in the new
				lock we are setting */
	ibool lock_is_on_supremum)  /* in: TRUE if we are setting the
				lock on the 'supremum' record of an
				index page: we know then that the lock
				request is really for a 'gap' type lock */
osku's avatar
osku committed
887 888
{
	ut_ad(trx && lock2);
889
	ut_ad(lock_get_type_low(lock2) == LOCK_REC);
osku's avatar
osku committed
890 891

	if (trx != lock2->trx
892 893
	    && !lock_mode_compatible(LOCK_MODE_MASK & type_mode,
				     lock_get_mode(lock2))) {
osku's avatar
osku committed
894 895 896 897 898

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

		if ((lock_is_on_supremum || (type_mode & LOCK_GAP))
899
		    && !(type_mode & LOCK_INSERT_INTENTION)) {
osku's avatar
osku committed
900 901

			/* Gap type locks without LOCK_INSERT_INTENTION flag
902 903
			do not need to wait for anything. This is because
			different users can have conflicting lock types
osku's avatar
osku committed
904
			on gaps. */
905

osku's avatar
osku committed
906 907
			return(FALSE);
		}
908

osku's avatar
osku committed
909
		if (!(type_mode & LOCK_INSERT_INTENTION)
910
		    && lock_rec_get_gap(lock2)) {
osku's avatar
osku committed
911 912 913 914 915 916 917 918

			/* 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)
919
		    && lock_rec_get_rec_not_gap(lock2)) {
920

osku's avatar
osku committed
921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938
			/* 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. */
939

osku's avatar
osku committed
940 941 942 943 944 945 946 947 948 949 950
			return(FALSE);
		}

		return(TRUE);
	}

	return(FALSE);
}

/*************************************************************************
Checks if a lock request lock1 has to wait for request lock2. */
951
UNIV_INTERN
osku's avatar
osku committed
952 953 954
ibool
lock_has_to_wait(
/*=============*/
955 956 957 958 959 960 961
				/* out: TRUE if lock1 has to wait for
				lock2 to be removed */
	const lock_t*	lock1,	/* in: waiting lock */
	const lock_t*	lock2)	/* in: another lock; NOTE that it is
				assumed that this has a lock bit set
				on the same record as in lock1 if the
				locks are record locks */
osku's avatar
osku committed
962 963 964 965
{
	ut_ad(lock1 && lock2);

	if (lock1->trx != lock2->trx
966 967
	    && !lock_mode_compatible(lock_get_mode(lock1),
				     lock_get_mode(lock2))) {
968 969
		if (lock_get_type_low(lock1) == LOCK_REC) {
			ut_ad(lock_get_type_low(lock2) == LOCK_REC);
osku's avatar
osku committed
970 971 972

			/* If this lock request is for a supremum record
			then the second bit on the lock bitmap is set */
973

osku's avatar
osku committed
974
			return(lock_rec_has_to_wait(lock1->trx,
975
						    lock1->type_mode, lock2,
976 977
						    lock_rec_get_nth_bit(
							    lock1, 1)));
osku's avatar
osku committed
978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993
		}

		return(TRUE);
	}

	return(FALSE);
}

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

/*************************************************************************
Gets the number of bits in a record lock bitmap. */
UNIV_INLINE
ulint
lock_rec_get_n_bits(
/*================*/
994 995
				/* out: number of bits */
	const lock_t*	lock)	/* in: record lock */
osku's avatar
osku committed
996 997 998 999 1000 1001 1002 1003 1004
{
	return(lock->un_member.rec_lock.n_bits);
}

/**************************************************************************
Sets the nth bit of a record lock to TRUE. */
UNIV_INLINE
void
lock_rec_set_nth_bit(
1005
/*=================*/
osku's avatar
osku committed
1006 1007 1008 1009 1010
	lock_t*	lock,	/* in: record lock */
	ulint	i)	/* in: index of the bit */
{
	ulint	byte_index;
	ulint	bit_index;
1011

osku's avatar
osku committed
1012
	ut_ad(lock);
1013
	ut_ad(lock_get_type_low(lock) == LOCK_REC);
osku's avatar
osku committed
1014
	ut_ad(i < lock->un_member.rec_lock.n_bits);
1015

osku's avatar
osku committed
1016 1017 1018
	byte_index = i / 8;
	bit_index = i % 8;

1019
	((byte*) &lock[1])[byte_index] |= 1 << bit_index;
1020
}
osku's avatar
osku committed
1021 1022 1023 1024

/**************************************************************************
Looks for a set bit in a record lock bitmap. Returns ULINT_UNDEFINED,
if none found. */
1025
UNIV_INTERN
osku's avatar
osku committed
1026 1027 1028
ulint
lock_rec_find_set_bit(
/*==================*/
1029 1030 1031
				/* out: bit index == heap number of
				the record, or ULINT_UNDEFINED if none found */
	const lock_t*	lock)	/* in: record lock with at least one bit set */
osku's avatar
osku committed
1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057
{
	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);
}

/**************************************************************************
Resets the nth bit of a record lock. */
UNIV_INLINE
void
lock_rec_reset_nth_bit(
/*===================*/
	lock_t*	lock,	/* in: record lock */
	ulint	i)	/* in: index of the bit which must be set to TRUE
			when this function is called */
{
	ulint	byte_index;
	ulint	bit_index;
1058

osku's avatar
osku committed
1059
	ut_ad(lock);
1060
	ut_ad(lock_get_type_low(lock) == LOCK_REC);
osku's avatar
osku committed
1061
	ut_ad(i < lock->un_member.rec_lock.n_bits);
1062

osku's avatar
osku committed
1063 1064 1065
	byte_index = i / 8;
	bit_index = i % 8;

1066
	((byte*) &lock[1])[byte_index] &= ~(1 << bit_index);
1067
}
osku's avatar
osku committed
1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081

/*************************************************************************
Gets the first or next record lock on a page. */
UNIV_INLINE
lock_t*
lock_rec_get_next_on_page(
/*======================*/
			/* out: next lock, NULL if none exists */
	lock_t*	lock)	/* in: a record lock */
{
	ulint	space;
	ulint	page_no;

	ut_ad(mutex_own(&kernel_mutex));
1082
	ut_ad(lock_get_type_low(lock) == LOCK_REC);
osku's avatar
osku committed
1083 1084 1085

	space = lock->un_member.rec_lock.space;
	page_no = lock->un_member.rec_lock.page_no;
1086

osku's avatar
osku committed
1087 1088 1089 1090 1091 1092 1093 1094
	for (;;) {
		lock = HASH_GET_NEXT(hash, lock);

		if (!lock) {

			break;
		}

1095
		if ((lock->un_member.rec_lock.space == space)
1096
		    && (lock->un_member.rec_lock.page_no == page_no)) {
osku's avatar
osku committed
1097 1098 1099 1100

			break;
		}
	}
1101

osku's avatar
osku committed
1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120
	return(lock);
}

/*************************************************************************
Gets the first record lock on a page, where the page is identified by its
file address. */
UNIV_INLINE
lock_t*
lock_rec_get_first_on_page_addr(
/*============================*/
			/* out: first lock, NULL if none exists */
	ulint	space,	/* in: space */
	ulint	page_no)/* in: page number */
{
	lock_t*	lock;

	ut_ad(mutex_own(&kernel_mutex));

	lock = HASH_GET_FIRST(lock_sys->rec_hash,
1121
			      lock_rec_hash(space, page_no));
osku's avatar
osku committed
1122
	while (lock) {
1123
		if ((lock->un_member.rec_lock.space == space)
1124
		    && (lock->un_member.rec_lock.page_no == page_no)) {
osku's avatar
osku committed
1125 1126 1127 1128 1129 1130 1131 1132 1133

			break;
		}

		lock = HASH_GET_NEXT(hash, lock);
	}

	return(lock);
}
1134

osku's avatar
osku committed
1135 1136
/*************************************************************************
Returns TRUE if there are explicit record locks on a page. */
1137
UNIV_INTERN
osku's avatar
osku committed
1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156
ibool
lock_rec_expl_exist_on_page(
/*========================*/
			/* out: TRUE if there are explicit record locks on
			the page */
	ulint	space,	/* in: space id */
	ulint	page_no)/* in: page number */
{
	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);
1157

osku's avatar
osku committed
1158 1159 1160 1161 1162 1163 1164 1165 1166 1167
	return(ret);
}

/*************************************************************************
Gets the first record lock on a page, where the page is identified by a
pointer to it. */
UNIV_INLINE
lock_t*
lock_rec_get_first_on_page(
/*=======================*/
1168 1169 1170
					/* out: first lock, NULL if
					none exists */
	const buf_block_t*	block)	/* in: buffer block */
osku's avatar
osku committed
1171 1172 1173
{
	ulint	hash;
	lock_t*	lock;
1174 1175
	ulint	space	= buf_block_get_space(block);
	ulint	page_no	= buf_block_get_page_no(block);
osku's avatar
osku committed
1176 1177

	ut_ad(mutex_own(&kernel_mutex));
1178

1179
	hash = buf_block_get_lock_hash_val(block);
osku's avatar
osku committed
1180 1181 1182 1183

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

	while (lock) {
1184
		if ((lock->un_member.rec_lock.space == space)
1185
		    && (lock->un_member.rec_lock.page_no == page_no)) {
osku's avatar
osku committed
1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202

			break;
		}

		lock = HASH_GET_NEXT(hash, lock);
	}

	return(lock);
}

/*************************************************************************
Gets the next explicit lock request on a record. */
UNIV_INLINE
lock_t*
lock_rec_get_next(
/*==============*/
			/* out: next lock, NULL if none exists */
marko's avatar
marko committed
1203
	ulint	heap_no,/* in: heap number of the record */
osku's avatar
osku committed
1204 1205 1206 1207
	lock_t*	lock)	/* in: lock */
{
	ut_ad(mutex_own(&kernel_mutex));

marko's avatar
marko committed
1208
	do {
1209
		ut_ad(lock_get_type_low(lock) == LOCK_REC);
marko's avatar
marko committed
1210 1211
		lock = lock_rec_get_next_on_page(lock);
	} while (lock && !lock_rec_get_nth_bit(lock, heap_no));
osku's avatar
osku committed
1212 1213 1214 1215 1216 1217 1218 1219 1220 1221

	return(lock);
}

/*************************************************************************
Gets the first explicit lock request on a record. */
UNIV_INLINE
lock_t*
lock_rec_get_first(
/*===============*/
1222 1223 1224 1225
					/* out: first lock, NULL if
					none exists */
	const buf_block_t*	block,	/* in: block containing the record */
	ulint			heap_no)/* in: heap number of the record */
osku's avatar
osku committed
1226 1227 1228 1229 1230
{
	lock_t*	lock;

	ut_ad(mutex_own(&kernel_mutex));

1231 1232 1233 1234 1235
	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;
		}
osku's avatar
osku committed
1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252
	}

	return(lock);
}

/*************************************************************************
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(
/*==================*/
	lock_t*	lock)	/* in: record lock */
{
	ulint	n_bytes;

1253
	ut_ad(lock_get_type_low(lock) == LOCK_REC);
osku's avatar
osku committed
1254 1255 1256 1257 1258 1259 1260

	/* 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);
1261

1262
	memset(&lock[1], 0, n_bytes);
osku's avatar
osku committed
1263 1264 1265 1266 1267 1268 1269 1270 1271
}

/*************************************************************************
Copies a record lock to heap. */
static
lock_t*
lock_rec_copy(
/*==========*/
				/* out: copy of lock */
1272
	const lock_t*	lock,	/* in: record lock */
osku's avatar
osku committed
1273 1274 1275 1276
	mem_heap_t*	heap)	/* in: memory heap */
{
	ulint	size;

1277
	ut_ad(lock_get_type_low(lock) == LOCK_REC);
osku's avatar
osku committed
1278

1279
	size = sizeof(lock_t) + lock_rec_get_n_bits(lock) / 8;
osku's avatar
osku committed
1280

1281
	return(mem_heap_dup(heap, lock, size));
osku's avatar
osku committed
1282 1283 1284 1285
}

/*************************************************************************
Gets the previous record lock set on a record. */
1286
UNIV_INTERN
1287
const lock_t*
osku's avatar
osku committed
1288 1289
lock_rec_get_prev(
/*==============*/
1290 1291 1292 1293
				/* out: previous lock on the same
				record, NULL if none exists */
	const lock_t*	in_lock,/* in: record lock */
	ulint		heap_no)/* in: heap number of the record */
osku's avatar
osku committed
1294
{
1295 1296 1297 1298
	lock_t*	lock;
	ulint	space;
	ulint	page_no;
	lock_t*	found_lock	= NULL;
osku's avatar
osku committed
1299 1300

	ut_ad(mutex_own(&kernel_mutex));
1301
	ut_ad(lock_get_type_low(in_lock) == LOCK_REC);
osku's avatar
osku committed
1302 1303 1304 1305 1306 1307 1308 1309

	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);
1310

osku's avatar
osku committed
1311 1312 1313 1314 1315 1316 1317 1318 1319 1320
		if (lock == in_lock) {

			return(found_lock);
		}

		if (lock_rec_get_nth_bit(lock, heap_no)) {

			found_lock = lock;
		}

1321
		lock = lock_rec_get_next_on_page(lock);
1322
	}
osku's avatar
osku committed
1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335
}

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

/*************************************************************************
Checks if a transaction has the specified table lock, or stronger. */
UNIV_INLINE
lock_t*
lock_table_has(
/*===========*/
				/* out: lock or NULL */
	trx_t*		trx,	/* in: transaction */
	dict_table_t*	table,	/* in: table */
1336
	enum lock_mode	mode)	/* in: lock mode */
osku's avatar
osku committed
1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348
{
	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
1349
		    && lock_mode_stronger_or_eq(lock_get_mode(lock), mode)) {
osku's avatar
osku committed
1350

1351
			/* The same trx already has locked the table in
osku's avatar
osku committed
1352 1353
			a mode stronger or equal to the mode given */

1354
			ut_ad(!lock_get_wait(lock));
osku's avatar
osku committed
1355 1356 1357 1358 1359 1360 1361 1362 1363

			return(lock);
		}

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

	return(NULL);
}
1364

osku's avatar
osku committed
1365 1366 1367 1368 1369 1370 1371 1372 1373
/*============= FUNCTIONS FOR ANALYZING RECORD LOCK QUEUE ================*/

/*************************************************************************
Checks if a transaction has a GRANTED explicit lock on rec stronger or equal
to precise_mode. */
UNIV_INLINE
lock_t*
lock_rec_has_expl(
/*==============*/
1374 1375 1376 1377 1378 1379 1380 1381 1382 1383
					/* out: lock or NULL */
	ulint			precise_mode,/* in: LOCK_S or LOCK_X
					possibly ORed to LOCK_GAP or
					LOCK_REC_NOT_GAP, for a
					supremum record we regard this
					always a gap type request */
	const buf_block_t*	block,	/* in: buffer block containing
					the record */
	ulint			heap_no,/* in: heap number of the record */
	trx_t*			trx)	/* in: transaction */
osku's avatar
osku committed
1384 1385 1386 1387 1388
{
	lock_t*	lock;

	ut_ad(mutex_own(&kernel_mutex));
	ut_ad((precise_mode & LOCK_MODE_MASK) == LOCK_S
1389
	      || (precise_mode & LOCK_MODE_MASK) == LOCK_X);
osku's avatar
osku committed
1390
	ut_ad(!(precise_mode & LOCK_INSERT_INTENTION));
1391

1392
	lock = lock_rec_get_first(block, heap_no);
osku's avatar
osku committed
1393 1394 1395

	while (lock) {
		if (lock->trx == trx
1396 1397 1398 1399 1400
		    && 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)
1401
			|| heap_no == PAGE_HEAP_NO_SUPREMUM)
1402 1403
		    && (!lock_rec_get_gap(lock)
			|| (precise_mode & LOCK_GAP)
1404
			|| heap_no == PAGE_HEAP_NO_SUPREMUM)
1405
		    && (!lock_rec_get_insert_intention(lock))) {
osku's avatar
osku committed
1406

1407
			return(lock);
osku's avatar
osku committed
1408 1409
		}

marko's avatar
marko committed
1410
		lock = lock_rec_get_next(heap_no, lock);
osku's avatar
osku committed
1411 1412 1413 1414
	}

	return(NULL);
}
1415

1416 1417
#ifdef UNIV_DEBUG
# ifndef UNIV_HOTBACKUP
osku's avatar
osku committed
1418 1419 1420 1421 1422 1423
/*************************************************************************
Checks if some other transaction has a lock request in the queue. */
static
lock_t*
lock_rec_other_has_expl_req(
/*========================*/
1424
					/* out: lock or NULL */
1425
	enum lock_mode		mode,	/* in: LOCK_S or LOCK_X */
1426 1427 1428 1429 1430 1431
	ulint			gap,	/* in: LOCK_GAP if also gap
					locks are taken into account,
					or 0 if not */
	ulint			wait,	/* in: LOCK_WAIT if also
					waiting locks are taken into
					account, or 0 if not */
1432 1433
	const buf_block_t*	block,	/* in: buffer block containing
					the record */
1434
	ulint			heap_no,/* in: heap number of the record */
1435
	const trx_t*		trx)	/* in: transaction, or NULL if
1436 1437
					requests by all transactions
					are taken into account */
osku's avatar
osku committed
1438 1439
{
	lock_t*	lock;
1440

osku's avatar
osku committed
1441 1442 1443 1444 1445
	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);

1446
	lock = lock_rec_get_first(block, heap_no);
osku's avatar
osku committed
1447 1448 1449

	while (lock) {
		if (lock->trx != trx
1450 1451
		    && (gap
			|| !(lock_rec_get_gap(lock)
1452
			     || heap_no == PAGE_HEAP_NO_SUPREMUM))
1453 1454
		    && (wait || !lock_get_wait(lock))
		    && lock_mode_stronger_or_eq(lock_get_mode(lock), mode)) {
osku's avatar
osku committed
1455

1456
			return(lock);
osku's avatar
osku committed
1457 1458
		}

marko's avatar
marko committed
1459
		lock = lock_rec_get_next(heap_no, lock);
osku's avatar
osku committed
1460 1461 1462 1463
	}

	return(NULL);
}
1464 1465
# endif /* !UNIV_HOTBACKUP */
#endif /* UNIV_DEBUG */
osku's avatar
osku committed
1466 1467 1468 1469 1470 1471 1472 1473

