lock0lock.c 127 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
/******************************************************
The transaction lock system

(c) 1996 Innobase Oy

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

#include "lock0lock.h"

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

#include "usr0sess.h"
16
#include "trx0purge.h"
unknown's avatar
unknown committed
17 18
#include "dict0mem.h"
#include "trx0sys.h"
19

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

/* 2 function prototypes copied from ha_innodb.cc: */

/*****************************************************************
If you want to print a thd that is not associated with the current thread,
you must call this function before reserving the InnoDB kernel_mutex, to
protect MySQL from setting thd->query NULL. If you print a thd of the current
thread, we know that MySQL cannot modify thd->query, and it is not necessary
to call this. Call innobase_mysql_end_print_arbitrary_thd() after you release
the kernel_mutex.
NOTE that /mysql/innobase/lock/lock0lock.c must contain the prototype for this
function! */

void
innobase_mysql_prepare_print_arbitrary_thd(void);
/*============================================*/

/*****************************************************************
Relases the mutex reserved by innobase_mysql_prepare_print_arbitrary_thd().
NOTE that /mysql/innobase/lock/lock0lock.c must contain the prototype for this
function! */

void
innobase_mysql_end_print_arbitrary_thd(void);
/*========================================*/

46 47 48 49
/* Restricts the length of search we will do in the waits-for
graph of transactions */
#define LOCK_MAX_N_STEPS_IN_DEADLOCK_CHECK 1000000

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

#define LOCK_RELEASE_KERNEL_INTERVAL	1000

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

#define LOCK_PAGE_BITMAP_MARGIN		64

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

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

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

Different transaction can have conflicting locks set on the gap at the
same time. The locks on the gap are purely inhibitive: an insert cannot
unknown's avatar
unknown committed
84
be made, or a select cursor may have to wait if a different transaction
85
has a conflicting lock on the gap. An x-lock on the gap does not give
unknown's avatar
unknown committed
86
the right to insert into the gap.
87 88 89 90 91 92 93 94 95 96 97 98

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.

unknown's avatar
unknown committed
99 100 101 102 103
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.

104 105 106 107 108 109 110 111 112 113 114 115 116 117 118
-------------------------------------------------------------------------
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
unknown's avatar
unknown committed
119
and page splits new gap type locks can be created by the database manager
120 121 122 123 124 125 126
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
unknown's avatar
unknown committed
127 128 129
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
130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323
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. */

ibool	lock_print_waits	= FALSE;

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

/* A table lock */
typedef struct lock_table_struct	lock_table_t;
struct lock_table_struct{
	dict_table_t*	table;	/* database table in dictionary cache */
	UT_LIST_NODE_T(lock_t)
			locks; 	/* list of locks on the same table */
};

/* Record lock for a page */
typedef struct lock_rec_struct		lock_rec_t;
struct lock_rec_struct{
	ulint	space;		/* space id */
	ulint	page_no;	/* page number */
	ulint	n_bits;		/* number of bits in the lock bitmap */
				/* NOTE: the lock bitmap is placed immediately
				after the lock struct */
};

/* Lock struct */
struct lock_struct{
	trx_t*		trx;		/* transaction owning the lock */
	UT_LIST_NODE_T(lock_t)		
			trx_locks;	/* list of the locks of the
					transaction */
unknown's avatar
unknown committed
324 325 326
	ulint		type_mode;	/* lock type, mode, LOCK_GAP or
					LOCK_REC_NOT_GAP,
					LOCK_INSERT_INTENTION,
327 328 329 330 331 332 333 334 335
					wait flag, ORed */
	hash_node_t	hash;		/* hash chain node for a record lock */
	dict_index_t*	index;		/* index for a record lock */
	union {
		lock_table_t	tab_lock;/* table lock */
		lock_rec_t	rec_lock;/* record lock */
	} un_member;
};

unknown's avatar
unknown committed
336 337 338
/* We store info on the latest deadlock error to this buffer. InnoDB
Monitor will then fetch it and print */
ibool	lock_deadlock_found = FALSE;
339
FILE*	lock_latest_err_file;
unknown's avatar
unknown committed
340

unknown's avatar
unknown committed
341 342 343 344
/* Flags for recursive deadlock search */
#define LOCK_VICTIM_IS_START	1
#define LOCK_VICTIM_IS_OTHER	2

345 346 347 348 349 350 351 352 353 354 355 356 357 358 359
/************************************************************************
Checks if a lock request results in a deadlock. */
static
ibool
lock_deadlock_occurs(
/*=================*/
			/* out: TRUE if a deadlock was detected */
	lock_t*	lock,	/* in: lock the transaction is requesting */
	trx_t*	trx);	/* in: transaction */
/************************************************************************
Looks recursively for a deadlock. */
static
ibool
lock_deadlock_recursive(
/*====================*/
360 361
				/* out: TRUE if a deadlock was detected
				or the calculation took too long */
362 363
	trx_t*	start,		/* in: recursion starting point */
	trx_t*	trx,		/* in: a transaction waiting for a lock */
364 365 366 367
	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 TRUE */
unknown's avatar
unknown committed
368 369 370 371

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

unknown's avatar
unknown committed
372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395
/*************************************************************************
Checks that a transaction id is sensible, i.e., not in the future. */

ibool
lock_check_trx_id_sanity(
/*=====================*/
					/* out: TRUE if ok */
	dulint		trx_id,		/* in: trx id */
	rec_t*		rec,		/* in: user record */
	dict_index_t*	index,		/* in: clustered index */
	ibool		has_kernel_mutex)/* in: TRUE if the caller owns the
					kernel mutex */
{
	ibool	is_ok		= TRUE;
	
	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);
396 397 398 399 400 401 402
		fputs("  InnoDB: Error: transaction id associated"
			" with record\n",
			stderr);
		rec_print(stderr, rec);
		fputs("InnoDB: in ", stderr);
		dict_index_name_print(stderr, index);
		fprintf(stderr, "\n"
unknown's avatar
unknown committed
403 404
"InnoDB: is %lu %lu which is higher than the global trx id counter %lu %lu!\n"
"InnoDB: The table is corrupt. You have to do dump + drop + reimport.\n",
405 406 407 408
			       (ulong) ut_dulint_get_high(trx_id),
			       (ulong) ut_dulint_get_low(trx_id),
			       (ulong) ut_dulint_get_high(trx_sys->max_trx_id),
			       (ulong) ut_dulint_get_low(trx_sys->max_trx_id));
unknown's avatar
unknown committed
409 410 411 412 413 414 415 416 417 418 419

		is_ok = FALSE;
	}
	
	if (!has_kernel_mutex) {
		mutex_exit(&kernel_mutex);
	}

	return(is_ok);
}

420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437
/*************************************************************************
Checks that a record is seen in a consistent read. */

ibool
lock_clust_rec_cons_read_sees(
/*==========================*/
				/* out: TRUE if sees, or FALSE if an earlier
				version of the record should be retrieved */
	rec_t*		rec,	/* in: user record which should be read or
				passed over by a read cursor */
	dict_index_t*	index,	/* in: clustered index */
	read_view_t*	view)	/* in: consistent read view */
{
	dulint	trx_id;

	ut_ad(index->type & DICT_CLUSTERED);
	ut_ad(page_rec_is_user_rec(rec));

unknown's avatar
unknown committed
438 439 440 441
	/* 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! */

442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467
	trx_id = row_get_rec_trx_id(rec, index);
	
	if (read_view_sees_trx_id(view, trx_id)) {

		return(TRUE);
	}

	return(FALSE);
}

/*************************************************************************
Checks that a non-clustered index record is seen in a consistent read. */

ulint
lock_sec_rec_cons_read_sees(
/*========================*/
				/* 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 */
	rec_t*		rec,	/* in: user record which should be read or
				passed over by a read cursor */
unknown's avatar
unknown committed
468
	dict_index_t*	index,	/* in: non-clustered index */
469 470 471
	read_view_t*	view)	/* in: consistent read view */
{
	dulint	max_trx_id;
unknown's avatar
unknown committed
472 473 474
	
	UT_NOT_USED(index);
	
475 476 477
	ut_ad(!(index->type & DICT_CLUSTERED));
	ut_ad(page_rec_is_user_rec(rec));

unknown's avatar
unknown committed
478 479 480 481
	/* 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! */

482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509
	if (recv_recovery_is_on()) {

		return(FALSE);
	}

	max_trx_id = page_get_max_trx_id(buf_frame_align(rec));

	if (ut_dulint_cmp(max_trx_id, view->up_limit_id) >= 0) {

		return(FALSE);
	}

	return(TRUE);
}

/*************************************************************************
Creates the lock system at database start. */

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); */
unknown's avatar
unknown committed
510

511
	lock_latest_err_file = os_file_create_tmpfile();
512
	ut_a(lock_latest_err_file);
513 514
}

515 516 517 518 519 520 521 522 523 524 525
/*************************************************************************
Gets the size of a lock struct. */

ulint
lock_get_size(void)
/*===============*/
			/* out: size in bytes */
{
	return((ulint)sizeof(lock_t));
}

526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545
/*************************************************************************
Gets the mode of a lock. */
UNIV_INLINE
ulint
lock_get_mode(
/*==========*/
			/* out: mode */
	lock_t*	lock)	/* in: lock */
{
	ut_ad(lock);

	return(lock->type_mode & LOCK_MODE_MASK);
}

/*************************************************************************
Gets the type of a lock. */
UNIV_INLINE
ulint
lock_get_type(
/*==========*/
unknown's avatar
unknown committed
546
			/* out: LOCK_TABLE or LOCK_REC */
547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627
	lock_t*	lock)	/* in: lock */
{
	ut_ad(lock);

	return(lock->type_mode & LOCK_TYPE_MASK);
}

/*************************************************************************
Gets the wait flag of a lock. */
UNIV_INLINE
ibool
lock_get_wait(
/*==========*/
			/* out: TRUE if waiting */
	lock_t*	lock)	/* in: lock */
{
	ut_ad(lock);

	if (lock->type_mode & LOCK_WAIT) {

		return(TRUE);
	}

	return(FALSE);
}

/*************************************************************************
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);
	
	trx->wait_lock = lock;
 	lock->type_mode = lock->type_mode | LOCK_WAIT;
}

/**************************************************************************
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;
 	lock->type_mode = lock->type_mode & ~LOCK_WAIT;
}

/*************************************************************************
Gets the gap flag of a record lock. */
UNIV_INLINE
ibool
lock_rec_get_gap(
/*=============*/
			/* out: TRUE if gap flag set */
	lock_t*	lock)	/* in: record lock */
{
	ut_ad(lock);
	ut_ad(lock_get_type(lock) == LOCK_REC);

	if (lock->type_mode & LOCK_GAP) {

		return(TRUE);
	}

	return(FALSE);
}

/*************************************************************************
unknown's avatar
unknown committed
628
Gets the LOCK_REC_NOT_GAP flag of a record lock. */
629
UNIV_INLINE
unknown's avatar
unknown committed
630 631 632 633 634
ibool
lock_rec_get_rec_not_gap(
/*=====================*/
			/* out: TRUE if LOCK_REC_NOT_GAP flag set */
	lock_t*	lock)	/* in: record lock */
635 636 637 638
{
	ut_ad(lock);
	ut_ad(lock_get_type(lock) == LOCK_REC);

unknown's avatar
unknown committed
639 640 641
	if (lock->type_mode & LOCK_REC_NOT_GAP) {

		return(TRUE);
642
	}
unknown's avatar
unknown committed
643 644

	return(FALSE);
645 646
}

unknown's avatar
unknown committed
647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666
/*************************************************************************
Gets the waiting insert flag of a record lock. */
UNIV_INLINE
ibool
lock_rec_get_insert_intention(
/*==========================*/
			/* out: TRUE if gap flag set */
	lock_t*	lock)	/* in: record lock */
{
	ut_ad(lock);
	ut_ad(lock_get_type(lock) == LOCK_REC);

	if (lock->type_mode & LOCK_INSERT_INTENTION) {

		return(TRUE);
	}

	return(FALSE);
}

667 668 669 670 671 672 673 674 675 676
/*************************************************************************
Calculates if lock mode 1 is stronger or equal to lock mode 2. */
UNIV_INLINE
ibool
lock_mode_stronger_or_eq(
/*=====================*/
			/* out: TRUE if mode1 stronger or equal to mode2 */
	ulint	mode1,	/* in: lock mode */
	ulint	mode2)	/* in: lock mode */
{
unknown's avatar
unknown committed
677
	ut_ad(mode1 == LOCK_X || mode1 == LOCK_S || mode1 == LOCK_IX
678
				|| mode1 == LOCK_IS || mode1 == LOCK_AUTO_INC);
unknown's avatar
unknown committed
679
	ut_ad(mode2 == LOCK_X || mode2 == LOCK_S || mode2 == LOCK_IX
680
				|| mode2 == LOCK_IS || mode2 == LOCK_AUTO_INC);
681 682 683 684
	if (mode1 == LOCK_X) {

		return(TRUE);

685 686 687 688
	} else if (mode1 == LOCK_AUTO_INC && mode2 == LOCK_AUTO_INC) {

		return(TRUE);

unknown's avatar
unknown committed
689 690
	} else if (mode1 == LOCK_S
				&& (mode2 == LOCK_S || mode2 == LOCK_IS)) {
691 692
		return(TRUE);

unknown's avatar
unknown committed
693
	} else if (mode1 == LOCK_IS && mode2 == LOCK_IS) {
694 695 696

		return(TRUE);

unknown's avatar
unknown committed
697 698
	} else if (mode1 == LOCK_IX && (mode2 == LOCK_IX
						|| mode2 == LOCK_IS)) {
699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714
		return(TRUE);
	}

	return(FALSE);
}

/*************************************************************************
Calculates if lock mode 1 is compatible with lock mode 2. */
UNIV_INLINE
ibool
lock_mode_compatible(
/*=================*/
			/* out: TRUE if mode1 compatible with mode2 */
	ulint	mode1,	/* in: lock mode */
	ulint	mode2)	/* in: lock mode */
{
unknown's avatar
unknown committed
715
	ut_ad(mode1 == LOCK_X || mode1 == LOCK_S || mode1 == LOCK_IX
716
				|| mode1 == LOCK_IS || mode1 == LOCK_AUTO_INC);
unknown's avatar
unknown committed
717
	ut_ad(mode2 == LOCK_X || mode2 == LOCK_S || mode2 == LOCK_IX
718
				|| mode2 == LOCK_IS || mode2 == LOCK_AUTO_INC);
719

unknown's avatar
unknown committed
720
	if (mode1 == LOCK_S && (mode2 == LOCK_IS || mode2 == LOCK_S)) {
721 722 723 724 725 726 727

		return(TRUE);

	} else if (mode1 == LOCK_X) {

		return(FALSE);

728 729 730 731
	} else if (mode1 == LOCK_AUTO_INC && (mode2 == LOCK_IS
					  	|| mode2 == LOCK_IX)) {
		return(TRUE);

unknown's avatar
unknown committed
732 733
	} else if (mode1 == LOCK_IS && (mode2 == LOCK_IS
					  	|| mode2 == LOCK_IX
734
					  	|| mode2 == LOCK_AUTO_INC
unknown's avatar
unknown committed
735
					  	|| mode2 == LOCK_S)) {
736 737
		return(TRUE);

unknown's avatar
unknown committed
738
	} else if (mode1 == LOCK_IX && (mode2 == LOCK_IS
739
					  	|| mode2 == LOCK_AUTO_INC
unknown's avatar
unknown committed
740
					  	|| mode2 == LOCK_IX)) {
741 742 743 744 745 746 747
		return(TRUE);
	}

	return(FALSE);
}

/*************************************************************************
unknown's avatar
unknown committed
748
Checks if a lock request for a new lock has to wait for request lock2. */
749
UNIV_INLINE
unknown's avatar
unknown committed
750 751 752 753 754 755
ibool
lock_rec_has_to_wait(
/*=================*/
			/* out: TRUE if new lock has to wait for lock2 to be
			removed */
	trx_t*	trx,	/* in: trx of new lock */
unknown's avatar
unknown committed
756 757 758
	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 */
759
	lock_t*	lock2,	/* in: another record lock; NOTE that it is assumed
unknown's avatar
unknown committed
760
			that this has a lock bit set on the same record as
unknown's avatar
unknown committed
761
			in the new lock we are setting */
762 763 764 765
	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 */
766
{
unknown's avatar
unknown committed
767 768 769
	ut_ad(trx && lock2);
	ut_ad(lock_get_type(lock2) == LOCK_REC);

unknown's avatar
unknown committed
770 771
	if (trx != lock2->trx
	    && !lock_mode_compatible(LOCK_MODE_MASK & type_mode,
unknown's avatar
unknown committed
772 773
				     		lock_get_mode(lock2))) {

unknown's avatar
unknown committed
774 775
		/* We have somewhat complex rules when gap type record locks
		cause waits */
unknown's avatar
unknown committed
776

