Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * predicate.c
4 : * POSTGRES predicate locking
5 : * to support full serializable transaction isolation
6 : *
7 : *
8 : * The approach taken is to implement Serializable Snapshot Isolation (SSI)
9 : * as initially described in this paper:
10 : *
11 : * Michael J. Cahill, Uwe Röhm, and Alan D. Fekete. 2008.
12 : * Serializable isolation for snapshot databases.
13 : * In SIGMOD '08: Proceedings of the 2008 ACM SIGMOD
14 : * international conference on Management of data,
15 : * pages 729-738, New York, NY, USA. ACM.
16 : * http://doi.acm.org/10.1145/1376616.1376690
17 : *
18 : * and further elaborated in Cahill's doctoral thesis:
19 : *
20 : * Michael James Cahill. 2009.
21 : * Serializable Isolation for Snapshot Databases.
22 : * Sydney Digital Theses.
23 : * University of Sydney, School of Information Technologies.
24 : * http://hdl.handle.net/2123/5353
25 : *
26 : *
27 : * Predicate locks for Serializable Snapshot Isolation (SSI) are SIREAD
28 : * locks, which are so different from normal locks that a distinct set of
29 : * structures is required to handle them. They are needed to detect
30 : * rw-conflicts when the read happens before the write. (When the write
31 : * occurs first, the reading transaction can check for a conflict by
32 : * examining the MVCC data.)
33 : *
34 : * (1) Besides tuples actually read, they must cover ranges of tuples
35 : * which would have been read based on the predicate. This will
36 : * require modelling the predicates through locks against database
37 : * objects such as pages, index ranges, or entire tables.
38 : *
39 : * (2) They must be kept in RAM for quick access. Because of this, it
40 : * isn't possible to always maintain tuple-level granularity -- when
41 : * the space allocated to store these approaches exhaustion, a
42 : * request for a lock may need to scan for situations where a single
43 : * transaction holds many fine-grained locks which can be coalesced
44 : * into a single coarser-grained lock.
45 : *
46 : * (3) They never block anything; they are more like flags than locks
47 : * in that regard; although they refer to database objects and are
48 : * used to identify rw-conflicts with normal write locks.
49 : *
50 : * (4) While they are associated with a transaction, they must survive
51 : * a successful COMMIT of that transaction, and remain until all
52 : * overlapping transactions complete. This even means that they
53 : * must survive termination of the transaction's process. If a
54 : * top level transaction is rolled back, however, it is immediately
55 : * flagged so that it can be ignored, and its SIREAD locks can be
56 : * released any time after that.
57 : *
58 : * (5) The only transactions which create SIREAD locks or check for
59 : * conflicts with them are serializable transactions.
60 : *
61 : * (6) When a write lock for a top level transaction is found to cover
62 : * an existing SIREAD lock for the same transaction, the SIREAD lock
63 : * can be deleted.
64 : *
65 : * (7) A write from a serializable transaction must ensure that an xact
66 : * record exists for the transaction, with the same lifespan (until
67 : * all concurrent transaction complete or the transaction is rolled
68 : * back) so that rw-dependencies to that transaction can be
69 : * detected.
70 : *
71 : * We use an optimization for read-only transactions. Under certain
72 : * circumstances, a read-only transaction's snapshot can be shown to
73 : * never have conflicts with other transactions. This is referred to
74 : * as a "safe" snapshot (and one known not to be is "unsafe").
75 : * However, it can't be determined whether a snapshot is safe until
76 : * all concurrent read/write transactions complete.
77 : *
78 : * Once a read-only transaction is known to have a safe snapshot, it
79 : * can release its predicate locks and exempt itself from further
80 : * predicate lock tracking. READ ONLY DEFERRABLE transactions run only
81 : * on safe snapshots, waiting as necessary for one to be available.
82 : *
83 : *
84 : * Lightweight locks to manage access to the predicate locking shared
85 : * memory objects must be taken in this order, and should be released in
86 : * reverse order:
87 : *
88 : * SerializableFinishedListLock
89 : * - Protects the list of transactions which have completed but which
90 : * may yet matter because they overlap still-active transactions.
91 : *
92 : * SerializablePredicateListLock
93 : * - Protects the linked list of locks held by a transaction. Note
94 : * that the locks themselves are also covered by the partition
95 : * locks of their respective lock targets; this lock only affects
96 : * the linked list connecting the locks related to a transaction.
97 : * - All transactions share this single lock (with no partitioning).
98 : * - There is never a need for a process other than the one running
99 : * an active transaction to walk the list of locks held by that
100 : * transaction, except parallel query workers sharing the leader's
101 : * transaction. In the parallel case, an extra per-sxact lock is
102 : * taken; see below.
103 : * - It is relatively infrequent that another process needs to
104 : * modify the list for a transaction, but it does happen for such
105 : * things as index page splits for pages with predicate locks and
106 : * freeing of predicate locked pages by a vacuum process. When
107 : * removing a lock in such cases, the lock itself contains the
108 : * pointers needed to remove it from the list. When adding a
109 : * lock in such cases, the lock can be added using the anchor in
110 : * the transaction structure. Neither requires walking the list.
111 : * - Cleaning up the list for a terminated transaction is sometimes
112 : * not done on a retail basis, in which case no lock is required.
113 : * - Due to the above, a process accessing its active transaction's
114 : * list always uses a shared lock, regardless of whether it is
115 : * walking or maintaining the list. This improves concurrency
116 : * for the common access patterns.
117 : * - A process which needs to alter the list of a transaction other
118 : * than its own active transaction must acquire an exclusive
119 : * lock.
120 : *
121 : * SERIALIZABLEXACT's member 'perXactPredicateListLock'
122 : * - Protects the linked list of predicate locks held by a transaction.
123 : * Only needed for parallel mode, where multiple backends share the
124 : * same SERIALIZABLEXACT object. Not needed if
125 : * SerializablePredicateListLock is held exclusively.
126 : *
127 : * PredicateLockHashPartitionLock(hashcode)
128 : * - The same lock protects a target, all locks on that target, and
129 : * the linked list of locks on the target.
130 : * - When more than one is needed, acquire in ascending address order.
131 : * - When all are needed (rare), acquire in ascending index order with
132 : * PredicateLockHashPartitionLockByIndex(index).
133 : *
134 : * SerializableXactHashLock
135 : * - Protects both PredXact and SerializableXidHash.
136 : *
137 : * SerialControlLock
138 : * - Protects SerialControlData members
139 : *
140 : * SLRU per-bank locks
141 : * - Protects SerialSlruCtl
142 : *
143 : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
144 : * Portions Copyright (c) 1994, Regents of the University of California
145 : *
146 : *
147 : * IDENTIFICATION
148 : * src/backend/storage/lmgr/predicate.c
149 : *
150 : *-------------------------------------------------------------------------
151 : */
152 : /*
153 : * INTERFACE ROUTINES
154 : *
155 : * predicate lock reporting
156 : * GetPredicateLockStatusData(void)
157 : * PageIsPredicateLocked(Relation relation, BlockNumber blkno)
158 : *
159 : * predicate lock maintenance
160 : * GetSerializableTransactionSnapshot(Snapshot snapshot)
161 : * SetSerializableTransactionSnapshot(Snapshot snapshot,
162 : * VirtualTransactionId *sourcevxid)
163 : * RegisterPredicateLockingXid(void)
164 : * PredicateLockRelation(Relation relation, Snapshot snapshot)
165 : * PredicateLockPage(Relation relation, BlockNumber blkno,
166 : * Snapshot snapshot)
167 : * PredicateLockTID(Relation relation, const ItemPointerData *tid, Snapshot snapshot,
168 : * TransactionId tuple_xid)
169 : * PredicateLockPageSplit(Relation relation, BlockNumber oldblkno,
170 : * BlockNumber newblkno)
171 : * PredicateLockPageCombine(Relation relation, BlockNumber oldblkno,
172 : * BlockNumber newblkno)
173 : * TransferPredicateLocksToHeapRelation(Relation relation)
174 : * ReleasePredicateLocks(bool isCommit, bool isReadOnlySafe)
175 : *
176 : * conflict detection (may also trigger rollback)
177 : * CheckForSerializableConflictOut(Relation relation, TransactionId xid,
178 : * Snapshot snapshot)
179 : * CheckForSerializableConflictIn(Relation relation, const ItemPointerData *tid,
180 : * BlockNumber blkno)
181 : * CheckTableForSerializableConflictIn(Relation relation)
182 : *
183 : * final rollback checking
184 : * PreCommit_CheckForSerializationFailure(void)
185 : *
186 : * two-phase commit support
187 : * AtPrepare_PredicateLocks(void);
188 : * PostPrepare_PredicateLocks(TransactionId xid);
189 : * PredicateLockTwoPhaseFinish(FullTransactionId fxid, bool isCommit);
190 : * predicatelock_twophase_recover(FullTransactionId fxid, uint16 info,
191 : * void *recdata, uint32 len);
192 : */
193 :
194 : #include "postgres.h"
195 :
196 : #include "access/parallel.h"
197 : #include "access/slru.h"
198 : #include "access/transam.h"
199 : #include "access/twophase.h"
200 : #include "access/twophase_rmgr.h"
201 : #include "access/xact.h"
202 : #include "access/xlog.h"
203 : #include "miscadmin.h"
204 : #include "pgstat.h"
205 : #include "port/pg_lfind.h"
206 : #include "storage/predicate.h"
207 : #include "storage/predicate_internals.h"
208 : #include "storage/proc.h"
209 : #include "storage/procarray.h"
210 : #include "storage/shmem.h"
211 : #include "storage/subsystems.h"
212 : #include "utils/guc_hooks.h"
213 : #include "utils/rel.h"
214 : #include "utils/snapmgr.h"
215 : #include "utils/wait_event.h"
216 :
217 : /* Uncomment the next line to test the graceful degradation code. */
218 : /* #define TEST_SUMMARIZE_SERIAL */
219 :
220 : /*
221 : * Test the most selective fields first, for performance.
222 : *
223 : * a is covered by b if all of the following hold:
224 : * 1) a.database = b.database
225 : * 2) a.relation = b.relation
226 : * 3) b.offset is invalid (b is page-granularity or higher)
227 : * 4) either of the following:
228 : * 4a) a.offset is valid (a is tuple-granularity) and a.page = b.page
229 : * or 4b) a.offset is invalid and b.page is invalid (a is
230 : * page-granularity and b is relation-granularity
231 : */
232 : #define TargetTagIsCoveredBy(covered_target, covering_target) \
233 : ((GET_PREDICATELOCKTARGETTAG_RELATION(covered_target) == /* (2) */ \
234 : GET_PREDICATELOCKTARGETTAG_RELATION(covering_target)) \
235 : && (GET_PREDICATELOCKTARGETTAG_OFFSET(covering_target) == \
236 : InvalidOffsetNumber) /* (3) */ \
237 : && (((GET_PREDICATELOCKTARGETTAG_OFFSET(covered_target) != \
238 : InvalidOffsetNumber) /* (4a) */ \
239 : && (GET_PREDICATELOCKTARGETTAG_PAGE(covering_target) == \
240 : GET_PREDICATELOCKTARGETTAG_PAGE(covered_target))) \
241 : || ((GET_PREDICATELOCKTARGETTAG_PAGE(covering_target) == \
242 : InvalidBlockNumber) /* (4b) */ \
243 : && (GET_PREDICATELOCKTARGETTAG_PAGE(covered_target) \
244 : != InvalidBlockNumber))) \
245 : && (GET_PREDICATELOCKTARGETTAG_DB(covered_target) == /* (1) */ \
246 : GET_PREDICATELOCKTARGETTAG_DB(covering_target)))
247 :
248 : /*
249 : * The predicate locking target and lock shared hash tables are partitioned to
250 : * reduce contention. To determine which partition a given target belongs to,
251 : * compute the tag's hash code with PredicateLockTargetTagHashCode(), then
252 : * apply one of these macros.
253 : * NB: NUM_PREDICATELOCK_PARTITIONS must be a power of 2!
254 : */
255 : #define PredicateLockHashPartition(hashcode) \
256 : ((hashcode) % NUM_PREDICATELOCK_PARTITIONS)
257 : #define PredicateLockHashPartitionLock(hashcode) \
258 : (&MainLWLockArray[PREDICATELOCK_MANAGER_LWLOCK_OFFSET + \
259 : PredicateLockHashPartition(hashcode)].lock)
260 : #define PredicateLockHashPartitionLockByIndex(i) \
261 : (&MainLWLockArray[PREDICATELOCK_MANAGER_LWLOCK_OFFSET + (i)].lock)
262 :
263 : #define NPREDICATELOCKTARGETENTS() \
264 : mul_size(max_predicate_locks_per_xact, add_size(MaxBackends, max_prepared_xacts))
265 :
266 : #define SxactIsOnFinishedList(sxact) (!dlist_node_is_detached(&(sxact)->finishedLink))
267 :
268 : /*
269 : * Note that a sxact is marked "prepared" once it has passed
270 : * PreCommit_CheckForSerializationFailure, even if it isn't using
271 : * 2PC. This is the point at which it can no longer be aborted.
272 : *
273 : * The PREPARED flag remains set after commit, so SxactIsCommitted
274 : * implies SxactIsPrepared.
275 : */
276 : #define SxactIsCommitted(sxact) (((sxact)->flags & SXACT_FLAG_COMMITTED) != 0)
277 : #define SxactIsPrepared(sxact) (((sxact)->flags & SXACT_FLAG_PREPARED) != 0)
278 : #define SxactIsRolledBack(sxact) (((sxact)->flags & SXACT_FLAG_ROLLED_BACK) != 0)
279 : #define SxactIsDoomed(sxact) (((sxact)->flags & SXACT_FLAG_DOOMED) != 0)
280 : #define SxactIsReadOnly(sxact) (((sxact)->flags & SXACT_FLAG_READ_ONLY) != 0)
281 : #define SxactHasSummaryConflictIn(sxact) (((sxact)->flags & SXACT_FLAG_SUMMARY_CONFLICT_IN) != 0)
282 : #define SxactHasSummaryConflictOut(sxact) (((sxact)->flags & SXACT_FLAG_SUMMARY_CONFLICT_OUT) != 0)
283 : /*
284 : * The following macro actually means that the specified transaction has a
285 : * conflict out *to a transaction which committed ahead of it*. It's hard
286 : * to get that into a name of a reasonable length.
287 : */
288 : #define SxactHasConflictOut(sxact) (((sxact)->flags & SXACT_FLAG_CONFLICT_OUT) != 0)
289 : #define SxactIsDeferrableWaiting(sxact) (((sxact)->flags & SXACT_FLAG_DEFERRABLE_WAITING) != 0)
290 : #define SxactIsROSafe(sxact) (((sxact)->flags & SXACT_FLAG_RO_SAFE) != 0)
291 : #define SxactIsROUnsafe(sxact) (((sxact)->flags & SXACT_FLAG_RO_UNSAFE) != 0)
292 : #define SxactIsPartiallyReleased(sxact) (((sxact)->flags & SXACT_FLAG_PARTIALLY_RELEASED) != 0)
293 :
294 : /*
295 : * Compute the hash code associated with a PREDICATELOCKTARGETTAG.
296 : *
297 : * To avoid unnecessary recomputations of the hash code, we try to do this
298 : * just once per function, and then pass it around as needed. Aside from
299 : * passing the hashcode to hash_search_with_hash_value(), we can extract
300 : * the lock partition number from the hashcode.
301 : */
302 : #define PredicateLockTargetTagHashCode(predicatelocktargettag) \
303 : get_hash_value(PredicateLockTargetHash, predicatelocktargettag)
304 :
305 : /*
306 : * Given a predicate lock tag, and the hash for its target,
307 : * compute the lock hash.
308 : *
309 : * To make the hash code also depend on the transaction, we xor the sxid
310 : * struct's address into the hash code, left-shifted so that the
311 : * partition-number bits don't change. Since this is only a hash, we
312 : * don't care if we lose high-order bits of the address; use an
313 : * intermediate variable to suppress cast-pointer-to-int warnings.
314 : */
315 : #define PredicateLockHashCodeFromTargetHashCode(predicatelocktag, targethash) \
316 : ((targethash) ^ ((uint32) PointerGetDatum((predicatelocktag)->myXact)) \
317 : << LOG2_NUM_PREDICATELOCK_PARTITIONS)
318 :
319 :
320 : /*
321 : * The SLRU buffer area through which we access the old xids.
322 : */
323 : static bool SerialPagePrecedesLogically(int64 page1, int64 page2);
324 : static int serial_errdetail_for_io_error(const void *opaque_data);
325 :
326 : static SlruDesc SerialSlruDesc;
327 :
328 : #define SerialSlruCtl (&SerialSlruDesc)
329 :
330 : #define SERIAL_PAGESIZE BLCKSZ
331 : #define SERIAL_ENTRYSIZE sizeof(SerCommitSeqNo)
332 : #define SERIAL_ENTRIESPERPAGE (SERIAL_PAGESIZE / SERIAL_ENTRYSIZE)
333 :
334 : /*
335 : * Set maximum pages based on the number needed to track all transactions.