/*************************************************************************
Checks if some other transaction has a conflicting explicit lock request
in the queue, so that we have to wait. */
static
lock_t*
lock_rec_other_has_conflicting(
/*===========================*/
1474
					/* out: lock or NULL */
1475
	enum lock_mode		mode,	/* in: LOCK_S or LOCK_X,
1476 1477 1478 1479 1480 1481 1482
					possibly ORed to LOCK_GAP or
					LOC_REC_NOT_GAP,
					LOCK_INSERT_INTENTION */
	const buf_block_t*	block,	/* in: buffer block containing
					the record */
	ulint			heap_no,/* in: heap number of the record */
	trx_t*			trx)	/* in: our transaction */
osku's avatar
osku committed
1483 1484
{
	lock_t*	lock;
1485

osku's avatar
osku committed
1486 1487
	ut_ad(mutex_own(&kernel_mutex));

1488
	lock = lock_rec_get_first(block, heap_no);
osku's avatar
osku committed
1489

marko's avatar
marko committed
1490
	if (UNIV_LIKELY_NULL(lock)) {
1491
		if (UNIV_UNLIKELY(heap_no == PAGE_HEAP_NO_SUPREMUM)) {
osku's avatar
osku committed
1492

marko's avatar
marko committed
1493 1494
			do {
				if (lock_rec_has_to_wait(trx, mode, lock,
1495
							 TRUE)) {
marko's avatar
marko committed
1496 1497 1498 1499 1500 1501 1502 1503 1504
					return(lock);
				}

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

			do {
				if (lock_rec_has_to_wait(trx, mode, lock,
1505
							 FALSE)) {
marko's avatar
marko committed
1506 1507 1508 1509 1510
					return(lock);
				}

				lock = lock_rec_get_next(heap_no, lock);
			} while (lock);
osku's avatar
osku committed
1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524
		}
	}

	return(NULL);
}

/*************************************************************************
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:
no new struct is needed, if a suitable old is found. */
UNIV_INLINE
lock_t*
lock_rec_find_similar_on_page(
/*==========================*/
1525 1526 1527 1528 1529
					/* out: lock or NULL */
	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 */
osku's avatar
osku committed
1530 1531 1532 1533 1534
{
	ut_ad(mutex_own(&kernel_mutex));

	while (lock != NULL) {
		if (lock->trx == trx
1535 1536
		    && lock->type_mode == type_mode
		    && lock_rec_get_n_bits(lock) > heap_no) {
1537

osku's avatar
osku committed
1538 1539
			return(lock);
		}
1540

osku's avatar
osku committed
1541 1542 1543 1544 1545 1546 1547 1548 1549
		lock = lock_rec_get_next_on_page(lock);
	}

	return(NULL);
}

/*************************************************************************
Checks if some transaction has an implicit x-lock on a record in a secondary
index. */
1550
static
osku's avatar
osku committed
1551 1552 1553 1554 1555
trx_t*
lock_sec_rec_some_has_impl_off_kernel(
/*==================================*/
				/* out: transaction which has the x-lock, or
				NULL */
1556
	const rec_t*	rec,	/* in: user record */
osku's avatar
osku committed
1557 1558 1559
	dict_index_t*	index,	/* in: secondary index */
	const ulint*	offsets)/* in: rec_get_offsets(rec, index) */
{
1560
	const page_t*	page = page_align(rec);
1561

osku's avatar
osku committed
1562
	ut_ad(mutex_own(&kernel_mutex));
1563
	ut_ad(!dict_index_is_clust(index));
osku's avatar
osku committed
1564 1565 1566 1567 1568 1569 1570 1571 1572 1573
	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),
1574 1575
			    trx_list_get_min_trx_id()) >= 0)
	    && !recv_recovery_is_on()) {
osku's avatar
osku committed
1576 1577 1578 1579 1580 1581

		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. */
1582

osku's avatar
osku committed
1583
	if (!lock_check_trx_id_sanity(page_get_max_trx_id(page),
1584
				      rec, index, offsets, TRUE)) {
1585
		buf_page_print(page, 0);
1586

osku's avatar
osku committed
1587 1588 1589 1590 1591 1592 1593 1594
		/* 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));
}

1595 1596 1597 1598
/*************************************************************************
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. */
1599
UNIV_INTERN
1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612
ulint
lock_number_of_rows_locked(
/*=======================*/
	trx_t*	trx)	/* in: transaction */
{
	lock_t*	lock;
	ulint   n_records = 0;
	ulint	n_bits;
	ulint	n_bit;

	lock = UT_LIST_GET_FIRST(trx->trx_locks);

	while (lock) {
1613
		if (lock_get_type_low(lock) == LOCK_REC) {
1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628
			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);
}

osku's avatar
osku committed
1629 1630 1631 1632 1633 1634 1635 1636 1637
/*============== RECORD LOCK CREATION AND QUEUE MANAGEMENT =============*/

/*************************************************************************
Creates a new record lock and inserts it to the lock queue. Does NOT check
for deadlocks or lock compatibility! */
static
lock_t*
lock_rec_create(
/*============*/
1638 1639 1640 1641 1642 1643 1644 1645 1646
					/* out: created lock */
	ulint			type_mode,/* in: lock mode and wait
					flag, type is ignored and
					replaced by LOCK_REC */
	const buf_block_t*	block,	/* in: buffer block containing
					the record */
	ulint			heap_no,/* in: heap number of the record */
	dict_index_t*		index,	/* in: index of record */
	trx_t*			trx)	/* in: transaction */
osku's avatar
osku committed
1647
{
1648 1649 1650 1651 1652 1653
	lock_t*		lock;
	ulint		page_no;
	ulint		space;
	ulint		n_bits;
	ulint		n_bytes;
	const page_t*	page;
1654

osku's avatar
osku committed
1655 1656
	ut_ad(mutex_own(&kernel_mutex));

1657 1658 1659
	space = buf_block_get_space(block);
	page_no	= buf_block_get_page_no(block);
	page = block->frame;
osku's avatar
osku committed
1660

1661
	ut_ad(!!page_is_comp(page) == dict_table_is_comp(index->table));
osku's avatar
osku committed
1662 1663 1664 1665 1666

	/* 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 */

1667
	if (UNIV_UNLIKELY(heap_no == PAGE_HEAP_NO_SUPREMUM)) {
osku's avatar
osku committed
1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684
		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;
1685

osku's avatar
osku committed
1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698
	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,
1699
		    lock_rec_fold(space, page_no), lock);
1700
	if (UNIV_UNLIKELY(type_mode & LOCK_WAIT)) {
osku's avatar
osku committed
1701 1702 1703

		lock_set_lock_and_trx_wait(lock, trx);
	}
1704

osku's avatar
osku committed
1705 1706 1707 1708 1709 1710 1711 1712 1713 1714
	return(lock);
}

/*************************************************************************
Enqueues a waiting request for a lock which cannot be granted immediately.
Checks for deadlocks. */
static
ulint
lock_rec_enqueue_waiting(
/*=====================*/
1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737
					/* out: 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 */
	ulint			type_mode,/* in: lock mode this
					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 */
	const buf_block_t*	block,	/* in: buffer block containing
					the record */
	ulint			heap_no,/* in: heap number of the record */
	dict_index_t*		index,	/* in: index of record */
	que_thr_t*		thr)	/* in: query thread */
osku's avatar
osku committed
1738 1739 1740
{
	lock_t*	lock;
	trx_t*	trx;
1741

osku's avatar
osku committed
1742 1743 1744 1745 1746 1747
	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 */

marko's avatar
marko committed
1748
	if (UNIV_UNLIKELY(que_thr_stop(thr))) {
osku's avatar
osku committed
1749 1750 1751 1752 1753

		ut_error;

		return(DB_QUE_THR_SUSPENDED);
	}
1754

osku's avatar
osku committed
1755 1756
	trx = thr_get_trx(thr);

1757 1758 1759 1760 1761
	switch (trx_get_dict_operation(trx)) {
	case TRX_DICT_OP_NONE:
		break;
	case TRX_DICT_OP_TABLE:
	case TRX_DICT_OP_INDEX:
osku's avatar
osku committed
1762
		ut_print_timestamp(stderr);
1763 1764
		fputs("  InnoDB: Error: a record lock wait happens"
		      " in a dictionary operation!\n"
1765 1766
		      "InnoDB: ", stderr);
		dict_index_name_print(stderr, trx, index);
osku's avatar
osku committed
1767
		fputs(".\n"
1768 1769 1770
		      "InnoDB: Submit a detailed bug report"
		      " to http://bugs.mysql.com\n",
		      stderr);
osku's avatar
osku committed
1771
	}
marko's avatar
marko committed
1772

osku's avatar
osku committed
1773
	/* Enqueue the lock request that will wait to be granted */
1774 1775
	lock = lock_rec_create(type_mode | LOCK_WAIT,
			       block, heap_no, index, trx);
osku's avatar
osku committed
1776 1777 1778

	/* Check if a deadlock occurs: if yes, remove the lock request and
	return an error code */
1779

marko's avatar
marko committed
1780
	if (UNIV_UNLIKELY(lock_deadlock_occurs(lock, trx))) {
osku's avatar
osku committed
1781 1782

		lock_reset_lock_and_trx_wait(lock);
marko's avatar
marko committed
1783
		lock_rec_reset_nth_bit(lock, heap_no);
osku's avatar
osku committed
1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805

		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);

	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));
1806
		ut_print_name(stderr, trx, FALSE, index->name);
osku's avatar
osku committed
1807 1808
	}
#endif /* UNIV_DEBUG */
1809 1810

	return(DB_LOCK_WAIT);
osku's avatar
osku committed
1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823
}

/*************************************************************************
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
which does NOT check for deadlocks or lock compatibility! */
static
lock_t*
lock_rec_add_to_queue(
/*==================*/
1824 1825 1826 1827 1828 1829 1830 1831 1832
					/* out: lock where the bit was set */
	ulint			type_mode,/* in: lock mode, wait, gap
					etc. flags; type is ignored
					and replaced by LOCK_REC */
	const buf_block_t*	block,	/* in: buffer block containing
					the record */
	ulint			heap_no,/* in: heap number of the record */
	dict_index_t*		index,	/* in: index of record */
	trx_t*			trx)	/* in: transaction */
osku's avatar
osku committed
1833 1834
{
	lock_t*	lock;
1835

osku's avatar
osku committed
1836
	ut_ad(mutex_own(&kernel_mutex));
1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857
#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;
osku's avatar
osku committed
1858 1859 1860 1861 1862 1863

	/* 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 */

1864
	if (UNIV_UNLIKELY(heap_no == PAGE_HEAP_NO_SUPREMUM)) {
osku's avatar
osku committed
1865 1866 1867 1868
		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 */
1869

osku's avatar
osku committed
1870 1871 1872 1873 1874
		type_mode = type_mode & ~(LOCK_GAP | LOCK_REC_NOT_GAP);
	}

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

1875
	lock = lock_rec_get_first_on_page(block);
osku's avatar
osku committed
1876 1877 1878

	while (lock != NULL) {
		if (lock_get_wait(lock)
1879
		    && (lock_rec_get_nth_bit(lock, heap_no))) {
osku's avatar
osku committed
1880

marko's avatar
marko committed
1881
			goto somebody_waits;
osku's avatar
osku committed
1882 1883 1884 1885 1886
		}

		lock = lock_rec_get_next_on_page(lock);
	}

1887
	if (UNIV_LIKELY(!(type_mode & LOCK_WAIT))) {
osku's avatar
osku committed
1888

marko's avatar
marko committed
1889 1890 1891
		/* 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 */
osku's avatar
osku committed
1892

1893 1894
		lock = lock_rec_find_similar_on_page(
			type_mode, heap_no,
1895
			lock_rec_get_first_on_page(block), trx);
osku's avatar
osku committed
1896

marko's avatar
marko committed
1897
		if (lock) {
osku's avatar
osku committed
1898

marko's avatar
marko committed
1899 1900 1901 1902
			lock_rec_set_nth_bit(lock, heap_no);

			return(lock);
		}
osku's avatar
osku committed
1903 1904
	}

marko's avatar
marko committed
1905
somebody_waits:
1906
	return(lock_rec_create(type_mode, block, heap_no, index, trx));
osku's avatar
osku committed
1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919
}

/*************************************************************************
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
a page supremum record, a gap type lock. */
UNIV_INLINE
ibool
lock_rec_lock_fast(
/*===============*/
1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932
					/* out: TRUE if locking succeeded */
	ibool			impl,	/* in: if TRUE, no lock is set
					if no wait is necessary: we
					assume that the caller will
					set an implicit lock */
	ulint			mode,	/* in: lock mode: LOCK_X or
					LOCK_S possibly ORed to either
					LOCK_GAP or LOCK_REC_NOT_GAP */
	const buf_block_t*	block,	/* in: buffer block containing
					the record */
	ulint			heap_no,/* in: heap number of record */
	dict_index_t*		index,	/* in: index of record */
	que_thr_t*		thr)	/* in: query thread */
osku's avatar
osku committed
1933 1934
{
	lock_t*	lock;
1935
	trx_t*	trx;
osku's avatar
osku committed
1936 1937 1938

	ut_ad(mutex_own(&kernel_mutex));
	ut_ad((LOCK_MODE_MASK & mode) != LOCK_S
1939
	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IS));
osku's avatar
osku committed
1940
	ut_ad((LOCK_MODE_MASK & mode) != LOCK_X
1941
	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IX));
osku's avatar
osku committed
1942
	ut_ad((LOCK_MODE_MASK & mode) == LOCK_S
1943
	      || (LOCK_MODE_MASK & mode) == LOCK_X);
osku's avatar
osku committed
1944
	ut_ad(mode - (LOCK_MODE_MASK & mode) == LOCK_GAP
1945 1946
	      || mode - (LOCK_MODE_MASK & mode) == 0
	      || mode - (LOCK_MODE_MASK & mode) == LOCK_REC_NOT_GAP);
1947

1948
	lock = lock_rec_get_first_on_page(block);
osku's avatar
osku committed
1949 1950 1951 1952 1953

	trx = thr_get_trx(thr);

	if (lock == NULL) {
		if (!impl) {
1954
			lock_rec_create(mode, block, heap_no, index, trx);
1955

1956
			if (srv_locks_unsafe_for_binlog
1957 1958
			    || trx->isolation_level
			    == TRX_ISO_READ_COMMITTED) {
osku's avatar
osku committed
1959 1960 1961
				trx_register_new_rec_lock(trx, index);
			}
		}
1962

osku's avatar
osku committed
1963 1964
		return(TRUE);
	}
1965

osku's avatar
osku committed
1966 1967 1968 1969 1970 1971
	if (lock_rec_get_next_on_page(lock)) {

		return(FALSE);
	}

	if (lock->trx != trx
1972 1973
	    || lock->type_mode != (mode | LOCK_REC)
	    || lock_rec_get_n_bits(lock) <= heap_no) {
osku's avatar
osku committed
1974

1975
		return(FALSE);
osku's avatar
osku committed
1976 1977 1978 1979 1980 1981 1982 1983
	}

	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);
1984
			if (srv_locks_unsafe_for_binlog
1985 1986
			    || trx->isolation_level
			    == TRX_ISO_READ_COMMITTED) {
osku's avatar
osku committed
1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
				trx_register_new_rec_lock(trx, index);
			}
		}
	}

	return(TRUE);
}

/*************************************************************************
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
lock, or in the case of a page supremum record, a gap type lock. */
static
ulint
lock_rec_lock_slow(
/*===============*/
2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017
					/* out: DB_SUCCESS,
					DB_LOCK_WAIT, or error code */
	ibool			impl,	/* in: if TRUE, no lock is set
					if no wait is necessary: we
					assume that the caller will
					set an implicit lock */
	ulint			mode,	/* in: lock mode: LOCK_X or
					LOCK_S possibly ORed to either
					LOCK_GAP or LOCK_REC_NOT_GAP */
	const buf_block_t*	block,	/* in: buffer block containing
					the record */
	ulint			heap_no,/* in: heap number of record */
	dict_index_t*		index,	/* in: index of record */
	que_thr_t*		thr)	/* in: query thread */
osku's avatar
osku committed
2018 2019 2020 2021 2022 2023
{
	trx_t*	trx;
	ulint	err;

	ut_ad(mutex_own(&kernel_mutex));
	ut_ad((LOCK_MODE_MASK & mode) != LOCK_S
2024
	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IS));
osku's avatar
osku committed
2025
	ut_ad((LOCK_MODE_MASK & mode) != LOCK_X
2026
	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IX));
osku's avatar
osku committed
2027
	ut_ad((LOCK_MODE_MASK & mode) == LOCK_S
2028
	      || (LOCK_MODE_MASK & mode) == LOCK_X);
osku's avatar
osku committed
2029
	ut_ad(mode - (LOCK_MODE_MASK & mode) == LOCK_GAP
2030 2031
	      || mode - (LOCK_MODE_MASK & mode) == 0
	      || mode - (LOCK_MODE_MASK & mode) == LOCK_REC_NOT_GAP);
2032

osku's avatar
osku committed
2033
	trx = thr_get_trx(thr);
2034

2035
	if (lock_rec_has_expl(mode, block, heap_no, trx)) {
osku's avatar
osku committed
2036 2037 2038 2039
		/* The trx already has a strong enough lock on rec: do
		nothing */

		err = DB_SUCCESS;
2040
	} else if (lock_rec_other_has_conflicting(mode, block, heap_no, trx)) {
osku's avatar
osku committed
2041 2042 2043 2044

		/* 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. */
2045

2046 2047
		err = lock_rec_enqueue_waiting(mode, block, heap_no,
					       index, thr);
osku's avatar
osku committed
2048

2049
		if (srv_locks_unsafe_for_binlog
2050
		    || trx->isolation_level == TRX_ISO_READ_COMMITTED) {
osku's avatar
osku committed
2051 2052 2053 2054 2055 2056
			trx_register_new_rec_lock(trx, index);
		}
	} else {
		if (!impl) {
			/* Set the requested lock on the record */

2057
			lock_rec_add_to_queue(LOCK_REC | mode, block,
2058
					      heap_no, index, trx);
2059
			if (srv_locks_unsafe_for_binlog
2060 2061
			    || trx->isolation_level
			    == TRX_ISO_READ_COMMITTED) {
osku's avatar
osku committed
2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081
				trx_register_new_rec_lock(trx, index);
			}
		}

		err = DB_SUCCESS;
	}

	return(err);
}

/*************************************************************************
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
of a page supremum record, a gap type lock. */
static
ulint
lock_rec_lock(
/*==========*/
2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095
					/* out: DB_SUCCESS,
					DB_LOCK_WAIT, or error code */
	ibool			impl,	/* in: if TRUE, no lock is set
					if no wait is necessary: we
					assume that the caller will
					set an implicit lock */
	ulint			mode,	/* in: lock mode: LOCK_X or
					LOCK_S possibly ORed to either
					LOCK_GAP or LOCK_REC_NOT_GAP */
	const buf_block_t*	block,	/* in: buffer block containing
					the record */
	ulint			heap_no,/* in: heap number of record */
	dict_index_t*		index,	/* in: index of record */
	que_thr_t*		thr)	/* in: query thread */
osku's avatar
osku committed
2096 2097 2098 2099 2100
{
	ulint	err;

	ut_ad(mutex_own(&kernel_mutex));
	ut_ad((LOCK_MODE_MASK & mode) != LOCK_S
2101
	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IS));
osku's avatar
osku committed
2102
	ut_ad((LOCK_MODE_MASK & mode) != LOCK_X
2103
	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IX));
osku's avatar
osku committed
2104
	ut_ad((LOCK_MODE_MASK & mode) == LOCK_S
2105
	      || (LOCK_MODE_MASK & mode) == LOCK_X);
osku's avatar
osku committed
2106
	ut_ad(mode - (LOCK_MODE_MASK & mode) == LOCK_GAP
2107 2108
	      || mode - (LOCK_MODE_MASK & mode) == LOCK_REC_NOT_GAP
	      || mode - (LOCK_MODE_MASK & mode) == 0);
2109