777
		if ((lock_is_on_supremum || (type_mode & LOCK_GAP))
778
			&& !(type_mode & LOCK_INSERT_INTENTION)) {
779

780
			/* Gap type locks without LOCK_INSERT_INTENTION flag
781 782 783
			do not need to wait for anything. This is because 
			different users can have conflicting lock types 
			on gaps. */
784 785 786 787
						  
			return(FALSE);
		}
		
788
		if (!(type_mode & LOCK_INSERT_INTENTION)
unknown's avatar
unknown committed
789
						&& lock_rec_get_gap(lock2)) {
790 791 792

			/* Record lock (LOCK_ORDINARY or LOCK_REC_NOT_GAP
			does not need to wait for a gap type lock */
unknown's avatar
unknown committed
793 794 795 796 797 798 799 800 801

			return(FALSE);
		}

		if ((type_mode & LOCK_GAP)
					&& lock_rec_get_rec_not_gap(lock2)) {
		
			/* Lock on gap does not need to wait for
			a LOCK_REC_NOT_GAP type lock */
802

unknown's avatar
unknown committed
803 804 805
			return(FALSE);
		}

unknown's avatar
unknown committed
806
		if (lock_rec_get_insert_intention(lock2)) {
unknown's avatar
unknown committed
807

unknown's avatar
unknown committed
808 809 810 811 812 813 814
			/* 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.
unknown's avatar
unknown committed
815

unknown's avatar
unknown committed
816 817 818
			Also, insert intention locks do not disturb each
			other. */
				
unknown's avatar
unknown committed
819 820
			return(FALSE);
		}
821

unknown's avatar
unknown committed
822
		return(TRUE);
823 824
	}

unknown's avatar
unknown committed
825
	return(FALSE);
826 827 828
}

/*************************************************************************
unknown's avatar
unknown committed
829 830
Checks if a lock request lock1 has to wait for request lock2. */
static
831 832 833
ibool
lock_has_to_wait(
/*=============*/
unknown's avatar
unknown committed
834 835 836
			/* out: TRUE if lock1 has to wait for lock2 to be
			removed */
	lock_t*	lock1,	/* in: waiting lock */
837
	lock_t*	lock2)	/* in: another lock; NOTE that it is assumed that this
unknown's avatar
unknown committed
838 839
			has a lock bit set on the same record as in lock1 if
			the locks are record locks */
840
{
unknown's avatar
unknown committed
841 842
	ut_ad(lock1 && lock2);

unknown's avatar
unknown committed
843
	if (lock1->trx != lock2->trx
844 845
			&& !lock_mode_compatible(lock_get_mode(lock1),
				     		lock_get_mode(lock2))) {
unknown's avatar
unknown committed
846 847
		if (lock_get_type(lock1) == LOCK_REC) {
			ut_ad(lock_get_type(lock2) == LOCK_REC);
848

849 850 851
			/* If this lock request is for a supremum record
			then the second bit on the lock bitmap is set */
			
unknown's avatar
unknown committed
852
			return(lock_rec_has_to_wait(lock1->trx,
853 854
					lock1->type_mode, lock2,
					lock_rec_get_nth_bit(lock1,1)));
unknown's avatar
unknown committed
855 856
		}

857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003
		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(
/*================*/
			/* out: number of bits */
	lock_t*	lock)	/* in: record lock */
{
	return(lock->un_member.rec_lock.n_bits);
}

/*************************************************************************
Gets the nth bit of a record lock. */
UNIV_INLINE
ibool
lock_rec_get_nth_bit(
/*=================*/
			/* out: TRUE if bit set */
	lock_t*	lock,	/* in: record lock */
	ulint	i)	/* in: index of the bit */
{
	ulint	byte_index;
	ulint	bit_index;
	ulint	b;

	ut_ad(lock);
	ut_ad(lock_get_type(lock) == LOCK_REC);

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

		return(FALSE);
	}

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

	b = (ulint)*((byte*)lock + sizeof(lock_t) + byte_index);

	return(ut_bit_get_nth(b, bit_index));
}	

/**************************************************************************
Sets the nth bit of a record lock to TRUE. */
UNIV_INLINE
void
lock_rec_set_nth_bit(
/*==================*/
	lock_t*	lock,	/* in: record lock */
	ulint	i)	/* in: index of the bit */
{
	ulint	byte_index;
	ulint	bit_index;
	byte*	ptr;
	ulint	b;
	
	ut_ad(lock);
	ut_ad(lock_get_type(lock) == LOCK_REC);
	ut_ad(i < lock->un_member.rec_lock.n_bits);
	
	byte_index = i / 8;
	bit_index = i % 8;

	ptr = (byte*)lock + sizeof(lock_t) + byte_index;
		
	b = (ulint)*ptr;

	b = ut_bit_set_nth(b, bit_index, TRUE);

	*ptr = (byte)b;
}	

/**************************************************************************
Looks for a set bit in a record lock bitmap. Returns ULINT_UNDEFINED,
if none found. */
static
ulint
lock_rec_find_set_bit(
/*==================*/
			/* out: bit index == heap number of the record, or
			ULINT_UNDEFINED if none found */
	lock_t*	lock)	/* in: record lock with at least one bit set */
{
	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;
	byte*	ptr;
	ulint	b;
	
	ut_ad(lock);
	ut_ad(lock_get_type(lock) == LOCK_REC);
	ut_ad(i < lock->un_member.rec_lock.n_bits);
	
	byte_index = i / 8;
	bit_index = i % 8;

	ptr = (byte*)lock + sizeof(lock_t) + byte_index;
		
	b = (ulint)*ptr;

	b = ut_bit_set_nth(b, bit_index, FALSE);

	*ptr = (byte)b;
}	

/*************************************************************************
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;

1004
#ifdef UNIV_SYNC_DEBUG
1005
	ut_ad(mutex_own(&kernel_mutex));
1006
#endif /* UNIV_SYNC_DEBUG */
unknown's avatar
unknown committed
1007
	ut_ad(lock_get_type(lock) == LOCK_REC);
1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042

	space = lock->un_member.rec_lock.space;
	page_no = lock->un_member.rec_lock.page_no;
	
	for (;;) {
		lock = HASH_GET_NEXT(hash, lock);

		if (!lock) {

			break;
		}

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

			break;
		}
	}
	
	return(lock);
}

/*************************************************************************
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;

1043
#ifdef UNIV_SYNC_DEBUG
1044
	ut_ad(mutex_own(&kernel_mutex));
1045
#endif /* UNIV_SYNC_DEBUG */
1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102

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

			break;
		}

		lock = HASH_GET_NEXT(hash, lock);
	}

	return(lock);
}
	
/*************************************************************************
Returns TRUE if there are explicit record locks on a page. */

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);
	
	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(
/*=======================*/
			/* out: first lock, NULL if none exists */
	byte*	ptr)	/* in: pointer to somewhere on the page */
{
	ulint	hash;
	lock_t*	lock;
	ulint	space;
	ulint	page_no;

1103
#ifdef UNIV_SYNC_DEBUG
1104
	ut_ad(mutex_own(&kernel_mutex));
1105
#endif /* UNIV_SYNC_DEBUG */
1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136
	
	hash = buf_frame_get_lock_hash_val(ptr);

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

	while (lock) {
		space = buf_frame_get_space_id(ptr);
		page_no = buf_frame_get_page_no(ptr);

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

			break;
		}

		lock = HASH_GET_NEXT(hash, lock);
	}

	return(lock);
}

/*************************************************************************
Gets the next explicit lock request on a record. */
UNIV_INLINE
lock_t*
lock_rec_get_next(
/*==============*/
			/* out: next lock, NULL if none exists */
	rec_t*	rec,	/* in: record on a page */
	lock_t*	lock)	/* in: lock */
{
1137
#ifdef UNIV_SYNC_DEBUG
1138
	ut_ad(mutex_own(&kernel_mutex));
1139
#endif /* UNIV_SYNC_DEBUG */
unknown's avatar
unknown committed
1140
	ut_ad(lock_get_type(lock) == LOCK_REC);
1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167

	for (;;) {
		lock = lock_rec_get_next_on_page(lock);

		if (lock == NULL) {

			return(NULL);
		}

		if (lock_rec_get_nth_bit(lock, rec_get_heap_no(rec))) {

			return(lock);
		}
	}
}

/*************************************************************************
Gets the first explicit lock request on a record. */
UNIV_INLINE
lock_t*
lock_rec_get_first(
/*===============*/
			/* out: first lock, NULL if none exists */
	rec_t*	rec)	/* in: record on a page */
{
	lock_t*	lock;

1168
#ifdef UNIV_SYNC_DEBUG
1169
	ut_ad(mutex_own(&kernel_mutex));
1170
#endif /* UNIV_SYNC_DEBUG */
1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199

	lock = lock_rec_get_first_on_page(rec);

	while (lock) {
		if (lock_rec_get_nth_bit(lock, rec_get_heap_no(rec))) {

			break;
		}

		lock = lock_rec_get_next_on_page(lock);
	}

	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 */
{
	byte*	ptr;
	ulint	n_bytes;
	ulint	i;

unknown's avatar
unknown committed
1200 1201
	ut_ad(lock_get_type(lock) == LOCK_REC);

1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230
	/* Reset to zero the bitmap which resides immediately after the lock
	struct */

	ptr = (byte*)lock + sizeof(lock_t);

	n_bytes = lock_rec_get_n_bits(lock) / 8;

	ut_ad((lock_rec_get_n_bits(lock) % 8) == 0);
	
	for (i = 0; i < n_bytes; i++) {

		*ptr = 0;
		ptr++;
	}
}

/*************************************************************************
Copies a record lock to heap. */
static
lock_t*
lock_rec_copy(
/*==========*/
				/* out: copy of lock */
	lock_t*		lock,	/* in: record lock */
	mem_heap_t*	heap)	/* in: memory heap */
{
	lock_t*	dupl_lock;
	ulint	size;

unknown's avatar
unknown committed
1231 1232
	ut_ad(lock_get_type(lock) == LOCK_REC);

1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257
	size = sizeof(lock_t) + lock_rec_get_n_bits(lock) / 8;	

	dupl_lock = mem_heap_alloc(heap, size);

	ut_memcpy(dupl_lock, lock, size);

	return(dupl_lock);
}

/*************************************************************************
Gets the previous record lock set on a record. */
static
lock_t*
lock_rec_get_prev(
/*==============*/
			/* out: previous lock on the same record, NULL if
			none exists */
	lock_t*	in_lock,/* in: record lock */
	ulint	heap_no)/* in: heap number of the record */
{
	lock_t*	lock;
	ulint	space;
	ulint	page_no;
	lock_t*	found_lock 	= NULL;

1258
#ifdef UNIV_SYNC_DEBUG
1259
	ut_ad(mutex_own(&kernel_mutex));
1260
#endif /* UNIV_SYNC_DEBUG */
1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299
	ut_ad(lock_get_type(in_lock) == LOCK_REC);

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

	lock = lock_rec_get_first_on_page_addr(space, page_no);

	for (;;) {
		ut_ad(lock);
		
		if (lock == in_lock) {

			return(found_lock);
		}

		if (lock_rec_get_nth_bit(lock, heap_no)) {

			found_lock = lock;
		}

		lock = lock_rec_get_next_on_page(lock);
	}	
}

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

/*************************************************************************
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 */
	ulint		mode)	/* in: lock mode */
{
	lock_t*	lock;

1300
#ifdef UNIV_SYNC_DEBUG
1301
	ut_ad(mutex_own(&kernel_mutex));
1302
#endif /* UNIV_SYNC_DEBUG */
1303 1304 1305 1306 1307 1308 1309

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

	lock = UT_LIST_GET_LAST(table->locks);

	while (lock != NULL) {

unknown's avatar
unknown committed
1310 1311
		if (lock->trx == trx
		    && lock_mode_stronger_or_eq(lock_get_mode(lock), mode)) {
1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329

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

			ut_ad(!lock_get_wait(lock)); 

			return(lock);
		}

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

	return(NULL);
}
	
/*============= FUNCTIONS FOR ANALYZING RECORD LOCK QUEUE ================*/

/*************************************************************************
unknown's avatar
unknown committed
1330 1331
Checks if a transaction has a GRANTED explicit lock on rec stronger or equal
to precise_mode. */
1332 1333 1334 1335 1336
UNIV_INLINE
lock_t*
lock_rec_has_expl(
/*==============*/
			/* out: lock or NULL */
unknown's avatar
unknown committed
1337 1338 1339 1340
	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 */
1341 1342 1343 1344 1345
	rec_t*	rec,	/* in: record */
	trx_t*	trx)	/* in: transaction */
{
	lock_t*	lock;

1346
#ifdef UNIV_SYNC_DEBUG
unknown's avatar
unknown committed
1347
	ut_ad(mutex_own(&kernel_mutex));
1348
#endif /* UNIV_SYNC_DEBUG */
unknown's avatar
unknown committed
1349 1350 1351 1352
	ut_ad((precise_mode & LOCK_MODE_MASK) == LOCK_S
	      || (precise_mode & LOCK_MODE_MASK) == LOCK_X);
	ut_ad(!(precise_mode & LOCK_INSERT_INTENTION));
	
1353 1354 1355
	lock = lock_rec_get_first(rec);

	while (lock) {
unknown's avatar
unknown committed
1356
		if (lock->trx == trx
unknown's avatar
unknown committed
1357 1358
		    && lock_mode_stronger_or_eq(lock_get_mode(lock),
		    				precise_mode & LOCK_MODE_MASK)
1359
		    && !lock_get_wait(lock)
unknown's avatar
unknown committed
1360 1361 1362 1363 1364 1365 1366
		    && (!lock_rec_get_rec_not_gap(lock)
		    		|| (precise_mode & LOCK_REC_NOT_GAP)
		    		|| page_rec_is_supremum(rec))
		    && (!lock_rec_get_gap(lock)
				|| (precise_mode & LOCK_GAP)
				|| page_rec_is_supremum(rec))
		    && (!lock_rec_get_insert_intention(lock))) {
unknown's avatar
unknown committed
1367

1368 1369 1370 1371 1372 1373 1374 1375
		    	return(lock);
		}

		lock = lock_rec_get_next(rec, lock);
	}

	return(NULL);
}
1376
			
1377
/*************************************************************************
unknown's avatar
unknown committed
1378 1379
Checks if some other transaction has a lock request in the queue. */
static
1380 1381 1382 1383
lock_t*
lock_rec_other_has_expl_req(
/*========================*/
			/* out: lock or NULL */
unknown's avatar
unknown committed
1384
	ulint	mode,	/* in: LOCK_S or LOCK_X */
1385 1386 1387 1388 1389
	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 */
	rec_t*	rec,	/* in: record to look at */	
unknown's avatar
unknown committed
1390 1391
	trx_t*	trx)	/* in: transaction, or NULL if requests by all
			transactions are taken into account */
1392 1393 1394
{
	lock_t*	lock;
	
1395
#ifdef UNIV_SYNC_DEBUG
1396
	ut_ad(mutex_own(&kernel_mutex));
1397
#endif /* UNIV_SYNC_DEBUG */
unknown's avatar
unknown committed
1398 1399 1400
	ut_ad(mode == LOCK_X || mode == LOCK_S);
	ut_ad(gap == 0 || gap == LOCK_GAP);
	ut_ad(wait == 0 || wait == LOCK_WAIT);
unknown's avatar
unknown committed
1401