336 : */
337 : #define SERIAL_MAX_PAGE (MaxTransactionId / SERIAL_ENTRIESPERPAGE)
338 :
339 : #define SerialNextPage(page) (((page) >= SERIAL_MAX_PAGE) ? 0 : (page) + 1)
340 :
341 : #define SerialValue(slotno, xid) (*((SerCommitSeqNo *) \
342 : (SerialSlruCtl->shared->page_buffer[slotno] + \
343 : ((((uint32) (xid)) % SERIAL_ENTRIESPERPAGE) * SERIAL_ENTRYSIZE))))
344 :
345 : #define SerialPage(xid) (((uint32) (xid)) / SERIAL_ENTRIESPERPAGE)
346 :
347 : typedef struct SerialControlData
348 : {
349 : int64 headPage; /* newest initialized page */
350 : TransactionId headXid; /* newest valid Xid in the SLRU */
351 : TransactionId tailXid; /* oldest xmin we might be interested in */
352 : } SerialControlData;
353 :
354 : typedef struct SerialControlData *SerialControl;
355 :
356 : static SerialControl serialControl;
357 :
358 : /*
359 : * When the oldest committed transaction on the "finished" list is moved to
360 : * SLRU, its predicate locks will be moved to this "dummy" transaction,
361 : * collapsing duplicate targets. When a duplicate is found, the later
362 : * commitSeqNo is used.
363 : */
364 : static SERIALIZABLEXACT *OldCommittedSxact;
365 :
366 :
367 : /*
368 : * These configuration variables are used to set the predicate lock table size
369 : * and to control promotion of predicate locks to coarser granularity in an
370 : * attempt to degrade performance (mostly as false positive serialization
371 : * failure) gracefully in the face of memory pressure.
372 : */
373 : int max_predicate_locks_per_xact; /* in guc_tables.c */
374 : int max_predicate_locks_per_relation; /* in guc_tables.c */
375 : int max_predicate_locks_per_page; /* in guc_tables.c */
376 :
377 : /*
378 : * This provides a list of objects in order to track transactions
379 : * participating in predicate locking. Entries in the list are fixed size,
380 : * and reside in shared memory. The memory address of an entry must remain
381 : * fixed during its lifetime. The list will be protected from concurrent
382 : * update externally; no provision is made in this code to manage that. The
383 : * number of entries in the list, and the size allowed for each entry is
384 : * fixed upon creation.
385 : */
386 : static PredXactList PredXact;
387 :
388 : static void PredicateLockShmemRequest(void *arg);
389 : static void PredicateLockShmemInit(void *arg);
390 : static void PredicateLockShmemAttach(void *arg);
391 :
392 : const ShmemCallbacks PredicateLockShmemCallbacks = {
393 : .request_fn = PredicateLockShmemRequest,
394 : .init_fn = PredicateLockShmemInit,
395 : .attach_fn = PredicateLockShmemAttach,
396 : };
397 :
398 :
399 : /*
400 : * This provides a pool of RWConflict data elements to use in conflict lists
401 : * between transactions.
402 : */
403 : static RWConflictPoolHeader RWConflictPool;
404 :
405 : /*
406 : * The predicate locking hash tables are in shared memory.
407 : * Each backend keeps pointers to them.
408 : */
409 : static HTAB *SerializableXidHash;
410 : static HTAB *PredicateLockTargetHash;
411 : static HTAB *PredicateLockHash;
412 : static dlist_head *FinishedSerializableTransactions;
413 :
414 : /*
415 : * Tag for a dummy entry in PredicateLockTargetHash. By temporarily removing
416 : * this entry, you can ensure that there's enough scratch space available for
417 : * inserting one entry in the hash table. This is an otherwise-invalid tag.
418 : */
419 : static const PREDICATELOCKTARGETTAG ScratchTargetTag = {0, 0, 0, 0};
420 : static uint32 ScratchTargetTagHash;
421 : static LWLock *ScratchPartitionLock;
422 :
423 : /*
424 : * The local hash table used to determine when to combine multiple fine-
425 : * grained locks into a single courser-grained lock.
426 : */
427 : static HTAB *LocalPredicateLockHash = NULL;
428 :
429 : /*
430 : * Keep a pointer to the currently-running serializable transaction (if any)
431 : * for quick reference. Also, remember if we have written anything that could
432 : * cause a rw-conflict.
433 : */
434 : static SERIALIZABLEXACT *MySerializableXact = InvalidSerializableXact;
435 : static bool MyXactDidWrite = false;
436 :
437 : /*
438 : * The SXACT_FLAG_RO_UNSAFE optimization might lead us to release
439 : * MySerializableXact early. If that happens in a parallel query, the leader
440 : * needs to defer the destruction of the SERIALIZABLEXACT until end of
441 : * transaction, because the workers still have a reference to it. In that
442 : * case, the leader stores it here.
443 : */
444 : static SERIALIZABLEXACT *SavedSerializableXact = InvalidSerializableXact;
445 :
446 : static int64 max_serializable_xacts;
447 :
448 : /* local functions */
449 :
450 : static SERIALIZABLEXACT *CreatePredXact(void);
451 : static void ReleasePredXact(SERIALIZABLEXACT *sxact);
452 :
453 : static bool RWConflictExists(const SERIALIZABLEXACT *reader, const SERIALIZABLEXACT *writer);
454 : static void SetRWConflict(SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer);
455 : static void SetPossibleUnsafeConflict(SERIALIZABLEXACT *roXact, SERIALIZABLEXACT *activeXact);
456 : static void ReleaseRWConflict(RWConflict conflict);
457 : static void FlagSxactUnsafe(SERIALIZABLEXACT *sxact);
458 :
459 : static void SerialAdd(TransactionId xid, SerCommitSeqNo minConflictCommitSeqNo);
460 : static SerCommitSeqNo SerialGetMinConflictCommitSeqNo(TransactionId xid);
461 : static void SerialSetActiveSerXmin(TransactionId xid);
462 :
463 : static uint32 predicatelock_hash(const void *key, Size keysize);
464 :
465 : static void SummarizeOldestCommittedSxact(void);
466 : static Snapshot GetSafeSnapshot(Snapshot origSnapshot);
467 : static Snapshot GetSerializableTransactionSnapshotInt(Snapshot snapshot,
468 : VirtualTransactionId *sourcevxid,
469 : int sourcepid);
470 : static bool PredicateLockExists(const PREDICATELOCKTARGETTAG *targettag);
471 : static bool GetParentPredicateLockTag(const PREDICATELOCKTARGETTAG *tag,
472 : PREDICATELOCKTARGETTAG *parent);
473 : static bool CoarserLockCovers(const PREDICATELOCKTARGETTAG *newtargettag);
474 : static void RemoveScratchTarget(bool lockheld);
475 : static void RestoreScratchTarget(bool lockheld);
476 : static void RemoveTargetIfNoLongerUsed(PREDICATELOCKTARGET *target,
477 : uint32 targettaghash);
478 : static void DeleteChildTargetLocks(const PREDICATELOCKTARGETTAG *newtargettag);
479 : static int MaxPredicateChildLocks(const PREDICATELOCKTARGETTAG *tag);
480 : static bool CheckAndPromotePredicateLockRequest(const PREDICATELOCKTARGETTAG *reqtag);
481 : static void DecrementParentLocks(const PREDICATELOCKTARGETTAG *targettag);
482 : static void CreatePredicateLock(const PREDICATELOCKTARGETTAG *targettag,
483 : uint32 targettaghash,
484 : SERIALIZABLEXACT *sxact);
485 : static void DeleteLockTarget(PREDICATELOCKTARGET *target, uint32 targettaghash);
486 : static bool TransferPredicateLocksToNewTarget(PREDICATELOCKTARGETTAG oldtargettag,
487 : PREDICATELOCKTARGETTAG newtargettag,
488 : bool removeOld);
489 : static void PredicateLockAcquire(const PREDICATELOCKTARGETTAG *targettag);
490 : static void DropAllPredicateLocksFromTable(Relation relation,
491 : bool transfer);
492 : static void SetNewSxactGlobalXmin(void);
493 : static void ClearOldPredicateLocks(void);
494 : static void ReleaseOneSerializableXact(SERIALIZABLEXACT *sxact, bool partial,
495 : bool summarize);
496 : static bool XidIsConcurrent(TransactionId xid);
497 : static void CheckTargetForConflictsIn(PREDICATELOCKTARGETTAG *targettag);
498 : static void FlagRWConflict(SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer);
499 : static void OnConflict_CheckForSerializationFailure(const SERIALIZABLEXACT *reader,
500 : SERIALIZABLEXACT *writer);
501 : static void CreateLocalPredicateLockHash(void);
502 : static void ReleasePredicateLocksLocal(void);
503 :
504 :
505 : /*------------------------------------------------------------------------*/
506 :
507 : /*
508 : * Does this relation participate in predicate locking? Temporary and system
509 : * relations are exempt.
510 : */
511 : static inline bool
512 143521 : PredicateLockingNeededForRelation(Relation relation)
513 : {
514 183362 : return !(relation->rd_id < FirstUnpinnedObjectId ||
515 39841 : RelationUsesLocalBuffers(relation));
516 : }
517 :
518 : /*
519 : * When a public interface method is called for a read, this is the test to
520 : * see if we should do a quick return.
521 : *
522 : * Note: this function has side-effects! If this transaction has been flagged
523 : * as RO-safe since the last call, we release all predicate locks and reset
524 : * MySerializableXact. That makes subsequent calls to return quickly.
525 : *
526 : * This is marked as 'inline' to eliminate the function call overhead in the
527 : * common case that serialization is not needed.
528 : */
529 : static inline bool
530 83238796 : SerializationNeededForRead(Relation relation, Snapshot snapshot)
531 : {
532 : /* Nothing to do if this is not a serializable transaction */
533 83238796 : if (MySerializableXact == InvalidSerializableXact)
534 83101430 : return false;
535 :
536 : /*
537 : * Don't acquire locks or conflict when scanning with a special snapshot.
538 : * This excludes things like CLUSTER and REINDEX. They use the wholesale
539 : * functions TransferPredicateLocksToHeapRelation() and
540 : * CheckTableForSerializableConflictIn() to participate in serialization,
541 : * but the scans involved don't need serialization.
542 : */
543 137366 : if (!IsMVCCSnapshot(snapshot))
544 1855 : return false;
545 :
546 : /*
547 : * Check if we have just become "RO-safe". If we have, immediately release
548 : * all locks as they're not needed anymore. This also resets
549 : * MySerializableXact, so that subsequent calls to this function can exit
550 : * quickly.
551 : *
552 : * A transaction is flagged as RO_SAFE if all concurrent R/W transactions
553 : * commit without having conflicts out to an earlier snapshot, thus
554 : * ensuring that no conflicts are possible for this transaction.
555 : */
556 135511 : if (SxactIsROSafe(MySerializableXact))
557 : {
558 33 : ReleasePredicateLocks(false, true);
559 33 : return false;
560 : }
561 :
562 : /* Check if the relation doesn't participate in predicate locking */
563 135478 : if (!PredicateLockingNeededForRelation(relation))
564 100375 : return false;
565 :
566 35103 : return true; /* no excuse to skip predicate locking */
567 : }
568 :
569 : /*
570 : * Like SerializationNeededForRead(), but called on writes.
571 : * The logic is the same, but there is no snapshot and we can't be RO-safe.
572 : */
573 : static inline bool
574 24569247 : SerializationNeededForWrite(Relation relation)
575 : {
576 : /* Nothing to do if this is not a serializable transaction */
577 24569247 : if (MySerializableXact == InvalidSerializableXact)
578 24561285 : return false;
579 :
580 : /* Check if the relation doesn't participate in predicate locking */
581 7962 : if (!PredicateLockingNeededForRelation(relation))
582 3436 : return false;
583 :
584 4526 : return true; /* no excuse to skip predicate locking */
585 : }
586 :
587 :
588 : /*------------------------------------------------------------------------*/
589 :
590 : /*
591 : * These functions are a simple implementation of a list for this specific
592 : * type of struct. If there is ever a generalized shared memory list, we
593 : * should probably switch to that.
594 : */
595 : static SERIALIZABLEXACT *
596 2927 : CreatePredXact(void)
597 : {
598 : SERIALIZABLEXACT *sxact;
599 :
600 2927 : if (dlist_is_empty(&PredXact->availableList))
601 0 : return NULL;
602 :
603 2927 : sxact = dlist_container(SERIALIZABLEXACT, xactLink,
604 : dlist_pop_head_node(&PredXact->availableList));
605 2927 : dlist_push_tail(&PredXact->activeList, &sxact->xactLink);
606 2927 : return sxact;
607 : }
608 :
609 : static void
610 1692 : ReleasePredXact(SERIALIZABLEXACT *sxact)
611 : {
612 : Assert(ShmemAddrIsValid(sxact));
613 :
614 1692 : dlist_delete(&sxact->xactLink);
615 1692 : dlist_push_tail(&PredXact->availableList, &sxact->xactLink);
616 1692 : }
617 :
618 : /*------------------------------------------------------------------------*/
619 :
620 : /*
621 : * These functions manage primitive access to the RWConflict pool and lists.
622 : */
623 : static bool
624 1890 : RWConflictExists(const SERIALIZABLEXACT *reader, const SERIALIZABLEXACT *writer)
625 : {
626 : dlist_iter iter;
627 :
628 : Assert(reader != writer);
629 :
630 : /* Check the ends of the purported conflict first. */
631 1890 : if (SxactIsDoomed(reader)
632 1890 : || SxactIsDoomed(writer)
633 1890 : || dlist_is_empty(&reader->outConflicts)
634 569 : || dlist_is_empty(&writer->inConflicts))
635 1361 : return false;
636 :
637 : /*
638 : * A conflict is possible; walk the list to find out.
639 : *
640 : * The unconstify is needed as we have no const version of
641 : * dlist_foreach().
642 : */
643 545 : dlist_foreach(iter, &unconstify(SERIALIZABLEXACT *, reader)->outConflicts)
644 : {
645 529 : RWConflict conflict =
646 : dlist_container(RWConflictData, outLink, iter.cur);
647 :
648 529 : if (conflict->sxactIn == writer)
649 513 : return true;
650 : }
651 :
652 : /* No conflict found. */
653 16 : return false;
654 : }
655 :
656 : static void
657 792 : SetRWConflict(SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer)
658 : {
659 : RWConflict conflict;
660 :
661 : Assert(reader != writer);
662 : Assert(!RWConflictExists(reader, writer));
663 :
664 792 : if (dlist_is_empty(&RWConflictPool->availableList))
665 0 : ereport(ERROR,
666 : (errcode(ERRCODE_OUT_OF_MEMORY),
667 : errmsg("not enough elements in RWConflictPool to record a read/write conflict"),
668 : errhint("You might need to run fewer transactions at a time or increase \"max_connections\".")));
669 :
670 792 : conflict = dlist_head_element(RWConflictData, outLink, &RWConflictPool->availableList);
671 792 : dlist_delete(&conflict->outLink);
672 :
673 792 : conflict->sxactOut = reader;
674 792 : conflict->sxactIn = writer;
675 792 : dlist_push_tail(&reader->outConflicts, &conflict->outLink);
676 792 : dlist_push_tail(&writer->inConflicts, &conflict->inLink);
677 792 : }
678 :
679 : static void
680 134 : SetPossibleUnsafeConflict(SERIALIZABLEXACT *roXact,
681 : SERIALIZABLEXACT *activeXact)
682 : {
683 : RWConflict conflict;
684 :
685 : Assert(roXact != activeXact);
686 : Assert(SxactIsReadOnly(roXact));
687 : Assert(!SxactIsReadOnly(activeXact));
688 :
689 134 : if (dlist_is_empty(&RWConflictPool->availableList))
690 0 : ereport(ERROR,
691 : (errcode(ERRCODE_OUT_OF_MEMORY),
692 : errmsg("not enough elements in RWConflictPool to record a potential read/write conflict"),
693 : errhint("You might need to run fewer transactions at a time or increase \"max_connections\".")));
694 :
695 134 : conflict = dlist_head_element(RWConflictData, outLink, &RWConflictPool->availableList);
696 134 : dlist_delete(&conflict->outLink);
697 :
698 134 : conflict->sxactOut = activeXact;
699 134 : conflict->sxactIn = roXact;
700 134 : dlist_push_tail(&activeXact->possibleUnsafeConflicts, &conflict->outLink);
701 134 : dlist_push_tail(&roXact->possibleUnsafeConflicts, &conflict->inLink);
702 134 : }
703 :
704 : static void
705 926 : ReleaseRWConflict(RWConflict conflict)
706 : {
707 926 : dlist_delete(&conflict->inLink);
708 926 : dlist_delete(&conflict->outLink);
709 926 : dlist_push_tail(&RWConflictPool->availableList, &conflict->outLink);
710 926 : }
711 :
712 : static void
713 3 : FlagSxactUnsafe(SERIALIZABLEXACT *sxact)
714 : {
715 : dlist_mutable_iter iter;
716 :
717 : Assert(SxactIsReadOnly(sxact));
718 : Assert(!SxactIsROSafe(sxact));
719 :
720 3 : sxact->flags |= SXACT_FLAG_RO_UNSAFE;
721 :
722 : /*
723 : * We know this isn't a safe snapshot, so we can stop looking for other
724 : * potential conflicts.