2110
	if (lock_rec_lock_fast(impl, mode, block, heap_no, index, thr)) {
osku's avatar
osku committed
2111 2112 2113 2114 2115 2116

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

		err = DB_SUCCESS;
	} else {
2117 2118
		err = lock_rec_lock_slow(impl, mode, block,
					 heap_no, index, thr);
osku's avatar
osku committed
2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138
	}

	return(err);
}

/*************************************************************************
Checks if a waiting record lock request still has to wait in a queue. */
static
ibool
lock_rec_has_to_wait_in_queue(
/*==========================*/
				/* out: TRUE if still has to wait */
	lock_t*	wait_lock)	/* in: waiting record lock */
{
	lock_t*	lock;
	ulint	space;
	ulint	page_no;
	ulint	heap_no;

	ut_ad(mutex_own(&kernel_mutex));
2139
	ut_ad(lock_get_wait(wait_lock));
2140
	ut_ad(lock_get_type_low(wait_lock) == LOCK_REC);
2141

osku's avatar
osku committed
2142 2143 2144 2145 2146 2147 2148 2149 2150
	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)
2151
		    && lock_has_to_wait(wait_lock, lock)) {
osku's avatar
osku committed
2152

2153
			return(TRUE);
osku's avatar
osku committed
2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168
		}

		lock = lock_rec_get_next_on_page(lock);
	}

	return(FALSE);
}

/*****************************************************************
Grants a lock to a waiting lock request and releases the waiting
transaction. */
static
void
lock_grant(
/*=======*/
2169
	lock_t*	lock)	/* in/out: waiting lock request */
osku's avatar
osku committed
2170 2171 2172 2173 2174
{
	ut_ad(mutex_own(&kernel_mutex));

	lock_reset_lock_and_trx_wait(lock);

2175
	if (lock_get_mode(lock) == LOCK_AUTO_INC) {
2176 2177
		trx_t*		trx = lock->trx;
		dict_table_t*	table = lock->un_member.tab_lock.table;
osku's avatar
osku committed
2178

2179
		if (table->autoinc_trx == trx) {
2180
			fprintf(stderr,
2181 2182
				"InnoDB: Error: trx already had"
				" an AUTO-INC lock!\n");
2183 2184
		} else {
			table->autoinc_trx = trx;
osku's avatar
osku committed
2185

2186 2187
			ib_vector_push(trx->autoinc_locks, lock);
		}
2188
	}
osku's avatar
osku committed
2189 2190 2191 2192

#ifdef UNIV_DEBUG
	if (lock_print_waits) {
		fprintf(stderr, "Lock wait for trx %lu ends\n",
2193
			(ulong) ut_dulint_get_low(lock->trx->id));
osku's avatar
osku committed
2194 2195 2196 2197 2198 2199 2200
	}
#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 */
2201 2202

	if (lock->trx->que_state == TRX_QUE_LOCK_WAIT) {
osku's avatar
osku committed
2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217
		trx_end_lock_wait(lock->trx);
	}
}

/*****************************************************************
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(
/*============*/
	lock_t*	lock)	/* in: waiting record lock request */
{
	ut_ad(mutex_own(&kernel_mutex));
2218
	ut_ad(lock_get_type_low(lock) == LOCK_REC);
osku's avatar
osku committed
2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230

	/* 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);
}
2231

osku's avatar
osku committed
2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248
/*****************************************************************
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(
/*=======================*/
	lock_t*	in_lock)/* in: record lock object: all record locks which
			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;
2249

osku's avatar
osku committed
2250
	ut_ad(mutex_own(&kernel_mutex));
2251
	ut_ad(lock_get_type_low(in_lock) == LOCK_REC);
osku's avatar
osku committed
2252 2253 2254 2255 2256 2257 2258

	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,
2259
		    lock_rec_fold(space, page_no), in_lock);
osku's avatar
osku committed
2260 2261 2262 2263 2264 2265 2266 2267

	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);

2268
	while (lock != NULL) {
osku's avatar
osku committed
2269
		if (lock_get_wait(lock)
2270
		    && !lock_rec_has_to_wait_in_queue(lock)) {
osku's avatar
osku committed
2271 2272 2273 2274 2275 2276 2277

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

		lock = lock_rec_get_next_on_page(lock);
	}
2278
}
osku's avatar
osku committed
2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291

/*****************************************************************
Removes a record lock request, waiting or granted, from the queue. */
static
void
lock_rec_discard(
/*=============*/
	lock_t*	in_lock)/* in: record lock object: all record locks which
			are contained in this lock object are removed */
{
	ulint	space;
	ulint	page_no;
	trx_t*	trx;
2292

osku's avatar
osku committed
2293
	ut_ad(mutex_own(&kernel_mutex));
2294
	ut_ad(lock_get_type_low(in_lock) == LOCK_REC);
osku's avatar
osku committed
2295 2296

	trx = in_lock->trx;
2297

osku's avatar
osku committed
2298 2299 2300 2301
	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,
2302
		    lock_rec_fold(space, page_no), in_lock);
osku's avatar
osku committed
2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314

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

/*****************************************************************
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(
/*================================*/
2315
	const buf_block_t*	block)	/* in: page to be discarded */
osku's avatar
osku committed
2316 2317 2318 2319 2320
{
	ulint	space;
	ulint	page_no;
	lock_t*	lock;
	lock_t*	next_lock;
2321

osku's avatar
osku committed
2322
	ut_ad(mutex_own(&kernel_mutex));
2323

2324 2325
	space = buf_block_get_space(block);
	page_no = buf_block_get_page_no(block);
osku's avatar
osku committed
2326 2327 2328 2329 2330 2331 2332 2333

	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);
2334

osku's avatar
osku committed
2335
		lock_rec_discard(lock);
2336

osku's avatar
osku committed
2337 2338
		lock = next_lock;
	}
2339
}
osku's avatar
osku committed
2340 2341 2342 2343 2344 2345

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

/*****************************************************************
Resets the lock bits for a single record. Releases transactions waiting for
lock requests here. */
2346
static
osku's avatar
osku committed
2347 2348 2349
void
lock_rec_reset_and_release_wait(
/*============================*/
2350 2351 2352
	const buf_block_t*	block,	/* in: buffer block containing
					the record */
	ulint			heap_no)/* in: heap number of record */
osku's avatar
osku committed
2353 2354 2355 2356 2357
{
	lock_t*	lock;

	ut_ad(mutex_own(&kernel_mutex));

2358
	lock = lock_rec_get_first(block, heap_no);
osku's avatar
osku committed
2359 2360 2361 2362 2363 2364 2365 2366

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

marko's avatar
marko committed
2367
		lock = lock_rec_get_next(heap_no, lock);
osku's avatar
osku committed
2368
	}
2369
}
osku's avatar
osku committed
2370 2371 2372 2373 2374 2375

/*****************************************************************
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. */
2376
static
osku's avatar
osku committed
2377 2378 2379
void
lock_rec_inherit_to_gap(
/*====================*/
2380 2381 2382 2383 2384 2385 2386 2387 2388 2389
	const buf_block_t*	heir_block,	/* in: block containing the
						record which inherits */
	const buf_block_t*	block,		/* in: block containing the
						record from which inherited;
						does NOT reset the locks on
						this record */
	ulint			heir_heap_no,	/* in: heap_no of the
						inheriting record */
	ulint			heap_no)	/* in: heap_no of the
						donating record */
osku's avatar
osku committed
2390 2391
{
	lock_t*	lock;
2392

osku's avatar
osku committed
2393
	ut_ad(mutex_own(&kernel_mutex));
2394

2395
	lock = lock_rec_get_first(block, heap_no);
osku's avatar
osku committed
2396

2397 2398
	/* If srv_locks_unsafe_for_binlog is TRUE or session is using
	READ COMMITTED isolation level, we do not want locks set
osku's avatar
osku committed
2399 2400 2401 2402 2403 2404
	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)
2405 2406 2407 2408 2409 2410 2411
		    && !((srv_locks_unsafe_for_binlog
			  || lock->trx->isolation_level
			  == TRX_ISO_READ_COMMITTED)
			 && lock_get_mode(lock) == LOCK_X)) {

			lock_rec_add_to_queue(LOCK_REC | LOCK_GAP
					      | lock_get_mode(lock),
2412
					      heir_block, heir_heap_no,
2413
					      lock->index, lock->trx);
2414 2415
		}

marko's avatar
marko committed
2416
		lock = lock_rec_get_next(heap_no, lock);
osku's avatar
osku committed
2417
	}
2418
}
osku's avatar
osku committed
2419 2420 2421 2422 2423 2424 2425 2426 2427

/*****************************************************************
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(
/*================================*/
2428 2429 2430 2431 2432 2433 2434
	const buf_block_t*	block,		/* in: buffer block */
	ulint			heir_heap_no,	/* in: heap_no of
						record which inherits */
	ulint			heap_no)	/* in: heap_no of record
						from which inherited;
						does NOT reset the locks
						on this record */
osku's avatar
osku committed
2435 2436
{
	lock_t*	lock;
2437

osku's avatar
osku committed
2438
	ut_ad(mutex_own(&kernel_mutex));
2439

2440
	lock = lock_rec_get_first(block, heap_no);
osku's avatar
osku committed
2441 2442 2443

	while (lock != NULL) {
		if (!lock_rec_get_insert_intention(lock)
2444
		    && (heap_no == PAGE_HEAP_NO_SUPREMUM
2445
			|| !lock_rec_get_rec_not_gap(lock))) {
2446

2447 2448
			lock_rec_add_to_queue(LOCK_REC | LOCK_GAP
					      | lock_get_mode(lock),
2449
					      block, heir_heap_no,
2450
					      lock->index, lock->trx);
2451
		}
osku's avatar
osku committed
2452

marko's avatar
marko committed
2453
		lock = lock_rec_get_next(heap_no, lock);
osku's avatar
osku committed
2454
	}
2455
}
osku's avatar
osku committed
2456 2457 2458 2459 2460 2461 2462 2463

/*****************************************************************
Moves the locks of a record to another record and resets the lock bits of
the donating record. */
static
void
lock_rec_move(
/*==========*/
2464 2465 2466 2467 2468 2469 2470 2471 2472 2473
	const buf_block_t*	receiver,	/* in: buffer block containing
						the receiving record */
	const buf_block_t*	donator,	/* in: buffer block containing
						the donating record */
	ulint			receiver_heap_no,/* in: heap_no of the record
						which gets the locks; there
						must be no lock requests
						on it! */
	ulint			donator_heap_no)/* in: heap_no of the record
						which gives the locks */
osku's avatar
osku committed
2474 2475
{
	lock_t*	lock;
2476

osku's avatar
osku committed
2477 2478
	ut_ad(mutex_own(&kernel_mutex));

marko's avatar
marko committed
2479
	lock = lock_rec_get_first(donator, donator_heap_no);
osku's avatar
osku committed
2480

marko's avatar
marko committed
2481
	ut_ad(lock_rec_get_first(receiver, receiver_heap_no) == NULL);
osku's avatar
osku committed
2482 2483

	while (lock != NULL) {
2484
		const ulint	type_mode = lock->type_mode;
2485

marko's avatar
marko committed
2486
		lock_rec_reset_nth_bit(lock, donator_heap_no);
osku's avatar
osku committed
2487

2488
		if (UNIV_UNLIKELY(type_mode & LOCK_WAIT)) {
osku's avatar
osku committed
2489
			lock_reset_lock_and_trx_wait(lock);
2490
		}
osku's avatar
osku committed
2491 2492 2493 2494

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

marko's avatar
marko committed
2495
		lock_rec_add_to_queue(type_mode, receiver, receiver_heap_no,
2496
				      lock->index, lock->trx);
marko's avatar
marko committed
2497
		lock = lock_rec_get_next(donator_heap_no, lock);
osku's avatar
osku committed
2498 2499
	}

marko's avatar
marko committed
2500
	ut_ad(lock_rec_get_first(donator, donator_heap_no) == NULL);
2501
}
osku's avatar
osku committed
2502 2503 2504 2505 2506 2507

/*****************************************************************
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. */
2508
UNIV_INTERN
osku's avatar
osku committed
2509 2510 2511
void
lock_move_reorganize_page(
/*======================*/
2512 2513 2514 2515
	const buf_block_t*	block,	/* in: old index page, now
					reorganized */
	const buf_block_t*	oblock)	/* in: copy of the old, not
					reorganized page */
osku's avatar
osku committed
2516 2517 2518 2519 2520 2521 2522 2523
{
	lock_t*		lock;
	UT_LIST_BASE_NODE_T(lock_t)	old_locks;
	mem_heap_t*	heap		= NULL;
	ulint		comp;

	lock_mutex_enter_kernel();

2524
	lock = lock_rec_get_first_on_page(block);
osku's avatar
osku committed
2525 2526 2527 2528 2529 2530 2531 2532

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

		return;
	}

	heap = mem_heap_create(256);
2533

osku's avatar
osku committed
2534 2535 2536 2537 2538
	/* 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);
2539

2540
	do {
osku's avatar
osku committed
2541
		/* Make a copy of the lock */
2542
		lock_t*	old_lock = lock_rec_copy(lock, heap);
osku's avatar
osku committed
2543 2544 2545 2546 2547 2548 2549 2550

		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);
2551
		}
osku's avatar
osku committed
2552 2553

		lock = lock_rec_get_next_on_page(lock);
2554
	} while (lock != NULL);
osku's avatar
osku committed
2555

2556 2557
	comp = page_is_comp(block->frame);
	ut_ad(comp == page_is_comp(oblock->frame));
osku's avatar
osku committed
2558

2559 2560
	for (lock = UT_LIST_GET_FIRST(old_locks); lock;
	     lock = UT_LIST_GET_NEXT(trx_locks, lock)) {
osku's avatar
osku committed
2561 2562 2563 2564
		/* 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 */
2565 2566
		page_cur_t	cur1;
		page_cur_t	cur2;
2567

2568 2569
		page_cur_set_before_first(block, &cur1);
		page_cur_set_before_first(oblock, &cur2);
osku's avatar
osku committed
2570 2571 2572

		/* Set locks according to old locks */
		for (;;) {
2573 2574 2575
			ulint	old_heap_no;
			ulint	new_heap_no;

2576 2577
			ut_ad(comp || !memcmp(page_cur_get_rec(&cur1),
					      page_cur_get_rec(&cur2),
2578 2579 2580
					      rec_get_data_size_old(
						      page_cur_get_rec(
							      &cur2))));
marko's avatar
marko committed
2581
			if (UNIV_LIKELY(comp)) {
2582 2583 2584 2585
				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));
marko's avatar
marko committed
2586
			} else {
2587 2588 2589 2590
				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));
marko's avatar
marko committed
2591
			}
osku's avatar
osku committed
2592 2593 2594

			if (lock_rec_get_nth_bit(lock, old_heap_no)) {

2595 2596 2597 2598
				/* Clear the bit in old_lock. */
				ut_d(lock_rec_reset_nth_bit(lock,
							    old_heap_no));

osku's avatar
osku committed
2599 2600 2601
				/* NOTE that the old lock bitmap could be too
				small for the new heap number! */

2602
				lock_rec_add_to_queue(lock->type_mode, block,
2603 2604
						      new_heap_no,
						      lock->index, lock->trx);
osku's avatar
osku committed
2605

2606
				/* if (new_heap_no == PAGE_HEAP_NO_SUPREMUM
2607 2608
				&& lock_get_wait(lock)) {
				fprintf(stderr,
osku's avatar
osku committed
2609
				"---\n--\n!!!Lock reorg: supr type %lu\n",
2610
				lock->type_mode);
osku's avatar
osku committed
2611 2612 2613
				} */
			}

2614 2615
			if (UNIV_UNLIKELY
			    (new_heap_no == PAGE_HEAP_NO_SUPREMUM)) {
osku's avatar
osku committed
2616

2617
				ut_ad(old_heap_no == PAGE_HEAP_NO_SUPREMUM);
osku's avatar
osku committed
2618 2619 2620 2621 2622
				break;
			}

			page_cur_move_to_next(&cur1);
			page_cur_move_to_next(&cur2);
2623
		}
osku's avatar
osku committed
2624

2625
#ifdef UNIV_DEBUG
2626 2627 2628 2629 2630 2631 2632 2633 2634 2635
		{
			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;
2636 2637
			}
		}
2638
#endif /* UNIV_DEBUG */
osku's avatar
osku committed
2639 2640 2641 2642 2643 2644
	}

	lock_mutex_exit_kernel();

	mem_heap_free(heap);

2645
#ifdef UNIV_DEBUG_LOCK_VALIDATE
2646 2647
	ut_ad(lock_rec_validate_page(buf_block_get_space(block),
				     buf_block_get_page_no(block)));
2648
#endif
2649
}
osku's avatar
osku committed
2650 2651 2652 2653

/*****************************************************************
Moves the explicit locks on user records to another page if a record
list end is moved to another page. */
2654
UNIV_INTERN
osku's avatar
osku committed
2655 2656 2657
void
lock_move_rec_list_end(
/*===================*/
2658 2659 2660 2661
	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
						is the first record moved */
osku's avatar
osku committed
2662 2663
{
	lock_t*		lock;
2664
	const ulint	comp	= page_rec_is_comp(rec);
osku's avatar
osku committed
2665 2666 2667 2668 2669 2670 2671 2672 2673

	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. */

2674 2675
	for (lock = lock_rec_get_first_on_page(block); lock;
	     lock = lock_rec_get_next_on_page(lock)) {
2676 2677 2678
		page_cur_t	cur1;
		page_cur_t	cur2;
		const ulint	type_mode = lock->type_mode;
2679

2680
		page_cur_position(rec, block, &cur1);
osku's avatar
osku committed
2681 2682 2683 2684 2685