1402 1403 1404
	lock = lock_rec_get_first(rec);

	while (lock) {
unknown's avatar
unknown committed
1405
		if (lock->trx != trx
1406 1407 1408 1409 1410
		    && (gap ||
			!(lock_rec_get_gap(lock) || page_rec_is_supremum(rec)))
		    && (wait || !lock_get_wait(lock))
		    && lock_mode_stronger_or_eq(lock_get_mode(lock), mode)) {

unknown's avatar
unknown committed
1411
		    	return(lock);
1412 1413 1414 1415 1416 1417 1418 1419
		}

		lock = lock_rec_get_next(rec, lock);
	}

	return(NULL);
}

unknown's avatar
unknown committed
1420 1421 1422 1423 1424 1425 1426 1427
/*************************************************************************
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(
/*===========================*/
			/* out: lock or NULL */
unknown's avatar
unknown committed
1428 1429 1430
	ulint	mode,	/* in: LOCK_S or LOCK_X,
			possibly ORed to LOCK_GAP or LOC_REC_NOT_GAP,
			LOCK_INSERT_INTENTION */
unknown's avatar
unknown committed
1431 1432 1433 1434 1435
	rec_t*	rec,	/* in: record to look at */	
	trx_t*	trx)	/* in: our transaction */
{
	lock_t*	lock;
	
1436
#ifdef UNIV_SYNC_DEBUG
unknown's avatar
unknown committed
1437
	ut_ad(mutex_own(&kernel_mutex));
1438
#endif /* UNIV_SYNC_DEBUG */
unknown's avatar
unknown committed
1439

unknown's avatar
unknown committed
1440 1441 1442
	lock = lock_rec_get_first(rec);

	while (lock) {
1443 1444
		if (lock_rec_has_to_wait(trx, mode, lock,
			page_rec_is_supremum(rec))) {
unknown's avatar
unknown committed
1445

unknown's avatar
unknown committed
1446 1447 1448 1449 1450 1451 1452 1453 1454
			return(lock);
		}
		
		lock = lock_rec_get_next(rec, lock);
	}

	return(NULL);
}

1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470
/*************************************************************************
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(
/*==========================*/
				/* out: lock or NULL */
	ulint	type_mode,	/* in: lock type_mode field */
	rec_t*	rec,		/* in: record */
	trx_t*	trx)		/* in: transaction */
{
	lock_t*	lock;
	ulint	heap_no;

1471
#ifdef UNIV_SYNC_DEBUG
1472
	ut_ad(mutex_own(&kernel_mutex));
1473
#endif /* UNIV_SYNC_DEBUG */
1474 1475 1476 1477 1478 1479

	heap_no = rec_get_heap_no(rec);
	
	lock = lock_rec_get_first_on_page(rec);

	while (lock != NULL) {
unknown's avatar
unknown committed
1480 1481 1482
		if (lock->trx == trx
		    && lock->type_mode == type_mode
		    && lock_rec_get_n_bits(lock) > heap_no) {
1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506
		    	
			return(lock);
		}
		
		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. */

trx_t*
lock_sec_rec_some_has_impl_off_kernel(
/*==================================*/
				/* out: transaction which has the x-lock, or
				NULL */
	rec_t*		rec,	/* in: user record */
	dict_index_t*	index)	/* in: secondary index */
{
	page_t*	page;
	
1507
#ifdef UNIV_SYNC_DEBUG
1508
	ut_ad(mutex_own(&kernel_mutex));
1509
#endif /* UNIV_SYNC_DEBUG */
1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530
	ut_ad(!(index->type & DICT_CLUSTERED));
	ut_ad(page_rec_is_user_rec(rec));

	page = buf_frame_align(rec);

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

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

		return(NULL);
	}

	/* Ok, in this case it is possible that some transaction has an
	implicit x-lock. We have to look in the clustered index. */
			
unknown's avatar
unknown committed
1531 1532 1533 1534 1535 1536 1537 1538 1539
	if (!lock_check_trx_id_sanity(page_get_max_trx_id(page), rec, index,
								     TRUE)) {
		buf_page_print(page);
		
		/* The page is corrupt: try to avoid a crash by returning
		NULL */
		return(NULL);
	}

1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566
	return(row_vers_impl_x_locked_off_kernel(rec, index));
}

/*============== 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(
/*============*/
				/* out: created lock, NULL if out of memory */
	ulint		type_mode,/* in: lock mode and wait flag, type is
				ignored and replaced by LOCK_REC */
	rec_t*		rec,	/* in: record on page */
	dict_index_t*	index,	/* in: index of record */
	trx_t*		trx)	/* in: transaction */
{
	page_t*	page;
	lock_t*	lock;
	ulint	page_no;
	ulint	heap_no;
	ulint	space;
	ulint	n_bits;
	ulint	n_bytes;
	
1567
#ifdef UNIV_SYNC_DEBUG
1568
	ut_ad(mutex_own(&kernel_mutex));
1569
#endif /* UNIV_SYNC_DEBUG */
1570 1571 1572 1573 1574 1575

	page = buf_frame_align(rec);
	space = buf_frame_get_space_id(page);
	page_no	= buf_frame_get_page_no(page);
	heap_no = rec_get_heap_no(rec);

unknown's avatar
unknown committed
1576 1577 1578
	/* 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 */
1579 1580

	if (rec == page_get_supremum_rec(page)) {
unknown's avatar
unknown committed
1581
		ut_ad(!(type_mode & LOCK_REC_NOT_GAP));
1582

unknown's avatar
unknown committed
1583
		type_mode = type_mode & ~(LOCK_GAP | LOCK_REC_NOT_GAP);
1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634
	}

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

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

		return(NULL);
	}

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

	lock->trx = trx;

	lock->type_mode = (type_mode & ~LOCK_TYPE_MASK) | LOCK_REC;
	lock->index = index;
	
	lock->un_member.rec_lock.space = space;
	lock->un_member.rec_lock.page_no = page_no;
	lock->un_member.rec_lock.n_bits = n_bytes * 8;

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

	lock_rec_bitmap_reset(lock);

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

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

		lock_set_lock_and_trx_wait(lock, trx);
	}
	
	return(lock);
}

/*************************************************************************
Enqueues a waiting request for a lock which cannot be granted immediately.
Checks for deadlocks. */
static
ulint
lock_rec_enqueue_waiting(
/*=====================*/
				/* out: DB_LOCK_WAIT, DB_DEADLOCK, or
unknown's avatar
unknown committed
1635 1636 1637 1638 1639
				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 */
1640
	ulint		type_mode,/* in: lock mode this transaction is
unknown's avatar
unknown committed
1641 1642
				requesting: LOCK_S or LOCK_X, possibly ORed
				with LOCK_GAP or LOCK_REC_NOT_GAP, ORed
unknown's avatar
unknown committed
1643
				with LOCK_INSERT_INTENTION if this waiting
unknown's avatar
unknown committed
1644 1645
				lock request is set when performing an
				insert of an index record */
1646 1647 1648 1649 1650 1651 1652
	rec_t*		rec,	/* in: record */
	dict_index_t*	index,	/* in: index of record */
	que_thr_t*	thr)	/* in: query thread */
{
	lock_t*	lock;
	trx_t*	trx;
	
1653
#ifdef UNIV_SYNC_DEBUG
1654
	ut_ad(mutex_own(&kernel_mutex));
1655
#endif /* UNIV_SYNC_DEBUG */
1656 1657 1658 1659 1660 1661 1662

	/* 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)) {

1663
		ut_error;
unknown's avatar
unknown committed
1664

1665 1666 1667 1668 1669
		return(DB_QUE_THR_SUSPENDED);
	}
		
	trx = thr_get_trx(thr);

unknown's avatar
unknown committed
1670 1671
	if (trx->dict_operation) {
		ut_print_timestamp(stderr);
1672
		fputs(
unknown's avatar
unknown committed
1673
"  InnoDB: Error: a record lock wait happens in a dictionary operation!\n"
1674 1675
"InnoDB: Table name ", stderr);
		ut_print_name(stderr, index->table_name);
1676 1677
		fputs(".\n"
"InnoDB: Submit a detailed bug report to http://bugs.mysql.com\n",
1678
			stderr);
unknown's avatar
unknown committed
1679 1680
	}
	
1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694
	/* Enqueue the lock request that will wait to be granted */
	lock = lock_rec_create(type_mode | LOCK_WAIT, rec, index, 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_rec_reset_nth_bit(lock, rec_get_heap_no(rec));

		return(DB_DEADLOCK);
	}

unknown's avatar
unknown committed
1695 1696 1697 1698 1699 1700 1701 1702
	/* 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);
	}

1703
	trx->que_state = TRX_QUE_LOCK_WAIT;
unknown's avatar
unknown committed
1704
	trx->was_chosen_as_deadlock_victim = FALSE;
unknown's avatar
unknown committed
1705
	trx->wait_started = time(NULL);
1706 1707 1708 1709

	ut_a(que_thr_stop(thr));

	if (lock_print_waits) {
1710
		fprintf(stderr, "Lock wait for trx %lu in index ",
unknown's avatar
unknown committed
1711
			(ulong) ut_dulint_get_low(trx->id));
1712
		ut_print_name(stderr, index->name);
1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730
	}
	
	return(DB_LOCK_WAIT);	
}

/*************************************************************************
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(
/*==================*/
				/* out: lock where the bit was set, NULL if out
				of memory */
unknown's avatar
unknown committed
1731 1732
	ulint		type_mode,/* in: lock mode, wait, gap etc. flags;
				type is ignored and replaced by LOCK_REC */
1733 1734 1735 1736 1737 1738 1739 1740 1741 1742
	rec_t*		rec,	/* in: record on page */
	dict_index_t*	index,	/* in: index of record */
	trx_t*		trx)	/* in: transaction */
{
	lock_t*	lock;
	lock_t*	similar_lock	= NULL;
	ulint	heap_no;
	page_t*	page;
	ibool	somebody_waits	= FALSE;
	
1743
#ifdef UNIV_SYNC_DEBUG
1744
	ut_ad(mutex_own(&kernel_mutex));
1745
#endif /* UNIV_SYNC_DEBUG */
1746 1747
	ut_ad((type_mode & (LOCK_WAIT | LOCK_GAP))
	      || ((type_mode & LOCK_MODE_MASK) != LOCK_S)
unknown's avatar
unknown committed
1748
	      || !lock_rec_other_has_expl_req(LOCK_X, 0, LOCK_WAIT, rec, trx));
1749 1750
	ut_ad((type_mode & (LOCK_WAIT | LOCK_GAP))
	      || ((type_mode & LOCK_MODE_MASK) != LOCK_X)
unknown's avatar
unknown committed
1751 1752
	      || !lock_rec_other_has_expl_req(LOCK_S, 0, LOCK_WAIT, rec, trx));

1753 1754 1755 1756 1757 1758 1759 1760 1761 1762
	type_mode = type_mode | LOCK_REC;

	page = buf_frame_align(rec);

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

	if (rec == page_get_supremum_rec(page)) {
unknown's avatar
unknown committed
1763
		ut_ad(!(type_mode & LOCK_REC_NOT_GAP));
1764

unknown's avatar
unknown committed
1765 1766 1767 1768
		/* There should never be LOCK_REC_NOT_GAP on a supremum
		record, but let us play safe */
		
		type_mode = type_mode & ~(LOCK_GAP | LOCK_REC_NOT_GAP);
1769 1770
	}

unknown's avatar
unknown committed
1771
	/* Look for a waiting lock request on the same record or on a gap */
1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785

	heap_no = rec_get_heap_no(rec);
	lock = lock_rec_get_first_on_page(rec);

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

			somebody_waits = TRUE;
		}

		lock = lock_rec_get_next_on_page(lock);
	}

unknown's avatar
unknown committed
1786 1787 1788
	/* 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 */

1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805
	similar_lock = lock_rec_find_similar_on_page(type_mode, rec, trx);

	if (similar_lock && !somebody_waits && !(type_mode & LOCK_WAIT)) {

		lock_rec_set_nth_bit(similar_lock, heap_no);

		return(similar_lock);
	}

	return(lock_rec_create(type_mode, rec, index, trx));
}

/*************************************************************************
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
unknown's avatar
unknown committed
1806 1807
explicit locks. This function sets a normal next-key lock, or in the case of
a page supremum record, a gap type lock. */
1808 1809 1810 1811 1812 1813 1814 1815
UNIV_INLINE
ibool
lock_rec_lock_fast(
/*===============*/
				/* 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 */
unknown's avatar
unknown committed
1816 1817
	ulint		mode,	/* in: lock mode: LOCK_X or LOCK_S possibly
				ORed to either LOCK_GAP or LOCK_REC_NOT_GAP */
1818 1819 1820 1821 1822 1823 1824
	rec_t*		rec,	/* in: record */
	dict_index_t*	index,	/* in: index of record */
	que_thr_t* 	thr)	/* in: query thread */
{
	lock_t*	lock;
	ulint	heap_no;

1825
#ifdef UNIV_SYNC_DEBUG
1826
	ut_ad(mutex_own(&kernel_mutex));
1827
#endif /* UNIV_SYNC_DEBUG */
unknown's avatar
unknown committed
1828 1829 1830 1831 1832 1833 1834 1835 1836 1837
	ut_ad((LOCK_MODE_MASK & mode) != LOCK_S
		|| lock_table_has(thr_get_trx(thr), index->table, LOCK_IS));
	ut_ad((LOCK_MODE_MASK & mode) != LOCK_X
		|| lock_table_has(thr_get_trx(thr), index->table, LOCK_IX));
	ut_ad((LOCK_MODE_MASK & mode) == LOCK_S
		|| (LOCK_MODE_MASK & mode) == LOCK_X);
	ut_ad(mode - (LOCK_MODE_MASK & mode) == LOCK_GAP
			|| mode - (LOCK_MODE_MASK & mode) == 0
			|| mode - (LOCK_MODE_MASK & mode) == LOCK_REC_NOT_GAP);
			
1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854
	heap_no = rec_get_heap_no(rec);
	
	lock = lock_rec_get_first_on_page(rec);

	if (lock == NULL) {
		if (!impl) {
			lock_rec_create(mode, rec, index, thr_get_trx(thr));
		}
		
		return(TRUE);
	}
	
	if (lock_rec_get_next_on_page(lock)) {

		return(FALSE);
	}

unknown's avatar
unknown committed
1855 1856 1857
	if (lock->trx != thr_get_trx(thr)
				|| lock->type_mode != (mode | LOCK_REC)
				|| lock_rec_get_n_bits(lock) <= heap_no) {
1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870
	    	return(FALSE);
	}

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

	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
unknown's avatar
unknown committed
1871 1872
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. */
1873 1874 1875 1876 1877 1878 1879 1880 1881
static
ulint
lock_rec_lock_slow(
/*===============*/
				/* 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 */
unknown's avatar
unknown committed
1882 1883
	ulint		mode,	/* in: lock mode: LOCK_X or LOCK_S possibly
				ORed to either LOCK_GAP or LOCK_REC_NOT_GAP */
1884 1885 1886 1887 1888 1889 1890
	rec_t*		rec,	/* in: record */
	dict_index_t*	index,	/* in: index of record */
	que_thr_t* 	thr)	/* in: query thread */
{
	trx_t*	trx;
	ulint	err;

1891
#ifdef UNIV_SYNC_DEBUG
1892
	ut_ad(mutex_own(&kernel_mutex));
1893
#endif /* UNIV_SYNC_DEBUG */
unknown's avatar
unknown committed
1894 1895 1896 1897 1898 1899 1900 1901 1902 1903
	ut_ad((LOCK_MODE_MASK & mode) != LOCK_S
		|| lock_table_has(thr_get_trx(thr), index->table, LOCK_IS));
	ut_ad((LOCK_MODE_MASK & mode) != LOCK_X
		|| lock_table_has(thr_get_trx(thr), index->table, LOCK_IX));
	ut_ad((LOCK_MODE_MASK & mode) == LOCK_S
		|| (LOCK_MODE_MASK & mode) == LOCK_X);
	ut_ad(mode - (LOCK_MODE_MASK & mode) == LOCK_GAP
			|| mode - (LOCK_MODE_MASK & mode) == 0
			|| mode - (LOCK_MODE_MASK & mode) == LOCK_REC_NOT_GAP);
			