725 : */
726 6 : dlist_foreach_modify(iter, &sxact->possibleUnsafeConflicts)
727 : {
728 3 : RWConflict conflict =
729 3 : dlist_container(RWConflictData, inLink, iter.cur);
730 :
731 : Assert(!SxactIsReadOnly(conflict->sxactOut));
732 : Assert(sxact == conflict->sxactIn);
733 :
734 3 : ReleaseRWConflict(conflict);
735 : }
736 3 : }
737 :
738 : /*------------------------------------------------------------------------*/
739 :
740 : /*
741 : * Decide whether a Serial page number is "older" for truncation purposes.
742 : * Analogous to CLOGPagePrecedes().
743 : */
744 : static bool
745 0 : SerialPagePrecedesLogically(int64 page1, int64 page2)
746 : {
747 : TransactionId xid1;
748 : TransactionId xid2;
749 :
750 0 : xid1 = ((TransactionId) page1) * SERIAL_ENTRIESPERPAGE;
751 0 : xid1 += FirstNormalTransactionId + 1;
752 0 : xid2 = ((TransactionId) page2) * SERIAL_ENTRIESPERPAGE;
753 0 : xid2 += FirstNormalTransactionId + 1;
754 :
755 0 : return (TransactionIdPrecedes(xid1, xid2) &&
756 0 : TransactionIdPrecedes(xid1, xid2 + SERIAL_ENTRIESPERPAGE - 1));
757 : }
758 :
759 : static int
760 0 : serial_errdetail_for_io_error(const void *opaque_data)
761 : {
762 0 : TransactionId xid = *(const TransactionId *) opaque_data;
763 :
764 0 : return errdetail("Could not access serializable CSN of transaction %u.", xid);
765 : }
766 :
767 : #ifdef USE_ASSERT_CHECKING
768 : static void
769 : SerialPagePrecedesLogicallyUnitTests(void)
770 : {
771 : int per_page = SERIAL_ENTRIESPERPAGE,
772 : offset = per_page / 2;
773 : int64 newestPage,
774 : oldestPage,
775 : headPage,
776 : targetPage;
777 : TransactionId newestXact,
778 : oldestXact;
779 :
780 : /* GetNewTransactionId() has assigned the last XID it can safely use. */
781 : newestPage = 2 * SLRU_PAGES_PER_SEGMENT - 1; /* nothing special */
782 : newestXact = newestPage * per_page + offset;
783 : Assert(newestXact / per_page == newestPage);
784 : oldestXact = newestXact + 1;
785 : oldestXact -= 1U << 31;
786 : oldestPage = oldestXact / per_page;
787 :
788 : /*
789 : * In this scenario, the SLRU headPage pertains to the last ~1000 XIDs
790 : * assigned. oldestXact finishes, ~2B XIDs having elapsed since it
791 : * started. Further transactions cause us to summarize oldestXact to
792 : * tailPage. Function must return false so SerialAdd() doesn't zero
793 : * tailPage (which may contain entries for other old, recently-finished
794 : * XIDs) and half the SLRU. Reaching this requires burning ~2B XIDs in
795 : * single-user mode, a negligible possibility.
796 : */
797 : headPage = newestPage;
798 : targetPage = oldestPage;
799 : Assert(!SerialPagePrecedesLogically(headPage, targetPage));
800 :
801 : /*
802 : * In this scenario, the SLRU headPage pertains to oldestXact. We're
803 : * summarizing an XID near newestXact. (Assume few other XIDs used
804 : * SERIALIZABLE, hence the minimal headPage advancement. Assume
805 : * oldestXact was long-running and only recently reached the SLRU.)
806 : * Function must return true to make SerialAdd() create targetPage.
807 : *
808 : * Today's implementation mishandles this case, but it doesn't matter
809 : * enough to fix. Verify that the defect affects just one page by
810 : * asserting correct treatment of its prior page. Reaching this case
811 : * requires burning ~2B XIDs in single-user mode, a negligible
812 : * possibility. Moreover, if it does happen, the consequence would be
813 : * mild, namely a new transaction failing in SimpleLruReadPage().
814 : */
815 : headPage = oldestPage;
816 : targetPage = newestPage;
817 : Assert(SerialPagePrecedesLogically(headPage, targetPage - 1));
818 : #if 0
819 : Assert(SerialPagePrecedesLogically(headPage, targetPage));
820 : #endif
821 : }
822 : #endif
823 :
824 : /*
825 : * GUC check_hook for serializable_buffers
826 : */
827 : bool
828 1279 : check_serial_buffers(int *newval, void **extra, GucSource source)
829 : {
830 1279 : return check_slru_buffers("serializable_buffers", newval);
831 : }
832 :
833 : /*
834 : * Record a committed read write serializable xid and the minimum
835 : * commitSeqNo of any transactions to which this xid had a rw-conflict out.
836 : * An invalid commitSeqNo means that there were no conflicts out from xid.
837 : */
838 : static void
839 0 : SerialAdd(TransactionId xid, SerCommitSeqNo minConflictCommitSeqNo)
840 : {
841 : TransactionId tailXid;
842 : int64 targetPage;
843 : int slotno;
844 : int64 firstZeroPage;
845 : bool isNewPage;
846 : LWLock *lock;
847 :
848 : Assert(TransactionIdIsValid(xid));
849 :
850 0 : targetPage = SerialPage(xid);
851 0 : lock = SimpleLruGetBankLock(SerialSlruCtl, targetPage);
852 :
853 : /*
854 : * In this routine, we must hold both SerialControlLock and the SLRU bank
855 : * lock simultaneously while making the SLRU data catch up with the new
856 : * state that we determine.
857 : */
858 0 : LWLockAcquire(SerialControlLock, LW_EXCLUSIVE);
859 :
860 : /*
861 : * If 'xid' is older than the global xmin (== tailXid), there's no need to
862 : * store it, after all. This can happen if the oldest transaction holding
863 : * back the global xmin just finished, making 'xid' uninteresting, but
864 : * ClearOldPredicateLocks() has not yet run.
865 : */
866 0 : tailXid = serialControl->tailXid;
867 0 : if (!TransactionIdIsValid(tailXid) || TransactionIdPrecedes(xid, tailXid))
868 : {
869 0 : LWLockRelease(SerialControlLock);
870 0 : return;
871 : }
872 :
873 : /*
874 : * If the SLRU is currently unused, zero out the whole active region from
875 : * tailXid to headXid before taking it into use. Otherwise zero out only
876 : * any new pages that enter the tailXid-headXid range as we advance
877 : * headXid.
878 : */
879 0 : if (serialControl->headPage < 0)
880 : {
881 0 : firstZeroPage = SerialPage(tailXid);
882 0 : isNewPage = true;
883 : }
884 : else
885 : {
886 0 : firstZeroPage = SerialNextPage(serialControl->headPage);
887 0 : isNewPage = SerialPagePrecedesLogically(serialControl->headPage,
888 : targetPage);
889 : }
890 :
891 0 : if (!TransactionIdIsValid(serialControl->headXid)
892 0 : || TransactionIdFollows(xid, serialControl->headXid))
893 0 : serialControl->headXid = xid;
894 0 : if (isNewPage)
895 0 : serialControl->headPage = targetPage;
896 :
897 0 : if (isNewPage)
898 : {
899 : /* Initialize intervening pages; might involve trading locks */
900 : for (;;)
901 : {
902 0 : lock = SimpleLruGetBankLock(SerialSlruCtl, firstZeroPage);
903 0 : LWLockAcquire(lock, LW_EXCLUSIVE);
904 0 : slotno = SimpleLruZeroPage(SerialSlruCtl, firstZeroPage);
905 0 : if (firstZeroPage == targetPage)
906 0 : break;
907 0 : firstZeroPage = SerialNextPage(firstZeroPage);
908 0 : LWLockRelease(lock);
909 : }
910 : }
911 : else
912 : {
913 0 : LWLockAcquire(lock, LW_EXCLUSIVE);
914 0 : slotno = SimpleLruReadPage(SerialSlruCtl, targetPage, true, &xid);
915 : }
916 :
917 0 : SerialValue(slotno, xid) = minConflictCommitSeqNo;
918 0 : SerialSlruCtl->shared->page_dirty[slotno] = true;
919 :
920 0 : LWLockRelease(lock);
921 0 : LWLockRelease(SerialControlLock);
922 : }
923 :
924 : /*
925 : * Get the minimum commitSeqNo for any conflict out for the given xid. For
926 : * a transaction which exists but has no conflict out, InvalidSerCommitSeqNo
927 : * will be returned.
928 : */
929 : static SerCommitSeqNo
930 21 : SerialGetMinConflictCommitSeqNo(TransactionId xid)
931 : {
932 : TransactionId headXid;
933 : TransactionId tailXid;
934 : SerCommitSeqNo val;
935 : int slotno;
936 :
937 : Assert(TransactionIdIsValid(xid));
938 :
939 21 : LWLockAcquire(SerialControlLock, LW_SHARED);
940 21 : headXid = serialControl->headXid;
941 21 : tailXid = serialControl->tailXid;
942 21 : LWLockRelease(SerialControlLock);
943 :
944 21 : if (!TransactionIdIsValid(headXid))
945 21 : return 0;
946 :
947 : Assert(TransactionIdIsValid(tailXid));
948 :
949 0 : if (TransactionIdPrecedes(xid, tailXid)
950 0 : || TransactionIdFollows(xid, headXid))
951 0 : return 0;
952 :
953 : /*
954 : * The following function must be called without holding SLRU bank lock,
955 : * but will return with that lock held, which must then be released.
956 : */
957 0 : slotno = SimpleLruReadPage_ReadOnly(SerialSlruCtl,
958 0 : SerialPage(xid), &xid);
959 0 : val = SerialValue(slotno, xid);
960 0 : LWLockRelease(SimpleLruGetBankLock(SerialSlruCtl, SerialPage(xid)));
961 0 : return val;
962 : }
963 :
964 : /*
965 : * Call this whenever there is a new xmin for active serializable
966 : * transactions. We don't need to keep information on transactions which
967 : * precede that. InvalidTransactionId means none active, so everything in
968 : * the SLRU can be discarded.
969 : */
970 : static void
971 1767 : SerialSetActiveSerXmin(TransactionId xid)
972 : {
973 1767 : LWLockAcquire(SerialControlLock, LW_EXCLUSIVE);
974 :
975 : /*
976 : * When no sxacts are active, nothing overlaps, set the xid values to
977 : * invalid to show that there are no valid entries. Don't clear headPage,
978 : * though. A new xmin might still land on that page, and we don't want to
979 : * repeatedly zero out the same page.
980 : */
981 1767 : if (!TransactionIdIsValid(xid))
982 : {
983 874 : serialControl->tailXid = InvalidTransactionId;
984 874 : serialControl->headXid = InvalidTransactionId;
985 874 : LWLockRelease(SerialControlLock);
986 874 : return;
987 : }
988 :
989 : /*
990 : * When we're recovering prepared transactions, the global xmin might move
991 : * backwards depending on the order they're recovered. Normally that's not
992 : * OK, but during recovery no serializable transactions will commit, so
993 : * the SLRU is empty and we can get away with it.
994 : */
995 893 : if (RecoveryInProgress())
996 : {
997 : Assert(serialControl->headPage < 0);
998 0 : if (!TransactionIdIsValid(serialControl->tailXid)
999 0 : || TransactionIdPrecedes(xid, serialControl->tailXid))
1000 : {
1001 0 : serialControl->tailXid = xid;
1002 : }
1003 0 : LWLockRelease(SerialControlLock);
1004 0 : return;
1005 : }
1006 :
1007 : Assert(!TransactionIdIsValid(serialControl->tailXid)
1008 : || TransactionIdFollows(xid, serialControl->tailXid));
1009 :
1010 893 : serialControl->tailXid = xid;
1011 :
1012 893 : LWLockRelease(SerialControlLock);
1013 : }
1014 :
1015 : /*
1016 : * Perform a checkpoint --- either during shutdown, or on-the-fly
1017 : *
1018 : * We don't have any data that needs to survive a restart, but this is a
1019 : * convenient place to truncate the SLRU.
1020 : */
1021 : void
1022 1938 : CheckPointPredicate(void)
1023 : {
1024 : int64 truncateCutoffPage;
1025 :
1026 1938 : LWLockAcquire(SerialControlLock, LW_EXCLUSIVE);
1027 :
1028 : /* Exit quickly if the SLRU is currently not in use. */
1029 1938 : if (serialControl->headPage < 0)
1030 : {
1031 1938 : LWLockRelease(SerialControlLock);
1032 1938 : return;
1033 : }
1034 :
1035 0 : if (TransactionIdIsValid(serialControl->tailXid))
1036 : {
1037 : int64 tailPage;
1038 :
1039 0 : tailPage = SerialPage(serialControl->tailXid);
1040 :
1041 : /*
1042 : * It is possible for the tailXid to be ahead of the headXid. This
1043 : * occurs if we checkpoint while there are in-progress serializable
1044 : * transaction(s) advancing the tail but we are yet to summarize the
1045 : * transactions. In this case, we cutoff up to the headPage and the
1046 : * next summary will advance the headXid.
1047 : */
1048 0 : if (SerialPagePrecedesLogically(tailPage, serialControl->headPage))
1049 : {
1050 : /* We can truncate the SLRU up to the page containing tailXid */
1051 0 : truncateCutoffPage = tailPage;
1052 : }
1053 : else
1054 0 : truncateCutoffPage = serialControl->headPage;
1055 : }
1056 : else
1057 : {
1058 : /*----------
1059 : * The SLRU is no longer needed. Truncate to head before we set head
1060 : * invalid.
1061 : *
1062 : * XXX: It's possible that the SLRU is not needed again until XID
1063 : * wrap-around has happened, so that the segment containing headPage
1064 : * that we leave behind will appear to be new again. In that case it
1065 : * won't be removed until XID horizon advances enough to make it
1066 : * current again.
1067 : *
1068 : * XXX: This should happen in vac_truncate_clog(), not in checkpoints.
1069 : * Consider this scenario, starting from a system with no in-progress
1070 : * transactions and VACUUM FREEZE having maximized oldestXact:
1071 : * - Start a SERIALIZABLE transaction.
1072 : * - Start, finish, and summarize a SERIALIZABLE transaction, creating
1073 : * one SLRU page.
1074 : * - Consume XIDs to reach xidStopLimit.
1075 : * - Finish all transactions. Due to the long-running SERIALIZABLE
1076 : * transaction, earlier checkpoints did not touch headPage. The
1077 : * next checkpoint will change it, but that checkpoint happens after
1078 : * the end of the scenario.
1079 : * - VACUUM to advance XID limits.
1080 : * - Consume ~2M XIDs, crossing the former xidWrapLimit.
1081 : * - Start, finish, and summarize a SERIALIZABLE transaction.
1082 : * SerialAdd() declines to create the targetPage, because headPage
1083 : * is not regarded as in the past relative to that targetPage. The
1084 : * transaction instigating the summarize fails in
1085 : * SimpleLruReadPage().
1086 : */
1087 0 : truncateCutoffPage = serialControl->headPage;
1088 0 : serialControl->headPage = -1;
1089 : }
1090 :
1091 0 : LWLockRelease(SerialControlLock);
1092 :
1093 : /*
1094 : * Truncate away pages that are no longer required. Note that no
1095 : * additional locking is required, because this is only called as part of
1096 : * a checkpoint, and the validity limits have already been determined.
1097 : */
1098 0 : SimpleLruTruncate(SerialSlruCtl, truncateCutoffPage);
1099 :
1100 : /*
1101 : * Write dirty SLRU pages to disk
1102 : *
1103 : * This is not actually necessary from a correctness point of view. We do
1104 : * it merely as a debugging aid.
1105 : *
1106 : * We're doing this after the truncation to avoid writing pages right
1107 : * before deleting the file in which they sit, which would be completely
1108 : * pointless.
1109 : */
1110 0 : SimpleLruWriteAll(SerialSlruCtl, true);
1111 : }
1112 :
1113 : /*------------------------------------------------------------------------*/
1114 :
1115 : /*
1116 : * PredicateLockShmemRequest -- Register the predicate locking data structures.
1117 : */
1118 : static void
1119 1238 : PredicateLockShmemRequest(void *arg)
1120 : {
1121 : int64 max_predicate_lock_targets;
1122 : int64 max_predicate_locks;
1123 : int64 max_rw_conflicts;
1124 :
1125 : /*
1126 : * Register hash table for PREDICATELOCKTARGET structs. This stores
1127 : * per-predicate-lock-target information.
1128 : */
1129 1238 : max_predicate_lock_targets = NPREDICATELOCKTARGETENTS();
1130 :
1131 1238 : ShmemRequestHash(.name = "PREDICATELOCKTARGET hash",
1132 : .nelems = max_predicate_lock_targets,
1133 : .ptr = &PredicateLockTargetHash,
1134 : .hash_info.keysize = sizeof(PREDICATELOCKTARGETTAG),
1135 : .hash_info.entrysize = sizeof(PREDICATELOCKTARGET),
1136 : .hash_info.num_partitions = NUM_PREDICATELOCK_PARTITIONS,
1137 : .hash_flags = HASH_ELEM | HASH_BLOBS | HASH_PARTITION | HASH_FIXED_SIZE,
1138 : );
1139 :
1140 : /*
1141 : * Allocate hash table for PREDICATELOCK structs. This stores per
1142 : * xact-lock-of-a-target information.