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

2686
		page_cur_set_before_first(new_block, &cur2);
osku's avatar
osku committed
2687
		page_cur_move_to_next(&cur2);
2688

osku's avatar
osku committed
2689 2690 2691
		/* Copy lock requests on user records to new page and
		reset the lock bits on the old */

marko's avatar
marko committed
2692
		while (!page_cur_is_after_last(&cur1)) {
2693 2694
			ulint	heap_no;

2695
			if (comp) {
2696 2697
				heap_no = rec_get_heap_no_new(
					page_cur_get_rec(&cur1));
marko's avatar
marko committed
2698
			} else {
2699 2700
				heap_no = rec_get_heap_no_old(
					page_cur_get_rec(&cur1));
2701 2702 2703 2704
				ut_ad(!memcmp(page_cur_get_rec(&cur1),
					 page_cur_get_rec(&cur2),
					 rec_get_data_size_old(
						 page_cur_get_rec(&cur2))));
marko's avatar
marko committed
2705
			}
osku's avatar
osku committed
2706 2707 2708 2709

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

2710
				if (UNIV_UNLIKELY(type_mode & LOCK_WAIT)) {
osku's avatar
osku committed
2711
					lock_reset_lock_and_trx_wait(lock);
2712
				}
osku's avatar
osku committed
2713

2714
				if (comp) {
2715 2716
					heap_no = rec_get_heap_no_new(
						page_cur_get_rec(&cur2));
marko's avatar
marko committed
2717
				} else {
2718 2719
					heap_no = rec_get_heap_no_old(
						page_cur_get_rec(&cur2));
marko's avatar
marko committed
2720 2721
				}

osku's avatar
osku committed
2722
				lock_rec_add_to_queue(type_mode,
2723
						      new_block, heap_no,
2724
						      lock->index, lock->trx);
osku's avatar
osku committed
2725 2726 2727 2728 2729 2730
			}

			page_cur_move_to_next(&cur1);
			page_cur_move_to_next(&cur2);
		}
	}
2731

osku's avatar
osku committed
2732 2733
	lock_mutex_exit_kernel();

2734
#ifdef UNIV_DEBUG_LOCK_VALIDATE
2735 2736 2737 2738
	ut_ad(lock_rec_validate_page(buf_block_get_space(block),
				     buf_block_get_page_no(block)));
	ut_ad(lock_rec_validate_page(buf_block_get_space(new_block),
				     buf_block_get_page_no(new_block)));
2739
#endif
2740
}
osku's avatar
osku committed
2741 2742 2743 2744

/*****************************************************************
Moves the explicit locks on user records to another page if a record
list start is moved to another page. */
2745
UNIV_INTERN
osku's avatar
osku committed
2746 2747 2748
void
lock_move_rec_list_start(
/*=====================*/
2749 2750 2751 2752 2753 2754 2755 2756 2757 2758
	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 is the first
						record NOT copied */
	const rec_t*		old_end)	/* in: old
						previous-to-last
						record on new_page
						before the records
						were copied */
osku's avatar
osku committed
2759 2760
{
	lock_t*		lock;
2761
	const ulint	comp	= page_rec_is_comp(rec);
osku's avatar
osku committed
2762

2763 2764
	ut_ad(block->frame == page_align(rec));
	ut_ad(new_block->frame == page_align(old_end));
osku's avatar
osku committed
2765 2766 2767

	lock_mutex_enter_kernel();

2768 2769
	for (lock = lock_rec_get_first_on_page(block); lock;
	     lock = lock_rec_get_next_on_page(lock)) {
2770 2771 2772
		page_cur_t	cur1;
		page_cur_t	cur2;
		const ulint	type_mode = lock->type_mode;
2773

2774
		page_cur_set_before_first(block, &cur1);
osku's avatar
osku committed
2775 2776
		page_cur_move_to_next(&cur1);

2777
		page_cur_position(old_end, new_block, &cur2);
osku's avatar
osku committed
2778 2779 2780 2781 2782 2783
		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) {
2784 2785
			ulint	heap_no;

2786
			if (comp) {
2787 2788
				heap_no = rec_get_heap_no_new(
					page_cur_get_rec(&cur1));
marko's avatar
marko committed
2789
			} else {
2790 2791
				heap_no = rec_get_heap_no_old(
					page_cur_get_rec(&cur1));
2792 2793 2794 2795 2796
				ut_ad(!memcmp(page_cur_get_rec(&cur1),
					      page_cur_get_rec(&cur2),
					      rec_get_data_size_old(
						      page_cur_get_rec(
							      &cur2))));
marko's avatar
marko committed
2797
			}
osku's avatar
osku committed
2798 2799 2800 2801

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

2802
				if (UNIV_UNLIKELY(type_mode & LOCK_WAIT)) {
osku's avatar
osku committed
2803
					lock_reset_lock_and_trx_wait(lock);
2804
				}
osku's avatar
osku committed
2805

2806
				if (comp) {
2807 2808
					heap_no = rec_get_heap_no_new(
						page_cur_get_rec(&cur2));
marko's avatar
marko committed
2809
				} else {
2810 2811
					heap_no = rec_get_heap_no_old(
						page_cur_get_rec(&cur2));
marko's avatar
marko committed
2812 2813
				}

osku's avatar
osku committed
2814
				lock_rec_add_to_queue(type_mode,
2815
						      new_block, heap_no,
2816
						      lock->index, lock->trx);
osku's avatar
osku committed
2817 2818 2819 2820 2821 2822
			}

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

2823 2824 2825 2826
#ifdef UNIV_DEBUG
		if (page_rec_is_supremum(rec)) {
			ulint	i;

2827 2828
			for (i = PAGE_HEAP_NO_USER_LOW;
			     i < lock_rec_get_n_bits(lock); i++) {
2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840
				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 */
osku's avatar
osku committed
2841
	}
2842

osku's avatar
osku committed
2843
	lock_mutex_exit_kernel();
2844

2845
#ifdef UNIV_DEBUG_LOCK_VALIDATE
2846 2847
	ut_ad(lock_rec_validate_page(buf_block_get_space(block),
				     buf_block_get_page_no(block)));
2848
#endif
2849
}
osku's avatar
osku committed
2850 2851 2852

/*****************************************************************
Updates the lock table when a page is split to the right. */
2853
UNIV_INTERN
osku's avatar
osku committed
2854 2855 2856
void
lock_update_split_right(
/*====================*/
2857 2858
	const buf_block_t*	right_block,	/* in: right page */
	const buf_block_t*	left_block)	/* in: left page */
osku's avatar
osku committed
2859
{
2860 2861
	ulint	heap_no = lock_get_min_heap_no(right_block);

osku's avatar
osku committed
2862 2863 2864 2865 2866
	lock_mutex_enter_kernel();

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

2867 2868
	lock_rec_move(right_block, left_block,
		      PAGE_HEAP_NO_SUPREMUM, PAGE_HEAP_NO_SUPREMUM);
2869

osku's avatar
osku committed
2870 2871 2872
	/* Inherit the locks to the supremum of left page from the successor
	of the infimum on right page */

2873 2874
	lock_rec_inherit_to_gap(left_block, right_block,
				PAGE_HEAP_NO_SUPREMUM, heap_no);
osku's avatar
osku committed
2875 2876

	lock_mutex_exit_kernel();
2877
}
osku's avatar
osku committed
2878 2879 2880

/*****************************************************************
Updates the lock table when a page is merged to the right. */
2881
UNIV_INTERN
osku's avatar
osku committed
2882 2883 2884
void
lock_update_merge_right(
/*====================*/
2885 2886 2887 2888 2889 2890 2891 2892 2893
	const buf_block_t*	right_block,	/* in: right page to
						which merged */
	const rec_t*		orig_succ,	/* in: original
						successor of infimum
						on the right page
						before merge */
	const buf_block_t*	left_block)	/* in: merged index
						page which will be
						discarded */
osku's avatar
osku committed
2894 2895
{
	lock_mutex_enter_kernel();
2896

osku's avatar
osku committed
2897 2898 2899 2900
	/* 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 */

2901 2902 2903
	lock_rec_inherit_to_gap(right_block, left_block,
				page_rec_get_heap_no(orig_succ),
				PAGE_HEAP_NO_SUPREMUM);
osku's avatar
osku committed
2904 2905 2906 2907

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

2908 2909
	lock_rec_reset_and_release_wait(left_block,
					PAGE_HEAP_NO_SUPREMUM);
2910

2911
	lock_rec_free_all_from_discard_page(left_block);
osku's avatar
osku committed
2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922

	lock_mutex_exit_kernel();
}

/*****************************************************************
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. */
2923
UNIV_INTERN
osku's avatar
osku committed
2924 2925 2926
void
lock_update_root_raise(
/*===================*/
2927 2928
	const buf_block_t*	block,	/* in: index page to which copied */
	const buf_block_t*	root)	/* in: root page */
osku's avatar
osku committed
2929 2930 2931 2932
{
	lock_mutex_enter_kernel();

	/* Move the locks on the supremum of the root to the supremum
2933
	of block */
osku's avatar
osku committed
2934

2935 2936
	lock_rec_move(block, root,
		      PAGE_HEAP_NO_SUPREMUM, PAGE_HEAP_NO_SUPREMUM);
osku's avatar
osku committed
2937 2938 2939 2940 2941 2942
	lock_mutex_exit_kernel();
}

/*****************************************************************
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! */
2943
UNIV_INTERN
osku's avatar
osku committed
2944 2945 2946
void
lock_update_copy_and_discard(
/*=========================*/
2947 2948 2949 2950
	const buf_block_t*	new_block,	/* in: index page to
						which copied */
	const buf_block_t*	block)		/* in: index page;
						NOT the root! */
osku's avatar
osku committed
2951 2952 2953 2954 2955 2956
{
	lock_mutex_enter_kernel();

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

2957 2958 2959
	lock_rec_move(new_block, block,
		      PAGE_HEAP_NO_SUPREMUM, PAGE_HEAP_NO_SUPREMUM);
	lock_rec_free_all_from_discard_page(block);
osku's avatar
osku committed
2960 2961 2962 2963 2964 2965

	lock_mutex_exit_kernel();
}

/*****************************************************************
Updates the lock table when a page is split to the left. */
2966
UNIV_INTERN
osku's avatar
osku committed
2967 2968 2969
void
lock_update_split_left(
/*===================*/
2970 2971
	const buf_block_t*	right_block,	/* in: right page */
	const buf_block_t*	left_block)	/* in: left page */
osku's avatar
osku committed
2972
{
2973 2974
	ulint	heap_no = lock_get_min_heap_no(right_block);

osku's avatar
osku committed
2975
	lock_mutex_enter_kernel();
2976

osku's avatar
osku committed
2977 2978 2979
	/* Inherit the locks to the supremum of the left page from the
	successor of the infimum on the right page */

2980 2981
	lock_rec_inherit_to_gap(left_block, right_block,
				PAGE_HEAP_NO_SUPREMUM, heap_no);
osku's avatar
osku committed
2982 2983 2984 2985 2986 2987

	lock_mutex_exit_kernel();
}

/*****************************************************************
Updates the lock table when a page is merged to the left. */
2988
UNIV_INTERN
osku's avatar
osku committed
2989 2990 2991
void
lock_update_merge_left(
/*===================*/
2992 2993 2994 2995 2996 2997 2998
	const buf_block_t*	left_block,	/* in: left page to
						which merged */
	const rec_t*		orig_pred,	/* in: original predecessor
						of supremum on the left page
						before merge */
	const buf_block_t*	right_block)	/* in: merged index page
						which will be discarded */
osku's avatar
osku committed
2999
{
3000 3001
	const rec_t*	left_next_rec;

3002
	ut_ad(left_block->frame == page_align(orig_pred));
3003

osku's avatar
osku committed
3004 3005
	lock_mutex_enter_kernel();

3006
	left_next_rec = page_rec_get_next_const(orig_pred);
osku's avatar
osku committed
3007

3008
	if (!page_rec_is_supremum(left_next_rec)) {
osku's avatar
osku committed
3009 3010 3011 3012

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

3013 3014 3015
		lock_rec_inherit_to_gap(left_block, left_block,
					page_rec_get_heap_no(left_next_rec),
					PAGE_HEAP_NO_SUPREMUM);
osku's avatar
osku committed
3016 3017 3018 3019

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

3020 3021
		lock_rec_reset_and_release_wait(left_block,
						PAGE_HEAP_NO_SUPREMUM);
osku's avatar
osku committed
3022 3023 3024 3025
	}

	/* Move the locks from the supremum of right page to the supremum
	of the left page */
3026

3027 3028
	lock_rec_move(left_block, right_block,
		      PAGE_HEAP_NO_SUPREMUM, PAGE_HEAP_NO_SUPREMUM);
osku's avatar
osku committed
3029

3030
	lock_rec_free_all_from_discard_page(right_block);
osku's avatar
osku committed
3031 3032 3033 3034 3035 3036 3037

	lock_mutex_exit_kernel();
}

/*****************************************************************
Resets the original locks on heir and replaces them with gap type locks
inherited from rec. */
3038
UNIV_INTERN
osku's avatar
osku committed
3039 3040 3041
void
lock_rec_reset_and_inherit_gap_locks(
/*=================================*/
3042 3043 3044 3045 3046 3047 3048 3049 3050 3051
	const buf_block_t*	heir_block,	/* in: block containing the
						record which inherits */
	const buf_block_t*	block,		/* in: block containing the
						record from which inherited;
						does NOT reset the locks on
						this record */
	ulint			heir_heap_no,	/* in: heap_no of the
						inheriting record */
	ulint			heap_no)	/* in: heap_no of the
						donating record */
osku's avatar
osku committed
3052
{
3053
	mutex_enter(&kernel_mutex);
osku's avatar
osku committed
3054

3055
	lock_rec_reset_and_release_wait(heir_block, heir_heap_no);
3056

3057
	lock_rec_inherit_to_gap(heir_block, block, heir_heap_no, heap_no);
osku's avatar
osku committed
3058

3059
	mutex_exit(&kernel_mutex);
osku's avatar
osku committed
3060 3061 3062 3063
}

/*****************************************************************
Updates the lock table when a page is discarded. */
3064
UNIV_INTERN
osku's avatar
osku committed
3065 3066 3067
void
lock_update_discard(
/*================*/
3068 3069 3070 3071 3072 3073
	const buf_block_t*	heir_block,	/* in: index page
						which will inherit the locks */
	ulint			heir_heap_no,	/* in: heap_no of the record
						which will inherit the locks */
	const buf_block_t*	block)		/* in: index page
						which will be discarded */
osku's avatar
osku committed
3074
{
3075 3076 3077
	const page_t*	page = block->frame;
	const rec_t*	rec;
	ulint		heap_no;
osku's avatar
osku committed
3078 3079

	lock_mutex_enter_kernel();
3080

3081
	if (!lock_rec_get_first_on_page(block)) {
osku's avatar
osku committed
3082 3083 3084 3085 3086 3087
		/* No locks exist on page, nothing to do */

		lock_mutex_exit_kernel();

		return;
	}
3088

osku's avatar
osku committed
3089 3090 3091
	/* Inherit all the locks on the page to the record and reset all
	the locks on the page */

3092 3093
	if (page_is_comp(page)) {
		rec = page + PAGE_NEW_INFIMUM;
osku's avatar
osku committed
3094

3095 3096
		do {
			heap_no = rec_get_heap_no_new(rec);
osku's avatar
osku committed
3097

3098 3099
			lock_rec_inherit_to_gap(heir_block, block,
						heir_heap_no, heap_no);
osku's avatar
osku committed
3100

3101
			lock_rec_reset_and_release_wait(block, heap_no);
osku's avatar
osku committed
3102

3103
			rec = page + rec_get_next_offs(rec, TRUE);
3104
		} while (heap_no != PAGE_HEAP_NO_SUPREMUM);
3105 3106
	} else {
		rec = page + PAGE_OLD_INFIMUM;
osku's avatar
osku committed
3107

3108 3109 3110 3111 3112
		do {
			heap_no = rec_get_heap_no_old(rec);

			lock_rec_inherit_to_gap(heir_block, block,
						heir_heap_no, heap_no);
3113

3114 3115 3116
			lock_rec_reset_and_release_wait(block, heap_no);

			rec = page + rec_get_next_offs(rec, FALSE);
3117
		} while (heap_no != PAGE_HEAP_NO_SUPREMUM);
osku's avatar
osku committed
3118 3119
	}

3120
	lock_rec_free_all_from_discard_page(block);
osku's avatar
osku committed
3121 3122 3123 3124 3125 3126

	lock_mutex_exit_kernel();
}

/*****************************************************************
Updates the lock table when a new user record is inserted. */
3127
UNIV_INTERN
osku's avatar
osku committed
3128 3129 3130
void
lock_update_insert(
/*===============*/
3131 3132
	const buf_block_t*	block,	/* in: buffer block containing rec */
	const rec_t*		rec)	/* in: the inserted record */
osku's avatar
osku committed
3133
{
3134 3135 3136
	ulint	receiver_heap_no;
	ulint	donator_heap_no;

3137
	ut_ad(block->frame == page_align(rec));
osku's avatar
osku committed
3138 3139 3140 3141

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

3142 3143 3144
	if (page_rec_is_comp(rec)) {
		receiver_heap_no = rec_get_heap_no_new(rec);
		donator_heap_no = rec_get_heap_no_new(
3145
			page_rec_get_next_low(rec, TRUE));
3146 3147 3148
	} else {
		receiver_heap_no = rec_get_heap_no_old(rec);
		donator_heap_no = rec_get_heap_no_old(
3149
			page_rec_get_next_low(rec, FALSE));
3150
	}
osku's avatar
osku committed
3151

3152 3153 3154
	lock_mutex_enter_kernel();
	lock_rec_inherit_to_gap_if_gap_lock(block,
					    receiver_heap_no, donator_heap_no);
osku's avatar
osku committed
3155 3156 3157 3158 3159
	lock_mutex_exit_kernel();
}

/*****************************************************************
Updates the lock table when a record is removed. */
3160
UNIV_INTERN
osku's avatar
osku committed
3161 3162 3163
void
lock_update_delete(
/*===============*/
3164 3165
	const buf_block_t*	block,	/* in: buffer block containing rec */
	const rec_t*		rec)	/* in: the record to be removed */
osku's avatar
osku committed
3166
{
3167 3168 3169 3170
	const page_t*	page = block->frame;
	ulint		heap_no;
	ulint		next_heap_no;

marko's avatar
marko committed
3171 3172
	ut_ad(page == page_align(rec));

3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184
	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));
	}

osku's avatar
osku committed
3185 3186 3187 3188
	lock_mutex_enter_kernel();

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

3189
	lock_rec_inherit_to_gap(block, block, next_heap_no, heap_no);
osku's avatar
osku committed
3190 3191 3192