1904
	trx = thr_get_trx(thr);
unknown's avatar
unknown committed
1905
		
1906 1907 1908 1909 1910
	if (lock_rec_has_expl(mode, rec, trx)) {
		/* The trx already has a strong enough lock on rec: do
		nothing */

		err = DB_SUCCESS;
unknown's avatar
unknown committed
1911
	} else if (lock_rec_other_has_conflicting(mode, rec, trx)) {
unknown's avatar
unknown committed
1912

1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935
		/* If another transaction has a non-gap conflicting request in
		the queue, as this transaction does not have a lock strong
		enough already granted on the record, we have to wait. */
    				
		err = lock_rec_enqueue_waiting(mode, rec, index, thr);
	} else {
		if (!impl) {
			/* Set the requested lock on the record */

			lock_rec_add_to_queue(LOCK_REC | mode, rec, index,
									trx);
		}

		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
unknown's avatar
unknown committed
1936 1937
explicit locks. This function sets a normal next-key lock, or in the case
of a page supremum record, a gap type lock. */
unknown's avatar
unknown committed
1938
static
1939 1940 1941 1942 1943 1944 1945 1946
ulint
lock_rec_lock(
/*==========*/
				/* 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 */
unknown's avatar
unknown committed
1947 1948
	ulint		mode,	/* in: lock mode: LOCK_X or LOCK_S possibly
				ORed to either LOCK_GAP or LOCK_REC_NOT_GAP */
1949 1950 1951 1952 1953 1954
	rec_t*		rec,	/* in: record */
	dict_index_t*	index,	/* in: index of record */
	que_thr_t* 	thr)	/* in: query thread */
{
	ulint	err;

1955
#ifdef UNIV_SYNC_DEBUG
1956
	ut_ad(mutex_own(&kernel_mutex));
1957
#endif /* UNIV_SYNC_DEBUG */
unknown's avatar
unknown committed
1958 1959 1960 1961 1962 1963 1964 1965 1966 1967
	ut_ad((LOCK_MODE_MASK & mode) != LOCK_S
		|| lock_table_has(thr_get_trx(thr), index->table, LOCK_IS));
	ut_ad((LOCK_MODE_MASK & mode) != LOCK_X
		|| lock_table_has(thr_get_trx(thr), index->table, LOCK_IX));
	ut_ad((LOCK_MODE_MASK & mode) == LOCK_S
		|| (LOCK_MODE_MASK & mode) == LOCK_X);
	ut_ad(mode - (LOCK_MODE_MASK & mode) == LOCK_GAP
			|| mode - (LOCK_MODE_MASK & mode) == LOCK_REC_NOT_GAP
			|| mode - (LOCK_MODE_MASK & mode) == 0);
			
1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981
	if (lock_rec_lock_fast(impl, mode, rec, index, thr)) {

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

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

	return(err);
}

/*************************************************************************
unknown's avatar
unknown committed
1982
Checks if a waiting record lock request still has to wait in a queue. */
1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994
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;

1995
#ifdef UNIV_SYNC_DEBUG
1996
	ut_ad(mutex_own(&kernel_mutex));
1997
#endif /* UNIV_SYNC_DEBUG */
1998
 	ut_ad(lock_get_wait(wait_lock));
unknown's avatar
unknown committed
1999
	ut_ad(lock_get_type(wait_lock) == LOCK_REC);
2000 2001 2002 2003 2004 2005 2006 2007 2008
 	
	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) {

unknown's avatar
unknown committed
2009 2010
		if (lock_rec_get_nth_bit(lock, heap_no)
		    && lock_has_to_wait(wait_lock, lock)) {
2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023

		    	return(TRUE);
		}

		lock = lock_rec_get_next_on_page(lock);
	}

	return(FALSE);
}

/*****************************************************************
Grants a lock to a waiting lock request and releases the waiting
transaction. */
unknown's avatar
unknown committed
2024
static
2025 2026 2027 2028 2029
void
lock_grant(
/*=======*/
	lock_t*	lock)	/* in: waiting lock request */
{
2030
#ifdef UNIV_SYNC_DEBUG
2031
	ut_ad(mutex_own(&kernel_mutex));
2032
#endif /* UNIV_SYNC_DEBUG */
2033 2034

	lock_reset_lock_and_trx_wait(lock);
unknown's avatar
unknown committed
2035

unknown's avatar
unknown committed
2036
        if (lock_get_mode(lock) == LOCK_AUTO_INC) {
unknown's avatar
unknown committed
2037

unknown's avatar
unknown committed
2038 2039 2040 2041
                if (lock->trx->auto_inc_lock != NULL) {
                        fprintf(stderr,
                   "InnoDB: Error: trx already had an AUTO-INC lock!\n");
                }
unknown's avatar
unknown committed
2042

unknown's avatar
unknown committed
2043 2044 2045 2046
                /* Store pointer to lock to trx so that we know to
                release it at the end of the SQL statement */

                lock->trx->auto_inc_lock = lock;
2047
        } else if (lock_get_type(lock) == LOCK_TABLE_EXP) {
unknown's avatar
unknown committed
2048
		ut_a(lock_get_mode(lock) == LOCK_S
2049 2050
			|| lock_get_mode(lock) == LOCK_X);
	}
2051 2052

	if (lock_print_waits) {
2053
		fprintf(stderr, "Lock wait for trx %lu ends\n",
2054
		       (ulong) ut_dulint_get_low(lock->trx->id));
2055
	}
unknown's avatar
unknown committed
2056 2057 2058 2059 2060

	/* 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 */
2061
	
unknown's avatar
unknown committed
2062 2063 2064
	if (lock->trx->que_state == TRX_QUE_LOCK_WAIT) {	
		trx_end_lock_wait(lock->trx);
	}
2065 2066 2067 2068 2069 2070
}

/*****************************************************************
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! */
2071
static
2072 2073 2074 2075 2076
void
lock_rec_cancel(
/*============*/
	lock_t*	lock)	/* in: waiting record lock request */
{
2077
#ifdef UNIV_SYNC_DEBUG
2078
	ut_ad(mutex_own(&kernel_mutex));
2079
#endif /* UNIV_SYNC_DEBUG */
unknown's avatar
unknown committed
2080
	ut_ad(lock_get_type(lock) == LOCK_REC);
2081

unknown's avatar
unknown committed
2082
	/* Reset the bit (there can be only one set bit) in the lock bitmap */
2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097
	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);
}
	
/*****************************************************************
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. */
unknown's avatar
unknown committed
2098
static
2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111
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;
	
2112
#ifdef UNIV_SYNC_DEBUG
2113
	ut_ad(mutex_own(&kernel_mutex));
2114
#endif /* UNIV_SYNC_DEBUG */
2115 2116 2117
	ut_ad(lock_get_type(in_lock) == LOCK_REC);

	trx = in_lock->trx;
unknown's avatar
unknown committed
2118

2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131
	space = in_lock->un_member.rec_lock.space;
	page_no = in_lock->un_member.rec_lock.page_no;

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

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

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

	lock = lock_rec_get_first_on_page_addr(space, page_no);

unknown's avatar
unknown committed
2132
	while (lock != NULL) {		
2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143
		if (lock_get_wait(lock)
				&& !lock_rec_has_to_wait_in_queue(lock)) {

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

		lock = lock_rec_get_next_on_page(lock);
	}
}	

unknown's avatar
unknown committed
2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156
/*****************************************************************
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;
	
2157
#ifdef UNIV_SYNC_DEBUG
unknown's avatar
unknown committed
2158
	ut_ad(mutex_own(&kernel_mutex));
2159
#endif /* UNIV_SYNC_DEBUG */
unknown's avatar
unknown committed
2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172
	ut_ad(lock_get_type(in_lock) == LOCK_REC);

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

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

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

2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187
/*****************************************************************
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(
/*================================*/
	page_t*	page)	/* in: page to be discarded */
{
	ulint	space;
	ulint	page_no;
	lock_t*	lock;
	lock_t*	next_lock;
	
2188
#ifdef UNIV_SYNC_DEBUG
2189
	ut_ad(mutex_own(&kernel_mutex));
2190
#endif /* UNIV_SYNC_DEBUG */
2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202
	
	space = buf_frame_get_space_id(page);
	page_no = buf_frame_get_page_no(page);

	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);
		
unknown's avatar
unknown committed
2203
		lock_rec_discard(lock);
2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222
		
		lock = next_lock;
	}
}	

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

/*****************************************************************
Resets the lock bits for a single record. Releases transactions waiting for
lock requests here. */

void
lock_rec_reset_and_release_wait(
/*============================*/
	rec_t*	rec)	/* in: record whose locks bits should be reset */
{
	lock_t*	lock;
	ulint	heap_no;
	
2223
#ifdef UNIV_SYNC_DEBUG
2224
	ut_ad(mutex_own(&kernel_mutex));
2225
#endif /* UNIV_SYNC_DEBUG */
2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242

	heap_no = rec_get_heap_no(rec);
	
	lock = lock_rec_get_first(rec);

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

		lock = lock_rec_get_next(rec, lock);
	}
}	

/*****************************************************************
unknown's avatar
unknown committed
2243 2244 2245 2246
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. */
2247 2248 2249 2250 2251 2252 2253 2254 2255 2256

void
lock_rec_inherit_to_gap(
/*====================*/
	rec_t*	heir,	/* in: record which inherits */
	rec_t*	rec)	/* in: record from which inherited; does NOT reset
			the locks on this record */
{
	lock_t*	lock;
	
2257
#ifdef UNIV_SYNC_DEBUG
2258
	ut_ad(mutex_own(&kernel_mutex));
2259
#endif /* UNIV_SYNC_DEBUG */
2260 2261 2262 2263
	
	lock = lock_rec_get_first(rec);

	while (lock != NULL) {
unknown's avatar
unknown committed
2264 2265 2266 2267
		if (!lock_rec_get_insert_intention(lock)) {
			
			lock_rec_add_to_queue(LOCK_REC | lock_get_mode(lock)
						| LOCK_GAP,
2268
	 			     		heir, lock->index, lock->trx);
unknown's avatar
unknown committed
2269 2270 2271 2272 2273 2274 2275 2276 2277 2278
	 	}
	 	
		lock = lock_rec_get_next(rec, lock);
	}
}	

/*****************************************************************
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. */
unknown's avatar
unknown committed
2279
static
unknown's avatar
unknown committed
2280 2281 2282 2283 2284 2285 2286 2287 2288
void
lock_rec_inherit_to_gap_if_gap_lock(
/*================================*/
	rec_t*	heir,	/* in: record which inherits */
	rec_t*	rec)	/* in: record from which inherited; does NOT reset
			the locks on this record */
{
	lock_t*	lock;
	
2289
#ifdef UNIV_SYNC_DEBUG
unknown's avatar
unknown committed
2290
	ut_ad(mutex_own(&kernel_mutex));
2291
#endif /* UNIV_SYNC_DEBUG */
unknown's avatar
unknown committed
2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304
	
	lock = lock_rec_get_first(rec);

	while (lock != NULL) {
		if (!lock_rec_get_insert_intention(lock)
		    && (page_rec_is_supremum(rec)
			|| !lock_rec_get_rec_not_gap(lock))) {
			
			lock_rec_add_to_queue(LOCK_REC | lock_get_mode(lock)
						| LOCK_GAP,
	 			     		heir, lock->index, lock->trx);
	 	}

2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323
		lock = lock_rec_get_next(rec, lock);
	}
}	

/*****************************************************************
Moves the locks of a record to another record and resets the lock bits of
the donating record. */
static
void
lock_rec_move(
/*==========*/
	rec_t*	receiver,	/* in: record which gets locks; this record
				must have no lock requests on it! */
	rec_t*	donator)	/* in: record which gives locks */
{
	lock_t*	lock;
	ulint	heap_no;
	ulint	type_mode;
	
2324
#ifdef UNIV_SYNC_DEBUG
2325
	ut_ad(mutex_own(&kernel_mutex));
2326
#endif /* UNIV_SYNC_DEBUG */
2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442

	heap_no = rec_get_heap_no(donator);
	
	lock = lock_rec_get_first(donator);

	ut_ad(lock_rec_get_first(receiver) == NULL);

	while (lock != NULL) {
		type_mode = lock->type_mode;
	
		lock_rec_reset_nth_bit(lock, heap_no);

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

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

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

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

/*****************************************************************
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. */

void
lock_move_reorganize_page(
/*======================*/
	page_t*	page,		/* in: old index page, now reorganized */
	page_t*	old_page)	/* in: copy of the old, not reorganized page */
{
	lock_t*		lock;
	lock_t*		old_lock;
	page_cur_t	cur1;
	page_cur_t	cur2;
	ulint		old_heap_no;
	UT_LIST_BASE_NODE_T(lock_t)	old_locks;
	mem_heap_t*	heap		= NULL;
	rec_t*		sup;

	lock_mutex_enter_kernel();

	lock = lock_rec_get_first_on_page(page);

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

		return;
	}

	heap = mem_heap_create(256);
	
	/* Copy first all the locks on the page to heap and reset the
	bitmaps in the original locks; chain the copies of the locks
	using the trx_locks field in them. */

	UT_LIST_INIT(old_locks);
	
	while (lock != NULL) {

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

		UT_LIST_ADD_LAST(trx_locks, old_locks, old_lock);

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

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

		lock = lock_rec_get_next_on_page(lock);
	}

	sup = page_get_supremum_rec(page);
	
	lock = UT_LIST_GET_FIRST(old_locks);

	while (lock) {
		/* NOTE: we copy also the locks set on the infimum and
		supremum of the page; the infimum may carry locks if an
		update of a record is occurring on the page, and its locks
		were temporarily stored on the infimum */
		
		page_cur_set_before_first(page, &cur1);
		page_cur_set_before_first(old_page, &cur2);

		/* Set locks according to old locks */
		for (;;) {
			ut_ad(0 == ut_memcmp(page_cur_get_rec(&cur1),
						page_cur_get_rec(&cur2),
						rec_get_data_size(
						   page_cur_get_rec(&cur2))));
		
			old_heap_no = rec_get_heap_no(page_cur_get_rec(&cur2));

			if (lock_rec_get_nth_bit(lock, old_heap_no)) {

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

				lock_rec_add_to_queue(lock->type_mode,
						page_cur_get_rec(&cur1),
						lock->index, lock->trx);

				/* if ((page_cur_get_rec(&cur1) == sup)
						&& lock_get_wait(lock)) {
2443
					fprintf(stderr,
2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560
				"---\n--\n!!!Lock reorg: supr type %lu\n",
					lock->type_mode);
				} */
			}

			if (page_cur_get_rec(&cur1) == sup) {

				break;
			}

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

		/* Remember that we chained old locks on the trx_locks field: */

		lock = UT_LIST_GET_NEXT(trx_locks, lock);
	}

	lock_mutex_exit_kernel();

	mem_heap_free(heap);

/* 	ut_ad(lock_rec_validate_page(buf_frame_get_space_id(page),
					buf_frame_get_page_no(page))); */
}	

/*****************************************************************
Moves the explicit locks on user records to another page if a record
list end is moved to another page. */

void
lock_move_rec_list_end(
/*===================*/
	page_t*	new_page,	/* in: index page to move to */
	page_t*	page,		/* in: index page */
	rec_t*	rec)		/* in: record on page: this is the
				first record moved */
{
	lock_t*		lock;
	page_cur_t	cur1;
	page_cur_t	cur2;
	ulint		heap_no;
	rec_t*		sup;
	ulint		type_mode;
	
	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. */

	sup = page_get_supremum_rec(page);
	
	lock = lock_rec_get_first_on_page(page);

	while (lock != NULL) {
		
		page_cur_position(rec, &cur1);

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

		page_cur_set_before_first(new_page, &cur2);
		page_cur_move_to_next(&cur2);
	