1143 : *
1144 : * Assume an average of 2 xacts per target.
1145 : */
1146 1238 : max_predicate_locks = max_predicate_lock_targets * 2;
1147 :
1148 1238 : ShmemRequestHash(.name = "PREDICATELOCK hash",
1149 : .nelems = max_predicate_locks,
1150 : .ptr = &PredicateLockHash,
1151 : .hash_info.keysize = sizeof(PREDICATELOCKTAG),
1152 : .hash_info.entrysize = sizeof(PREDICATELOCK),
1153 : .hash_info.hash = predicatelock_hash,
1154 : .hash_info.num_partitions = NUM_PREDICATELOCK_PARTITIONS,
1155 : .hash_flags = HASH_ELEM | HASH_FUNCTION | HASH_PARTITION | HASH_FIXED_SIZE,
1156 : );
1157 :
1158 : /*
1159 : * Compute size for serializable transaction hashtable.
1160 : *
1161 : * Assume an average of 10 predicate locking transactions per backend.
1162 : * This allows aggressive cleanup while detail is present before data must
1163 : * be summarized for storage in SLRU and the "dummy" transaction.
1164 : */
1165 1238 : max_serializable_xacts = (MaxBackends + max_prepared_xacts) * 10;
1166 :
1167 : /*
1168 : * Register a list to hold information on transactions participating in
1169 : * predicate locking.
1170 : */
1171 1238 : ShmemRequestStruct(.name = "PredXactList",
1172 : .size = add_size(PredXactListDataSize,
1173 : (mul_size((Size) max_serializable_xacts,
1174 : sizeof(SERIALIZABLEXACT)))),
1175 : .ptr = (void **) &PredXact,
1176 : );
1177 :
1178 : /*
1179 : * Register hash table for SERIALIZABLEXID structs. This stores per-xid
1180 : * information for serializable transactions which have accessed data.
1181 : */
1182 1238 : ShmemRequestHash(.name = "SERIALIZABLEXID hash",
1183 : .nelems = max_serializable_xacts,
1184 : .ptr = &SerializableXidHash,
1185 : .hash_info.keysize = sizeof(SERIALIZABLEXIDTAG),
1186 : .hash_info.entrysize = sizeof(SERIALIZABLEXID),
1187 : .hash_flags = HASH_ELEM | HASH_BLOBS | HASH_FIXED_SIZE,
1188 : );
1189 :
1190 : /*
1191 : * Allocate space for tracking rw-conflicts in lists attached to the
1192 : * transactions.
1193 : *
1194 : * Assume an average of 5 conflicts per transaction. Calculations suggest
1195 : * that this will prevent resource exhaustion in even the most pessimal
1196 : * loads up to max_connections = 200 with all 200 connections pounding the
1197 : * database with serializable transactions. Beyond that, there may be
1198 : * occasional transactions canceled when trying to flag conflicts. That's
1199 : * probably OK.
1200 : */
1201 1238 : max_rw_conflicts = max_serializable_xacts * 5;
1202 :
1203 1238 : ShmemRequestStruct(.name = "RWConflictPool",
1204 : .size = RWConflictPoolHeaderDataSize + mul_size((Size) max_rw_conflicts,
1205 : RWConflictDataSize),
1206 : .ptr = (void **) &RWConflictPool,
1207 : );
1208 :
1209 1238 : ShmemRequestStruct(.name = "FinishedSerializableTransactions",
1210 : .size = sizeof(dlist_head),
1211 : .ptr = (void **) &FinishedSerializableTransactions,
1212 : );
1213 :
1214 : /*
1215 : * Initialize the SLRU storage for old committed serializable
1216 : * transactions.
1217 : */
1218 1238 : SimpleLruRequest(.desc = &SerialSlruDesc,
1219 : .name = "serializable",
1220 : .Dir = "pg_serial",
1221 : .long_segment_names = false,
1222 :
1223 : .nslots = serializable_buffers,
1224 :
1225 : .sync_handler = SYNC_HANDLER_NONE,
1226 : .PagePrecedes = SerialPagePrecedesLogically,
1227 : .errdetail_for_io_error = serial_errdetail_for_io_error,
1228 :
1229 : .buffer_tranche_id = LWTRANCHE_SERIAL_BUFFER,
1230 : .bank_tranche_id = LWTRANCHE_SERIAL_SLRU,
1231 : );
1232 : #ifdef USE_ASSERT_CHECKING
1233 : SerialPagePrecedesLogicallyUnitTests();
1234 : #endif
1235 :
1236 1238 : ShmemRequestStruct(.name = "SerialControlData",
1237 : .size = sizeof(SerialControlData),
1238 : .ptr = (void **) &serialControl,
1239 : );
1240 1238 : }
1241 :
1242 : static void
1243 1235 : PredicateLockShmemInit(void *arg)
1244 : {
1245 : int max_rw_conflicts;
1246 : bool found;
1247 :
1248 : /*
1249 : * Reserve a dummy entry in the hash table; we use it to make sure there's
1250 : * always one entry available when we need to split or combine a page,
1251 : * because running out of space there could mean aborting a
1252 : * non-serializable transaction.
1253 : */
1254 1235 : (void) hash_search(PredicateLockTargetHash, &ScratchTargetTag,
1255 : HASH_ENTER, &found);
1256 : Assert(!found);
1257 :
1258 1235 : dlist_init(&PredXact->availableList);
1259 1235 : dlist_init(&PredXact->activeList);
1260 1235 : PredXact->SxactGlobalXmin = InvalidTransactionId;
1261 1235 : PredXact->SxactGlobalXminCount = 0;
1262 1235 : PredXact->WritableSxactCount = 0;
1263 1235 : PredXact->LastSxactCommitSeqNo = FirstNormalSerCommitSeqNo - 1;
1264 1235 : PredXact->CanPartialClearThrough = 0;
1265 1235 : PredXact->HavePartialClearedThrough = 0;
1266 1235 : PredXact->element
1267 1235 : = (SERIALIZABLEXACT *) ((char *) PredXact + PredXactListDataSize);
1268 : /* Add all elements to available list, clean. */
1269 1168055 : for (int i = 0; i < max_serializable_xacts; i++)
1270 : {
1271 1166820 : LWLockInitialize(&PredXact->element[i].perXactPredicateListLock,
1272 : LWTRANCHE_PER_XACT_PREDICATE_LIST);
1273 1166820 : dlist_push_tail(&PredXact->availableList, &PredXact->element[i].xactLink);
1274 : }
1275 1235 : PredXact->OldCommittedSxact = CreatePredXact();
1276 1235 : SetInvalidVirtualTransactionId(PredXact->OldCommittedSxact->vxid);
1277 1235 : PredXact->OldCommittedSxact->prepareSeqNo = 0;
1278 1235 : PredXact->OldCommittedSxact->commitSeqNo = 0;
1279 1235 : PredXact->OldCommittedSxact->SeqNo.lastCommitBeforeSnapshot = 0;
1280 1235 : dlist_init(&PredXact->OldCommittedSxact->outConflicts);
1281 1235 : dlist_init(&PredXact->OldCommittedSxact->inConflicts);
1282 1235 : dlist_init(&PredXact->OldCommittedSxact->predicateLocks);
1283 1235 : dlist_node_init(&PredXact->OldCommittedSxact->finishedLink);
1284 1235 : dlist_init(&PredXact->OldCommittedSxact->possibleUnsafeConflicts);
1285 1235 : PredXact->OldCommittedSxact->topXid = InvalidTransactionId;
1286 1235 : PredXact->OldCommittedSxact->finishedBefore = InvalidTransactionId;
1287 1235 : PredXact->OldCommittedSxact->xmin = InvalidTransactionId;
1288 1235 : PredXact->OldCommittedSxact->flags = SXACT_FLAG_COMMITTED;
1289 1235 : PredXact->OldCommittedSxact->pid = 0;
1290 1235 : PredXact->OldCommittedSxact->pgprocno = INVALID_PROC_NUMBER;
1291 :
1292 : /* Initialize the rw-conflict pool */
1293 1235 : dlist_init(&RWConflictPool->availableList);
1294 1235 : RWConflictPool->element = (RWConflict) ((char *) RWConflictPool +
1295 : RWConflictPoolHeaderDataSize);
1296 :
1297 1235 : max_rw_conflicts = max_serializable_xacts * 5;
1298 :
1299 : /* Add all elements to available list, clean. */
1300 5835335 : for (int i = 0; i < max_rw_conflicts; i++)
1301 : {
1302 5834100 : dlist_push_tail(&RWConflictPool->availableList,
1303 5834100 : &RWConflictPool->element[i].outLink);
1304 : }
1305 :
1306 : /* Initialize the list of finished serializable transactions */
1307 1235 : dlist_init(FinishedSerializableTransactions);
1308 :
1309 : /* Initialize SerialControl to reflect empty SLRU. */
1310 1235 : LWLockAcquire(SerialControlLock, LW_EXCLUSIVE);
1311 1235 : serialControl->headPage = -1;
1312 1235 : serialControl->headXid = InvalidTransactionId;
1313 1235 : serialControl->tailXid = InvalidTransactionId;
1314 1235 : LWLockRelease(SerialControlLock);
1315 :
1316 : SlruPagePrecedesUnitTests(SerialSlruCtl, SERIAL_ENTRIESPERPAGE);
1317 :
1318 : /* This never changes, so let's keep a local copy. */
1319 1235 : OldCommittedSxact = PredXact->OldCommittedSxact;
1320 :
1321 : /* Pre-calculate the hash and partition lock of the scratch entry */
1322 1235 : ScratchTargetTagHash = PredicateLockTargetTagHashCode(&ScratchTargetTag);
1323 1235 : ScratchPartitionLock = PredicateLockHashPartitionLock(ScratchTargetTagHash);
1324 1235 : }
1325 :
1326 : static void
1327 0 : PredicateLockShmemAttach(void *arg)
1328 : {
1329 : /* This never changes, so let's keep a local copy. */
1330 0 : OldCommittedSxact = PredXact->OldCommittedSxact;
1331 :
1332 : /* Pre-calculate the hash and partition lock of the scratch entry */
1333 0 : ScratchTargetTagHash = PredicateLockTargetTagHashCode(&ScratchTargetTag);
1334 0 : ScratchPartitionLock = PredicateLockHashPartitionLock(ScratchTargetTagHash);
1335 0 : }
1336 :
1337 : /*
1338 : * Compute the hash code associated with a PREDICATELOCKTAG.
1339 : *
1340 : * Because we want to use just one set of partition locks for both the
1341 : * PREDICATELOCKTARGET and PREDICATELOCK hash tables, we have to make sure
1342 : * that PREDICATELOCKs fall into the same partition number as their
1343 : * associated PREDICATELOCKTARGETs. dynahash.c expects the partition number
1344 : * to be the low-order bits of the hash code, and therefore a
1345 : * PREDICATELOCKTAG's hash code must have the same low-order bits as the
1346 : * associated PREDICATELOCKTARGETTAG's hash code. We achieve this with this
1347 : * specialized hash function.
1348 : */
1349 : static uint32
1350 0 : predicatelock_hash(const void *key, Size keysize)
1351 : {
1352 0 : const PREDICATELOCKTAG *predicatelocktag = (const PREDICATELOCKTAG *) key;
1353 : uint32 targethash;
1354 :
1355 : Assert(keysize == sizeof(PREDICATELOCKTAG));
1356 :
1357 : /* Look into the associated target object, and compute its hash code */
1358 0 : targethash = PredicateLockTargetTagHashCode(&predicatelocktag->myTarget->tag);
1359 :
1360 0 : return PredicateLockHashCodeFromTargetHashCode(predicatelocktag, targethash);
1361 : }
1362 :
1363 :
1364 : /*
1365 : * GetPredicateLockStatusData
1366 : * Return a table containing the internal state of the predicate
1367 : * lock manager for use in pg_lock_status.
1368 : *
1369 : * Like GetLockStatusData, this function tries to hold the partition LWLocks
1370 : * for as short a time as possible by returning two arrays that simply
1371 : * contain the PREDICATELOCKTARGETTAG and SERIALIZABLEXACT for each lock
1372 : * table entry. Multiple copies of the same PREDICATELOCKTARGETTAG and
1373 : * SERIALIZABLEXACT will likely appear.
1374 : */
1375 : PredicateLockData *
1376 414 : GetPredicateLockStatusData(void)
1377 : {
1378 : PredicateLockData *data;
1379 : int i;
1380 : int els,
1381 : el;
1382 : HASH_SEQ_STATUS seqstat;
1383 : PREDICATELOCK *predlock;
1384 :
1385 414 : data = palloc_object(PredicateLockData);
1386 :
1387 : /*
1388 : * To ensure consistency, take simultaneous locks on all partition locks
1389 : * in ascending order, then SerializableXactHashLock.
1390 : */
1391 7038 : for (i = 0; i < NUM_PREDICATELOCK_PARTITIONS; i++)
1392 6624 : LWLockAcquire(PredicateLockHashPartitionLockByIndex(i), LW_SHARED);
1393 414 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
1394 :
1395 : /* Get number of locks and allocate appropriately-sized arrays. */
1396 414 : els = hash_get_num_entries(PredicateLockHash);
1397 414 : data->nelements = els;
1398 414 : data->locktags = palloc_array(PREDICATELOCKTARGETTAG, els);
1399 414 : data->xacts = palloc_array(SERIALIZABLEXACT, els);
1400 :
1401 :
1402 : /* Scan through PredicateLockHash and copy contents */
1403 414 : hash_seq_init(&seqstat, PredicateLockHash);
1404 :
1405 414 : el = 0;
1406 :
1407 418 : while ((predlock = (PREDICATELOCK *) hash_seq_search(&seqstat)))
1408 : {
1409 4 : data->locktags[el] = predlock->tag.myTarget->tag;
1410 4 : data->xacts[el] = *predlock->tag.myXact;
1411 4 : el++;
1412 : }
1413 :
1414 : Assert(el == els);
1415 :
1416 : /* Release locks in reverse order */
1417 414 : LWLockRelease(SerializableXactHashLock);
1418 7038 : for (i = NUM_PREDICATELOCK_PARTITIONS - 1; i >= 0; i--)
1419 6624 : LWLockRelease(PredicateLockHashPartitionLockByIndex(i));
1420 :
1421 414 : return data;
1422 : }
1423 :
1424 : /*
1425 : * Free up shared memory structures by pushing the oldest sxact (the one at
1426 : * the front of the SummarizeOldestCommittedSxact queue) into summary form.
1427 : * Each call will free exactly one SERIALIZABLEXACT structure and may also
1428 : * free one or more of these structures: SERIALIZABLEXID, PREDICATELOCK,
1429 : * PREDICATELOCKTARGET, RWConflictData.
1430 : */
1431 : static void
1432 0 : SummarizeOldestCommittedSxact(void)
1433 : {
1434 : SERIALIZABLEXACT *sxact;
1435 :
1436 0 : LWLockAcquire(SerializableFinishedListLock, LW_EXCLUSIVE);
1437 :
1438 : /*
1439 : * This function is only called if there are no sxact slots available.
1440 : * Some of them must belong to old, already-finished transactions, so
1441 : * there should be something in FinishedSerializableTransactions list that
1442 : * we can summarize. However, there's a race condition: while we were not
1443 : * holding any locks, a transaction might have ended and cleaned up all
1444 : * the finished sxact entries already, freeing up their sxact slots. In
1445 : * that case, we have nothing to do here. The caller will find one of the
1446 : * slots released by the other backend when it retries.
1447 : */
1448 0 : if (dlist_is_empty(FinishedSerializableTransactions))
1449 : {
1450 0 : LWLockRelease(SerializableFinishedListLock);
1451 0 : return;
1452 : }
1453 :
1454 : /*
1455 : * Grab the first sxact off the finished list -- this will be the earliest
1456 : * commit. Remove it from the list.
1457 : */
1458 0 : sxact = dlist_head_element(SERIALIZABLEXACT, finishedLink,
1459 : FinishedSerializableTransactions);
1460 0 : dlist_delete_thoroughly(&sxact->finishedLink);
1461 :
1462 : /* Add to SLRU summary information. */
1463 0 : if (TransactionIdIsValid(sxact->topXid) && !SxactIsReadOnly(sxact))
1464 0 : SerialAdd(sxact->topXid, SxactHasConflictOut(sxact)
1465 : ? sxact->SeqNo.earliestOutConflictCommit : InvalidSerCommitSeqNo);
1466 :
1467 : /* Summarize and release the detail. */
1468 0 : ReleaseOneSerializableXact(sxact, false, true);
1469 :
1470 0 : LWLockRelease(SerializableFinishedListLock);
1471 : }
1472 :
1473 : /*
1474 : * GetSafeSnapshot
1475 : * Obtain and register a snapshot for a READ ONLY DEFERRABLE
1476 : * transaction. Ensures that the snapshot is "safe", i.e. a
1477 : * read-only transaction running on it can execute serializably
1478 : * without further checks. This requires waiting for concurrent
1479 : * transactions to complete, and retrying with a new snapshot if
1480 : * one of them could possibly create a conflict.