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

3193
	lock_rec_reset_and_release_wait(block, heap_no);
osku's avatar
osku committed
3194 3195 3196

	lock_mutex_exit_kernel();
}
3197

osku's avatar
osku committed
3198 3199 3200 3201 3202 3203 3204
/*************************************************************************
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. */
3205
UNIV_INTERN
osku's avatar
osku committed
3206 3207 3208
void
lock_rec_store_on_page_infimum(
/*===========================*/
3209 3210 3211 3212 3213 3214
	const buf_block_t*	block,	/* in: buffer block containing rec */
	const rec_t*		rec)	/* in: record whose lock state
					is stored on the infimum
					record of the same page; lock
					bits are reset on the
					record */
osku's avatar
osku committed
3215
{
3216 3217
	ulint	heap_no = page_rec_get_heap_no(rec);

3218
	ut_ad(block->frame == page_align(rec));
osku's avatar
osku committed
3219 3220

	lock_mutex_enter_kernel();
3221

3222
	lock_rec_move(block, block, PAGE_HEAP_NO_INFIMUM, heap_no);
osku's avatar
osku committed
3223

3224
	lock_mutex_exit_kernel();
osku's avatar
osku committed
3225 3226 3227 3228 3229
}

/*************************************************************************
Restores the state of explicit lock requests on a single record, where the
state was stored on the infimum of the page. */
3230
UNIV_INTERN
osku's avatar
osku committed
3231 3232 3233
void
lock_rec_restore_from_page_infimum(
/*===============================*/
3234 3235
	const buf_block_t*	block,	/* in: buffer block containing rec */
	const rec_t*		rec,	/* in: record whose lock state
marko's avatar
marko committed
3236
					is restored */
3237 3238 3239 3240 3241
	const buf_block_t*	donator)/* in: page (rec is not
					necessarily on this page)
					whose infimum stored the lock
					state; lock bits are reset on
					the infimum */
osku's avatar
osku committed
3242
{
3243 3244
	ulint	heap_no = page_rec_get_heap_no(rec);

osku's avatar
osku committed
3245 3246
	lock_mutex_enter_kernel();

3247
	lock_rec_move(block, donator, heap_no, PAGE_HEAP_NO_INFIMUM);
3248

osku's avatar
osku committed
3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272
	lock_mutex_exit_kernel();
}

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

/************************************************************************
Checks if a lock request results in a deadlock. */
static
ibool
lock_deadlock_occurs(
/*=================*/
			/* out: 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) */
	lock_t*	lock,	/* in: lock the transaction is requesting */
	trx_t*	trx)	/* in: transaction */
{
	dict_table_t*	table;
	dict_index_t*	index;
	trx_t*		mark_trx;
	ulint		ret;
	ulint		cost	= 0;

3273 3274
	ut_ad(trx);
	ut_ad(lock);
osku's avatar
osku committed
3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296
	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);

	if (ret == LOCK_VICTIM_IS_OTHER) {
		/* We chose some other trx as a victim: retry if there still
		is a deadlock */

		goto retry;
	}

marko's avatar
marko committed
3297
	if (UNIV_UNLIKELY(ret == LOCK_VICTIM_IS_START)) {
3298
		if (lock_get_type_low(lock) & LOCK_TABLE) {
osku's avatar
osku committed
3299 3300 3301 3302 3303 3304 3305 3306 3307 3308
			table = lock->un_member.tab_lock.table;
			index = NULL;
		} else {
			index = lock->index;
			table = index->table;
		}

		lock_deadlock_found = TRUE;

		fputs("*** WE ROLL BACK TRANSACTION (2)\n",
3309
		      lock_latest_err_file);
osku's avatar
osku committed
3310 3311 3312

		return(TRUE);
	}
3313

osku's avatar
osku committed
3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336
	return(FALSE);
}

/************************************************************************
Looks recursively for a deadlock. */
static
ulint
lock_deadlock_recursive(
/*====================*/
				/* out: 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
				deadlock! */
	trx_t*	start,		/* in: recursion starting point */
	trx_t*	trx,		/* in: a transaction waiting for a lock */
	lock_t*	wait_lock,	/* in: the lock trx is waiting to be granted */
	ulint*	cost,		/* in/out: number of calculation steps thus
				far: if this exceeds LOCK_MAX_N_STEPS_...
				we return LOCK_VICTIM_IS_START */
3337
	ulint	depth)		/* in: recursion depth: if this exceeds
osku's avatar
osku committed
3338 3339 3340 3341 3342 3343 3344
				LOCK_MAX_DEPTH_IN_DEADLOCK_CHECK, we
				return LOCK_VICTIM_IS_START */
{
	lock_t*	lock;
	ulint	bit_no		= ULINT_UNDEFINED;
	trx_t*	lock_trx;
	ulint	ret;
3345

3346 3347 3348
	ut_a(trx);
	ut_a(start);
	ut_a(wait_lock);
osku's avatar
osku committed
3349
	ut_ad(mutex_own(&kernel_mutex));
3350

osku's avatar
osku committed
3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361
	if (trx->deadlock_mark == 1) {
		/* We have already exhaustively searched the subtree starting
		from this trx */

		return(0);
	}

	*cost = *cost + 1;

	lock = wait_lock;

3362
	if (lock_get_type_low(wait_lock) == LOCK_REC) {
osku's avatar
osku committed
3363 3364 3365 3366 3367 3368 3369 3370 3371

		bit_no = lock_rec_find_set_bit(wait_lock);

		ut_a(bit_no != ULINT_UNDEFINED);
	}

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

	for (;;) {
3372
		if (lock_get_type_low(lock) & LOCK_TABLE) {
osku's avatar
osku committed
3373

3374 3375
			lock = UT_LIST_GET_PREV(un_member.tab_lock.locks,
						lock);
osku's avatar
osku committed
3376
		} else {
3377
			ut_ad(lock_get_type_low(lock) == LOCK_REC);
osku's avatar
osku committed
3378 3379
			ut_a(bit_no != ULINT_UNDEFINED);

3380
			lock = (lock_t*) lock_rec_get_prev(lock, bit_no);
osku's avatar
osku committed
3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391
		}

		if (lock == NULL) {
			/* We can mark this subtree as searched */
			trx->deadlock_mark = 1;

			return(FALSE);
		}

		if (lock_has_to_wait(wait_lock, lock)) {

3392 3393 3394 3395
			ibool	too_far
				= depth > LOCK_MAX_DEPTH_IN_DEADLOCK_CHECK
				|| *cost > LOCK_MAX_N_STEPS_IN_DEADLOCK_CHECK;

osku's avatar
osku committed
3396 3397
			lock_trx = lock->trx;

3398 3399
			if (lock_trx == start || too_far) {

osku's avatar
osku committed
3400
				/* We came back to the recursion starting
3401 3402 3403
				point: a deadlock detected; or we have
				searched the waits-for graph too long */

osku's avatar
osku committed
3404
				FILE*	ef = lock_latest_err_file;
3405

osku's avatar
osku committed
3406 3407 3408 3409 3410 3411 3412
				rewind(ef);
				ut_print_timestamp(ef);

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

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

3413 3414
				fputs("*** (1) WAITING FOR THIS LOCK"
				      " TO BE GRANTED:\n", ef);
3415

3416
				if (lock_get_type_low(wait_lock) == LOCK_REC) {
osku's avatar
osku committed
3417 3418 3419 3420
					lock_rec_print(ef, wait_lock);
				} else {
					lock_table_print(ef, wait_lock);
				}
3421

osku's avatar
osku committed
3422 3423 3424 3425 3426
				fputs("*** (2) TRANSACTION:\n", ef);

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

				fputs("*** (2) HOLDS THE LOCK(S):\n", ef);
3427

3428
				if (lock_get_type_low(lock) == LOCK_REC) {
osku's avatar
osku committed
3429 3430 3431 3432
					lock_rec_print(ef, lock);
				} else {
					lock_table_print(ef, lock);
				}
3433

3434 3435
				fputs("*** (2) WAITING FOR THIS LOCK"
				      " TO BE GRANTED:\n", ef);
3436

3437
				if (lock_get_type_low(start->wait_lock)
3438
				    == LOCK_REC) {
osku's avatar
osku committed
3439 3440 3441 3442 3443 3444
					lock_rec_print(ef, start->wait_lock);
				} else {
					lock_table_print(ef, start->wait_lock);
				}
#ifdef UNIV_DEBUG
				if (lock_print_waits) {
3445 3446 3447
					fputs("Deadlock detected"
					      " or too long search\n",
					      stderr);
osku's avatar
osku committed
3448 3449
				}
#endif /* UNIV_DEBUG */
3450 3451 3452 3453 3454 3455 3456 3457 3458
				if (too_far) {

					fputs("TOO DEEP OR LONG SEARCH"
					      " IN THE LOCK TABLE"
					      " WAITS-FOR GRAPH\n", ef);

					return(LOCK_VICTIM_IS_START);
				}

3459 3460
				if (trx_weight_cmp(wait_lock->trx,
						   start) >= 0) {
osku's avatar
osku committed
3461 3462 3463 3464 3465 3466
					/* Our recursion starting point
					transaction is 'smaller', let us
					choose 'start' as the victim and roll
					back it */

					return(LOCK_VICTIM_IS_START);
3467
				}
osku's avatar
osku committed
3468 3469 3470 3471 3472 3473

				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 */
3474

osku's avatar
osku committed
3475
				fputs("*** WE ROLL BACK TRANSACTION (1)\n",
3476
				      ef);
3477

osku's avatar
osku committed
3478
				wait_lock->trx->was_chosen_as_deadlock_victim
3479
					= TRUE;
3480

osku's avatar
osku committed
3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491
				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);
			}
3492

osku's avatar
osku committed
3493 3494 3495 3496 3497 3498
			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 */

3499 3500 3501
				ret = lock_deadlock_recursive(
					start, lock_trx,
					lock_trx->wait_lock, cost, depth + 1);
osku's avatar
osku committed
3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519
				if (ret != 0) {

					return(ret);
				}
			}
		}
	}/* end of the 'for (;;)'-loop */
}

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

/*************************************************************************
Creates a table lock object and adds it as the last in the lock queue
of the table. Does NOT check for deadlocks or lock compatibility. */
UNIV_INLINE
lock_t*
lock_table_create(
/*==============*/
3520
				/* out, own: new lock object */
osku's avatar
osku committed
3521 3522 3523 3524 3525 3526 3527 3528 3529 3530
	dict_table_t*	table,	/* in: database table in dictionary cache */
	ulint		type_mode,/* in: lock mode possibly ORed with
				LOCK_WAIT */
	trx_t*		trx)	/* in: trx */
{
	lock_t*	lock;

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

3531 3532 3533 3534
	if ((type_mode & LOCK_MODE_MASK) == LOCK_AUTO_INC) {
		++table->n_waiting_or_granted_auto_inc_locks;
	}

3535 3536 3537
	/* 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. */
osku's avatar
osku committed
3538 3539
	if (type_mode == LOCK_AUTO_INC) {

3540 3541 3542
		lock = table->autoinc_lock;

		table->autoinc_trx = trx;
osku's avatar
osku committed
3543

3544
		ib_vector_push(trx->autoinc_locks, lock);
osku's avatar
osku committed
3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557
	} 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);

3558
	if (UNIV_UNLIKELY(type_mode & LOCK_WAIT)) {
osku's avatar
osku committed
3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576

		lock_set_lock_and_trx_wait(lock, trx);
	}

	return(lock);
}

/*****************************************************************
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(
/*==================*/
	lock_t*	lock)	/* in: table lock */
{
	trx_t*		trx;
3577
	dict_table_t*	table;
osku's avatar
osku committed
3578 3579 3580 3581

	ut_ad(mutex_own(&kernel_mutex));

	trx = lock->trx;
3582 3583 3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597 3598 3599 3600 3601
	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;
		}

		/* 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. */
osku's avatar
osku committed
3602

3603 3604 3605 3606 3607 3608
		if (!ib_vector_is_empty(trx->autoinc_locks)) {
			lock_t*	autoinc_lock;

			autoinc_lock = ib_vector_pop(trx->autoinc_locks);
			ut_a(autoinc_lock == lock);
		}
3609 3610 3611

		ut_a(table->n_waiting_or_granted_auto_inc_locks > 0);
		--table->n_waiting_or_granted_auto_inc_locks;
osku's avatar
osku committed
3612 3613 3614 3615
	}

	UT_LIST_REMOVE(trx_locks, trx->trx_locks, lock);
	UT_LIST_REMOVE(un_member.tab_lock.locks, table->locks, lock);
3616
}
osku's avatar
osku committed
3617 3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636 3637

/*************************************************************************
Enqueues a waiting request for a table lock which cannot be granted
immediately. Checks for deadlocks. */
static
ulint
lock_table_enqueue_waiting(
/*=======================*/
				/* out: 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 */
	ulint		mode,	/* in: lock mode this transaction is
				requesting */
	dict_table_t*	table,	/* in: table */
	que_thr_t*	thr)	/* in: query thread */
{
	lock_t*	lock;
	trx_t*	trx;
3638

osku's avatar
osku committed
3639
	ut_ad(mutex_own(&kernel_mutex));
3640

osku's avatar
osku committed
3641 3642 3643 3644 3645 3646 3647 3648 3649 3650 3651 3652
	/* 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);

3653 3654 3655 3656 3657
	switch (trx_get_dict_operation(trx)) {
	case TRX_DICT_OP_NONE:
		break;
	case TRX_DICT_OP_TABLE:
	case TRX_DICT_OP_INDEX:
osku's avatar
osku committed
3658
		ut_print_timestamp(stderr);
3659 3660 3661
		fputs("  InnoDB: Error: a table lock wait happens"
		      " in a dictionary operation!\n"
		      "InnoDB: Table name ", stderr);
3662
		ut_print_name(stderr, trx, TRUE, table->name);
osku's avatar
osku committed
3663
		fputs(".\n"
3664 3665 3666
		      "InnoDB: Submit a detailed bug report"
		      " to http://bugs.mysql.com\n",
		      stderr);
osku's avatar
osku committed
3667
	}
3668

osku's avatar
osku committed
3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683 3684 3685 3686
	/* 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)) {

		lock_reset_lock_and_trx_wait(lock);
		lock_table_remove_low(lock);

		return(DB_DEADLOCK);
	}

	if (trx->wait_lock == NULL) {
		/* Deadlock resolution chose another transaction as a victim,
		and we accidentally got our lock granted! */
3687

osku's avatar
osku committed
3688 3689
		return(DB_SUCCESS);
	}
3690

osku's avatar
osku committed
3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711
	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);
}

/*************************************************************************
Checks if other transactions have an incompatible mode lock request in
the lock queue. */
UNIV_INLINE
ibool
lock_table_other_has_incompatible(
/*==============================*/
	trx_t*		trx,	/* in: transaction, or NULL if all
				transactions should be included */
	ulint		wait,	/* in: LOCK_WAIT if also waiting locks are
				taken into account, or 0 if not */
	dict_table_t*	table,	/* in: table */
3712
	enum lock_mode	mode)	/* in: lock mode */
osku's avatar
osku committed
3713 3714 3715 3716 3717 3718 3719 3720 3721
{
	lock_t*	lock;

	ut_ad(mutex_own(&kernel_mutex));

	lock = UT_LIST_GET_LAST(table->locks);

	while (lock != NULL) {

3722
		if ((lock->trx != trx)
3723 3724
		    && (!lock_mode_compatible(lock_get_mode(lock), mode))
		    && (wait || !(lock_get_wait(lock)))) {
osku's avatar
osku committed
3725

3726
			return(TRUE);
osku's avatar
osku committed
3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737
		}

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

	return(FALSE);
}

/*************************************************************************
Locks the specified database table in the mode given. If the lock cannot
be granted immediately, the query thread is put to wait. */
3738
UNIV_INTERN
osku's avatar
osku committed
3739 3740 3741 3742 3743 3744 3745 3746
ulint
lock_table(
/*=======*/
				/* out: DB_SUCCESS, DB_LOCK_WAIT,
				DB_DEADLOCK, or DB_QUE_THR_SUSPENDED */
	ulint		flags,	/* in: if BTR_NO_LOCKING_FLAG bit is set,
				does nothing */
	dict_table_t*	table,	/* in: database table in dictionary cache */
3747
	enum lock_mode	mode,	/* in: lock mode */
osku's avatar
osku committed
3748 3749 3750 3751
	que_thr_t*	thr)	/* in: query thread */
{
	trx_t*	trx;
	ulint	err;
3752

osku's avatar
osku committed
3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778
	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)) {
3779

osku's avatar
osku committed
3780 3781 3782 3783
		/* 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);
3784

osku's avatar
osku committed
3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800
		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);
}

/*************************************************************************
Checks if there are any locks set on the table. */
3801
UNIV_INTERN
osku's avatar
osku committed
3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836
ibool
lock_is_on_table(
/*=============*/
				/* out: TRUE if there are lock(s) */
	dict_table_t*	table)	/* in: database table in dictionary cache */
{
	ibool	ret;

	ut_ad(table);

	lock_mutex_enter_kernel();

	if (UT_LIST_GET_LAST(table->locks)) {
		ret = TRUE;
	} else {
		ret = FALSE;
	}

	lock_mutex_exit_kernel();

	return(ret);
}

/*************************************************************************
Checks if a waiting table lock request still has to wait in a queue. */
static
ibool
lock_table_has_to_wait_in_queue(
/*============================*/
				/* out: TRUE if still has to wait */
	lock_t*	wait_lock)	/* in: waiting table lock */
{
	dict_table_t*	table;
	lock_t*		lock;

3837 3838
	ut_ad(lock_get_wait(wait_lock));

osku's avatar
osku committed
3839 3840 3841 3842 3843 3844 3845 3846
	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)) {

3847
			return(TRUE);
osku's avatar
osku committed
3848 3849 3850 3851 3852 3853 3854 3855 3856 3857 3858 3859 3860 3861 3862 3863 3864 3865 3866 3867 3868
		}

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

	return(FALSE);
}

/*****************************************************************
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(
/*===============*/
	lock_t*	in_lock)/* in: table lock object; transactions waiting
			behind will get their lock requests granted, if
			they are now qualified to it */
{
	lock_t*	lock;
3869

osku's avatar
osku committed
3870
	ut_ad(mutex_own(&kernel_mutex));
3871
	ut_a(lock_get_type_low(in_lock) == LOCK_TABLE);
osku's avatar
osku committed
3872 3873 3874 3875 3876 3877 3878 3879 3880 3881 3882

	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)
3883
		    && !lock_table_has_to_wait_in_queue(lock)) {
osku's avatar
osku committed
3884 3885 3886 3887 3888 3889 3890

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

		lock = UT_LIST_GET_NEXT(un_member.tab_lock.locks, lock);
	}
3891
}
osku's avatar
osku committed
3892 3893 3894