		/* Copy lock requests on user records to new page and
		reset the lock bits on the old */

		while (page_cur_get_rec(&cur1) != sup) {

			ut_ad(0 == ut_memcmp(page_cur_get_rec(&cur1),
						page_cur_get_rec(&cur2),
						rec_get_data_size(
						   page_cur_get_rec(&cur2))));
		
			heap_no = rec_get_heap_no(page_cur_get_rec(&cur1));

			if (lock_rec_get_nth_bit(lock, heap_no)) {
				type_mode = lock->type_mode;

				lock_rec_reset_nth_bit(lock, heap_no);

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

				lock_rec_add_to_queue(type_mode,
						page_cur_get_rec(&cur2),
						lock->index, lock->trx);
			}

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

		lock = lock_rec_get_next_on_page(lock);
	}
	
	lock_mutex_exit_kernel();

/*	ut_ad(lock_rec_validate_page(buf_frame_get_space_id(page),
					buf_frame_get_page_no(page)));
	ut_ad(lock_rec_validate_page(buf_frame_get_space_id(new_page),
					buf_frame_get_page_no(new_page))); */
}	

/*****************************************************************
Moves the explicit locks on user records to another page if a record
list start is moved to another page. */

void
lock_move_rec_list_start(
/*=====================*/
unknown's avatar
unknown committed
2561
	page_t*	new_page,	/* in: index page to move to */
2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573
	page_t*	page,		/* in: index page */
	rec_t*	rec,		/* in: record on page: this is the
				first record NOT copied */
	rec_t*	old_end)	/* in: old previous-to-last record on
				new_page before the records were copied */
{
	lock_t*		lock;
	page_cur_t	cur1;
	page_cur_t	cur2;
	ulint		heap_no;
	ulint		type_mode;

unknown's avatar
unknown committed
2574
	ut_a(new_page);
2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864

	lock_mutex_enter_kernel();

	lock = lock_rec_get_first_on_page(page);

	while (lock != NULL) {
		
		page_cur_set_before_first(page, &cur1);
		page_cur_move_to_next(&cur1);

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

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

		while (page_cur_get_rec(&cur1) != rec) {

			ut_ad(0 == ut_memcmp(page_cur_get_rec(&cur1),
						page_cur_get_rec(&cur2),
						rec_get_data_size(
						   page_cur_get_rec(&cur2))));
		
			heap_no = rec_get_heap_no(page_cur_get_rec(&cur1));

			if (lock_rec_get_nth_bit(lock, heap_no)) {
				type_mode = lock->type_mode;

				lock_rec_reset_nth_bit(lock, heap_no);

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

				lock_rec_add_to_queue(type_mode,
						page_cur_get_rec(&cur2),
						lock->index, lock->trx);
			}

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

		lock = lock_rec_get_next_on_page(lock);
	}
	
	lock_mutex_exit_kernel();

/*	ut_ad(lock_rec_validate_page(buf_frame_get_space_id(page),
					buf_frame_get_page_no(page)));
	ut_ad(lock_rec_validate_page(buf_frame_get_space_id(new_page),
					buf_frame_get_page_no(new_page))); */
}	

/*****************************************************************
Updates the lock table when a page is split to the right. */

void
lock_update_split_right(
/*====================*/
	page_t*	right_page,	/* in: right page */
	page_t*	left_page)	/* in: left page */
{
	lock_mutex_enter_kernel();
	
	/* Move the locks on the supremum of the left page to the supremum
	of the right page */

	lock_rec_move(page_get_supremum_rec(right_page),
					page_get_supremum_rec(left_page));
	
	/* Inherit the locks to the supremum of left page from the successor
	of the infimum on right page */

	lock_rec_inherit_to_gap(page_get_supremum_rec(left_page),
			page_rec_get_next(page_get_infimum_rec(right_page)));

	lock_mutex_exit_kernel();
}	

/*****************************************************************
Updates the lock table when a page is merged to the right. */

void
lock_update_merge_right(
/*====================*/
	rec_t*	orig_succ,	/* in: original successor of infimum
				on the right page before merge */
	page_t*	left_page)	/* in: merged index page which will be
				discarded */
{
	lock_mutex_enter_kernel();
	
	/* Inherit the locks from the supremum of the left page to the
	original successor of infimum on the right page, to which the left
	page was merged */

	lock_rec_inherit_to_gap(orig_succ, page_get_supremum_rec(left_page));

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

	lock_rec_reset_and_release_wait(page_get_supremum_rec(left_page));
	
	lock_rec_free_all_from_discard_page(left_page);

	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. */

void
lock_update_root_raise(
/*===================*/
	page_t*	new_page,	/* in: index page to which copied */
	page_t*	root)		/* in: root page */
{
	lock_mutex_enter_kernel();
	
	/* Move the locks on the supremum of the root to the supremum
	of new_page */

	lock_rec_move(page_get_supremum_rec(new_page),
						page_get_supremum_rec(root));
	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! */

void
lock_update_copy_and_discard(
/*=========================*/
	page_t*	new_page,	/* in: index page to which copied */
	page_t*	page)		/* in: index page; NOT the root! */
{
	lock_mutex_enter_kernel();
	
	/* Move the locks on the supremum of the old page to the supremum
	of new_page */

	lock_rec_move(page_get_supremum_rec(new_page),
						page_get_supremum_rec(page));
	lock_rec_free_all_from_discard_page(page);

	lock_mutex_exit_kernel();
}

/*****************************************************************
Updates the lock table when a page is split to the left. */

void
lock_update_split_left(
/*===================*/
	page_t*	right_page,	/* in: right page */
	page_t*	left_page)	/* in: left page */
{
	lock_mutex_enter_kernel();
	
	/* Inherit the locks to the supremum of the left page from the
	successor of the infimum on the right page */

	lock_rec_inherit_to_gap(page_get_supremum_rec(left_page),
			page_rec_get_next(page_get_infimum_rec(right_page)));

	lock_mutex_exit_kernel();
}

/*****************************************************************
Updates the lock table when a page is merged to the left. */

void
lock_update_merge_left(
/*===================*/
	page_t*	left_page,	/* in: left page to which merged */
	rec_t*	orig_pred,	/* in: original predecessor of supremum
				on the left page before merge */
	page_t*	right_page)	/* in: merged index page which will be
				discarded */
{
	lock_mutex_enter_kernel();
	
	if (page_rec_get_next(orig_pred) != page_get_supremum_rec(left_page)) {

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

		lock_rec_inherit_to_gap(page_rec_get_next(orig_pred),
					page_get_supremum_rec(left_page));

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

		lock_rec_reset_and_release_wait(page_get_supremum_rec(
								left_page));
	}

	/* Move the locks from the supremum of right page to the supremum
	of the left page */
	
	lock_rec_move(page_get_supremum_rec(left_page),
					 page_get_supremum_rec(right_page));

	lock_rec_free_all_from_discard_page(right_page);

	lock_mutex_exit_kernel();
}

/*****************************************************************
Resets the original locks on heir and replaces them with gap type locks
inherited from rec. */

void
lock_rec_reset_and_inherit_gap_locks(
/*=================================*/
	rec_t*	heir,	/* in: heir record */
	rec_t*	rec)	/* in: record */
{
	mutex_enter(&kernel_mutex);	      				

	lock_rec_reset_and_release_wait(heir);
	
	lock_rec_inherit_to_gap(heir, rec);

	mutex_exit(&kernel_mutex);	      				
}

/*****************************************************************
Updates the lock table when a page is discarded. */

void
lock_update_discard(
/*================*/
	rec_t*	heir,	/* in: record which will inherit the locks */
	page_t*	page)	/* in: index page which will be discarded */
{
	rec_t*	rec;

	lock_mutex_enter_kernel();
	
	if (NULL == lock_rec_get_first_on_page(page)) {
		/* No locks exist on page, nothing to do */

		lock_mutex_exit_kernel();

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

	rec = page_get_infimum_rec(page);

	for (;;) {
		lock_rec_inherit_to_gap(heir, rec);

		/* Reset the locks on rec, releasing waiting transactions */

		lock_rec_reset_and_release_wait(rec);

		if (rec == page_get_supremum_rec(page)) {

			break;
		}
		
		rec = page_rec_get_next(rec);
	}

	lock_rec_free_all_from_discard_page(page);

	lock_mutex_exit_kernel();
}

/*****************************************************************
Updates the lock table when a new user record is inserted. */

void
lock_update_insert(
/*===============*/
	rec_t*	rec)	/* in: the inserted record */
{
	lock_mutex_enter_kernel();

unknown's avatar
unknown committed
2865 2866
	/* Inherit the gap-locking locks for rec, in gap mode, from the next
	record */
2867

unknown's avatar
unknown committed
2868
	lock_rec_inherit_to_gap_if_gap_lock(rec, page_rec_get_next(rec));
2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946

	lock_mutex_exit_kernel();
}	

/*****************************************************************
Updates the lock table when a record is removed. */

void
lock_update_delete(
/*===============*/
	rec_t*	rec)	/* in: the record to be removed */
{
	lock_mutex_enter_kernel();

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

	lock_rec_inherit_to_gap(page_rec_get_next(rec), rec);

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

	lock_rec_reset_and_release_wait(rec);	

	lock_mutex_exit_kernel();
}
 
/*************************************************************************
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. */

void
lock_rec_store_on_page_infimum(
/*===========================*/
	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 */
{
	page_t*	page;

	page = buf_frame_align(rec);

	lock_mutex_enter_kernel();
	
	lock_rec_move(page_get_infimum_rec(page), rec);

	lock_mutex_exit_kernel();	
}

/*************************************************************************
Restores the state of explicit lock requests on a single record, where the
state was stored on the infimum of the page. */

void
lock_rec_restore_from_page_infimum(
/*===============================*/
	rec_t*	rec,	/* in: record whose lock state is restored */
	page_t*	page)	/* in: page (rec is not necessarily on this page)
			whose infimum stored the lock state; lock bits are
			reset on the infimum */ 
{
	lock_mutex_enter_kernel();
	
	lock_rec_move(rec, page_get_infimum_rec(page));
	
	lock_mutex_exit_kernel();
}

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

/************************************************************************
Checks if a lock request results in a deadlock. */
static
ibool
lock_deadlock_occurs(
/*=================*/
unknown's avatar
unknown committed
2947 2948 2949 2950
			/* 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) */
2951 2952 2953 2954 2955
	lock_t*	lock,	/* in: lock the transaction is requesting */
	trx_t*	trx)	/* in: transaction */
{
	dict_table_t*	table;
	dict_index_t*	index;
2956
	trx_t*		mark_trx;
unknown's avatar
unknown committed
2957
	ulint		ret;
2958
	ulint		cost	= 0;
2959 2960

	ut_ad(trx && lock);
2961
#ifdef UNIV_SYNC_DEBUG
2962
	ut_ad(mutex_own(&kernel_mutex));
2963
#endif /* UNIV_SYNC_DEBUG */
unknown's avatar
unknown committed
2964
retry:
2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976
	/* 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);
2977

unknown's avatar
unknown committed
2978 2979 2980 2981 2982 2983 2984 2985
	if (ret == LOCK_VICTIM_IS_OTHER) {
		/* We chose some other trx as a victim: retry if there still
		is a deadlock */

		goto retry;
	}

	if (ret == LOCK_VICTIM_IS_START) {
2986
		if (lock_get_type(lock) & LOCK_TABLE) {
2987 2988 2989 2990 2991 2992
			table = lock->un_member.tab_lock.table;
			index = NULL;
		} else {
			index = lock->index;
			table = index->table;
		}
unknown's avatar
unknown committed
2993 2994 2995

		lock_deadlock_found = TRUE;

2996 2997
		fputs("*** WE ROLL BACK TRANSACTION (2)\n",
			lock_latest_err_file);
2998

unknown's avatar
unknown committed
2999
		return(TRUE);
3000 3001
	}
	
unknown's avatar
unknown committed
3002
	return(FALSE);
3003 3004 3005 3006 3007
}

/************************************************************************
Looks recursively for a deadlock. */
static
unknown's avatar
unknown committed
3008
ulint
3009 3010
lock_deadlock_recursive(
/*====================*/
unknown's avatar
unknown committed
3011 3012 3013 3014 3015 3016 3017 3018
				/* 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! */
3019 3020
	trx_t*	start,		/* in: recursion starting point */
	trx_t*	trx,		/* in: a transaction waiting for a lock */
3021 3022 3023
	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_...
unknown's avatar
unknown committed
3024
				we return LOCK_VICTIM_IS_START */
3025 3026
{
	lock_t*	lock;
unknown's avatar
unknown committed
3027
	ulint	bit_no		= ULINT_UNDEFINED;
3028
	trx_t*	lock_trx;
unknown's avatar
unknown committed
3029
	ulint	ret;
3030 3031
	
	ut_a(trx && start && wait_lock);
3032
#ifdef UNIV_SYNC_DEBUG
3033
	ut_ad(mutex_own(&kernel_mutex));
3034
#endif /* UNIV_SYNC_DEBUG */
3035
	
3036 3037 3038 3039
	if (trx->deadlock_mark == 1) {
		/* We have already exhaustively searched the subtree starting
		from this trx */

unknown's avatar
unknown committed
3040
		return(0);
3041 3042 3043 3044 3045 3046
	}

	*cost = *cost + 1;

	if (*cost > LOCK_MAX_N_STEPS_IN_DEADLOCK_CHECK) {

unknown's avatar
unknown committed
3047
		return(LOCK_VICTIM_IS_START);
3048 3049
	}

3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061
	lock = wait_lock;

	if (lock_get_type(wait_lock) == LOCK_REC) {

		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 (;;) {
3062
		if (lock_get_type(lock) & LOCK_TABLE) {
3063 3064 3065 3066

			lock = UT_LIST_GET_PREV(un_member.tab_lock.locks, lock);
		} else {
			ut_ad(lock_get_type(lock) == LOCK_REC);
unknown's avatar
unknown committed
3067
			ut_a(bit_no != ULINT_UNDEFINED);
3068 3069 3070 3071 3072

			lock = lock_rec_get_prev(lock, bit_no);
		}

		if (lock == NULL) {
3073 3074
			/* We can mark this subtree as searched */
			trx->deadlock_mark = 1;
3075 3076 3077 3078 3079 3080 3081 3082 3083

			return(FALSE);
		}

		if (lock_has_to_wait(wait_lock, lock)) {

			lock_trx = lock->trx;

			if (lock_trx == start) {
unknown's avatar
unknown committed
3084 3085
				/* We came back to the recursion starting
				point: a deadlock detected */
3086
				FILE*	ef = lock_latest_err_file;
unknown's avatar
unknown committed
3087
				
3088 3089
				rewind(ef);
				ut_print_timestamp(ef);
unknown's avatar
unknown committed
3090

3091
				fputs("\n*** (1) TRANSACTION:\n", ef);
unknown's avatar
unknown committed
3092

3093
				trx_print(ef, wait_lock->trx);
unknown's avatar
unknown committed
3094

3095 3096
				fputs(
			"*** (1) WAITING FOR THIS LOCK TO BE GRANTED:\n", ef);
unknown's avatar
unknown committed
3097 3098
			
				if (lock_get_type(wait_lock) == LOCK_REC) {
3099
					lock_rec_print(ef, wait_lock);
unknown's avatar
unknown committed
3100
				} else {
3101
					lock_table_print(ef, wait_lock);
unknown's avatar
unknown committed
3102 3103
				}
			
3104
				fputs("*** (2) TRANSACTION:\n", ef);
unknown's avatar
unknown committed
3105

3106
				trx_print(ef, lock->trx);
unknown's avatar
unknown committed
3107

3108
				fputs("*** (2) HOLDS THE LOCK(S):\n", ef);
unknown's avatar
unknown committed
3109 3110
			
				if (lock_get_type(lock) == LOCK_REC) {
3111
					lock_rec_print(ef, lock);
unknown's avatar
unknown committed
3112
				} else {
3113
					lock_table_print(ef, lock);
unknown's avatar
unknown committed
3114 3115
				}
			