1481 : *
1482 : * As with GetSerializableTransactionSnapshot (which this is a subroutine
1483 : * for), the passed-in Snapshot pointer should reference a static data
1484 : * area that can safely be passed to GetSnapshotData.
1485 : */
1486 : static Snapshot
1487 7 : GetSafeSnapshot(Snapshot origSnapshot)
1488 : {
1489 : Snapshot snapshot;
1490 :
1491 : Assert(XactReadOnly && XactDeferrable);
1492 :
1493 : while (true)
1494 : {
1495 : /*
1496 : * GetSerializableTransactionSnapshotInt is going to call
1497 : * GetSnapshotData, so we need to provide it the static snapshot area
1498 : * our caller passed to us. The pointer returned is actually the same
1499 : * one passed to it, but we avoid assuming that here.
1500 : */
1501 8 : snapshot = GetSerializableTransactionSnapshotInt(origSnapshot,
1502 : NULL, InvalidPid);
1503 :
1504 8 : if (MySerializableXact == InvalidSerializableXact)
1505 5 : return snapshot; /* no concurrent r/w xacts; it's safe */
1506 :
1507 3 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
1508 :
1509 : /*
1510 : * Wait for concurrent transactions to finish. Stop early if one of
1511 : * them marked us as conflicted.
1512 : */
1513 3 : MySerializableXact->flags |= SXACT_FLAG_DEFERRABLE_WAITING;
1514 6 : while (!(dlist_is_empty(&MySerializableXact->possibleUnsafeConflicts) ||
1515 3 : SxactIsROUnsafe(MySerializableXact)))
1516 : {
1517 3 : LWLockRelease(SerializableXactHashLock);
1518 3 : ProcWaitForSignal(WAIT_EVENT_SAFE_SNAPSHOT);
1519 3 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
1520 : }
1521 3 : MySerializableXact->flags &= ~SXACT_FLAG_DEFERRABLE_WAITING;
1522 :
1523 3 : if (!SxactIsROUnsafe(MySerializableXact))
1524 : {
1525 2 : LWLockRelease(SerializableXactHashLock);
1526 2 : break; /* success */
1527 : }
1528 :
1529 1 : LWLockRelease(SerializableXactHashLock);
1530 :
1531 : /* else, need to retry... */
1532 1 : ereport(DEBUG2,
1533 : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1534 : errmsg_internal("deferrable snapshot was unsafe; trying a new one")));
1535 1 : ReleasePredicateLocks(false, false);
1536 : }
1537 :
1538 : /*
1539 : * Now we have a safe snapshot, so we don't need to do any further checks.
1540 : */
1541 : Assert(SxactIsROSafe(MySerializableXact));
1542 2 : ReleasePredicateLocks(false, true);
1543 :
1544 2 : return snapshot;
1545 : }
1546 :
1547 : /*
1548 : * GetSafeSnapshotBlockingPids
1549 : * If the specified process is currently blocked in GetSafeSnapshot,
1550 : * write the process IDs of all processes that it is blocked by
1551 : * into the caller-supplied buffer output[]. The list is truncated at
1552 : * output_size, and the number of PIDs written into the buffer is
1553 : * returned. Returns zero if the given PID is not currently blocked
1554 : * in GetSafeSnapshot.
1555 : */
1556 : int
1557 409 : GetSafeSnapshotBlockingPids(int blocked_pid, int *output, int output_size)
1558 : {
1559 409 : int num_written = 0;
1560 : dlist_iter iter;
1561 409 : SERIALIZABLEXACT *blocking_sxact = NULL;
1562 :
1563 409 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
1564 :
1565 : /* Find blocked_pid's SERIALIZABLEXACT by linear search. */
1566 958 : dlist_foreach(iter, &PredXact->activeList)
1567 : {
1568 628 : SERIALIZABLEXACT *sxact =
1569 628 : dlist_container(SERIALIZABLEXACT, xactLink, iter.cur);
1570 :
1571 628 : if (sxact->pid == blocked_pid)
1572 : {
1573 79 : blocking_sxact = sxact;
1574 79 : break;
1575 : }
1576 : }
1577 :
1578 : /* Did we find it, and is it currently waiting in GetSafeSnapshot? */
1579 409 : if (blocking_sxact != NULL && SxactIsDeferrableWaiting(blocking_sxact))
1580 : {
1581 : /* Traverse the list of possible unsafe conflicts collecting PIDs. */
1582 2 : dlist_foreach(iter, &blocking_sxact->possibleUnsafeConflicts)
1583 : {
1584 2 : RWConflict possibleUnsafeConflict =
1585 2 : dlist_container(RWConflictData, inLink, iter.cur);
1586 :
1587 2 : output[num_written++] = possibleUnsafeConflict->sxactOut->pid;
1588 :
1589 2 : if (num_written >= output_size)
1590 2 : break;
1591 : }
1592 : }
1593 :
1594 409 : LWLockRelease(SerializableXactHashLock);
1595 :
1596 409 : return num_written;
1597 : }
1598 :
1599 : /*
1600 : * Acquire a snapshot that can be used for the current transaction.
1601 : *
1602 : * Make sure we have a SERIALIZABLEXACT reference in MySerializableXact.
1603 : * It should be current for this process and be contained in PredXact.
1604 : *
1605 : * The passed-in Snapshot pointer should reference a static data area that
1606 : * can safely be passed to GetSnapshotData. The return value is actually
1607 : * always this same pointer; no new snapshot data structure is allocated
1608 : * within this function.
1609 : */
1610 : Snapshot
1611 1691 : GetSerializableTransactionSnapshot(Snapshot snapshot)
1612 : {
1613 : Assert(IsolationIsSerializable());
1614 :
1615 : /*
1616 : * Can't use serializable mode while recovery is still active, as it is,
1617 : * for example, on a hot standby. We could get here despite the check in
1618 : * check_transaction_isolation() if default_transaction_isolation is set
1619 : * to serializable, so phrase the hint accordingly.
1620 : */
1621 1691 : if (RecoveryInProgress())
1622 0 : ereport(ERROR,
1623 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1624 : errmsg("cannot use serializable mode in a hot standby"),
1625 : errdetail("\"default_transaction_isolation\" is set to \"serializable\"."),
1626 : errhint("You can use \"SET default_transaction_isolation = 'repeatable read'\" to change the default.")));
1627 :
1628 : /*
1629 : * A special optimization is available for SERIALIZABLE READ ONLY
1630 : * DEFERRABLE transactions -- we can wait for a suitable snapshot and
1631 : * thereby avoid all SSI overhead once it's running.
1632 : */
1633 1691 : if (XactReadOnly && XactDeferrable)
1634 7 : return GetSafeSnapshot(snapshot);
1635 :
1636 1684 : return GetSerializableTransactionSnapshotInt(snapshot,
1637 : NULL, InvalidPid);
1638 : }
1639 :
1640 : /*
1641 : * Import a snapshot to be used for the current transaction.
1642 : *
1643 : * This is nearly the same as GetSerializableTransactionSnapshot, except that
1644 : * we don't take a new snapshot, but rather use the data we're handed.
1645 : *
1646 : * The caller must have verified that the snapshot came from a serializable
1647 : * transaction; and if we're read-write, the source transaction must not be
1648 : * read-only.
1649 : */
1650 : void
1651 13 : SetSerializableTransactionSnapshot(Snapshot snapshot,
1652 : VirtualTransactionId *sourcevxid,
1653 : int sourcepid)
1654 : {
1655 : Assert(IsolationIsSerializable());
1656 :
1657 : /*
1658 : * If this is called by parallel.c in a parallel worker, we don't want to
1659 : * create a SERIALIZABLEXACT just yet because the leader's
1660 : * SERIALIZABLEXACT will be installed with AttachSerializableXact(). We
1661 : * also don't want to reject SERIALIZABLE READ ONLY DEFERRABLE in this
1662 : * case, because the leader has already determined that the snapshot it
1663 : * has passed us is safe. So there is nothing for us to do.
1664 : */
1665 13 : if (IsParallelWorker())
1666 13 : return;
1667 :
1668 : /*
1669 : * We do not allow SERIALIZABLE READ ONLY DEFERRABLE transactions to
1670 : * import snapshots, since there's no way to wait for a safe snapshot when
1671 : * we're using the snap we're told to. (XXX instead of throwing an error,
1672 : * we could just ignore the XactDeferrable flag?)
1673 : */
1674 0 : if (XactReadOnly && XactDeferrable)
1675 0 : ereport(ERROR,
1676 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1677 : errmsg("a snapshot-importing transaction must not be READ ONLY DEFERRABLE")));
1678 :
1679 0 : (void) GetSerializableTransactionSnapshotInt(snapshot, sourcevxid,
1680 : sourcepid);
1681 : }
1682 :
1683 : /*
1684 : * Guts of GetSerializableTransactionSnapshot
1685 : *
1686 : * If sourcevxid is valid, this is actually an import operation and we should
1687 : * skip calling GetSnapshotData, because the snapshot contents are already
1688 : * loaded up. HOWEVER: to avoid race conditions, we must check that the
1689 : * source xact is still running after we acquire SerializableXactHashLock.
1690 : * We do that by calling ProcArrayInstallImportedXmin.
1691 : */
1692 : static Snapshot
1693 1692 : GetSerializableTransactionSnapshotInt(Snapshot snapshot,
1694 : VirtualTransactionId *sourcevxid,
1695 : int sourcepid)
1696 : {
1697 : PGPROC *proc;
1698 : VirtualTransactionId vxid;
1699 : SERIALIZABLEXACT *sxact,
1700 : *othersxact;
1701 :
1702 : /* We only do this for serializable transactions. Once. */
1703 : Assert(MySerializableXact == InvalidSerializableXact);
1704 :
1705 : Assert(!RecoveryInProgress());
1706 :
1707 : /*
1708 : * Since all parts of a serializable transaction must use the same
1709 : * snapshot, it is too late to establish one after a parallel operation
1710 : * has begun.
1711 : */
1712 1692 : if (IsInParallelMode())
1713 0 : elog(ERROR, "cannot establish serializable snapshot during a parallel operation");
1714 :
1715 1692 : proc = MyProc;
1716 : Assert(proc != NULL);
1717 1692 : GET_VXID_FROM_PGPROC(vxid, *proc);
1718 :
1719 : /*
1720 : * First we get the sxact structure, which may involve looping and access
1721 : * to the "finished" list to free a structure for use.
1722 : *
1723 : * We must hold SerializableXactHashLock when taking/checking the snapshot
1724 : * to avoid race conditions, for much the same reasons that
1725 : * GetSnapshotData takes the ProcArrayLock. Since we might have to
1726 : * release SerializableXactHashLock to call SummarizeOldestCommittedSxact,
1727 : * this means we have to create the sxact first, which is a bit annoying
1728 : * (in particular, an elog(ERROR) in procarray.c would cause us to leak
1729 : * the sxact). Consider refactoring to avoid this.
1730 : */
1731 : #ifdef TEST_SUMMARIZE_SERIAL
1732 : SummarizeOldestCommittedSxact();
1733 : #endif
1734 1692 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
1735 : do
1736 : {
1737 1692 : sxact = CreatePredXact();
1738 : /* If null, push out committed sxact to SLRU summary & retry. */
1739 1692 : if (!sxact)
1740 : {
1741 0 : LWLockRelease(SerializableXactHashLock);
1742 0 : SummarizeOldestCommittedSxact();
1743 0 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
1744 : }
1745 1692 : } while (!sxact);
1746 :
1747 : /* Get the snapshot, or check that it's safe to use */
1748 1692 : if (!sourcevxid)
1749 1692 : snapshot = GetSnapshotData(snapshot);
1750 0 : else if (!ProcArrayInstallImportedXmin(snapshot->xmin, sourcevxid))
1751 : {
1752 0 : ReleasePredXact(sxact);
1753 0 : LWLockRelease(SerializableXactHashLock);
1754 0 : ereport(ERROR,
1755 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1756 : errmsg("could not import the requested snapshot"),
1757 : errdetail("The source process with PID %d is not running anymore.",
1758 : sourcepid)));
1759 : }
1760 :
1761 : /*
1762 : * If there are no serializable transactions which are not read-only, we
1763 : * can "opt out" of predicate locking and conflict checking for a
1764 : * read-only transaction.
1765 : *
1766 : * The reason this is safe is that a read-only transaction can only become
1767 : * part of a dangerous structure if it overlaps a writable transaction
1768 : * which in turn overlaps a writable transaction which committed before
1769 : * the read-only transaction started. A new writable transaction can
1770 : * overlap this one, but it can't meet the other condition of overlapping
1771 : * a transaction which committed before this one started.
1772 : */
1773 1692 : if (XactReadOnly && PredXact->WritableSxactCount == 0)
1774 : {
1775 115 : ReleasePredXact(sxact);
1776 115 : LWLockRelease(SerializableXactHashLock);
1777 115 : return snapshot;
1778 : }
1779 :
1780 : /* Initialize the structure. */
1781 1577 : sxact->vxid = vxid;
1782 1577 : sxact->SeqNo.lastCommitBeforeSnapshot = PredXact->LastSxactCommitSeqNo;
1783 1577 : sxact->prepareSeqNo = InvalidSerCommitSeqNo;
1784 1577 : sxact->commitSeqNo = InvalidSerCommitSeqNo;
1785 1577 : dlist_init(&(sxact->outConflicts));
1786 1577 : dlist_init(&(sxact->inConflicts));
1787 1577 : dlist_init(&(sxact->possibleUnsafeConflicts));
1788 1577 : sxact->topXid = GetTopTransactionIdIfAny();
1789 1577 : sxact->finishedBefore = InvalidTransactionId;
1790 1577 : sxact->xmin = snapshot->xmin;
1791 1577 : sxact->pid = MyProcPid;
1792 1577 : sxact->pgprocno = MyProcNumber;
1793 1577 : dlist_init(&sxact->predicateLocks);
1794 1577 : dlist_node_init(&sxact->finishedLink);
1795 1577 : sxact->flags = 0;
1796 1577 : if (XactReadOnly)
1797 : {
1798 : dlist_iter iter;
1799 :
1800 108 : sxact->flags |= SXACT_FLAG_READ_ONLY;
1801 :
1802 : /*
1803 : * Register all concurrent r/w transactions as possible conflicts; if
1804 : * all of them commit without any outgoing conflicts to earlier
1805 : * transactions then this snapshot can be deemed safe (and we can run
1806 : * without tracking predicate locks).
1807 : */
1808 472 : dlist_foreach(iter, &PredXact->activeList)
1809 : {
1810 364 : othersxact = dlist_container(SERIALIZABLEXACT, xactLink, iter.cur);
1811 :
1812 364 : if (!SxactIsCommitted(othersxact)
1813 243 : && !SxactIsDoomed(othersxact)
1814 243 : && !SxactIsReadOnly(othersxact))
1815 : {
1816 134 : SetPossibleUnsafeConflict(sxact, othersxact);
1817 : }
1818 : }
1819 :
1820 : /*
1821 : * If we didn't find any possibly unsafe conflicts because every
1822 : * uncommitted writable transaction turned out to be doomed, then we
1823 : * can "opt out" immediately. See comments above the earlier check
1824 : * for PredXact->WritableSxactCount == 0.
1825 : */
1826 108 : if (dlist_is_empty(&sxact->possibleUnsafeConflicts))
1827 : {
1828 0 : ReleasePredXact(sxact);
1829 0 : LWLockRelease(SerializableXactHashLock);
1830 0 : return snapshot;
1831 : }
1832 : }
1833 : else
1834 : {
1835 1469 : ++(PredXact->WritableSxactCount);
1836 : Assert(PredXact->WritableSxactCount <=
1837 : (MaxBackends + max_prepared_xacts));
1838 : }
1839 :
1840 : /* Maintain serializable global xmin info. */
1841 1577 : if (!TransactionIdIsValid(PredXact->SxactGlobalXmin))
1842 : {
1843 : Assert(PredXact->SxactGlobalXminCount == 0);
1844 874 : PredXact->SxactGlobalXmin = snapshot->xmin;
1845 874 : PredXact->SxactGlobalXminCount = 1;
1846 874 : SerialSetActiveSerXmin(snapshot->xmin);
1847 : }
1848 703 : else if (TransactionIdEquals(snapshot->xmin, PredXact->SxactGlobalXmin))
1849 : {
1850 : Assert(PredXact->SxactGlobalXminCount > 0);
1851 666 : PredXact->SxactGlobalXminCount++;
1852 : }
1853 : else
1854 : {
1855 : Assert(TransactionIdFollows(snapshot->xmin, PredXact->SxactGlobalXmin));
1856 : }
1857 :
1858 1577 : MySerializableXact = sxact;
1859 1577 : MyXactDidWrite = false; /* haven't written anything yet */
1860 :
1861 1577 : LWLockRelease(SerializableXactHashLock);
1862 :
1863 1577 : CreateLocalPredicateLockHash();
1864 :
1865 1577 : return snapshot;
1866 : }
1867 :
1868 : static void
1869 1590 : CreateLocalPredicateLockHash(void)
1870 : {
1871 : HASHCTL hash_ctl;
1872 :
1873 : /* Initialize the backend-local hash table of parent locks */
1874 : Assert(LocalPredicateLockHash == NULL);
1875 1590 : hash_ctl.keysize = sizeof(PREDICATELOCKTARGETTAG);
1876 1590 : hash_ctl.entrysize = sizeof(LOCALPREDICATELOCK);
1877 1590 : LocalPredicateLockHash = hash_create("Local predicate lock",
1878 : max_predicate_locks_per_xact,
1879 : &hash_ctl,
1880 : HASH_ELEM | HASH_BLOBS);
1881 1590 : }
1882 :
1883 : /*
1884 : * Register the top level XID in SerializableXidHash.