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

3895 3896 3897 3898
/*****************************************************************
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. */
3899
UNIV_INTERN
3900 3901 3902
void
lock_rec_unlock(
/*============*/
3903 3904 3905 3906
	trx_t*			trx,	/* in: transaction that has
					set a record lock */
	const buf_block_t*	block,	/* in: buffer block containing rec */
	const rec_t*		rec,	/* in: record */
3907
	enum lock_mode		lock_mode)/* in: LOCK_S or LOCK_X */
3908 3909 3910 3911 3912 3913
{
	lock_t*	lock;
	lock_t*	release_lock	= NULL;
	ulint	heap_no;

	ut_ad(trx && rec);
3914
	ut_ad(block->frame == page_align(rec));
3915

3916
	heap_no = page_rec_get_heap_no(rec);
3917

3918
	mutex_enter(&kernel_mutex);
3919

3920
	lock = lock_rec_get_first(block, heap_no);
3921 3922 3923 3924 3925 3926 3927 3928 3929 3930 3931 3932 3933 3934 3935 3936 3937 3938 3939 3940 3941

	/* 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,
3942 3943 3944
			"  InnoDB: Error: unlock row could not"
			" find a %lu mode lock on the record\n",
			(ulong) lock_mode);
3945 3946 3947 3948 3949 3950

		return;
	}

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

3951
	lock = lock_rec_get_first(block, heap_no);
3952 3953 3954

	while (lock != NULL) {
		if (lock_get_wait(lock)
3955
		    && !lock_rec_has_to_wait_in_queue(lock)) {
3956 3957 3958 3959 3960 3961 3962 3963 3964

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

		lock = lock_rec_get_next(heap_no, lock);
	}

	mutex_exit(&kernel_mutex);
3965
}
3966

osku's avatar
osku committed
3967 3968 3969
/*************************************************************************
Releases a table lock.
Releases possible other transactions waiting for this lock. */
3970
UNIV_INTERN
osku's avatar
osku committed
3971 3972 3973 3974 3975 3976 3977 3978 3979 3980 3981 3982 3983 3984 3985
void
lock_table_unlock(
/*==============*/
	lock_t*	lock)	/* in: lock */
{
	mutex_enter(&kernel_mutex);

	lock_table_dequeue(lock);

	mutex_exit(&kernel_mutex);
}

/*************************************************************************
Releases transaction locks, and releases possible other transactions waiting
because of these locks. */
3986
UNIV_INTERN
osku's avatar
osku committed
3987 3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998
void
lock_release_off_kernel(
/*====================*/
	trx_t*	trx)	/* in: transaction */
{
	dict_table_t*	table;
	ulint		count;
	lock_t*		lock;

	ut_ad(mutex_own(&kernel_mutex));

	lock = UT_LIST_GET_LAST(trx->trx_locks);
3999

osku's avatar
osku committed
4000 4001 4002 4003 4004 4005
	count = 0;

	while (lock != NULL) {

		count++;

4006
		if (lock_get_type_low(lock) == LOCK_REC) {
4007

osku's avatar
osku committed
4008 4009
			lock_rec_dequeue_from_page(lock);
		} else {
4010
			ut_ad(lock_get_type_low(lock) & LOCK_TABLE);
osku's avatar
osku committed
4011 4012

			if (lock_get_mode(lock) != LOCK_IS
4013
			    && !ut_dulint_is_zero(trx->undo_no)) {
osku's avatar
osku committed
4014

4015 4016 4017
				/* The trx may have modified the table. We
				block the use of the MySQL query cache for
				all currently active transactions. */
osku's avatar
osku committed
4018 4019

				table = lock->un_member.tab_lock.table;
4020

4021 4022
				table->query_cache_inv_trx_id
					= trx_sys->max_trx_id;
osku's avatar
osku committed
4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036
			}

			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;
4037
		}
osku's avatar
osku committed
4038 4039 4040 4041

		lock = UT_LIST_GET_LAST(trx->trx_locks);
	}

4042
	ut_a(ib_vector_size(trx->autoinc_locks) == 0);
osku's avatar
osku committed
4043

4044
	mem_heap_empty(trx->lock_heap);
osku's avatar
osku committed
4045 4046 4047 4048 4049
}

/*************************************************************************
Cancels a waiting lock request and releases possible other transactions
waiting behind it. */
4050
UNIV_INTERN
osku's avatar
osku committed
4051 4052 4053 4054 4055 4056 4057
void
lock_cancel_waiting_and_release(
/*============================*/
	lock_t*	lock)	/* in: waiting lock request */
{
	ut_ad(mutex_own(&kernel_mutex));

4058
	if (lock_get_type_low(lock) == LOCK_REC) {
4059

osku's avatar
osku committed
4060 4061
		lock_rec_dequeue_from_page(lock);
	} else {
4062
		ut_ad(lock_get_type_low(lock) & LOCK_TABLE);
osku's avatar
osku committed
4063

4064 4065 4066 4067 4068
		if (lock->trx->autoinc_locks != NULL) {
			/* Release the transaction's AUTOINC locks/ */
			lock_release_autoinc_locks(lock->trx);
		}

osku's avatar
osku committed
4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080
		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);
}

4081 4082 4083 4084 4085 4086
/* 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)


osku's avatar
osku committed
4087
/*************************************************************************
4088 4089 4090 4091
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. */
osku's avatar
osku committed
4092 4093
static
void
4094 4095 4096 4097 4098 4099
lock_remove_all_on_table_for_trx(
/*=============================*/
	dict_table_t*	table,			/* in: table to be dropped */
	trx_t*		trx,			/* in: a transaction */
	ibool		remove_also_table_sx_locks)/* in: also removes
						table S and X locks */
osku's avatar
osku committed
4100 4101 4102 4103 4104 4105 4106
{
	lock_t*	lock;
	lock_t*	prev_lock;

	ut_ad(mutex_own(&kernel_mutex));

	lock = UT_LIST_GET_LAST(trx->trx_locks);
4107

osku's avatar
osku committed
4108 4109
	while (lock != NULL) {
		prev_lock = UT_LIST_GET_PREV(trx_locks, lock);
4110

4111
		if (lock_get_type_low(lock) == LOCK_REC
4112
		    && lock->index->table == table) {
osku's avatar
osku committed
4113
			ut_a(!lock_get_wait(lock));
4114

osku's avatar
osku committed
4115
			lock_rec_discard(lock);
4116
		} else if (lock_get_type_low(lock) & LOCK_TABLE
4117 4118 4119
			   && lock->un_member.tab_lock.table == table
			   && (remove_also_table_sx_locks
			       || !IS_LOCK_S_OR_X(lock))) {
osku's avatar
osku committed
4120 4121

			ut_a(!lock_get_wait(lock));
4122

osku's avatar
osku committed
4123 4124 4125 4126 4127 4128 4129 4130
			lock_table_remove_low(lock);
		}

		lock = prev_lock;
	}
}

/*************************************************************************
4131 4132 4133 4134
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. */
4135
UNIV_INTERN
osku's avatar
osku committed
4136
void
4137 4138 4139 4140 4141 4142
lock_remove_all_on_table(
/*=====================*/
	dict_table_t*	table,			/* in: table to be dropped
						or truncated */
	ibool		remove_also_table_sx_locks)/* in: also removes
						table S and X locks */
osku's avatar
osku committed
4143 4144
{
	lock_t*	lock;
4145
	lock_t*	prev_lock;
osku's avatar
osku committed
4146 4147 4148 4149 4150

	mutex_enter(&kernel_mutex);

	lock = UT_LIST_GET_FIRST(table->locks);

4151 4152 4153 4154
	while (lock != NULL) {

		prev_lock = UT_LIST_GET_PREV(un_member.tab_lock.locks,
					     lock);
osku's avatar
osku committed
4155

4156 4157 4158 4159 4160 4161
		/* 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))) {
osku's avatar
osku committed
4162

4163 4164 4165 4166 4167 4168 4169 4170 4171 4172 4173 4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189
			ut_a(!lock_get_wait(lock));
		}

		lock_remove_all_on_table_for_trx(table, lock->trx,
						 remove_also_table_sx_locks);

		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);
		}
osku's avatar
osku committed
4190 4191 4192 4193 4194 4195 4196 4197 4198
	}

	mutex_exit(&kernel_mutex);
}

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

/*************************************************************************
Prints info of a table lock. */
4199
UNIV_INTERN
osku's avatar
osku committed
4200 4201 4202
void
lock_table_print(
/*=============*/
4203 4204
	FILE*		file,	/* in: file where to print */
	const lock_t*	lock)	/* in: table type lock */
osku's avatar
osku committed
4205 4206
{
	ut_ad(mutex_own(&kernel_mutex));
4207
	ut_a(lock_get_type_low(lock) == LOCK_TABLE);
osku's avatar
osku committed
4208 4209

	fputs("TABLE LOCK table ", file);
4210
	ut_print_name(file, lock->trx, TRUE,
4211
		      lock->un_member.tab_lock.table->name);
vasil's avatar
vasil committed
4212 4213
	fprintf(file, " trx id " TRX_ID_FMT,
		TRX_ID_PREP_PRINTF(lock->trx->id));
osku's avatar
osku committed
4214 4215 4216 4217 4218 4219 4220 4221 4222 4223 4224 4225

	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 {
4226 4227
		fprintf(file, " unknown lock mode %lu",
			(ulong) lock_get_mode(lock));
osku's avatar
osku committed
4228 4229 4230 4231 4232 4233 4234
	}

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

	putc('\n', file);
4235 4236
}

osku's avatar
osku committed
4237 4238
/*************************************************************************
Prints info of a record lock. */
4239
UNIV_INTERN
osku's avatar
osku committed
4240 4241 4242
void
lock_rec_print(
/*===========*/
4243 4244
	FILE*		file,	/* in: file where to print */
	const lock_t*	lock)	/* in: record type lock */
osku's avatar
osku committed
4245
{
4246 4247 4248 4249 4250 4251 4252 4253
	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_;
4254
	rec_offs_init(offsets_);
osku's avatar
osku committed
4255 4256

	ut_ad(mutex_own(&kernel_mutex));
4257
	ut_a(lock_get_type_low(lock) == LOCK_REC);
osku's avatar
osku committed
4258 4259

	space = lock->un_member.rec_lock.space;
4260
	page_no = lock->un_member.rec_lock.page_no;
osku's avatar
osku committed
4261 4262

	fprintf(file, "RECORD LOCKS space id %lu page no %lu n bits %lu ",
4263 4264
		(ulong) space, (ulong) page_no,
		(ulong) lock_rec_get_n_bits(lock));
osku's avatar
osku committed
4265
	dict_index_name_print(file, lock->trx, lock->index);
vasil's avatar
vasil committed
4266 4267
	fprintf(file, " trx id " TRX_ID_FMT,
		TRX_ID_PREP_PRINTF(lock->trx->id));
osku's avatar
osku committed
4268 4269 4270 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

	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);

vasil's avatar
vasil committed
4297
	block = buf_page_try_get(space, page_no, &mtr);
4298

4299
	if (block) {
4300
		for (i = 0; i < lock_rec_get_n_bits(lock); i++) {
osku's avatar
osku committed
4301

4302
			if (lock_rec_get_nth_bit(lock, i)) {
osku's avatar
osku committed
4303

4304
				const rec_t*	rec
4305 4306
					= page_find_rec_with_heap_no(
						buf_block_get_frame(block), i);
4307 4308 4309
				offsets = rec_get_offsets(
					rec, lock->index, offsets,
					ULINT_UNDEFINED, &heap);
4310 4311 4312

				fprintf(file, "Record lock, heap no %lu ",
					(ulong) i);
osku's avatar
osku committed
4313
				rec_print_new(file, rec, offsets);
4314
				putc('\n', file);
osku's avatar
osku committed
4315
			}
4316 4317 4318 4319
		}
	} else {
		for (i = 0; i < lock_rec_get_n_bits(lock); i++) {
			fprintf(file, "Record lock, heap no %lu\n", (ulong) i);
osku's avatar
osku committed
4320 4321 4322 4323 4324 4325 4326 4327 4328
		}
	}

	mtr_commit(&mtr);
	if (UNIV_LIKELY_NULL(heap)) {
		mem_heap_free(heap);
	}
}

4329
#ifndef UNIV_HOTBACKUP
4330 4331 4332 4333 4334 4335 4336 4337 4338

#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
osku's avatar
osku committed
4339 4340 4341 4342 4343 4344 4345 4346 4347 4348 4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359 4360 4361 4362 4363 4364
/*************************************************************************
Calculates the number of record lock structs in the record lock hash table. */
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);
}
4365
#endif /* PRINT_NUM_OF_LOCK_STRUCTS */
osku's avatar
osku committed
4366 4367 4368

/*************************************************************************
Prints info of locks for all transactions. */
4369
UNIV_INTERN
osku's avatar
osku committed
4370 4371 4372 4373 4374 4375 4376 4377 4378 4379 4380 4381 4382
void
lock_print_info_summary(
/*====================*/
	FILE*	file)	/* in: file where to print */
{
	/* We must protect the MySQL thd->query field with a MySQL mutex, and
	because the MySQL mutex must be reserved before the kernel_mutex of
	InnoDB, we call innobase_mysql_prepare_print_arbitrary_thd() here. */

	innobase_mysql_prepare_print_arbitrary_thd();
	lock_mutex_enter_kernel();

	if (lock_deadlock_found) {
4383 4384 4385
		fputs("------------------------\n"
		      "LATEST DETECTED DEADLOCK\n"
		      "------------------------\n", file);
osku's avatar
osku committed
4386 4387 4388 4389

		ut_copy_file(file, lock_latest_err_file);
	}

4390 4391 4392
	fputs("------------\n"
	      "TRANSACTIONS\n"
	      "------------\n", file);
osku's avatar
osku committed
4393

vasil's avatar
vasil committed
4394 4395
	fprintf(file, "Trx id counter " TRX_ID_FMT "\n",
		TRX_ID_PREP_PRINTF(trx_sys->max_trx_id));
osku's avatar
osku committed
4396 4397

	fprintf(file,
vasil's avatar
vasil committed
4398 4399 4400 4401
		"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));
osku's avatar
osku committed
4402 4403

	fprintf(file,
4404 4405
		"History list length %lu\n",
		(ulong) trx_sys->rseg_history_len);
osku's avatar
osku committed
4406

4407
#ifdef PRINT_NUM_OF_LOCK_STRUCTS
osku's avatar
osku committed
4408 4409
	fprintf(file,
		"Total number of lock structs in row lock hash table %lu\n",
4410
		(ulong) lock_get_n_rec_locks());
4411
#endif /* PRINT_NUM_OF_LOCK_STRUCTS */
osku's avatar
osku committed
4412 4413 4414 4415
}

/*************************************************************************
Prints info of locks for each transaction. */
4416
UNIV_INTERN
osku's avatar
osku committed
4417 4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430 4431 4432 4433 4434 4435 4436 4437 4438 4439 4440
void
lock_print_info_all_transactions(
/*=============================*/
	FILE*	file)	/* in: file where to print */
{
	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);
		}
4441

osku's avatar
osku committed
4442 4443 4444 4445 4446 4447 4448 4449 4450 4451 4452 4453
		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. */
4454

osku's avatar
osku committed
4455 4456 4457 4458 4459 4460 4461 4462 4463 4464 4465 4466 4467 4468 4469 4470 4471
	while (trx && (i < nth_trx)) {
		trx = UT_LIST_GET_NEXT(trx_list, trx);
		i++;
	}

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

		ut_ad(lock_validate());

		return;
	}

	if (nth_lock == 0) {
		fputs("---", file);
		trx_print(file, trx, 600);
4472 4473

		if (trx->read_view) {
osku's avatar
osku committed
4474
			fprintf(file,
4475
				"Trx read view will not see trx with"
vasil's avatar
vasil committed
4476 4477 4478
				" id >= " TRX_ID_FMT
				", sees < " TRX_ID_FMT "\n",
				TRX_ID_PREP_PRINTF(
4479
					trx->read_view->low_limit_id),
vasil's avatar
vasil committed
4480
				TRX_ID_PREP_PRINTF(
4481
					trx->read_view->up_limit_id));
4482
		}
osku's avatar
osku committed
4483 4484 4485

		if (trx->que_state == TRX_QUE_LOCK_WAIT) {
			fprintf(file,
4486 4487 4488 4489
				"------- TRX HAS BEEN WAITING %lu SEC"
				" FOR THIS LOCK TO BE GRANTED:\n",
				(ulong) difftime(time(NULL),
						 trx->wait_started));
osku's avatar
osku committed
4490

4491
			if (lock_get_type_low(trx->wait_lock) == LOCK_REC) {
osku's avatar
osku committed
4492 4493 4494 4495 4496 4497 4498 4499 4500 4501
				lock_rec_print(file, trx->wait_lock);
			} else {
				lock_table_print(file, trx->wait_lock);
			}

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

	if (!srv_print_innodb_lock_monitor) {
4502 4503
		nth_trx++;
		goto loop;
osku's avatar
osku committed
4504 4505 4506 4507 4508 4509
	}

	i = 0;

	/* Look at the note about the trx loop above why we loop here:
	lock may be an obsolete pointer now. */
4510

osku's avatar
osku committed
4511
	lock = UT_LIST_GET_FIRST(trx->trx_locks);
4512

osku's avatar
osku committed
4513 4514 4515 4516 4517 4518 4519 4520 4521 4522 4523 4524
	while (lock && (i < nth_lock)) {
		lock = UT_LIST_GET_NEXT(trx_locks, lock);
		i++;
	}