3116 3117
				fputs(
			"*** (2) WAITING FOR THIS LOCK TO BE GRANTED:\n", ef);
unknown's avatar
unknown committed
3118 3119 3120
			
				if (lock_get_type(start->wait_lock)
								== LOCK_REC) {
3121
					lock_rec_print(ef, start->wait_lock);
unknown's avatar
unknown committed
3122
				} else {
3123
					lock_table_print(ef, start->wait_lock);
unknown's avatar
unknown committed
3124 3125
				}

3126
				if (lock_print_waits) {
3127
					fputs("Deadlock detected\n", stderr);
3128 3129
				}

unknown's avatar
unknown committed
3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145
				if (ut_dulint_cmp(wait_lock->trx->undo_no,
							start->undo_no) >= 0) {
					/* Our recursion starting point
					transaction is 'smaller', let us
					choose 'start' as the victim and roll
					back it */

					return(LOCK_VICTIM_IS_START);
				}		

				lock_deadlock_found = TRUE;

				/* Let us choose the transaction of wait_lock
				as a victim to try to avoid deadlocking our
				recursion starting point transaction */
				
3146 3147
				fputs("*** WE ROLL BACK TRANSACTION (1)\n",
					ef);
unknown's avatar
unknown committed
3148
				
unknown's avatar
unknown committed
3149 3150
				wait_lock->trx->was_chosen_as_deadlock_victim
								= TRUE;
unknown's avatar
unknown committed
3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161
				
				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);
3162 3163 3164 3165 3166 3167 3168 3169
			}
	
			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 */

unknown's avatar
unknown committed
3170 3171 3172
				ret = lock_deadlock_recursive(start, lock_trx,
						lock_trx->wait_lock, cost);
				if (ret != 0) {
3173

unknown's avatar
unknown committed
3174
					return(ret);
3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199
				}
			}
		}
	}/* 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(
/*==============*/
				/* out, own: new lock object, or NULL if
				out of memory */
	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);
3200
#ifdef UNIV_SYNC_DEBUG
3201
	ut_ad(mutex_own(&kernel_mutex));
3202
#endif /* UNIV_SYNC_DEBUG */
3203

3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215
	if (type_mode == LOCK_AUTO_INC) {
		/* Only one trx can have the lock on the table
		at a time: we may use the memory preallocated
		to the table object */

		lock = table->auto_inc_lock;

		ut_a(trx->auto_inc_lock == NULL);
		trx->auto_inc_lock = lock;
	} else {
		lock = mem_heap_alloc(trx->lock_heap, sizeof(lock_t));
	}
3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226

	if (lock == NULL) {

		return(NULL);
	}

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

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

unknown's avatar
unknown committed
3227
	if (lock_get_type(lock) == LOCK_TABLE_EXP) {
unknown's avatar
unknown committed
3228 3229 3230
		lock->trx->n_lock_table_exp++;
	}

3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255
	lock->un_member.tab_lock.table = table;

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

	if (type_mode & LOCK_WAIT) {

		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 */
{
	dict_table_t*	table;
	trx_t*		trx;

3256
#ifdef UNIV_SYNC_DEBUG
3257
	ut_ad(mutex_own(&kernel_mutex));
3258
#endif /* UNIV_SYNC_DEBUG */
3259 3260 3261 3262

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

3263 3264 3265
	if (lock == trx->auto_inc_lock) {
		trx->auto_inc_lock = NULL;
	}
unknown's avatar
unknown committed
3266 3267 3268 3269 3270

	if (lock_get_type(lock) == LOCK_TABLE_EXP) {
		lock->trx->n_lock_table_exp--;
	}

3271 3272 3273 3274 3275 3276 3277
	UT_LIST_REMOVE(trx_locks, trx->trx_locks, lock);
	UT_LIST_REMOVE(un_member.tab_lock.locks, table->locks, lock);
}	

/*************************************************************************
Enqueues a waiting request for a table lock which cannot be granted
immediately. Checks for deadlocks. */
unknown's avatar
unknown committed
3278
static
3279 3280 3281 3282
ulint
lock_table_enqueue_waiting(
/*=======================*/
				/* out: DB_LOCK_WAIT, DB_DEADLOCK, or
unknown's avatar
unknown committed
3283 3284 3285 3286 3287
				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 */
3288 3289 3290 3291 3292 3293 3294 3295
	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;
	
3296
#ifdef UNIV_SYNC_DEBUG
3297
	ut_ad(mutex_own(&kernel_mutex));
3298
#endif /* UNIV_SYNC_DEBUG */
unknown's avatar
unknown committed
3299
	
3300 3301 3302 3303 3304
	/* 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)) {
3305
		ut_error;
3306 3307 3308 3309 3310

		return(DB_QUE_THR_SUSPENDED);
	}

	trx = thr_get_trx(thr);
unknown's avatar
unknown committed
3311 3312 3313

	if (trx->dict_operation) {
		ut_print_timestamp(stderr);
3314
		fputs(
unknown's avatar
unknown committed
3315
"  InnoDB: Error: a table lock wait happens in a dictionary operation!\n"
3316 3317
"InnoDB: Table name ", stderr);
		ut_print_name(stderr, table->name);
3318 3319
		fputs(".\n"
"InnoDB: Submit a detailed bug report to http://bugs.mysql.com\n",
3320
			stderr);
unknown's avatar
unknown committed
3321
	}
3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337
	
	/* 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);
	}

unknown's avatar
unknown committed
3338 3339 3340 3341 3342 3343 3344
	if (trx->wait_lock == NULL) {
		/* Deadlock resolution chose another transaction as a victim,
		and we accidentally got our lock granted! */
	
		return(DB_SUCCESS);
	}
	
3345
	trx->que_state = TRX_QUE_LOCK_WAIT;
unknown's avatar
unknown committed
3346
	trx->was_chosen_as_deadlock_victim = FALSE;
unknown's avatar
unknown committed
3347
	trx->wait_started = time(NULL);
3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369

	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 */
	ulint		mode)	/* in: lock mode */
{
	lock_t*	lock;

3370
#ifdef UNIV_SYNC_DEBUG
3371
	ut_ad(mutex_own(&kernel_mutex));
3372
#endif /* UNIV_SYNC_DEBUG */
3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400

	lock = UT_LIST_GET_LAST(table->locks);

	while (lock != NULL) {

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

		    	return(TRUE);
		}

		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. */

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,
3401 3402 3403
				does nothing;
				if LOCK_TABLE_EXP bits are set,
				creates an explicit table lock */
3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417
	dict_table_t*	table,	/* in: database table in dictionary cache */
	ulint		mode,	/* in: lock mode */
	que_thr_t*	thr)	/* in: query thread */
{
	trx_t*	trx;
	ulint	err;
	
	ut_ad(table && thr);

	if (flags & BTR_NO_LOCKING_FLAG) {

		return(DB_SUCCESS);
	}

unknown's avatar
unknown committed
3418
	ut_a(flags == 0 || flags == LOCK_TABLE_EXP);
3419

3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437
	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)) {
	
3438
		/* Another trx has a request on the table in an incompatible
unknown's avatar
unknown committed
3439
		mode: this trx may have to wait */
3440 3441 3442 3443 3444 3445 3446 3447

		err = lock_table_enqueue_waiting(mode, table, thr);
			
		lock_mutex_exit_kernel();

		return(err);
	}

3448 3449
	lock_table_create(table, mode | flags, trx);

unknown's avatar
unknown committed
3450
	ut_a(!flags || mode == LOCK_S || mode == LOCK_X);
3451 3452 3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517

	lock_mutex_exit_kernel();

	return(DB_SUCCESS);
}

/*************************************************************************
Checks if there are any locks set on the table. */

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;

 	ut_ad(lock_get_wait(wait_lock));
 	
	table = wait_lock->un_member.tab_lock.table;

	lock = UT_LIST_GET_FIRST(table->locks);

	while (lock != wait_lock) {

		if (lock_has_to_wait(wait_lock, lock)) {

		    	return(TRUE);
		}

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

	return(FALSE);
}

/*****************************************************************
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. */
unknown's avatar
unknown committed
3518
static
3519 3520 3521 3522 3523 3524 3525 3526 3527
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;
	
3528
#ifdef UNIV_SYNC_DEBUG
3529
	ut_ad(mutex_own(&kernel_mutex));
3530
#endif /* UNIV_SYNC_DEBUG */
unknown's avatar
unknown committed
3531
	ut_a(lock_get_type(in_lock) == LOCK_TABLE ||
3532
		lock_get_type(in_lock) == LOCK_TABLE_EXP);
3533 3534 3535 3536

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

	lock_table_remove_low(in_lock);
3537

3538 3539 3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555
	/* Check if waiting locks in the queue can now be granted: grant
	locks if there are no conflicting locks ahead. */

	while (lock != NULL) {

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

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

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

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

3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571
/*************************************************************************
Releases a table lock.
Releases possible other transactions waiting for this lock. */

void
lock_table_unlock(
/*==============*/
	lock_t*	lock)	/* in: lock */
{
	mutex_enter(&kernel_mutex);

	lock_table_dequeue(lock);

	mutex_exit(&kernel_mutex);
}

3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589
/*************************************************************************
Releases an auto-inc lock a transaction possibly has on a table.
Releases possible other transactions waiting for this lock. */

void
lock_table_unlock_auto_inc(
/*=======================*/
	trx_t*	trx)	/* in: transaction */
{
	if (trx->auto_inc_lock) {
		mutex_enter(&kernel_mutex);

		lock_table_dequeue(trx->auto_inc_lock);

		mutex_exit(&kernel_mutex);
	}
}

3590 3591 3592 3593 3594 3595 3596 3597 3598
/*************************************************************************
Releases transaction locks, and releases possible other transactions waiting
because of these locks. */

void
lock_release_off_kernel(
/*====================*/
	trx_t*	trx)	/* in: transaction */
{
unknown's avatar
unknown committed
3599 3600 3601
	dict_table_t*	table;
	ulint		count;
	lock_t*		lock;
3602

3603
#ifdef UNIV_SYNC_DEBUG
3604
	ut_ad(mutex_own(&kernel_mutex));
3605
#endif /* UNIV_SYNC_DEBUG */
3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618

	lock = UT_LIST_GET_LAST(trx->trx_locks);
	
	count = 0;

	while (lock != NULL) {

		count++;

		if (lock_get_type(lock) == LOCK_REC) {
			
			lock_rec_dequeue_from_page(lock);
		} else {
3619
			ut_ad(lock_get_type(lock) & LOCK_TABLE);
3620

unknown's avatar
unknown committed
3621
			if (lock_get_mode(lock) != LOCK_IS
unknown's avatar
unknown committed
3622 3623
			    && 0 != ut_dulint_cmp(trx->undo_no,
						  ut_dulint_zero)) {
unknown's avatar
unknown committed
3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634

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

				table = lock->un_member.tab_lock.table;
			
				table->query_cache_inv_trx_id =
							trx_sys->max_trx_id;
			}

3635
			lock_table_dequeue(lock);
3636
			if (lock_get_type(lock) == LOCK_TABLE_EXP) {
unknown's avatar
unknown committed
3637
				ut_a(lock_get_mode(lock) == LOCK_S
3638 3639
					|| lock_get_mode(lock) == LOCK_X);
			}
3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655 3656
		}

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

			lock_mutex_exit_kernel();

			lock_mutex_enter_kernel();

			count = 0;
		}	

		lock = UT_LIST_GET_LAST(trx->trx_locks);
	}

	mem_heap_empty(trx->lock_heap);
3657 3658

	ut_a(trx->auto_inc_lock == NULL);
unknown's avatar
unknown committed
3659
	ut_a(trx->n_lock_table_exp == 0);
3660 3661 3662
}

/*************************************************************************
unknown's avatar
unknown committed
3663 3664
Releases table locks explicitly requested with LOCK TABLES (indicated by
lock type LOCK_TABLE_EXP), and releases possible other transactions waiting
3665 3666 3667 3668 3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683 3684 3685 3686 3687 3688
because of these locks. */

void
lock_release_tables_off_kernel(
/*===========================*/
	trx_t*	trx)	/* in: transaction */
{
	dict_table_t*	table;
	ulint		count;
	lock_t*		lock;

#ifdef UNIV_SYNC_DEBUG
	ut_ad(mutex_own(&kernel_mutex));
#endif /* UNIV_SYNC_DEBUG */

	lock = UT_LIST_GET_LAST(trx->trx_locks);

	count = 0;

	while (lock != NULL) {

		count++;

		if (lock_get_type(lock) == LOCK_TABLE_EXP) {
unknown's avatar
unknown committed
3689
			ut_a(lock_get_mode(lock) == LOCK_S
3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704
				|| lock_get_mode(lock) == LOCK_X);
			if (trx->insert_undo || trx->update_undo) {

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

				table = lock->un_member.tab_lock.table;

				table->query_cache_inv_trx_id =
							trx_sys->max_trx_id;
			}

			lock_table_dequeue(lock);
unknown's avatar
unknown committed
3705

3706 3707 3708 3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723
			lock = UT_LIST_GET_LAST(trx->trx_locks);
			continue;
		}

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

			lock_mutex_exit_kernel();

			lock_mutex_enter_kernel();

			count = 0;
		}

		lock = UT_LIST_GET_PREV(trx_locks, lock);
	}

unknown's avatar
unknown committed
3724
	ut_a(trx->n_lock_table_exp == 0);
3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735
}

/*************************************************************************
Cancels a waiting lock request and releases possible other transactions
waiting behind it. */

void
lock_cancel_waiting_and_release(
/*============================*/
	lock_t*	lock)	/* in: waiting lock request */
{
3736
#ifdef UNIV_SYNC_DEBUG
3737
	ut_ad(mutex_own(&kernel_mutex));
3738
#endif /* UNIV_SYNC_DEBUG */
3739 3740 3741 3742 3743

	if (lock_get_type(lock) == LOCK_REC) {
			
		lock_rec_dequeue_from_page(lock);
	} else {
3744
		ut_ad(lock_get_type(lock) & LOCK_TABLE);
3745 3746 3747 3748 3749 3750 3751 3752 3753 3754 3755

		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);
3756 3757
}

unknown's avatar
unknown committed
3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770
/*************************************************************************
Resets all record and table locks of a transaction on a table to be dropped.
No lock is allowed to be a wait lock. */
static
void
lock_reset_all_on_table_for_trx(
/*============================*/
	dict_table_t*	table,	/* in: table to be dropped */
	trx_t*		trx)	/* in: a transaction */
{
	lock_t*	lock;
	lock_t*	prev_lock;

3771
#ifdef UNIV_SYNC_DEBUG
unknown's avatar
unknown committed
3772
	ut_ad(mutex_own(&kernel_mutex));
3773
#endif /* UNIV_SYNC_DEBUG */
unknown's avatar
unknown committed
3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784

	lock = UT_LIST_GET_LAST(trx->trx_locks);
	
	while (lock != NULL) {
		prev_lock = UT_LIST_GET_PREV(trx_locks, lock);
		
		if (lock_get_type(lock) == LOCK_REC
				&& lock->index->table == table) {
			ut_a(!lock_get_wait(lock));
			
			lock_rec_discard(lock);
3785
		} else if (lock_get_type(lock) & LOCK_TABLE
unknown's avatar
unknown committed
3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822
				&& lock->un_member.tab_lock.table == table) {

			ut_a(!lock_get_wait(lock));
			
			lock_table_remove_low(lock);
		}

		lock = prev_lock;
	}
}

/*************************************************************************
Resets all locks, both table and record locks, on a table to be dropped.
No lock is allowed to be a wait lock. */

void
lock_reset_all_on_table(
/*====================*/
	dict_table_t*	table)	/* in: table to be dropped */
{
	lock_t*	lock;

	mutex_enter(&kernel_mutex);

	lock = UT_LIST_GET_FIRST(table->locks);

	while (lock) {	
		ut_a(!lock_get_wait(lock));

		lock_reset_all_on_table_for_trx(table, lock->trx);

		lock = UT_LIST_GET_FIRST(table->locks);
	}

	mutex_exit(&kernel_mutex);
}

3823 3824 3825 3826 3827 3828 3829 3830
/*===================== VALIDATION AND DEBUGGING  ====================*/

/*************************************************************************
Prints info of a table lock. */

void
lock_table_print(
/*=============*/
3831
	FILE*	file,	/* in: file where to print */
3832 3833
	lock_t*	lock)	/* in: table type lock */
{
3834
#ifdef UNIV_SYNC_DEBUG
3835
	ut_ad(mutex_own(&kernel_mutex));
3836
#endif /* UNIV_SYNC_DEBUG */
3837 3838
	ut_a(lock_get_type(lock) == LOCK_TABLE ||
		lock_get_type(lock) == LOCK_TABLE_EXP);
3839