1885 : * Also store it for easy reference in MySerializableXact.
1886 : */
1887 : void
1888 167581 : RegisterPredicateLockingXid(TransactionId xid)
1889 : {
1890 : SERIALIZABLEXIDTAG sxidtag;
1891 : SERIALIZABLEXID *sxid;
1892 : bool found;
1893 :
1894 : /*
1895 : * If we're not tracking predicate lock data for this transaction, we
1896 : * should ignore the request and return quickly.
1897 : */
1898 167581 : if (MySerializableXact == InvalidSerializableXact)
1899 166279 : return;
1900 :
1901 : /* We should have a valid XID and be at the top level. */
1902 : Assert(TransactionIdIsValid(xid));
1903 :
1904 1302 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
1905 :
1906 : /* This should only be done once per transaction. */
1907 : Assert(MySerializableXact->topXid == InvalidTransactionId);
1908 :
1909 1302 : MySerializableXact->topXid = xid;
1910 :
1911 1302 : sxidtag.xid = xid;
1912 1302 : sxid = (SERIALIZABLEXID *) hash_search(SerializableXidHash,
1913 : &sxidtag,
1914 : HASH_ENTER, &found);
1915 : Assert(!found);
1916 :
1917 : /* Initialize the structure. */
1918 1302 : sxid->myXact = MySerializableXact;
1919 1302 : LWLockRelease(SerializableXactHashLock);
1920 : }
1921 :
1922 :
1923 : /*
1924 : * Check whether there are any predicate locks held by any transaction
1925 : * for the page at the given block number.
1926 : *
1927 : * Note that the transaction may be completed but not yet subject to
1928 : * cleanup due to overlapping serializable transactions. This must
1929 : * return valid information regardless of transaction isolation level.
1930 : *
1931 : * Also note that this doesn't check for a conflicting relation lock,
1932 : * just a lock specifically on the given page.
1933 : *
1934 : * One use is to support proper behavior during GiST index vacuum.
1935 : */
1936 : bool
1937 0 : PageIsPredicateLocked(Relation relation, BlockNumber blkno)
1938 : {
1939 : PREDICATELOCKTARGETTAG targettag;
1940 : uint32 targettaghash;
1941 : LWLock *partitionLock;
1942 : PREDICATELOCKTARGET *target;
1943 :
1944 0 : SET_PREDICATELOCKTARGETTAG_PAGE(targettag,
1945 : relation->rd_locator.dbOid,
1946 : relation->rd_id,
1947 : blkno);
1948 :
1949 0 : targettaghash = PredicateLockTargetTagHashCode(&targettag);
1950 0 : partitionLock = PredicateLockHashPartitionLock(targettaghash);
1951 0 : LWLockAcquire(partitionLock, LW_SHARED);
1952 : target = (PREDICATELOCKTARGET *)
1953 0 : hash_search_with_hash_value(PredicateLockTargetHash,
1954 : &targettag, targettaghash,
1955 : HASH_FIND, NULL);
1956 0 : LWLockRelease(partitionLock);
1957 :
1958 0 : return (target != NULL);
1959 : }
1960 :
1961 :
1962 : /*
1963 : * Check whether a particular lock is held by this transaction.
1964 : *
1965 : * Important note: this function may return false even if the lock is
1966 : * being held, because it uses the local lock table which is not
1967 : * updated if another transaction modifies our lock list (e.g. to
1968 : * split an index page). It can also return true when a coarser
1969 : * granularity lock that covers this target is being held. Be careful
1970 : * to only use this function in circumstances where such errors are
1971 : * acceptable!
1972 : */
1973 : static bool
1974 41896 : PredicateLockExists(const PREDICATELOCKTARGETTAG *targettag)
1975 : {
1976 : LOCALPREDICATELOCK *lock;
1977 :
1978 : /* check local hash table */
1979 41896 : lock = (LOCALPREDICATELOCK *) hash_search(LocalPredicateLockHash,
1980 : targettag,
1981 : HASH_FIND, NULL);
1982 :
1983 41896 : if (!lock)
1984 12752 : return false;
1985 :
1986 : /*
1987 : * Found entry in the table, but still need to check whether it's actually
1988 : * held -- it could just be a parent of some held lock.
1989 : */
1990 29144 : return lock->held;
1991 : }
1992 :
1993 : /*
1994 : * Return the parent lock tag in the lock hierarchy: the next coarser
1995 : * lock that covers the provided tag.
1996 : *
1997 : * Returns true and sets *parent to the parent tag if one exists,
1998 : * returns false if none exists.
1999 : */
2000 : static bool
2001 25390 : GetParentPredicateLockTag(const PREDICATELOCKTARGETTAG *tag,
2002 : PREDICATELOCKTARGETTAG *parent)
2003 : {
2004 25390 : switch (GET_PREDICATELOCKTARGETTAG_TYPE(*tag))
2005 : {
2006 8967 : case PREDLOCKTAG_RELATION:
2007 : /* relation locks have no parent lock */
2008 8967 : return false;
2009 :
2010 7553 : case PREDLOCKTAG_PAGE:
2011 : /* parent lock is relation lock */
2012 7553 : SET_PREDICATELOCKTARGETTAG_RELATION(*parent,
2013 : GET_PREDICATELOCKTARGETTAG_DB(*tag),
2014 : GET_PREDICATELOCKTARGETTAG_RELATION(*tag));
2015 :
2016 7553 : return true;
2017 :
2018 8870 : case PREDLOCKTAG_TUPLE:
2019 : /* parent lock is page lock */
2020 8870 : SET_PREDICATELOCKTARGETTAG_PAGE(*parent,
2021 : GET_PREDICATELOCKTARGETTAG_DB(*tag),
2022 : GET_PREDICATELOCKTARGETTAG_RELATION(*tag),
2023 : GET_PREDICATELOCKTARGETTAG_PAGE(*tag));
2024 8870 : return true;
2025 : }
2026 :
2027 : /* not reachable */
2028 : Assert(false);
2029 0 : return false;
2030 : }
2031 :
2032 : /*
2033 : * Check whether the lock we are considering is already covered by a
2034 : * coarser lock for our transaction.
2035 : *
2036 : * Like PredicateLockExists, this function might return a false
2037 : * negative, but it will never return a false positive.
2038 : */
2039 : static bool
2040 8608 : CoarserLockCovers(const PREDICATELOCKTARGETTAG *newtargettag)
2041 : {
2042 : PREDICATELOCKTARGETTAG targettag,
2043 : parenttag;
2044 :
2045 8608 : targettag = *newtargettag;
2046 :
2047 : /* check parents iteratively until no more */
2048 13416 : while (GetParentPredicateLockTag(&targettag, &parenttag))
2049 : {
2050 9469 : targettag = parenttag;
2051 9469 : if (PredicateLockExists(&targettag))
2052 4661 : return true;
2053 : }
2054 :
2055 : /* no more parents to check; lock is not covered */
2056 3947 : return false;
2057 : }
2058 :
2059 : /*
2060 : * Remove the dummy entry from the predicate lock target hash, to free up some
2061 : * scratch space. The caller must be holding SerializablePredicateListLock,
2062 : * and must restore the entry with RestoreScratchTarget() before releasing the
2063 : * lock.
2064 : *
2065 : * If lockheld is true, the caller is already holding the partition lock
2066 : * of the partition containing the scratch entry.
2067 : */
2068 : static void
2069 45 : RemoveScratchTarget(bool lockheld)
2070 : {
2071 : bool found;
2072 :
2073 : Assert(LWLockHeldByMe(SerializablePredicateListLock));
2074 :
2075 45 : if (!lockheld)
2076 0 : LWLockAcquire(ScratchPartitionLock, LW_EXCLUSIVE);
2077 45 : hash_search_with_hash_value(PredicateLockTargetHash,
2078 : &ScratchTargetTag,
2079 : ScratchTargetTagHash,
2080 : HASH_REMOVE, &found);
2081 : Assert(found);
2082 45 : if (!lockheld)
2083 0 : LWLockRelease(ScratchPartitionLock);
2084 45 : }
2085 :
2086 : /*
2087 : * Re-insert the dummy entry in predicate lock target hash.
2088 : */
2089 : static void
2090 45 : RestoreScratchTarget(bool lockheld)
2091 : {
2092 : bool found;
2093 :
2094 : Assert(LWLockHeldByMe(SerializablePredicateListLock));
2095 :
2096 45 : if (!lockheld)
2097 0 : LWLockAcquire(ScratchPartitionLock, LW_EXCLUSIVE);
2098 45 : hash_search_with_hash_value(PredicateLockTargetHash,
2099 : &ScratchTargetTag,
2100 : ScratchTargetTagHash,
2101 : HASH_ENTER, &found);
2102 : Assert(!found);
2103 45 : if (!lockheld)
2104 0 : LWLockRelease(ScratchPartitionLock);
2105 45 : }
2106 :
2107 : /*
2108 : * Check whether the list of related predicate locks is empty for a
2109 : * predicate lock target, and remove the target if it is.
2110 : */
2111 : static void
2112 3941 : RemoveTargetIfNoLongerUsed(PREDICATELOCKTARGET *target, uint32 targettaghash)
2113 : {
2114 : PREDICATELOCKTARGET *rmtarget PG_USED_FOR_ASSERTS_ONLY;
2115 :
2116 : Assert(LWLockHeldByMe(SerializablePredicateListLock));
2117 :
2118 : /* Can't remove it until no locks at this target. */
2119 3941 : if (!dlist_is_empty(&target->predicateLocks))
2120 973 : return;
2121 :
2122 : /* Actually remove the target. */
2123 2968 : rmtarget = hash_search_with_hash_value(PredicateLockTargetHash,
2124 2968 : &target->tag,
2125 : targettaghash,
2126 : HASH_REMOVE, NULL);
2127 : Assert(rmtarget == target);
2128 : }
2129 :
2130 : /*
2131 : * Delete child target locks owned by this process.
2132 : * This implementation is assuming that the usage of each target tag field
2133 : * is uniform. No need to make this hard if we don't have to.
2134 : *
2135 : * We acquire an LWLock in the case of parallel mode, because worker
2136 : * backends have access to the leader's SERIALIZABLEXACT. Otherwise,
2137 : * we aren't acquiring LWLocks for the predicate lock or lock
2138 : * target structures associated with this transaction unless we're going
2139 : * to modify them, because no other process is permitted to modify our
2140 : * locks.
2141 : */
2142 : static void
2143 2379 : DeleteChildTargetLocks(const PREDICATELOCKTARGETTAG *newtargettag)
2144 : {
2145 : SERIALIZABLEXACT *sxact;
2146 : PREDICATELOCK *predlock;
2147 : dlist_mutable_iter iter;
2148 :
2149 2379 : LWLockAcquire(SerializablePredicateListLock, LW_SHARED);
2150 2379 : sxact = MySerializableXact;
2151 2379 : if (IsInParallelMode())
2152 11 : LWLockAcquire(&sxact->perXactPredicateListLock, LW_EXCLUSIVE);
2153 :
2154 7527 : dlist_foreach_modify(iter, &sxact->predicateLocks)
2155 : {
2156 : PREDICATELOCKTAG oldlocktag;
2157 : PREDICATELOCKTARGET *oldtarget;
2158 : PREDICATELOCKTARGETTAG oldtargettag;
2159 :
2160 5148 : predlock = dlist_container(PREDICATELOCK, xactLink, iter.cur);
2161 :
2162 5148 : oldlocktag = predlock->tag;
2163 : Assert(oldlocktag.myXact == sxact);
2164 5148 : oldtarget = oldlocktag.myTarget;
2165 5148 : oldtargettag = oldtarget->tag;
2166 :
2167 5148 : if (TargetTagIsCoveredBy(oldtargettag, *newtargettag))
2168 : {
2169 : uint32 oldtargettaghash;
2170 : LWLock *partitionLock;
2171 : PREDICATELOCK *rmpredlock PG_USED_FOR_ASSERTS_ONLY;
2172 :
2173 679 : oldtargettaghash = PredicateLockTargetTagHashCode(&oldtargettag);
2174 679 : partitionLock = PredicateLockHashPartitionLock(oldtargettaghash);
2175 :
2176 679 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
2177 :
2178 679 : dlist_delete(&predlock->xactLink);
2179 679 : dlist_delete(&predlock->targetLink);
2180 679 : rmpredlock = hash_search_with_hash_value
2181 : (PredicateLockHash,
2182 : &oldlocktag,
2183 679 : PredicateLockHashCodeFromTargetHashCode(&oldlocktag,
2184 : oldtargettaghash),
2185 : HASH_REMOVE, NULL);
2186 : Assert(rmpredlock == predlock);
2187 :
2188 679 : RemoveTargetIfNoLongerUsed(oldtarget, oldtargettaghash);
2189 :
2190 679 : LWLockRelease(partitionLock);
2191 :
2192 679 : DecrementParentLocks(&oldtargettag);
2193 : }
2194 : }
2195 2379 : if (IsInParallelMode())
2196 11 : LWLockRelease(&sxact->perXactPredicateListLock);
2197 2379 : LWLockRelease(SerializablePredicateListLock);
2198 2379 : }
2199 :
2200 : /*
2201 : * Returns the promotion limit for a given predicate lock target. This is the
2202 : * max number of descendant locks allowed before promoting to the specified
2203 : * tag. Note that the limit includes non-direct descendants (e.g., both tuples
2204 : * and pages for a relation lock).
2205 : *
2206 : * Currently the default limit is 2 for a page lock, and half of the value of
2207 : * max_pred_locks_per_transaction - 1 for a relation lock, to match behavior
2208 : * of earlier releases when upgrading.
2209 : *
2210 : * TODO SSI: We should probably add additional GUCs to allow a maximum ratio
2211 : * of page and tuple locks based on the pages in a relation, and the maximum
2212 : * ratio of tuple locks to tuples in a page. This would provide more
2213 : * generally "balanced" allocation of locks to where they are most useful,
2214 : * while still allowing the absolute numbers to prevent one relation from
2215 : * tying up all predicate lock resources.
2216 : */
2217 : static int
2218 4808 : MaxPredicateChildLocks(const PREDICATELOCKTARGETTAG *tag)
2219 : {
2220 4808 : switch (GET_PREDICATELOCKTARGETTAG_TYPE(*tag))
2221 : {
2222 3240 : case PREDLOCKTAG_RELATION:
2223 3240 : return max_predicate_locks_per_relation < 0
2224 : ? (max_predicate_locks_per_xact
2225 3240 : / (-max_predicate_locks_per_relation)) - 1
2226 3240 : : max_predicate_locks_per_relation;
2227 :
2228 1568 : case PREDLOCKTAG_PAGE:
2229 1568 : return max_predicate_locks_per_page;
2230 :
2231 0 : case PREDLOCKTAG_TUPLE:
2232 :
2233 : /*
2234 : * not reachable: nothing is finer-granularity than a tuple, so we
2235 : * should never try to promote to it.
2236 : */
2237 : Assert(false);
2238 0 : return 0;
2239 : }
2240 :
2241 : /* not reachable */
2242 : Assert(false);
2243 0 : return 0;
2244 : }
2245 :
2246 : /*
2247 : * For all ancestors of a newly-acquired predicate lock, increment
2248 : * their child count in the parent hash table. If any of them have
2249 : * more descendants than their promotion threshold, acquire the
2250 : * coarsest such lock.
2251 : *
2252 : * Returns true if a parent lock was acquired and false otherwise.
2253 : */
2254 : static bool
2255 3947 : CheckAndPromotePredicateLockRequest(const PREDICATELOCKTARGETTAG *reqtag)
2256 : {
2257 : PREDICATELOCKTARGETTAG targettag,
2258 : nexttag,
2259 : promotiontag;
2260 : LOCALPREDICATELOCK *parentlock;
2261 : bool found,
2262 : promote;
2263 :
2264 3947 : promote = false;
2265 :
2266 3947 : targettag = *reqtag;
2267 :
2268 : /* check parents iteratively */
2269 12702 : while (GetParentPredicateLockTag(&targettag, &nexttag))
2270 : {
2271 4808 : targettag = nexttag;
2272 4808 : parentlock = (LOCALPREDICATELOCK *) hash_search(LocalPredicateLockHash,
2273 : &targettag,
2274 : HASH_ENTER,
2275 : &found);
2276 4808 : if (!found)
2277 : {
2278 3387 : parentlock->held = false;
2279 3387 : parentlock->childLocks = 1;
2280 : }
2281 : else
2282 1421 : parentlock->childLocks++;
2283 :
2284 4808 : if (parentlock->childLocks >
2285 4808 : MaxPredicateChildLocks(&targettag))
2286 : {
2287 : /*
2288 : * We should promote to this parent lock. Continue to check its
2289 : * ancestors, however, both to get their child counts right and to
2290 : * check whether we should just go ahead and promote to one of
2291 : * them.