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

		goto loop;
	}

4525
	if (lock_get_type_low(lock) == LOCK_REC) {
4526
		if (load_page_first) {
4527 4528 4529 4530
			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;

osku's avatar
osku committed
4531 4532 4533 4534
			lock_mutex_exit_kernel();
			innobase_mysql_end_print_arbitrary_thd();

			mtr_start(&mtr);
4535

4536 4537
			buf_page_get_with_no_latch(space, zip_size,
						   page_no, &mtr);
osku's avatar
osku committed
4538 4539 4540 4541 4542 4543 4544 4545 4546 4547

			mtr_commit(&mtr);

			load_page_first = FALSE;

			innobase_mysql_prepare_print_arbitrary_thd();
			lock_mutex_enter_kernel();

			goto loop;
		}
4548

osku's avatar
osku committed
4549 4550
		lock_rec_print(file, lock);
	} else {
4551
		ut_ad(lock_get_type_low(lock) & LOCK_TABLE);
4552

osku's avatar
osku committed
4553 4554 4555 4556 4557 4558 4559 4560
		lock_table_print(file, lock);
	}

	load_page_first = TRUE;

	nth_lock++;

	if (nth_lock >= 10) {
4561 4562 4563
		fputs("10 LOCKS PRINTED FOR THIS TRX:"
		      " SUPPRESSING FURTHER PRINTS\n",
		      file);
4564

osku's avatar
osku committed
4565 4566 4567 4568 4569 4570 4571 4572 4573
		nth_trx++;
		nth_lock = 0;

		goto loop;
	}

	goto loop;
}

4574
# ifdef UNIV_DEBUG
osku's avatar
osku committed
4575 4576
/*************************************************************************
Validates the lock queue on a table. */
4577
static
osku's avatar
osku committed
4578 4579 4580 4581 4582 4583 4584 4585 4586 4587 4588 4589 4590 4591
ibool
lock_table_queue_validate(
/*======================*/
				/* out: TRUE if ok */
	dict_table_t*	table)	/* in: table */
{
	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)
4592 4593
		     || ((lock->trx)->conc_state == TRX_PREPARED)
		     || ((lock->trx)->conc_state == TRX_COMMITTED_IN_MEMORY));
4594

osku's avatar
osku committed
4595 4596
		if (!lock_get_wait(lock)) {

4597 4598 4599
			ut_a(!lock_table_other_has_incompatible(
				     lock->trx, 0, table,
				     lock_get_mode(lock)));
osku's avatar
osku committed
4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612
		} else {

			ut_a(lock_table_has_to_wait_in_queue(lock));
		}

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

	return(TRUE);
}

/*************************************************************************
Validates the lock queue on a single record. */
4613
static
osku's avatar
osku committed
4614 4615 4616
ibool
lock_rec_queue_validate(
/*====================*/
4617 4618 4619 4620 4621
					/* out: TRUE if ok */
	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) */
osku's avatar
osku committed
4622
{
4623
	trx_t*	impl_trx;
osku's avatar
osku committed
4624
	lock_t*	lock;
marko's avatar
marko committed
4625
	ulint	heap_no;
osku's avatar
osku committed
4626 4627

	ut_a(rec);
4628
	ut_a(block->frame == page_align(rec));
osku's avatar
osku committed
4629 4630 4631
	ut_ad(rec_offs_validate(rec, index, offsets));
	ut_ad(!page_rec_is_comp(rec) == !rec_offs_comp(offsets));

4632
	heap_no = page_rec_get_heap_no(rec);
osku's avatar
osku committed
4633

4634
	lock_mutex_enter_kernel();
marko's avatar
marko committed
4635

osku's avatar
osku committed
4636 4637
	if (!page_rec_is_user_rec(rec)) {

4638
		lock = lock_rec_get_first(block, heap_no);
osku's avatar
osku committed
4639 4640

		while (lock) {
4641 4642 4643 4644 4645 4646 4647 4648
			switch(lock->trx->conc_state) {
			case TRX_ACTIVE:
			case TRX_PREPARED:
			case TRX_COMMITTED_IN_MEMORY:
				break;
			default:
				ut_error;
			}
4649

osku's avatar
osku committed
4650
			ut_a(trx_in_trx_list(lock->trx));
4651

osku's avatar
osku committed
4652 4653 4654 4655 4656 4657 4658 4659
			if (lock_get_wait(lock)) {
				ut_a(lock_rec_has_to_wait_in_queue(lock));
			}

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

marko's avatar
marko committed
4660
			lock = lock_rec_get_next(heap_no, lock);
osku's avatar
osku committed
4661 4662 4663 4664
		}

		lock_mutex_exit_kernel();

4665
		return(TRUE);
osku's avatar
osku committed
4666 4667
	}

marko's avatar
marko committed
4668
	if (!index);
4669
	else if (dict_index_is_clust(index)) {
4670

osku's avatar
osku committed
4671 4672
		impl_trx = lock_clust_rec_some_has_impl(rec, index, offsets);

4673 4674 4675
		if (impl_trx
		    && lock_rec_other_has_expl_req(LOCK_S, 0, LOCK_WAIT,
						   block, heap_no, impl_trx)) {
osku's avatar
osku committed
4676

4677
			ut_a(lock_rec_has_expl(LOCK_X | LOCK_REC_NOT_GAP,
4678
					       block, heap_no, impl_trx));
osku's avatar
osku committed
4679
		}
marko's avatar
marko committed
4680
	} else {
4681

osku's avatar
osku committed
4682 4683 4684
		/* The kernel mutex may get released temporarily in the
		next function call: we have to release lock table mutex
		to obey the latching order */
4685

4686 4687
		impl_trx = lock_sec_rec_some_has_impl_off_kernel(
			rec, index, offsets);
osku's avatar
osku committed
4688

4689 4690 4691
		if (impl_trx
		    && lock_rec_other_has_expl_req(LOCK_S, 0, LOCK_WAIT,
						   block, heap_no, impl_trx)) {
osku's avatar
osku committed
4692 4693

			ut_a(lock_rec_has_expl(LOCK_X | LOCK_REC_NOT_GAP,
4694
					       block, heap_no, impl_trx));
osku's avatar
osku committed
4695 4696 4697
		}
	}

4698
	lock = lock_rec_get_first(block, heap_no);
osku's avatar
osku committed
4699 4700 4701

	while (lock) {
		ut_a(lock->trx->conc_state == TRX_ACTIVE
4702 4703
		     || lock->trx->conc_state == TRX_PREPARED
		     || lock->trx->conc_state == TRX_COMMITTED_IN_MEMORY);
osku's avatar
osku committed
4704
		ut_a(trx_in_trx_list(lock->trx));
4705

osku's avatar
osku committed
4706 4707 4708 4709 4710 4711
		if (index) {
			ut_a(lock->index == index);
		}

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

4712
			enum lock_mode	mode;
4713

osku's avatar
osku committed
4714 4715 4716 4717 4718
			if (lock_get_mode(lock) == LOCK_S) {
				mode = LOCK_X;
			} else {
				mode = LOCK_S;
			}
4719
			ut_a(!lock_rec_other_has_expl_req(
4720
				     mode, 0, 0, block, heap_no, lock->trx));
osku's avatar
osku committed
4721 4722 4723 4724 4725 4726

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

			ut_a(lock_rec_has_to_wait_in_queue(lock));
		}

marko's avatar
marko committed
4727
		lock = lock_rec_get_next(heap_no, lock);
osku's avatar
osku committed
4728 4729 4730 4731 4732 4733 4734 4735 4736
	}

	lock_mutex_exit_kernel();

	return(TRUE);
}

/*************************************************************************
Validates the record lock queues on a page. */
4737
static
osku's avatar
osku committed
4738 4739 4740 4741 4742 4743 4744 4745
ibool
lock_rec_validate_page(
/*===================*/
			/* out: TRUE if ok */
	ulint	space,	/* in: space id */
	ulint	page_no)/* in: page number */
{
	dict_index_t*	index;
4746
	buf_block_t*	block;
4747
	const page_t*	page;
4748
	lock_t*		lock;
4749
	const rec_t*	rec;
4750 4751 4752 4753
	ulint		nth_lock	= 0;
	ulint		nth_bit		= 0;
	ulint		i;
	mtr_t		mtr;
osku's avatar
osku committed
4754 4755 4756
	mem_heap_t*	heap		= NULL;
	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
	ulint*		offsets		= offsets_;
4757
	rec_offs_init(offsets_);
osku's avatar
osku committed
4758 4759 4760 4761

	ut_ad(!mutex_own(&kernel_mutex));

	mtr_start(&mtr);
4762

4763 4764
	block = buf_page_get(space, fil_space_get_zip_size(space),
			     page_no, RW_X_LATCH, &mtr);
4765
	buf_block_dbg_add_level(block, SYNC_NO_ORDER_CHECK);
4766

4767
	page = block->frame;
osku's avatar
osku committed
4768 4769

	lock_mutex_enter_kernel();
4770
loop:
osku's avatar
osku committed
4771 4772 4773 4774 4775 4776 4777 4778 4779 4780 4781 4782 4783 4784 4785 4786 4787
	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
4788 4789
	     || lock->trx->conc_state == TRX_PREPARED
	     || lock->trx->conc_state == TRX_COMMITTED_IN_MEMORY);
4790

osku's avatar
osku committed
4791 4792 4793 4794 4795 4796
	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);
4797
			ut_a(rec);
osku's avatar
osku committed
4798
			offsets = rec_get_offsets(rec, index, offsets,
4799
						  ULINT_UNDEFINED, &heap);
osku's avatar
osku committed
4800 4801

			fprintf(stderr,
4802 4803
				"Validating %lu %lu\n",
				(ulong) space, (ulong) page_no);
osku's avatar
osku committed
4804 4805 4806

			lock_mutex_exit_kernel();

4807
			lock_rec_queue_validate(block, rec, index, offsets);
osku's avatar
osku committed
4808 4809 4810 4811 4812 4813 4814 4815 4816 4817 4818 4819 4820 4821 4822 4823 4824 4825 4826 4827 4828 4829 4830

			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);
4831 4832
}

osku's avatar
osku committed
4833 4834
/*************************************************************************
Validates the lock system. */
4835
static
osku's avatar
osku committed
4836 4837 4838 4839 4840 4841 4842 4843 4844 4845 4846 4847 4848
ibool
lock_validate(void)
/*===============*/
			/* out: TRUE if ok */
{
	lock_t*	lock;
	trx_t*	trx;
	dulint	limit;
	ulint	space;
	ulint	page_no;
	ulint	i;

	lock_mutex_enter_kernel();
4849

osku's avatar
osku committed
4850 4851 4852 4853
	trx = UT_LIST_GET_FIRST(trx_sys->trx_list);

	while (trx) {
		lock = UT_LIST_GET_FIRST(trx->trx_locks);
4854

osku's avatar
osku committed
4855
		while (lock) {
4856
			if (lock_get_type_low(lock) & LOCK_TABLE) {
4857

4858 4859
				lock_table_queue_validate(
					lock->un_member.tab_lock.table);
osku's avatar
osku committed
4860
			}
4861

osku's avatar
osku committed
4862 4863
			lock = UT_LIST_GET_NEXT(trx_locks, lock);
		}
4864

osku's avatar
osku committed
4865 4866 4867 4868 4869 4870 4871 4872 4873 4874 4875 4876 4877 4878 4879
		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;
4880

4881 4882 4883
				if (ut_dulint_cmp(
					    ut_dulint_create(space, page_no),
					    limit) >= 0) {
osku's avatar
osku committed
4884 4885 4886 4887 4888 4889 4890 4891 4892 4893
					break;
				}

				lock = HASH_GET_NEXT(hash, lock);
			}

			if (!lock) {

				break;
			}
4894

osku's avatar
osku committed
4895 4896 4897 4898 4899 4900 4901 4902 4903 4904 4905 4906 4907 4908
			lock_mutex_exit_kernel();

			lock_rec_validate_page(space, page_no);

			lock_mutex_enter_kernel();

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

	lock_mutex_exit_kernel();

	return(TRUE);
}
4909
# endif /* UNIV_DEBUG */
osku's avatar
osku committed
4910 4911 4912 4913 4914 4915 4916 4917 4918
#endif /* !UNIV_HOTBACKUP */
/*============ RECORD LOCK CHECKS FOR ROW OPERATIONS ====================*/

/*************************************************************************
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
for a gap x-lock to the lock queue. */
4919
UNIV_INTERN
osku's avatar
osku committed
4920 4921 4922 4923 4924
ulint
lock_rec_insert_check_and_lock(
/*===========================*/
				/* out: DB_SUCCESS, DB_LOCK_WAIT,
				DB_DEADLOCK, or DB_QUE_THR_SUSPENDED */
4925 4926
	ulint		flags,	/* in: if BTR_NO_LOCKING_FLAG bit is
				set, does nothing */
osku's avatar
osku committed
4927
	rec_t*		rec,	/* in: record after which to insert */
4928
	buf_block_t*	block,	/* in/out: buffer block of rec */
osku's avatar
osku committed
4929 4930
	dict_index_t*	index,	/* in: index */
	que_thr_t*	thr,	/* in: query thread */
4931 4932 4933 4934
	ibool*		inherit)/* out: set to TRUE if the new
				inserted record maybe should inherit
				LOCK_GAP type locks from the successor
				record */
osku's avatar
osku committed
4935
{
4936 4937 4938 4939 4940
	const rec_t*	next_rec;
	trx_t*		trx;
	lock_t*		lock;
	ulint		err;
	ulint		next_rec_heap_no;
osku's avatar
osku committed
4941

4942 4943
	ut_ad(block->frame == page_align(rec));

osku's avatar
osku committed
4944 4945 4946 4947 4948 4949 4950
	if (flags & BTR_NO_LOCKING_FLAG) {

		return(DB_SUCCESS);
	}

	trx = thr_get_trx(thr);
	next_rec = page_rec_get_next(rec);
marko's avatar
marko committed
4951
	next_rec_heap_no = page_rec_get_heap_no(next_rec);
osku's avatar
osku committed
4952 4953 4954

	lock_mutex_enter_kernel();

4955 4956
	/* 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
marko's avatar
marko committed
4957
	the table must be at least S-locked. */
4958 4959
	ut_ad(lock_table_has(trx, index->table, LOCK_IX)
	      || (*index->name == TEMP_INDEX_PREFIX
marko's avatar
marko committed
4960
		  && lock_table_has(trx, index->table, LOCK_S)));
osku's avatar
osku committed
4961

4962
	lock = lock_rec_get_first(block, next_rec_heap_no);
marko's avatar
marko committed
4963 4964

	if (UNIV_LIKELY(lock == NULL)) {
osku's avatar
osku committed
4965 4966 4967 4968
		/* We optimize CPU time usage in the simplest case */

		lock_mutex_exit_kernel();

4969
		if (!dict_index_is_clust(index)) {
osku's avatar
osku committed
4970
			/* Update the page max trx id field */
marko's avatar
marko committed
4971
			page_update_max_trx_id(block,
4972
					       buf_block_get_page_zip(block),
4973
					       trx->id);
osku's avatar
osku committed
4974
		}
4975

marko's avatar
marko committed
4976 4977
		*inherit = FALSE;

osku's avatar
osku committed
4978 4979 4980 4981 4982 4983 4984 4985 4986 4987 4988 4989 4990 4991 4992
		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. */

4993 4994
	if (lock_rec_other_has_conflicting(
		    LOCK_X | LOCK_GAP | LOCK_INSERT_INTENTION,
4995
		    block, next_rec_heap_no, trx)) {
osku's avatar
osku committed
4996 4997 4998

		/* Note that we may get DB_SUCCESS also here! */
		err = lock_rec_enqueue_waiting(LOCK_X | LOCK_GAP
4999
					       | LOCK_INSERT_INTENTION,
5000 5001
					       block, next_rec_heap_no,
					       index, thr);
osku's avatar
osku committed
5002 5003 5004 5005 5006 5007
	} else {
		err = DB_SUCCESS;
	}

	lock_mutex_exit_kernel();

5008
	if ((err == DB_SUCCESS) && !dict_index_is_clust(index)) {
osku's avatar
osku committed
5009
		/* Update the page max trx id field */
marko's avatar
marko committed
5010
		page_update_max_trx_id(block,
5011
				       buf_block_get_page_zip(block),
5012
				       trx->id);
osku's avatar
osku committed
5013 5014 5015 5016 5017 5018 5019
	}

#ifdef UNIV_DEBUG
	{
		mem_heap_t*	heap		= NULL;
		ulint		offsets_[REC_OFFS_NORMAL_SIZE];
		const ulint*	offsets;
5020
		rec_offs_init(offsets_);
osku's avatar
osku committed
5021 5022

		offsets = rec_get_offsets(next_rec, index, offsets_,
5023
					  ULINT_UNDEFINED, &heap);
5024 5025
		ut_ad(lock_rec_queue_validate(block,
					      next_rec, index, offsets));
osku's avatar
osku committed
5026 5027 5028 5029 5030 5031 5032 5033 5034 5035 5036 5037 5038 5039 5040 5041 5042
		if (UNIV_LIKELY_NULL(heap)) {
			mem_heap_free(heap);
		}
	}
#endif /* UNIV_DEBUG */

	return(err);
}

/*************************************************************************
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(
/*==========================*/
5043 5044 5045 5046
	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) */
osku's avatar
osku committed
5047 5048 5049 5050 5051 5052 5053 5054
{
	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));

5055
	if (dict_index_is_clust(index)) {
osku's avatar
osku committed
5056 5057
		impl_trx = lock_clust_rec_some_has_impl(rec, index, offsets);
	} else {
5058 5059
		impl_trx = lock_sec_rec_some_has_impl_off_kernel(
			rec, index, offsets);
osku's avatar
osku committed
5060 5061 5062
	}

	if (impl_trx) {
5063 5064
		ulint	heap_no = page_rec_get_heap_no(rec);

osku's avatar
osku committed
5065 5066 5067
		/* If the transaction has no explicit x-lock set on the
		record, set one for it */

5068
		if (!lock_rec_has_expl(LOCK_X | LOCK_REC_NOT_GAP, block,
5069
				       heap_no, impl_trx)) {
osku's avatar
osku committed
5070

5071 5072
			lock_rec_add_to_queue(
				LOCK_REC | LOCK_X | LOCK_REC_NOT_GAP,
5073
				block, heap_no, index, impl_trx);
osku's avatar
osku committed
5074 5075 5076 5077 5078 5079 5080 5081 5082 5083 5084
		}
	}
}

/*************************************************************************
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
lock queue. */
5085
UNIV_INTERN
osku's avatar
osku committed
5086 5087 5088
ulint
lock_clust_rec_modify_check_and_lock(
/*=================================*/
5089 5090 5091 5092 5093 5094 5095 5096 5097 5098 5099
					/* out: DB_SUCCESS,
					DB_LOCK_WAIT, DB_DEADLOCK, or
					DB_QUE_THR_SUSPENDED */
	ulint			flags,	/* in: if BTR_NO_LOCKING_FLAG
					bit is set, does nothing */
	const buf_block_t*	block,	/* in: buffer block of rec */
	const rec_t*		rec,	/* in: record which should be
					modified */
	dict_index_t*		index,	/* in: clustered index */
	const ulint*		offsets,/* in: rec_get_offsets(rec, index) */
	que_thr_t*		thr)	/* in: query thread */
osku's avatar
osku committed
5100 5101
{
	ulint	err;
5102
	ulint	heap_no;
osku's avatar
osku committed
5103 5104

	ut_ad(rec_offs_validate(rec, index, offsets));
5105
	ut_ad(dict_index_is_clust(index));
5106
	ut_ad(block->frame == page_align(rec));
osku's avatar
osku committed
5107 5108 5109 5110 5111 5112

	if (flags & BTR_NO_LOCKING_FLAG) {

		return(DB_SUCCESS);
	}

5113 5114 5115 5116
	heap_no = rec_offs_comp(offsets)
		? rec_get_heap_no_new(rec)
		: rec_get_heap_no_old(rec);

osku's avatar
osku committed
5117 5118 5119 5120 5121 5122 5123
	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 */