3840 3841 3842
	if (lock_get_type(lock) == LOCK_TABLE_EXP) {
		fputs("EXPLICIT ", file);
	}
3843 3844 3845
	fputs("TABLE LOCK table ", file);
	ut_print_name(file, lock->un_member.tab_lock.table->name);
	fprintf(file, " trx id %lu %lu",
3846
		(ulong) (lock->trx)->id.high, (ulong) (lock->trx)->id.low);
3847 3848

	if (lock_get_mode(lock) == LOCK_S) {
3849
		fputs(" lock mode S", file);
3850
	} else if (lock_get_mode(lock) == LOCK_X) {
3851
		fputs(" lock mode X", file);
3852
	} else if (lock_get_mode(lock) == LOCK_IS) {
3853
		fputs(" lock mode IS", file);
3854
	} else if (lock_get_mode(lock) == LOCK_IX) {
3855
		fputs(" lock mode IX", file);
3856
	} else if (lock_get_mode(lock) == LOCK_AUTO_INC) {
3857
		fputs(" lock mode AUTO-INC", file);
3858
	} else {
unknown's avatar
unknown committed
3859
		fprintf(file, " unknown lock mode %lu", (ulong) lock_get_mode(lock));
3860 3861 3862
	}

	if (lock_get_wait(lock)) {
3863
		fputs(" waiting", file);
3864 3865
	}

3866
	putc('\n', file);
3867 3868 3869 3870 3871 3872 3873 3874
}						
				
/*************************************************************************
Prints info of a record lock. */

void
lock_rec_print(
/*===========*/
3875
	FILE*	file,	/* in: file where to print */
3876 3877 3878 3879 3880 3881 3882 3883
	lock_t*	lock)	/* in: record type lock */
{
	page_t*	page;
	ulint	space;
	ulint	page_no;
	ulint	i;
	mtr_t	mtr;

3884
#ifdef UNIV_SYNC_DEBUG
3885
	ut_ad(mutex_own(&kernel_mutex));
3886
#endif /* UNIV_SYNC_DEBUG */
3887 3888 3889 3890 3891
	ut_a(lock_get_type(lock) == LOCK_REC);

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

3892
	fprintf(file, "RECORD LOCKS space id %lu page no %lu n bits %lu ",
3893 3894
		       (ulong) space, (ulong) page_no,
		       (ulong) lock_rec_get_n_bits(lock));
3895 3896
	dict_index_name_print(file, lock->index);
	fprintf(file, " trx id %lu %lu",
3897 3898
		       (ulong) (lock->trx)->id.high,
		       (ulong) (lock->trx)->id.low);
3899 3900

	if (lock_get_mode(lock) == LOCK_S) {
3901
		fputs(" lock mode S", file);
3902
	} else if (lock_get_mode(lock) == LOCK_X) {
3903
		fputs(" lock_mode X", file);
3904 3905 3906 3907 3908
	} else {
		ut_error;
	}

	if (lock_rec_get_gap(lock)) {
3909
		fputs(" locks gap before rec", file);
unknown's avatar
unknown committed
3910 3911 3912
	}

	if (lock_rec_get_rec_not_gap(lock)) {
3913
		fputs(" locks rec but not gap", file);
3914 3915
	}

unknown's avatar
unknown committed
3916
	if (lock_rec_get_insert_intention(lock)) {
3917
		fputs(" insert intention", file);
unknown's avatar
unknown committed
3918 3919
	}

3920
	if (lock_get_wait(lock)) {
3921
		fputs(" waiting", file);
3922 3923 3924 3925
	}

	mtr_start(&mtr);

3926
	putc('\n', file);
3927

3928 3929 3930 3931 3932 3933
	/* If the page is not in the buffer pool, we cannot load it
	because we have the kernel mutex and ibuf operations would
	break the latching order */
	
	page = buf_page_get_gen(space, page_no, RW_NO_LATCH,
					NULL, BUF_GET_IF_IN_POOL,
3934
					__FILE__, __LINE__, &mtr);
3935 3936
	if (page) {
		page = buf_page_get_nowait(space, page_no, RW_S_LATCH, &mtr);
unknown's avatar
unknown committed
3937 3938 3939 3940 3941 3942 3943 3944 3945

		if (!page) {
			/* Let us try to get an X-latch. If the current thread
			is holding an X-latch on the page, we cannot get an
			S-latch. */
			
			page = buf_page_get_nowait(space, page_no, RW_X_LATCH,
									&mtr);
		}
3946 3947 3948
	}
				
	if (page) {
3949
#ifdef UNIV_SYNC_DEBUG
3950
		buf_page_dbg_add_level(page, SYNC_NO_ORDER_CHECK);
3951
#endif /* UNIV_SYNC_DEBUG */
3952 3953 3954 3955 3956 3957
	}

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

		if (lock_rec_get_nth_bit(lock, i)) {

unknown's avatar
unknown committed
3958
			fprintf(file, "Record lock, heap no %lu ", (ulong) i);
3959 3960

			if (page) {
3961
				rec_print(file,
3962
				      page_find_rec_with_heap_no(page, i));
3963 3964
			}

3965
			putc('\n', file);
3966
		}
3967
	}
unknown's avatar
unknown committed
3968

3969 3970 3971 3972 3973 3974 3975 3976 3977 3978 3979 3980 3981 3982
	mtr_commit(&mtr);
}						
				
/*************************************************************************
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;

3983
#ifdef UNIV_SYNC_DEBUG
3984
	ut_ad(mutex_own(&kernel_mutex));
3985
#endif /* UNIV_SYNC_DEBUG */
3986 3987 3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004

	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);
}
	
/*************************************************************************
Prints info of locks for all transactions. */

void
unknown's avatar
unknown committed
4005 4006
lock_print_info(
/*============*/
4007
	FILE*	file)	/* in: file where to print */
4008 4009 4010 4011 4012 4013 4014 4015 4016 4017 4018
{
	lock_t*	lock;
	trx_t*	trx;
	ulint	space;
	ulint	page_no;
	page_t*	page;
	ibool	load_page_first = TRUE;
	ulint	nth_trx		= 0;
	ulint	nth_lock	= 0;
	ulint	i;
	mtr_t	mtr;
4019

4020 4021 4022 4023 4024
	/* 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();
unknown's avatar
Merge  
unknown committed
4025
	lock_mutex_enter_kernel();
4026

unknown's avatar
unknown committed
4027
	if (lock_deadlock_found) {
4028
		fputs(
unknown's avatar
unknown committed
4029 4030
"------------------------\n" 
"LATEST DETECTED DEADLOCK\n"
4031
"------------------------\n", file);
unknown's avatar
unknown committed
4032

4033
		ut_copy_file(file, lock_latest_err_file);
unknown's avatar
unknown committed
4034 4035
	}

4036
	fputs(
unknown's avatar
unknown committed
4037 4038
"------------\n" 
"TRANSACTIONS\n"
4039
"------------\n", file);
unknown's avatar
unknown committed
4040

4041
	fprintf(file, "Trx id counter %lu %lu\n",
4042 4043
		       (ulong) ut_dulint_get_high(trx_sys->max_trx_id),
		       (ulong) ut_dulint_get_low(trx_sys->max_trx_id));
unknown's avatar
unknown committed
4044

4045
	fprintf(file,
unknown's avatar
unknown committed
4046
	"Purge done for trx's n:o < %lu %lu undo n:o < %lu %lu\n",
4047 4048 4049 4050
		(ulong) ut_dulint_get_high(purge_sys->purge_trx_no),
		(ulong) ut_dulint_get_low(purge_sys->purge_trx_no),
		(ulong) ut_dulint_get_high(purge_sys->purge_undo_no),
		(ulong) ut_dulint_get_low(purge_sys->purge_undo_no));
unknown's avatar
unknown committed
4051

4052
	fprintf(file,
unknown's avatar
unknown committed
4053
		"Total number of lock structs in row lock hash table %lu\n",
4054
					 (ulong) lock_get_n_rec_locks());
unknown's avatar
unknown committed
4055

4056
	fprintf(file, "LIST OF TRANSACTIONS FOR EACH SESSION:\n");
4057 4058 4059 4060 4061 4062 4063

	/* 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) {
4064 4065
			fputs("---", file);
			trx_print(file, trx);
4066 4067 4068 4069 4070
		}
			
		trx = UT_LIST_GET_NEXT(mysql_trx_list, trx);
	}

4071 4072 4073 4074
loop:
	trx = UT_LIST_GET_FIRST(trx_sys->trx_list);

	i = 0;
4075 4076 4077 4078 4079

	/* 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. */
4080 4081 4082 4083 4084 4085 4086 4087
	
	while (trx && (i < nth_trx)) {
		trx = UT_LIST_GET_NEXT(trx_list, trx);
		i++;
	}

	if (trx == NULL) {
		lock_mutex_exit_kernel();
4088
		innobase_mysql_end_print_arbitrary_thd();
4089

4090
		ut_ad(lock_validate());
unknown's avatar
unknown committed
4091

unknown's avatar
unknown committed
4092 4093 4094
		return;
	}

4095
	if (nth_lock == 0) {
4096 4097
		fputs("---", file);
		trx_print(file, trx);
unknown's avatar
unknown committed
4098
		
4099
	        if (trx->read_view) {
4100
			fprintf(file,
4101
       "Trx read view will not see trx with id >= %lu %lu, sees < %lu %lu\n",
4102 4103 4104 4105
		      (ulong) ut_dulint_get_high(trx->read_view->low_limit_id),
       		      (ulong) ut_dulint_get_low(trx->read_view->low_limit_id),
       		      (ulong) ut_dulint_get_high(trx->read_view->up_limit_id),
       		      (ulong) ut_dulint_get_low(trx->read_view->up_limit_id));
4106 4107
	        }

4108
		if (trx->que_state == TRX_QUE_LOCK_WAIT) {
4109
			fprintf(file,
unknown's avatar
unknown committed
4110
 "------- TRX HAS BEEN WAITING %lu SEC FOR THIS LOCK TO BE GRANTED:\n",
4111
		   (ulong)difftime(time(NULL), trx->wait_started));
4112 4113

			if (lock_get_type(trx->wait_lock) == LOCK_REC) {
4114
				lock_rec_print(file, trx->wait_lock);
4115
			} else {
4116
				lock_table_print(file, trx->wait_lock);
4117 4118
			}

4119
			fputs("------------------\n", file);
4120
		}
4121 4122
	}

4123 4124 4125 4126 4127
	if (!srv_print_innodb_lock_monitor) {
	  	nth_trx++;
	  	goto loop;
	}

4128 4129
	i = 0;

4130 4131 4132
	/* Look at the note about the trx loop above why we loop here:
	lock may be an obsolete pointer now. */
	
4133 4134 4135 4136 4137 4138 4139 4140 4141 4142 4143 4144 4145 4146 4147 4148 4149 4150 4151 4152
	lock = UT_LIST_GET_FIRST(trx->trx_locks);
		
	while (lock && (i < nth_lock)) {
		lock = UT_LIST_GET_NEXT(trx_locks, lock);
		i++;
	}

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

		goto loop;
	}

	if (lock_get_type(lock) == LOCK_REC) {
		space = lock->un_member.rec_lock.space;
 		page_no = lock->un_member.rec_lock.page_no;

 		if (load_page_first) {
			lock_mutex_exit_kernel();
4153
			innobase_mysql_end_print_arbitrary_thd();
4154 4155 4156 4157 4158 4159 4160 4161 4162

			mtr_start(&mtr);
			
			page = buf_page_get_with_no_latch(space, page_no, &mtr);

			mtr_commit(&mtr);

			load_page_first = FALSE;

4163
			innobase_mysql_prepare_print_arbitrary_thd();
4164 4165 4166 4167 4168
			lock_mutex_enter_kernel();

			goto loop;
		}
		
4169
		lock_rec_print(file, lock);
4170
	} else {
4171
		ut_ad(lock_get_type(lock) & LOCK_TABLE);
4172
	
4173
		lock_table_print(file, lock);
4174 4175 4176 4177 4178 4179
	}

	load_page_first = TRUE;

	nth_lock++;

4180
	if (nth_lock >= 10) {
4181 4182 4183
		fputs(
		"10 LOCKS PRINTED FOR THIS TRX: SUPPRESSING FURTHER PRINTS\n",
			file);
4184 4185 4186 4187 4188 4189 4190
	
		nth_trx++;
		nth_lock = 0;

		goto loop;
	}

4191 4192 4193 4194 4195 4196 4197 4198 4199 4200 4201 4202 4203 4204 4205
	goto loop;
}

/*************************************************************************
Validates the lock queue on a table. */

ibool
lock_table_queue_validate(
/*======================*/
				/* out: TRUE if ok */
	dict_table_t*	table)	/* in: table */
{
	lock_t*	lock;
	ibool	is_waiting;

4206
#ifdef UNIV_SYNC_DEBUG
4207
	ut_ad(mutex_own(&kernel_mutex));
4208
#endif /* UNIV_SYNC_DEBUG */
4209 4210 4211 4212 4213 4214 4215 4216 4217 4218 4219 4220 4221 4222 4223 4224 4225 4226 4227 4228 4229 4230 4231 4232 4233 4234 4235 4236 4237 4238 4239 4240 4241 4242 4243 4244 4245 4246 4247 4248 4249 4250 4251 4252 4253 4254 4255 4256 4257 4258 4259

	is_waiting = FALSE;

	lock = UT_LIST_GET_FIRST(table->locks);

	while (lock) {
		ut_a(((lock->trx)->conc_state == TRX_ACTIVE)
		     || ((lock->trx)->conc_state == TRX_COMMITTED_IN_MEMORY));
	
		if (!lock_get_wait(lock)) {

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

			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. */

ibool
lock_rec_queue_validate(
/*====================*/
				/* out: TRUE if ok */
	rec_t*		rec,	/* in: record to look at */
	dict_index_t*	index)	/* in: index, or NULL if not known */
{
	trx_t*	impl_trx;	
	lock_t*	lock;
	
	ut_a(rec);

	lock_mutex_enter_kernel();

	if (page_rec_is_supremum(rec) || page_rec_is_infimum(rec)) {

		lock = lock_rec_get_first(rec);

		while (lock) {
			ut_a(lock->trx->conc_state == TRX_ACTIVE
		     	     || lock->trx->conc_state
unknown's avatar
unknown committed
4260
						== TRX_COMMITTED_IN_MEMORY);
4261 4262 4263 4264 4265 4266 4267 4268 4269 4270 4271 4272 4273 4274 4275 4276 4277 4278 4279
	
			ut_a(trx_in_trx_list(lock->trx));
			
			if (lock_get_wait(lock)) {
				ut_a(lock_rec_has_to_wait_in_queue(lock));
			}

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

			lock = lock_rec_get_next(rec, lock);
		}

		lock_mutex_exit_kernel();

	    	return(TRUE);
	}

unknown's avatar
unknown committed
4280
	if (index && (index->type & DICT_CLUSTERED)) {
4281 4282 4283 4284
	
		impl_trx = lock_clust_rec_some_has_impl(rec, index);

		if (impl_trx && lock_rec_other_has_expl_req(LOCK_S, 0,
unknown's avatar
unknown committed
4285
				LOCK_WAIT, rec, impl_trx)) {
4286

unknown's avatar
unknown committed
4287 4288
			ut_a(lock_rec_has_expl(LOCK_X | LOCK_REC_NOT_GAP, rec,
								impl_trx));
4289 4290 4291 4292 4293 4294 4295 4296 4297 4298 4299 4300
		}
	}

	if (index && !(index->type & DICT_CLUSTERED)) {
		
		/* The kernel mutex may get released temporarily in the
		next function call: we have to release lock table mutex
		to obey the latching order */
		
		impl_trx = lock_sec_rec_some_has_impl_off_kernel(rec, index);

		if (impl_trx && lock_rec_other_has_expl_req(LOCK_S, 0,
unknown's avatar
unknown committed
4301
				LOCK_WAIT, rec, impl_trx)) {
4302

unknown's avatar
unknown committed
4303 4304
			ut_a(lock_rec_has_expl(LOCK_X | LOCK_REC_NOT_GAP, rec,
								impl_trx));
4305 4306 4307 4308 4309 4310 4311 4312 4313 4314 4315 4316 4317 4318 4319 4320 4321 4322
		}
	}

	lock = lock_rec_get_first(rec);