2292 : */
2293 173 : promotiontag = targettag;
2294 173 : promote = true;
2295 : }
2296 : }
2297 :
2298 3947 : if (promote)
2299 : {
2300 : /* acquire coarsest ancestor eligible for promotion */
2301 173 : PredicateLockAcquire(&promotiontag);
2302 173 : return true;
2303 : }
2304 : else
2305 3774 : return false;
2306 : }
2307 :
2308 : /*
2309 : * When releasing a lock, decrement the child count on all ancestor
2310 : * locks.
2311 : *
2312 : * This is called only when releasing a lock via
2313 : * DeleteChildTargetLocks (i.e. when a lock becomes redundant because
2314 : * we've acquired its parent, possibly due to promotion) or when a new
2315 : * MVCC write lock makes the predicate lock unnecessary. There's no
2316 : * point in calling it when locks are released at transaction end, as
2317 : * this information is no longer needed.
2318 : */
2319 : static void
2320 1073 : DecrementParentLocks(const PREDICATELOCKTARGETTAG *targettag)
2321 : {
2322 : PREDICATELOCKTARGETTAG parenttag,
2323 : nexttag;
2324 :
2325 1073 : parenttag = *targettag;
2326 :
2327 3219 : while (GetParentPredicateLockTag(&parenttag, &nexttag))
2328 : {
2329 : uint32 targettaghash;
2330 : LOCALPREDICATELOCK *parentlock,
2331 : *rmlock PG_USED_FOR_ASSERTS_ONLY;
2332 :
2333 2146 : parenttag = nexttag;
2334 2146 : targettaghash = PredicateLockTargetTagHashCode(&parenttag);
2335 : parentlock = (LOCALPREDICATELOCK *)
2336 2146 : hash_search_with_hash_value(LocalPredicateLockHash,
2337 : &parenttag, targettaghash,
2338 : HASH_FIND, NULL);
2339 :
2340 : /*
2341 : * There's a small chance the parent lock doesn't exist in the lock
2342 : * table. This can happen if we prematurely removed it because an
2343 : * index split caused the child refcount to be off.
2344 : */
2345 2146 : if (parentlock == NULL)
2346 0 : continue;
2347 :
2348 2146 : parentlock->childLocks--;
2349 :
2350 : /*
2351 : * Under similar circumstances the parent lock's refcount might be
2352 : * zero. This only happens if we're holding that lock (otherwise we
2353 : * would have removed the entry).
2354 : */
2355 2146 : if (parentlock->childLocks < 0)
2356 : {
2357 : Assert(parentlock->held);
2358 0 : parentlock->childLocks = 0;
2359 : }
2360 :
2361 2146 : if ((parentlock->childLocks == 0) && (!parentlock->held))
2362 : {
2363 : rmlock = (LOCALPREDICATELOCK *)
2364 776 : hash_search_with_hash_value(LocalPredicateLockHash,
2365 : &parenttag, targettaghash,
2366 : HASH_REMOVE, NULL);
2367 : Assert(rmlock == parentlock);
2368 : }
2369 : }
2370 1073 : }
2371 :
2372 : /*
2373 : * Indicate that a predicate lock on the given target is held by the
2374 : * specified transaction. Has no effect if the lock is already held.
2375 : *
2376 : * This updates the lock table and the sxact's lock list, and creates
2377 : * the lock target if necessary, but does *not* do anything related to
2378 : * granularity promotion or the local lock table. See
2379 : * PredicateLockAcquire for that.
2380 : */
2381 : static void
2382 3947 : CreatePredicateLock(const PREDICATELOCKTARGETTAG *targettag,
2383 : uint32 targettaghash,
2384 : SERIALIZABLEXACT *sxact)
2385 : {
2386 : PREDICATELOCKTARGET *target;
2387 : PREDICATELOCKTAG locktag;
2388 : PREDICATELOCK *lock;
2389 : LWLock *partitionLock;
2390 : bool found;
2391 :
2392 3947 : partitionLock = PredicateLockHashPartitionLock(targettaghash);
2393 :
2394 3947 : LWLockAcquire(SerializablePredicateListLock, LW_SHARED);
2395 3947 : if (IsInParallelMode())
2396 16 : LWLockAcquire(&sxact->perXactPredicateListLock, LW_EXCLUSIVE);
2397 3947 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
2398 :
2399 : /* Make sure that the target is represented. */
2400 : target = (PREDICATELOCKTARGET *)
2401 3947 : hash_search_with_hash_value(PredicateLockTargetHash,
2402 : targettag, targettaghash,
2403 : HASH_ENTER_NULL, &found);
2404 3947 : if (!target)
2405 0 : ereport(ERROR,
2406 : (errcode(ERRCODE_OUT_OF_MEMORY),
2407 : errmsg("out of shared memory"),
2408 : errhint("You might need to increase \"%s\".", "max_pred_locks_per_transaction")));
2409 3947 : if (!found)
2410 2968 : dlist_init(&target->predicateLocks);
2411 :
2412 : /* We've got the sxact and target, make sure they're joined. */
2413 3947 : locktag.myTarget = target;
2414 3947 : locktag.myXact = sxact;
2415 : lock = (PREDICATELOCK *)
2416 3947 : hash_search_with_hash_value(PredicateLockHash, &locktag,
2417 3947 : PredicateLockHashCodeFromTargetHashCode(&locktag, targettaghash),
2418 : HASH_ENTER_NULL, &found);
2419 3947 : if (!lock)
2420 0 : ereport(ERROR,
2421 : (errcode(ERRCODE_OUT_OF_MEMORY),
2422 : errmsg("out of shared memory"),
2423 : errhint("You might need to increase \"%s\".", "max_pred_locks_per_transaction")));
2424 :
2425 3947 : if (!found)
2426 : {
2427 3941 : dlist_push_tail(&target->predicateLocks, &lock->targetLink);
2428 3941 : dlist_push_tail(&sxact->predicateLocks, &lock->xactLink);
2429 3941 : lock->commitSeqNo = InvalidSerCommitSeqNo;
2430 : }
2431 :
2432 3947 : LWLockRelease(partitionLock);
2433 3947 : if (IsInParallelMode())
2434 16 : LWLockRelease(&sxact->perXactPredicateListLock);
2435 3947 : LWLockRelease(SerializablePredicateListLock);
2436 3947 : }
2437 :
2438 : /*
2439 : * Acquire a predicate lock on the specified target for the current
2440 : * connection if not already held. This updates the local lock table
2441 : * and uses it to implement granularity promotion. It will consolidate
2442 : * multiple locks into a coarser lock if warranted, and will release
2443 : * any finer-grained locks covered by the new one.
2444 : */
2445 : static void
2446 26131 : PredicateLockAcquire(const PREDICATELOCKTARGETTAG *targettag)
2447 : {
2448 : uint32 targettaghash;
2449 : bool found;
2450 : LOCALPREDICATELOCK *locallock;
2451 :
2452 : /* Do we have the lock already, or a covering lock? */
2453 26131 : if (PredicateLockExists(targettag))
2454 22184 : return;
2455 :
2456 8608 : if (CoarserLockCovers(targettag))
2457 4661 : return;
2458 :
2459 : /* the same hash and LW lock apply to the lock target and the local lock. */
2460 3947 : targettaghash = PredicateLockTargetTagHashCode(targettag);
2461 :
2462 : /* Acquire lock in local table */
2463 : locallock = (LOCALPREDICATELOCK *)
2464 3947 : hash_search_with_hash_value(LocalPredicateLockHash,
2465 : targettag, targettaghash,
2466 : HASH_ENTER, &found);
2467 3947 : locallock->held = true;
2468 3947 : if (!found)
2469 3614 : locallock->childLocks = 0;
2470 :
2471 : /* Actually create the lock */
2472 3947 : CreatePredicateLock(targettag, targettaghash, MySerializableXact);
2473 :
2474 : /*
2475 : * Lock has been acquired. Check whether it should be promoted to a
2476 : * coarser granularity, or whether there are finer-granularity locks to
2477 : * clean up.
2478 : */
2479 3947 : if (CheckAndPromotePredicateLockRequest(targettag))
2480 : {
2481 : /*
2482 : * Lock request was promoted to a coarser-granularity lock, and that
2483 : * lock was acquired. It will delete this lock and any of its
2484 : * children, so we're done.
2485 : */
2486 : }
2487 : else
2488 : {
2489 : /* Clean up any finer-granularity locks */
2490 3774 : if (GET_PREDICATELOCKTARGETTAG_TYPE(*targettag) != PREDLOCKTAG_TUPLE)
2491 2379 : DeleteChildTargetLocks(targettag);
2492 : }
2493 : }
2494 :
2495 :
2496 : /*
2497 : * PredicateLockRelation
2498 : *
2499 : * Gets a predicate lock at the relation level.
2500 : * Skip if not in full serializable transaction isolation level.
2501 : * Skip if this is a temporary table.
2502 : * Clear any finer-grained predicate locks this session has on the relation.
2503 : */
2504 : void
2505 472979 : PredicateLockRelation(Relation relation, Snapshot snapshot)
2506 : {
2507 : PREDICATELOCKTARGETTAG tag;
2508 :
2509 472979 : if (!SerializationNeededForRead(relation, snapshot))
2510 472253 : return;
2511 :
2512 726 : SET_PREDICATELOCKTARGETTAG_RELATION(tag,
2513 : relation->rd_locator.dbOid,
2514 : relation->rd_id);
2515 726 : PredicateLockAcquire(&tag);
2516 : }
2517 :
2518 : /*
2519 : * PredicateLockPage
2520 : *
2521 : * Gets a predicate lock at the page level.
2522 : * Skip if not in full serializable transaction isolation level.
2523 : * Skip if this is a temporary table.
2524 : * Skip if a coarser predicate lock already covers this page.
2525 : * Clear any finer-grained predicate locks this session has on the relation.
2526 : */
2527 : void
2528 13314540 : PredicateLockPage(Relation relation, BlockNumber blkno, Snapshot snapshot)
2529 : {
2530 : PREDICATELOCKTARGETTAG tag;
2531 :
2532 13314540 : if (!SerializationNeededForRead(relation, snapshot))
2533 13295604 : return;
2534 :
2535 18936 : SET_PREDICATELOCKTARGETTAG_PAGE(tag,
2536 : relation->rd_locator.dbOid,
2537 : relation->rd_id,
2538 : blkno);
2539 18936 : PredicateLockAcquire(&tag);
2540 : }
2541 :
2542 : /*
2543 : * PredicateLockTID
2544 : *
2545 : * Gets a predicate lock at the tuple level.
2546 : * Skip if not in full serializable transaction isolation level.
2547 : * Skip if this is a temporary table.
2548 : */
2549 : void
2550 23569303 : PredicateLockTID(Relation relation, const ItemPointerData *tid, Snapshot snapshot,
2551 : TransactionId tuple_xid)
2552 : {
2553 : PREDICATELOCKTARGETTAG tag;
2554 :
2555 23569303 : if (!SerializationNeededForRead(relation, snapshot))
2556 23563007 : return;
2557 :
2558 : /*
2559 : * Return if this xact wrote it.
2560 : */
2561 6298 : if (relation->rd_index == NULL)
2562 : {
2563 : /* If we wrote it; we already have a write lock. */
2564 6298 : if (TransactionIdIsCurrentTransactionId(tuple_xid))
2565 2 : return;
2566 : }
2567 :
2568 : /*
2569 : * Do quick-but-not-definitive test for a relation lock first. This will
2570 : * never cause a return when the relation is *not* locked, but will
2571 : * occasionally let the check continue when there really *is* a relation
2572 : * level lock.
2573 : */
2574 6296 : SET_PREDICATELOCKTARGETTAG_RELATION(tag,
2575 : relation->rd_locator.dbOid,
2576 : relation->rd_id);
2577 6296 : if (PredicateLockExists(&tag))
2578 0 : return;
2579 :
2580 6296 : SET_PREDICATELOCKTARGETTAG_TUPLE(tag,
2581 : relation->rd_locator.dbOid,
2582 : relation->rd_id,
2583 : ItemPointerGetBlockNumber(tid),
2584 : ItemPointerGetOffsetNumber(tid));
2585 6296 : PredicateLockAcquire(&tag);
2586 : }
2587 :
2588 :
2589 : /*
2590 : * DeleteLockTarget
2591 : *
2592 : * Remove a predicate lock target along with any locks held for it.
2593 : *
2594 : * Caller must hold SerializablePredicateListLock and the
2595 : * appropriate hash partition lock for the target.
2596 : */
2597 : static void
2598 0 : DeleteLockTarget(PREDICATELOCKTARGET *target, uint32 targettaghash)
2599 : {
2600 : dlist_mutable_iter iter;
2601 :
2602 : Assert(LWLockHeldByMeInMode(SerializablePredicateListLock,
2603 : LW_EXCLUSIVE));
2604 : Assert(LWLockHeldByMe(PredicateLockHashPartitionLock(targettaghash)));
2605 :
2606 0 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
2607 :
2608 0 : dlist_foreach_modify(iter, &target->predicateLocks)
2609 : {
2610 0 : PREDICATELOCK *predlock =
2611 0 : dlist_container(PREDICATELOCK, targetLink, iter.cur);
2612 : bool found;
2613 :
2614 0 : dlist_delete(&(predlock->xactLink));
2615 0 : dlist_delete(&(predlock->targetLink));
2616 :
2617 0 : hash_search_with_hash_value
2618 : (PredicateLockHash,
2619 0 : &predlock->tag,
2620 0 : PredicateLockHashCodeFromTargetHashCode(&predlock->tag,
2621 : targettaghash),
2622 : HASH_REMOVE, &found);
2623 : Assert(found);
2624 : }
2625 0 : LWLockRelease(SerializableXactHashLock);
2626 :
2627 : /* Remove the target itself, if possible. */
2628 0 : RemoveTargetIfNoLongerUsed(target, targettaghash);
2629 0 : }
2630 :
2631 :
2632 : /*
2633 : * TransferPredicateLocksToNewTarget
2634 : *
2635 : * Move or copy all the predicate locks for a lock target, for use by
2636 : * index page splits/combines and other things that create or replace
2637 : * lock targets. If 'removeOld' is true, the old locks and the target
2638 : * will be removed.
2639 : *
2640 : * Returns true on success, or false if we ran out of shared memory to
2641 : * allocate the new target or locks. Guaranteed to always succeed if
2642 : * removeOld is set (by using the scratch entry in PredicateLockTargetHash
2643 : * for scratch space).
2644 : *
2645 : * Warning: the "removeOld" option should be used only with care,
2646 : * because this function does not (indeed, can not) update other
2647 : * backends' LocalPredicateLockHash. If we are only adding new
2648 : * entries, this is not a problem: the local lock table is used only
2649 : * as a hint, so missing entries for locks that are held are
2650 : * OK. Having entries for locks that are no longer held, as can happen
2651 : * when using "removeOld", is not in general OK. We can only use it
2652 : * safely when replacing a lock with a coarser-granularity lock that
2653 : * covers it, or if we are absolutely certain that no one will need to
2654 : * refer to that lock in the future.
2655 : *
2656 : * Caller must hold SerializablePredicateListLock exclusively.
2657 : */
2658 : static bool
2659 3 : TransferPredicateLocksToNewTarget(PREDICATELOCKTARGETTAG oldtargettag,
2660 : PREDICATELOCKTARGETTAG newtargettag,
2661 : bool removeOld)
2662 : {
2663 : uint32 oldtargettaghash;
2664 : LWLock *oldpartitionLock;
2665 : PREDICATELOCKTARGET *oldtarget;
2666 : uint32 newtargettaghash;
2667 : LWLock *newpartitionLock;
2668 : bool found;
2669 3 : bool outOfShmem = false;
2670 :
2671 : Assert(LWLockHeldByMeInMode(SerializablePredicateListLock,
2672 : LW_EXCLUSIVE));
2673 :
2674 3 : oldtargettaghash = PredicateLockTargetTagHashCode(&oldtargettag);
2675 3 : newtargettaghash = PredicateLockTargetTagHashCode(&newtargettag);
2676 3 : oldpartitionLock = PredicateLockHashPartitionLock(oldtargettaghash);
2677 3 : newpartitionLock = PredicateLockHashPartitionLock(newtargettaghash);
2678 :
2679 3 : if (removeOld)
2680 : {
2681 : /*
2682 : * Remove the dummy entry to give us scratch space, so we know we'll
2683 : * be able to create the new lock target.
2684 : */
2685 0 : RemoveScratchTarget(false);
2686 : }
2687 :
2688 : /*
2689 : * We must get the partition locks in ascending sequence to avoid
2690 : * deadlocks. If old and new partitions are the same, we must request the
2691 : * lock only once.