5124
	lock_rec_convert_impl_to_expl(block, rec, index, offsets);
osku's avatar
osku committed
5125

5126 5127
	err = lock_rec_lock(TRUE, LOCK_X | LOCK_REC_NOT_GAP,
			    block, heap_no, index, thr);
osku's avatar
osku committed
5128 5129 5130

	lock_mutex_exit_kernel();

5131
	ut_ad(lock_rec_queue_validate(block, rec, index, offsets));
osku's avatar
osku committed
5132 5133 5134 5135 5136 5137 5138

	return(err);
}

/*************************************************************************
Checks if locks of other transactions prevent an immediate modify (delete
mark or delete unmark) of a secondary index record. */
5139
UNIV_INTERN
osku's avatar
osku committed
5140 5141 5142 5143 5144
ulint
lock_sec_rec_modify_check_and_lock(
/*===============================*/
				/* out: DB_SUCCESS, DB_LOCK_WAIT,
				DB_DEADLOCK, or DB_QUE_THR_SUSPENDED */
5145 5146 5147 5148 5149 5150 5151 5152
	ulint		flags,	/* in: if BTR_NO_LOCKING_FLAG
				bit is set, does nothing */
	buf_block_t*	block,	/* in/out: buffer block of rec */
	rec_t*		rec,	/* in: record which should be
				modified; NOTE: as this is a secondary
				index, we always have to modify the
				clustered index record first: see the
				comment below */
osku's avatar
osku committed
5153 5154 5155 5156
	dict_index_t*	index,	/* in: secondary index */
	que_thr_t*	thr)	/* in: query thread */
{
	ulint	err;
5157
	ulint	heap_no;
5158

5159 5160 5161
	ut_ad(!dict_index_is_clust(index));
	ut_ad(block->frame == page_align(rec));

osku's avatar
osku committed
5162 5163 5164 5165 5166
	if (flags & BTR_NO_LOCKING_FLAG) {

		return(DB_SUCCESS);
	}

5167 5168
	heap_no = page_rec_get_heap_no(rec);

osku's avatar
osku committed
5169 5170 5171 5172 5173 5174 5175 5176 5177
	/* 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));

5178 5179
	err = lock_rec_lock(TRUE, LOCK_X | LOCK_REC_NOT_GAP,
			    block, heap_no, index, thr);
osku's avatar
osku committed
5180 5181 5182 5183 5184 5185 5186 5187

	lock_mutex_exit_kernel();

#ifdef UNIV_DEBUG
	{
		mem_heap_t*	heap		= NULL;
		ulint		offsets_[REC_OFFS_NORMAL_SIZE];
		const ulint*	offsets;
5188
		rec_offs_init(offsets_);
osku's avatar
osku committed
5189 5190

		offsets = rec_get_offsets(rec, index, offsets_,
5191
					  ULINT_UNDEFINED, &heap);
5192
		ut_ad(lock_rec_queue_validate(block, rec, index, offsets));
osku's avatar
osku committed
5193 5194 5195 5196 5197 5198 5199
		if (UNIV_LIKELY_NULL(heap)) {
			mem_heap_free(heap);
		}
	}
#endif /* UNIV_DEBUG */

	if (err == DB_SUCCESS) {
5200
		/* Update the page max trx id field */
marko's avatar
marko committed
5201
		page_update_max_trx_id(block,
5202
				       buf_block_get_page_zip(block),
5203
				       thr_get_trx(thr)->id);
osku's avatar
osku committed
5204 5205 5206 5207 5208 5209 5210 5211
	}

	return(err);
}

/*************************************************************************
Like the counterpart for a clustered index below, but now we read a
secondary index record. */
5212
UNIV_INTERN
osku's avatar
osku committed
5213 5214 5215
ulint
lock_sec_rec_read_check_and_lock(
/*=============================*/
5216 5217 5218 5219 5220 5221
					/* out: DB_SUCCESS,
					DB_LOCK_WAIT, DB_DEADLOCK, or
					DB_QUE_THR_SUSPENDED */
	ulint			flags,	/* in: if BTR_NO_LOCKING_FLAG
					bit is set, does nothing */
	const buf_block_t*	block,	/* in: buffer block of rec */
5222
	const rec_t*		rec,	/* in: user record or page
5223 5224 5225 5226 5227
					supremum record which should
					be read or passed over by a
					read cursor */
	dict_index_t*		index,	/* in: secondary index */
	const ulint*		offsets,/* in: rec_get_offsets(rec, index) */
5228
	enum lock_mode		mode,	/* in: mode of the lock which
5229 5230 5231 5232 5233 5234 5235
					the read cursor should set on
					records: LOCK_S or LOCK_X; the
					latter is possible in
					SELECT FOR UPDATE */
	ulint			gap_mode,/* in: LOCK_ORDINARY, LOCK_GAP, or
					LOCK_REC_NOT_GAP */
	que_thr_t*		thr)	/* in: query thread */
osku's avatar
osku committed
5236 5237
{
	ulint	err;
5238
	ulint	heap_no;
osku's avatar
osku committed
5239

5240
	ut_ad(!dict_index_is_clust(index));
5241
	ut_ad(block->frame == page_align(rec));
osku's avatar
osku committed
5242 5243
	ut_ad(page_rec_is_user_rec(rec) || page_rec_is_supremum(rec));
	ut_ad(rec_offs_validate(rec, index, offsets));
marko's avatar
marko committed
5244
	ut_ad(mode == LOCK_X || mode == LOCK_S);
osku's avatar
osku committed
5245 5246 5247 5248 5249 5250

	if (flags & BTR_NO_LOCKING_FLAG) {

		return(DB_SUCCESS);
	}

5251 5252
	heap_no = page_rec_get_heap_no(rec);

osku's avatar
osku committed
5253 5254 5255
	lock_mutex_enter_kernel();

	ut_ad(mode != LOCK_X
5256
	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IX));
osku's avatar
osku committed
5257
	ut_ad(mode != LOCK_S
5258
	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IS));
osku's avatar
osku committed
5259 5260 5261 5262 5263

	/* 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. */

5264
	if (((ut_dulint_cmp(page_get_max_trx_id(block->frame),
5265 5266 5267
			    trx_list_get_min_trx_id()) >= 0)
	     || recv_recovery_is_on())
	    && !page_rec_is_supremum(rec)) {
osku's avatar
osku committed
5268

5269
		lock_rec_convert_impl_to_expl(block, rec, index, offsets);
osku's avatar
osku committed
5270 5271
	}

5272 5273
	err = lock_rec_lock(FALSE, mode | gap_mode,
			    block, heap_no, index, thr);
osku's avatar
osku committed
5274 5275 5276

	lock_mutex_exit_kernel();

5277
	ut_ad(lock_rec_queue_validate(block, rec, index, offsets));
osku's avatar
osku committed
5278 5279 5280 5281 5282 5283 5284 5285 5286 5287 5288

	return(err);
}

/*************************************************************************
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. */
5289
UNIV_INTERN
osku's avatar
osku committed
5290 5291 5292
ulint
lock_clust_rec_read_check_and_lock(
/*===============================*/
5293 5294 5295 5296 5297 5298 5299 5300 5301 5302 5303 5304
					/* out: DB_SUCCESS,
					DB_LOCK_WAIT, DB_DEADLOCK, or
					DB_QUE_THR_SUSPENDED */
	ulint			flags,	/* in: if BTR_NO_LOCKING_FLAG
					bit is set, does nothing */
	const buf_block_t*	block,	/* in: buffer block of rec */
	const rec_t*		rec,	/* in: user record or page
					supremum record which should
					be read or passed over by a
					read cursor */
	dict_index_t*		index,	/* in: clustered index */
	const ulint*		offsets,/* in: rec_get_offsets(rec, index) */
5305
	enum lock_mode		mode,	/* in: mode of the lock which
5306 5307 5308 5309 5310 5311 5312
					the read cursor should set on
					records: LOCK_S or LOCK_X; the
					latter is possible in
					SELECT FOR UPDATE */
	ulint			gap_mode,/* in: LOCK_ORDINARY, LOCK_GAP, or
					LOCK_REC_NOT_GAP */
	que_thr_t*		thr)	/* in: query thread */
osku's avatar
osku committed
5313 5314
{
	ulint	err;
5315
	ulint	heap_no;
osku's avatar
osku committed
5316

5317
	ut_ad(dict_index_is_clust(index));
5318
	ut_ad(block->frame == page_align(rec));
osku's avatar
osku committed
5319 5320
	ut_ad(page_rec_is_user_rec(rec) || page_rec_is_supremum(rec));
	ut_ad(gap_mode == LOCK_ORDINARY || gap_mode == LOCK_GAP
5321
	      || gap_mode == LOCK_REC_NOT_GAP);
osku's avatar
osku committed
5322 5323 5324 5325 5326 5327 5328
	ut_ad(rec_offs_validate(rec, index, offsets));

	if (flags & BTR_NO_LOCKING_FLAG) {

		return(DB_SUCCESS);
	}

5329 5330
	heap_no = page_rec_get_heap_no(rec);

osku's avatar
osku committed
5331 5332 5333
	lock_mutex_enter_kernel();

	ut_ad(mode != LOCK_X
5334
	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IX));
osku's avatar
osku committed
5335
	ut_ad(mode != LOCK_S
5336
	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IS));
5337

5338
	if (UNIV_LIKELY(heap_no != PAGE_HEAP_NO_SUPREMUM)) {
5339

5340
		lock_rec_convert_impl_to_expl(block, rec, index, offsets);
osku's avatar
osku committed
5341 5342
	}

5343 5344
	err = lock_rec_lock(FALSE, mode | gap_mode,
			    block, heap_no, index, thr);
osku's avatar
osku committed
5345 5346 5347

	lock_mutex_exit_kernel();

5348
	ut_ad(lock_rec_queue_validate(block, rec, index, offsets));
osku's avatar
osku committed
5349 5350 5351 5352 5353 5354 5355 5356 5357 5358 5359 5360

	return(err);
}
/*************************************************************************
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
"offsets". */
5361
UNIV_INTERN
osku's avatar
osku committed
5362 5363 5364
ulint
lock_clust_rec_read_check_and_lock_alt(
/*===================================*/
5365 5366 5367 5368 5369 5370 5371 5372 5373 5374 5375
					/* out: DB_SUCCESS,
					DB_LOCK_WAIT, DB_DEADLOCK, or
					DB_QUE_THR_SUSPENDED */
	ulint			flags,	/* in: if BTR_NO_LOCKING_FLAG
					bit is set, does nothing */
	const buf_block_t*	block,	/* in: buffer block of rec */
	const rec_t*		rec,	/* in: user record or page
					supremum record which should
					be read or passed over by a
					read cursor */
	dict_index_t*		index,	/* in: clustered index */
5376
	enum lock_mode		mode,	/* in: mode of the lock which
5377 5378 5379 5380 5381 5382 5383
					the read cursor should set on
					records: LOCK_S or LOCK_X; the
					latter is possible in
					SELECT FOR UPDATE */
	ulint			gap_mode,/* in: LOCK_ORDINARY, LOCK_GAP, or
					LOCK_REC_NOT_GAP */
	que_thr_t*		thr)	/* in: query thread */
osku's avatar
osku committed
5384 5385 5386 5387 5388
{
	mem_heap_t*	tmp_heap	= NULL;
	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
	ulint*		offsets		= offsets_;
	ulint		ret;
5389
	rec_offs_init(offsets_);
osku's avatar
osku committed
5390 5391

	offsets = rec_get_offsets(rec, index, offsets,
5392
				  ULINT_UNDEFINED, &tmp_heap);
5393
	ret = lock_clust_rec_read_check_and_lock(flags, block, rec, index,
5394
						 offsets, mode, gap_mode, thr);
osku's avatar
osku committed
5395 5396 5397 5398 5399
	if (tmp_heap) {
		mem_heap_free(tmp_heap);
	}
	return(ret);
}
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 5436 5437 5438 5439 5440 5441 5442 5443 5444 5445 5446 5447 5448 5449 5450 5451 5452 5453 5454
/***********************************************************************
Release the last lock from the transaction's autoinc locks. */
UNIV_INLINE
void
lock_release_autoinc_last_lock(
/*===========================*/
	ib_vector_t*	autoinc_locks)	/* in/out: vector of AUTOINC locks */
{
	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);
}

/***********************************************************************
Release all the transaction's autoinc locks. */
UNIV_INTERN
void
lock_release_autoinc_locks(
/*=======================*/
	trx_t*		trx)		/* in/out: transaction */
{
	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));
}

5455 5456 5457
/***********************************************************************
Gets the type of a lock. Non-inline version for using outside of the
lock module. */
5458
UNIV_INTERN
5459 5460 5461 5462 5463 5464 5465 5466 5467
ulint
lock_get_type(
/*==========*/
				/* out: LOCK_TABLE or LOCK_REC */
	const lock_t*	lock)	/* in: lock */
{
	return(lock_get_type_low(lock));
}

5468 5469
/***********************************************************************
Gets the id of the transaction owning a lock. */
5470
UNIV_INTERN
5471 5472 5473 5474 5475 5476 5477 5478 5479 5480 5481 5482
ullint
lock_get_trx_id(
/*============*/
				/* out: transaction id */
	const lock_t*	lock)	/* in: lock */
{
	return(trx_get_id(lock->trx));
}

/***********************************************************************
Gets the mode of a lock in a human readable string.
The string should not be free()'d or modified. */
5483
UNIV_INTERN
5484 5485 5486 5487 5488 5489
const char*
lock_get_mode_str(
/*==============*/
				/* out: lock mode */
	const lock_t*	lock)	/* in: lock */
{
vasil's avatar
vasil committed
5490 5491 5492 5493 5494
	ibool	is_gap_lock;

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

5495 5496
	switch (lock_get_mode(lock)) {
	case LOCK_S:
vasil's avatar
vasil committed
5497 5498 5499 5500 5501
		if (is_gap_lock) {
			return("S,GAP");
		} else {
			return("S");
		}
5502
	case LOCK_X:
vasil's avatar
vasil committed
5503 5504 5505 5506 5507
		if (is_gap_lock) {
			return("X,GAP");
		} else {
			return("X");
		}
5508
	case LOCK_IS:
vasil's avatar
vasil committed
5509 5510 5511 5512 5513
		if (is_gap_lock) {
			return("IS,GAP");
		} else {
			return("IS");
		}
5514
	case LOCK_IX:
vasil's avatar
vasil committed
5515 5516 5517 5518 5519
		if (is_gap_lock) {
			return("IX,GAP");
		} else {
			return("IX");
		}
5520 5521 5522 5523 5524 5525 5526 5527 5528 5529
	case LOCK_AUTO_INC:
		return("AUTO_INC");
	default:
		return("UNKNOWN");
	}
}

/***********************************************************************
Gets the type of a lock in a human readable string.
The string should not be free()'d or modified. */
5530
UNIV_INTERN
5531 5532 5533 5534 5535 5536
const char*
lock_get_type_str(
/*==============*/
				/* out: lock type */
	const lock_t*	lock)	/* in: lock */
{
5537
	switch (lock_get_type_low(lock)) {
5538 5539 5540 5541 5542 5543 5544 5545 5546 5547 5548 5549 5550 5551 5552 5553 5554 5555
	case LOCK_REC:
		return("RECORD");
	case LOCK_TABLE:
		return("TABLE");
	default:
		return("UNKNOWN");
	}
}

/***********************************************************************
Gets the table on which the lock is. */
UNIV_INLINE
dict_table_t*
lock_get_table(
/*===========*/
				/* out: table */
	const lock_t*	lock)	/* in: lock */
{
5556
	switch (lock_get_type_low(lock)) {
5557 5558 5559 5560 5561 5562
	case LOCK_REC:
		return(lock->index->table);
	case LOCK_TABLE:
		return(lock->un_member.tab_lock.table);
	default:
		ut_error;
5563
		return(NULL);
5564 5565 5566 5567 5568
	}
}

/***********************************************************************
Gets the id of the table on which the lock is. */
5569
UNIV_INTERN
5570 5571 5572 5573 5574 5575 5576 5577 5578 5579 5580 5581 5582 5583 5584 5585
ullint
lock_get_table_id(
/*==============*/
				/* out: id of the table */
	const lock_t*	lock)	/* in: lock */
{
	dict_table_t*	table;

	table = lock_get_table(lock);

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

/***********************************************************************
Gets the name of the table on which the lock is.
The string should not be free()'d or modified. */
5586
UNIV_INTERN
5587 5588 5589 5590 5591 5592 5593 5594 5595 5596 5597 5598 5599
const char*
lock_get_table_name(
/*================*/
				/* out: name of the table */
	const lock_t*	lock)	/* in: lock */
{
	dict_table_t*	table;

	table = lock_get_table(lock);

	return(table->name);
}

vasil's avatar
vasil committed
5600 5601
/***********************************************************************
For a record lock, gets the index on which the lock is. */
5602
UNIV_INTERN
vasil's avatar
vasil committed
5603 5604 5605 5606 5607 5608 5609 5610 5611 5612 5613
const dict_index_t*
lock_rec_get_index(
/*===============*/
				/* out: index */
	const lock_t*	lock)	/* in: lock */
{
	ut_a(lock_get_type_low(lock) == LOCK_REC);

	return(lock->index);
}

5614 5615 5616
/***********************************************************************
For a record lock, gets the name of the index on which the lock is.
The string should not be free()'d or modified. */
5617
UNIV_INTERN
5618 5619 5620 5621 5622 5623
const char*
lock_rec_get_index_name(
/*====================*/
				/* out: name of the index */
	const lock_t*	lock)	/* in: lock */
{
5624
	ut_a(lock_get_type_low(lock) == LOCK_REC);
5625 5626 5627 5628 5629 5630

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

/***********************************************************************
For a record lock, gets the tablespace number on which the lock is. */
5631
UNIV_INTERN
5632 5633 5634 5635 5636 5637
ulint
lock_rec_get_space_id(
/*==================*/
				/* out: tablespace number */
	const lock_t*	lock)	/* in: lock */
{
5638
	ut_a(lock_get_type_low(lock) == LOCK_REC);
5639 5640 5641 5642 5643 5644

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

/***********************************************************************
For a record lock, gets the page number on which the lock is. */
5645
UNIV_INTERN
5646 5647 5648 5649 5650 5651
ulint
lock_rec_get_page_no(
/*=================*/
				/* out: page number */
	const lock_t*	lock)	/* in: lock */
{
5652
	ut_a(lock_get_type_low(lock) == LOCK_REC);
5653 5654 5655

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