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

		if (!lock_rec_get_gap(lock) && !lock_get_wait(lock)) {
		
			if (lock_get_mode(lock) == LOCK_S) {
				ut_a(!lock_rec_other_has_expl_req(LOCK_X,
unknown's avatar
unknown committed
4323
						0, 0, rec, lock->trx));
4324 4325
			} else {
				ut_a(!lock_rec_other_has_expl_req(LOCK_S,
unknown's avatar
unknown committed
4326
						0, 0, rec, lock->trx));
4327 4328 4329 4330 4331 4332 4333 4334 4335 4336 4337 4338 4339 4340 4341 4342 4343 4344 4345 4346 4347 4348 4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359 4360
			}

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

			ut_a(lock_rec_has_to_wait_in_queue(lock));
		}

		lock = lock_rec_get_next(rec, lock);
	}

	lock_mutex_exit_kernel();

	return(TRUE);
}

/*************************************************************************
Validates the record lock queues on a page. */

ibool
lock_rec_validate_page(
/*===================*/
			/* out: TRUE if ok */
	ulint	space,	/* in: space id */
	ulint	page_no)/* in: page number */
{
	dict_index_t*	index;
	page_t*	page;
	lock_t*	lock;
	rec_t*	rec;
	ulint	nth_lock	= 0;
	ulint	nth_bit		= 0;
	ulint	i;
	mtr_t	mtr;

4361
#ifdef UNIV_SYNC_DEBUG
4362
	ut_ad(!mutex_own(&kernel_mutex));
4363
#endif /* UNIV_SYNC_DEBUG */
4364 4365 4366 4367

	mtr_start(&mtr);
	
	page = buf_page_get(space, page_no, RW_X_LATCH, &mtr);
4368
#ifdef UNIV_SYNC_DEBUG
4369
	buf_page_dbg_add_level(page, SYNC_NO_ORDER_CHECK);
4370
#endif /* UNIV_SYNC_DEBUG */
4371 4372 4373 4374 4375 4376 4377 4378 4379 4380 4381 4382 4383 4384 4385 4386 4387 4388 4389

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

	if (!lock) {
		goto function_exit;
	}

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

		lock = lock_rec_get_next_on_page(lock);

		if (!lock) {
			goto function_exit;
		}
	}

	ut_a(trx_in_trx_list(lock->trx));
unknown's avatar
unknown committed
4390 4391
	ut_a(lock->trx->conc_state == TRX_ACTIVE
		     || lock->trx->conc_state == TRX_COMMITTED_IN_MEMORY);
4392 4393 4394
	
	for (i = nth_bit; i < lock_rec_get_n_bits(lock); i++) {

unknown's avatar
unknown committed
4395
		if (i == 1 || lock_rec_get_nth_bit(lock, i)) {
4396 4397 4398 4399

			index = lock->index;
			rec = page_find_rec_with_heap_no(page, i);

4400
			fprintf(stderr,
unknown's avatar
unknown committed
4401
				"Validating %lu %lu\n", (ulong) space, (ulong) page_no);
4402 4403 4404 4405 4406 4407 4408 4409 4410 4411 4412 4413 4414 4415 4416 4417 4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430 4431 4432 4433 4434 4435 4436 4437 4438 4439 4440 4441 4442 4443 4444 4445 4446 4447 4448 4449 4450

			lock_mutex_exit_kernel();

			lock_rec_queue_validate(rec, index);

			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);

	return(TRUE);
}						
				
/*************************************************************************
Validates the lock system. */

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

	while (trx) {
		lock = UT_LIST_GET_FIRST(trx->trx_locks);
		
		while (lock) {
4451
			if (lock_get_type(lock) & LOCK_TABLE) {
4452 4453 4454 4455 4456 4457 4458 4459 4460 4461 4462 4463 4464 4465 4466 4467 4468 4469 4470 4471 4472 4473 4474 4475 4476 4477 4478 4479 4480 4481 4482 4483 4484 4485 4486 4487 4488 4489 4490 4491 4492 4493 4494 4495 4496 4497 4498 4499 4500 4501 4502 4503 4504 4505 4506 4507 4508 4509 4510 4511 4512 4513 4514 4515 4516 4517 4518 4519 4520 4521 4522 4523 4524 4525 4526 4527 4528 4529 4530 4531 4532 4533 4534 4535 4536 4537 4538 4539 4540 4541 4542 4543 4544 4545 4546 4547 4548 4549 4550 4551 4552 4553 4554 4555 4556 4557 4558 4559 4560 4561 4562 4563 4564 4565 4566 4567
	
				lock_table_queue_validate(
					lock->un_member.tab_lock.table);
			}
	
			lock = UT_LIST_GET_NEXT(trx_locks, lock);
		}
	
		trx = UT_LIST_GET_NEXT(trx_list, trx);
	}

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

		limit = ut_dulint_zero;

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

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

				space = lock->un_member.rec_lock.space;
				page_no = lock->un_member.rec_lock.page_no;
		
				if (ut_dulint_cmp(
					ut_dulint_create(space, page_no),
							limit) >= 0) {
					break;
				}

				lock = HASH_GET_NEXT(hash, lock);
			}

			if (!lock) {

				break;
			}
			
			lock_mutex_exit_kernel();

			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);
}

/*============ 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. */

ulint
lock_rec_insert_check_and_lock(
/*===========================*/
				/* 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 */
	rec_t*		rec,	/* in: record after which to insert */
	dict_index_t*	index,	/* in: index */
	que_thr_t*	thr,	/* in: query thread */
	ibool*		inherit)/* out: set to TRUE if the new inserted
				record maybe should inherit LOCK_GAP type
				locks from the successor record */
{
	rec_t*	next_rec;
	trx_t*	trx;
	lock_t*	lock;
	ulint	err;

	if (flags & BTR_NO_LOCKING_FLAG) {

		return(DB_SUCCESS);
	}

	ut_ad(rec);

	trx = thr_get_trx(thr);
	next_rec = page_rec_get_next(rec);

	*inherit = FALSE;

	lock_mutex_enter_kernel();

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

	lock = lock_rec_get_first(next_rec);

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

		lock_mutex_exit_kernel();

		if (!(index->type & DICT_CLUSTERED)) {

			/* Update the page max trx id field */
			page_update_max_trx_id(buf_frame_align(rec),
							thr_get_trx(thr)->id);
		}
		
		return(DB_SUCCESS);
	}

	*inherit = TRUE;

unknown's avatar
unknown committed
4568 4569
	/* If another transaction has an explicit lock request which locks
	the gap, waiting or granted, on the successor, the insert has to wait.
unknown's avatar
unknown committed
4570 4571 4572 4573 4574 4575 4576 4577

	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. */

unknown's avatar
unknown committed
4578 4579 4580 4581
	if (lock_rec_other_has_conflicting(LOCK_X | LOCK_GAP
				| LOCK_INSERT_INTENTION, next_rec, trx)) {

		/* Note that we may get DB_SUCCESS also here! */
unknown's avatar
unknown committed
4582
		err = lock_rec_enqueue_waiting(LOCK_X | LOCK_GAP
unknown's avatar
unknown committed
4583 4584
						| LOCK_INSERT_INTENTION,
						next_rec, index, thr);
4585 4586 4587 4588 4589 4590 4591 4592 4593 4594 4595 4596 4597 4598 4599 4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613 4614 4615
	} else {
		err = DB_SUCCESS;
	}

	lock_mutex_exit_kernel();

	if (!(index->type & DICT_CLUSTERED) && (err == DB_SUCCESS)) {

		/* Update the page max trx id field */
		page_update_max_trx_id(buf_frame_align(rec),
							thr_get_trx(thr)->id);
	}
	
	ut_ad(lock_rec_queue_validate(next_rec, index));

	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(
/*==========================*/
	rec_t*		rec,	/* in: user record on page */
	dict_index_t*	index)	/* in: index of record */
{
	trx_t*	impl_trx;

4616
#ifdef UNIV_SYNC_DEBUG
4617
	ut_ad(mutex_own(&kernel_mutex));
4618
#endif /* UNIV_SYNC_DEBUG */
4619 4620 4621 4622 4623 4624 4625 4626 4627 4628 4629 4630
	ut_ad(page_rec_is_user_rec(rec));

	if (index->type & DICT_CLUSTERED) {
		impl_trx = lock_clust_rec_some_has_impl(rec, index);
	} else {
		impl_trx = lock_sec_rec_some_has_impl_off_kernel(rec, index);
	}

	if (impl_trx) {
		/* If the transaction has no explicit x-lock set on the
		record, set one for it */

unknown's avatar
unknown committed
4631 4632
		if (!lock_rec_has_expl(LOCK_X | LOCK_REC_NOT_GAP, rec,
								impl_trx)) {
4633

unknown's avatar
unknown committed
4634 4635
			lock_rec_add_to_queue(LOCK_REC | LOCK_X
					      | LOCK_REC_NOT_GAP, rec, index,
4636 4637 4638 4639 4640 4641 4642 4643 4644 4645 4646 4647 4648 4649 4650 4651 4652 4653 4654 4655 4656 4657 4658 4659 4660 4661 4662 4663 4664 4665 4666 4667 4668 4669 4670 4671 4672 4673 4674 4675 4676 4677
								impl_trx);
		}
	}
}

/*************************************************************************
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. */

ulint
lock_clust_rec_modify_check_and_lock(
/*=================================*/
				/* 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 */
	rec_t*		rec,	/* in: record which should be modified */
	dict_index_t*	index,	/* in: clustered index */
	que_thr_t*	thr)	/* in: query thread */
{
	ulint	err;
	
	if (flags & BTR_NO_LOCKING_FLAG) {

		return(DB_SUCCESS);
	}

	ut_ad(index->type & DICT_CLUSTERED);

	lock_mutex_enter_kernel();

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

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

	lock_rec_convert_impl_to_expl(rec, index);

unknown's avatar
unknown committed
4678
	err = lock_rec_lock(TRUE, LOCK_X | LOCK_REC_NOT_GAP, rec, index, thr);
4679 4680 4681 4682 4683 4684 4685 4686 4687 4688 4689 4690 4691 4692 4693 4694 4695 4696 4697 4698 4699 4700 4701 4702 4703 4704 4705 4706 4707 4708 4709 4710 4711 4712 4713 4714 4715 4716 4717 4718 4719 4720 4721 4722

	lock_mutex_exit_kernel();

	ut_ad(lock_rec_queue_validate(rec, index));

	return(err);
}

/*************************************************************************
Checks if locks of other transactions prevent an immediate modify (delete
mark or delete unmark) of a secondary index record. */

ulint
lock_sec_rec_modify_check_and_lock(
/*===============================*/
				/* 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 */
	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 */
	dict_index_t*	index,	/* in: secondary index */
	que_thr_t*	thr)	/* in: query thread */
{
	ulint	err;
	
	if (flags & BTR_NO_LOCKING_FLAG) {

		return(DB_SUCCESS);
	}

	ut_ad(!(index->type & DICT_CLUSTERED));

	/* 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));

unknown's avatar
unknown committed
4723
	err = lock_rec_lock(TRUE, LOCK_X | LOCK_REC_NOT_GAP, rec, index, thr);
4724 4725 4726 4727 4728 4729 4730 4731 4732 4733 4734 4735 4736 4737 4738 4739 4740 4741 4742 4743 4744 4745 4746 4747 4748 4749 4750 4751 4752 4753 4754 4755 4756

	lock_mutex_exit_kernel();
	
	ut_ad(lock_rec_queue_validate(rec, index));

	if (err == DB_SUCCESS) {
		/* Update the page max trx id field */

		page_update_max_trx_id(buf_frame_align(rec),
							thr_get_trx(thr)->id);
	}

	return(err);
}

/*************************************************************************
Like the counterpart for a clustered index below, but now we read a
secondary index record. */

ulint
lock_sec_rec_read_check_and_lock(
/*=============================*/
				/* 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 */
	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: secondary index */
	ulint		mode,	/* in: mode of the lock which the read cursor
				should set on records: LOCK_S or LOCK_X; the
				latter is possible in SELECT FOR UPDATE */
unknown's avatar
unknown committed
4757 4758
	ulint		gap_mode,/* in: LOCK_ORDINARY, LOCK_GAP, or
				LOCK_REC_NOT_GAP */
4759 4760 4761 4762 4763 4764 4765 4766 4767 4768 4769 4770 4771 4772
	que_thr_t*	thr)	/* in: query thread */
{
	ulint	err;

	ut_ad(!(index->type & DICT_CLUSTERED));
	ut_ad(page_rec_is_user_rec(rec) || page_rec_is_supremum(rec));

	if (flags & BTR_NO_LOCKING_FLAG) {

		return(DB_SUCCESS);
	}

	lock_mutex_enter_kernel();

unknown's avatar
unknown committed
4773
	ut_ad(mode != LOCK_X
4774
	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IX));
unknown's avatar
unknown committed
4775
	ut_ad(mode != LOCK_S
4776 4777 4778 4779 4780 4781 4782 4783 4784 4785 4786 4787 4788 4789
	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IS));

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

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

 		lock_rec_convert_impl_to_expl(rec, index);
	}

unknown's avatar
unknown committed
4790
	err = lock_rec_lock(FALSE, mode | gap_mode, rec, index, thr);
4791 4792 4793 4794 4795 4796 4797 4798 4799 4800 4801 4802 4803 4804 4805 4806 4807 4808 4809 4810 4811 4812 4813 4814 4815 4816 4817 4818 4819 4820

	lock_mutex_exit_kernel();

	ut_ad(lock_rec_queue_validate(rec, index));

	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. */

ulint
lock_clust_rec_read_check_and_lock(
/*===============================*/
				/* 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 */
	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 */
	ulint		mode,	/* in: mode of the lock which the read cursor
				should set on records: LOCK_S or LOCK_X; the
				latter is possible in SELECT FOR UPDATE */
unknown's avatar
unknown committed
4821 4822
	ulint		gap_mode,/* in: LOCK_ORDINARY, LOCK_GAP, or
				LOCK_REC_NOT_GAP */
4823 4824 4825 4826 4827 4828
	que_thr_t*	thr)	/* in: query thread */
{
	ulint	err;

	ut_ad(index->type & DICT_CLUSTERED);
	ut_ad(page_rec_is_user_rec(rec) || page_rec_is_supremum(rec));
unknown's avatar
unknown committed
4829 4830
	ut_ad(gap_mode == LOCK_ORDINARY || gap_mode == LOCK_GAP
					|| gap_mode == LOCK_REC_NOT_GAP);
4831 4832 4833 4834 4835 4836 4837
	if (flags & BTR_NO_LOCKING_FLAG) {

		return(DB_SUCCESS);
	}

	lock_mutex_enter_kernel();

unknown's avatar
unknown committed
4838
	ut_ad(mode != LOCK_X
4839
	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IX));
unknown's avatar
unknown committed
4840
	ut_ad(mode != LOCK_S
4841 4842 4843 4844 4845 4846 4847
	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IS));
	
	if (!page_rec_is_supremum(rec)) {
	      
		lock_rec_convert_impl_to_expl(rec, index);
	}

unknown's avatar
unknown committed
4848
	err = lock_rec_lock(FALSE, mode | gap_mode, rec, index, thr);
4849 4850 4851 4852 4853 4854 4855

	lock_mutex_exit_kernel();

	ut_ad(lock_rec_queue_validate(rec, index));
	
	return(err);
}