2692 : */
2693 3 : if (oldpartitionLock < newpartitionLock)
2694 : {
2695 3 : LWLockAcquire(oldpartitionLock,
2696 3 : (removeOld ? LW_EXCLUSIVE : LW_SHARED));
2697 3 : LWLockAcquire(newpartitionLock, LW_EXCLUSIVE);
2698 : }
2699 0 : else if (oldpartitionLock > newpartitionLock)
2700 : {
2701 0 : LWLockAcquire(newpartitionLock, LW_EXCLUSIVE);
2702 0 : LWLockAcquire(oldpartitionLock,
2703 0 : (removeOld ? LW_EXCLUSIVE : LW_SHARED));
2704 : }
2705 : else
2706 0 : LWLockAcquire(newpartitionLock, LW_EXCLUSIVE);
2707 :
2708 : /*
2709 : * Look for the old target. If not found, that's OK; no predicate locks
2710 : * are affected, so we can just clean up and return. If it does exist,
2711 : * walk its list of predicate locks and move or copy them to the new
2712 : * target.
2713 : */
2714 3 : oldtarget = hash_search_with_hash_value(PredicateLockTargetHash,
2715 : &oldtargettag,
2716 : oldtargettaghash,
2717 : HASH_FIND, NULL);
2718 :
2719 3 : if (oldtarget)
2720 : {
2721 : PREDICATELOCKTARGET *newtarget;
2722 : PREDICATELOCKTAG newpredlocktag;
2723 : dlist_mutable_iter iter;
2724 :
2725 0 : newtarget = hash_search_with_hash_value(PredicateLockTargetHash,
2726 : &newtargettag,
2727 : newtargettaghash,
2728 : HASH_ENTER_NULL, &found);
2729 :
2730 0 : if (!newtarget)
2731 : {
2732 : /* Failed to allocate due to insufficient shmem */
2733 0 : outOfShmem = true;
2734 0 : goto exit;
2735 : }
2736 :
2737 : /* If we created a new entry, initialize it */
2738 0 : if (!found)
2739 0 : dlist_init(&newtarget->predicateLocks);
2740 :
2741 0 : newpredlocktag.myTarget = newtarget;
2742 :
2743 : /*
2744 : * Loop through all the locks on the old target, replacing them with
2745 : * locks on the new target.
2746 : */
2747 0 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
2748 :
2749 0 : dlist_foreach_modify(iter, &oldtarget->predicateLocks)
2750 : {
2751 0 : PREDICATELOCK *oldpredlock =
2752 0 : dlist_container(PREDICATELOCK, targetLink, iter.cur);
2753 : PREDICATELOCK *newpredlock;
2754 0 : SerCommitSeqNo oldCommitSeqNo = oldpredlock->commitSeqNo;
2755 :
2756 0 : newpredlocktag.myXact = oldpredlock->tag.myXact;
2757 :
2758 0 : if (removeOld)
2759 : {
2760 0 : dlist_delete(&(oldpredlock->xactLink));
2761 0 : dlist_delete(&(oldpredlock->targetLink));
2762 :
2763 0 : hash_search_with_hash_value
2764 : (PredicateLockHash,
2765 0 : &oldpredlock->tag,
2766 0 : PredicateLockHashCodeFromTargetHashCode(&oldpredlock->tag,
2767 : oldtargettaghash),
2768 : HASH_REMOVE, &found);
2769 : Assert(found);
2770 : }
2771 :
2772 : newpredlock = (PREDICATELOCK *)
2773 0 : hash_search_with_hash_value(PredicateLockHash,
2774 : &newpredlocktag,
2775 0 : PredicateLockHashCodeFromTargetHashCode(&newpredlocktag,
2776 : newtargettaghash),
2777 : HASH_ENTER_NULL,
2778 : &found);
2779 0 : if (!newpredlock)
2780 : {
2781 : /* Out of shared memory. Undo what we've done so far. */
2782 0 : LWLockRelease(SerializableXactHashLock);
2783 0 : DeleteLockTarget(newtarget, newtargettaghash);
2784 0 : outOfShmem = true;
2785 0 : goto exit;
2786 : }
2787 0 : if (!found)
2788 : {
2789 0 : dlist_push_tail(&(newtarget->predicateLocks),
2790 : &(newpredlock->targetLink));
2791 0 : dlist_push_tail(&(newpredlocktag.myXact->predicateLocks),
2792 : &(newpredlock->xactLink));
2793 0 : newpredlock->commitSeqNo = oldCommitSeqNo;
2794 : }
2795 : else
2796 : {
2797 0 : if (newpredlock->commitSeqNo < oldCommitSeqNo)
2798 0 : newpredlock->commitSeqNo = oldCommitSeqNo;
2799 : }
2800 :
2801 : Assert(newpredlock->commitSeqNo != 0);
2802 : Assert((newpredlock->commitSeqNo == InvalidSerCommitSeqNo)
2803 : || (newpredlock->tag.myXact == OldCommittedSxact));
2804 : }
2805 0 : LWLockRelease(SerializableXactHashLock);
2806 :
2807 0 : if (removeOld)
2808 : {
2809 : Assert(dlist_is_empty(&oldtarget->predicateLocks));
2810 0 : RemoveTargetIfNoLongerUsed(oldtarget, oldtargettaghash);
2811 : }
2812 : }
2813 :
2814 :
2815 3 : exit:
2816 : /* Release partition locks in reverse order of acquisition. */
2817 3 : if (oldpartitionLock < newpartitionLock)
2818 : {
2819 3 : LWLockRelease(newpartitionLock);
2820 3 : LWLockRelease(oldpartitionLock);
2821 : }
2822 0 : else if (oldpartitionLock > newpartitionLock)
2823 : {
2824 0 : LWLockRelease(oldpartitionLock);
2825 0 : LWLockRelease(newpartitionLock);
2826 : }
2827 : else
2828 0 : LWLockRelease(newpartitionLock);
2829 :
2830 3 : if (removeOld)
2831 : {
2832 : /* We shouldn't run out of memory if we're moving locks */
2833 : Assert(!outOfShmem);
2834 :
2835 : /* Put the scratch entry back */
2836 0 : RestoreScratchTarget(false);
2837 : }
2838 :
2839 3 : return !outOfShmem;
2840 : }
2841 :
2842 : /*
2843 : * Drop all predicate locks of any granularity from the specified relation,
2844 : * which can be a heap relation or an index relation. If 'transfer' is true,
2845 : * acquire a relation lock on the heap for any transactions with any lock(s)
2846 : * on the specified relation.
2847 : *
2848 : * This requires grabbing a lot of LW locks and scanning the entire lock
2849 : * target table for matches. That makes this more expensive than most
2850 : * predicate lock management functions, but it will only be called for DDL
2851 : * type commands that are expensive anyway, and there are fast returns when
2852 : * no serializable transactions are active or the relation is temporary.
2853 : *
2854 : * We don't use the TransferPredicateLocksToNewTarget function because it
2855 : * acquires its own locks on the partitions of the two targets involved,
2856 : * and we'll already be holding all partition locks.
2857 : *
2858 : * We can't throw an error from here, because the call could be from a
2859 : * transaction which is not serializable.
2860 : *
2861 : * NOTE: This is currently only called with transfer set to true, but that may
2862 : * change. If we decide to clean up the locks from a table on commit of a
2863 : * transaction which executed DROP TABLE, the false condition will be useful.
2864 : */
2865 : static void
2866 22664 : DropAllPredicateLocksFromTable(Relation relation, bool transfer)
2867 : {
2868 : HASH_SEQ_STATUS seqstat;
2869 : PREDICATELOCKTARGET *oldtarget;
2870 : PREDICATELOCKTARGET *heaptarget;
2871 : Oid dbId;
2872 : Oid relId;
2873 : Oid heapId;
2874 : int i;
2875 : bool isIndex;
2876 : bool found;
2877 : uint32 heaptargettaghash;
2878 :
2879 : /*
2880 : * Bail out quickly if there are no serializable transactions running.
2881 : * It's safe to check this without taking locks because the caller is
2882 : * holding an ACCESS EXCLUSIVE lock on the relation. No new locks which
2883 : * would matter here can be acquired while that is held.
2884 : */
2885 22664 : if (!TransactionIdIsValid(PredXact->SxactGlobalXmin))
2886 22619 : return;
2887 :
2888 65 : if (!PredicateLockingNeededForRelation(relation))
2889 20 : return;
2890 :
2891 45 : dbId = relation->rd_locator.dbOid;
2892 45 : relId = relation->rd_id;
2893 45 : if (relation->rd_index == NULL)
2894 : {
2895 5 : isIndex = false;
2896 5 : heapId = relId;
2897 : }
2898 : else
2899 : {
2900 40 : isIndex = true;
2901 40 : heapId = relation->rd_index->indrelid;
2902 : }
2903 : Assert(heapId != InvalidOid);
2904 : Assert(transfer || !isIndex); /* index OID only makes sense with
2905 : * transfer */
2906 :
2907 : /* Retrieve first time needed, then keep. */
2908 45 : heaptargettaghash = 0;
2909 45 : heaptarget = NULL;
2910 :
2911 : /* Acquire locks on all lock partitions */
2912 45 : LWLockAcquire(SerializablePredicateListLock, LW_EXCLUSIVE);
2913 765 : for (i = 0; i < NUM_PREDICATELOCK_PARTITIONS; i++)
2914 720 : LWLockAcquire(PredicateLockHashPartitionLockByIndex(i), LW_EXCLUSIVE);
2915 45 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
2916 :
2917 : /*
2918 : * Remove the dummy entry to give us scratch space, so we know we'll be
2919 : * able to create the new lock target.
2920 : */
2921 45 : if (transfer)
2922 45 : RemoveScratchTarget(true);
2923 :
2924 : /* Scan through target map */
2925 45 : hash_seq_init(&seqstat, PredicateLockTargetHash);
2926 :
2927 90 : while ((oldtarget = (PREDICATELOCKTARGET *) hash_seq_search(&seqstat)))
2928 : {
2929 : dlist_mutable_iter iter;
2930 :
2931 : /*
2932 : * Check whether this is a target which needs attention.
2933 : */
2934 45 : if (GET_PREDICATELOCKTARGETTAG_RELATION(oldtarget->tag) != relId)
2935 45 : continue; /* wrong relation id */
2936 0 : if (GET_PREDICATELOCKTARGETTAG_DB(oldtarget->tag) != dbId)
2937 0 : continue; /* wrong database id */
2938 0 : if (transfer && !isIndex
2939 0 : && GET_PREDICATELOCKTARGETTAG_TYPE(oldtarget->tag) == PREDLOCKTAG_RELATION)
2940 0 : continue; /* already the right lock */
2941 :
2942 : /*
2943 : * If we made it here, we have work to do. We make sure the heap
2944 : * relation lock exists, then we walk the list of predicate locks for
2945 : * the old target we found, moving all locks to the heap relation lock
2946 : * -- unless they already hold that.
2947 : */
2948 :
2949 : /*
2950 : * First make sure we have the heap relation target. We only need to
2951 : * do this once.
2952 : */
2953 0 : if (transfer && heaptarget == NULL)
2954 : {
2955 : PREDICATELOCKTARGETTAG heaptargettag;
2956 :
2957 0 : SET_PREDICATELOCKTARGETTAG_RELATION(heaptargettag, dbId, heapId);
2958 0 : heaptargettaghash = PredicateLockTargetTagHashCode(&heaptargettag);
2959 0 : heaptarget = hash_search_with_hash_value(PredicateLockTargetHash,
2960 : &heaptargettag,
2961 : heaptargettaghash,
2962 : HASH_ENTER, &found);
2963 0 : if (!found)
2964 0 : dlist_init(&heaptarget->predicateLocks);
2965 : }
2966 :
2967 : /*
2968 : * Loop through all the locks on the old target, replacing them with
2969 : * locks on the new target.
2970 : */
2971 0 : dlist_foreach_modify(iter, &oldtarget->predicateLocks)
2972 : {
2973 0 : PREDICATELOCK *oldpredlock =
2974 0 : dlist_container(PREDICATELOCK, targetLink, iter.cur);
2975 : PREDICATELOCK *newpredlock;
2976 : SerCommitSeqNo oldCommitSeqNo;
2977 : SERIALIZABLEXACT *oldXact;
2978 :
2979 : /*
2980 : * Remove the old lock first. This avoids the chance of running
2981 : * out of lock structure entries for the hash table.
2982 : */
2983 0 : oldCommitSeqNo = oldpredlock->commitSeqNo;
2984 0 : oldXact = oldpredlock->tag.myXact;
2985 :
2986 0 : dlist_delete(&(oldpredlock->xactLink));
2987 :
2988 : /*
2989 : * No need for retail delete from oldtarget list, we're removing
2990 : * the whole target anyway.
2991 : */
2992 0 : hash_search(PredicateLockHash,
2993 0 : &oldpredlock->tag,
2994 : HASH_REMOVE, &found);
2995 : Assert(found);
2996 :
2997 0 : if (transfer)
2998 : {
2999 : PREDICATELOCKTAG newpredlocktag;
3000 :
3001 0 : newpredlocktag.myTarget = heaptarget;
3002 0 : newpredlocktag.myXact = oldXact;
3003 : newpredlock = (PREDICATELOCK *)
3004 0 : hash_search_with_hash_value(PredicateLockHash,
3005 : &newpredlocktag,
3006 0 : PredicateLockHashCodeFromTargetHashCode(&newpredlocktag,
3007 : heaptargettaghash),
3008 : HASH_ENTER,
3009 : &found);
3010 0 : if (!found)
3011 : {
3012 0 : dlist_push_tail(&(heaptarget->predicateLocks),
3013 : &(newpredlock->targetLink));
3014 0 : dlist_push_tail(&(newpredlocktag.myXact->predicateLocks),
3015 : &(newpredlock->xactLink));
3016 0 : newpredlock->commitSeqNo = oldCommitSeqNo;
3017 : }
3018 : else
3019 : {
3020 0 : if (newpredlock->commitSeqNo < oldCommitSeqNo)
3021 0 : newpredlock->commitSeqNo = oldCommitSeqNo;
3022 : }
3023 :
3024 : Assert(newpredlock->commitSeqNo != 0);
3025 : Assert((newpredlock->commitSeqNo == InvalidSerCommitSeqNo)
3026 : || (newpredlock->tag.myXact == OldCommittedSxact));
3027 : }
3028 : }
3029 :
3030 0 : hash_search(PredicateLockTargetHash, &oldtarget->tag, HASH_REMOVE,
3031 : &found);
3032 : Assert(found);
3033 : }
3034 :
3035 : /* Put the scratch entry back */
3036 45 : if (transfer)
3037 45 : RestoreScratchTarget(true);
3038 :
3039 : /* Release locks in reverse order */
3040 45 : LWLockRelease(SerializableXactHashLock);
3041 765 : for (i = NUM_PREDICATELOCK_PARTITIONS - 1; i >= 0; i--)
3042 720 : LWLockRelease(PredicateLockHashPartitionLockByIndex(i));
3043 45 : LWLockRelease(SerializablePredicateListLock);
3044 : }
3045 :
3046 : /*
3047 : * TransferPredicateLocksToHeapRelation
3048 : * For all transactions, transfer all predicate locks for the given
3049 : * relation to a single relation lock on the heap.
3050 : */
3051 : void
3052 22664 : TransferPredicateLocksToHeapRelation(Relation relation)
3053 : {
3054 22664 : DropAllPredicateLocksFromTable(relation, true);
3055 22664 : }
3056 :
3057 :
3058 : /*
3059 : * PredicateLockPageSplit
3060 : *
3061 : * Copies any predicate locks for the old page to the new page.
3062 : * Skip if this is a temporary table or toast table.
3063 : *
3064 : * NOTE: A page split (or overflow) affects all serializable transactions,
3065 : * even if it occurs in the context of another transaction isolation level.
3066 : *
3067 : * NOTE: This currently leaves the local copy of the locks without
3068 : * information on the new lock which is in shared memory. This could cause
3069 : * problems if enough page splits occur on locked pages without the processes
3070 : * which hold the locks getting in and noticing.
3071 : */
3072 : void
3073 37976 : PredicateLockPageSplit(Relation relation, BlockNumber oldblkno,
3074 : BlockNumber newblkno)
3075 : {
3076 : PREDICATELOCKTARGETTAG oldtargettag;
3077 : PREDICATELOCKTARGETTAG newtargettag;
3078 : bool success;
3079 :
3080 : /*
3081 : * Bail out quickly if there are no serializable transactions running.
3082 : *
3083 : * It's safe to do this check without taking any additional locks. Even if
3084 : * a serializable transaction starts concurrently, we know it can't take
3085 : * any SIREAD locks on the page being split because the caller is holding
3086 : * the associated buffer page lock. Memory reordering isn't an issue; the
3087 : * memory barrier in the LWLock acquisition guarantees that this read
3088 : * occurs while the buffer page lock is held.
3089 : */
3090 37976 : if (!TransactionIdIsValid(PredXact->SxactGlobalXmin))
3091 37973 : return;
3092 :
3093 16 : if (!PredicateLockingNeededForRelation(relation))
3094 13 : return;
3095 :
3096 : Assert(oldblkno != newblkno);
3097 : Assert(BlockNumberIsValid(oldblkno));
3098 : Assert(BlockNumberIsValid(newblkno));
3099 :
3100 3 : SET_PREDICATELOCKTARGETTAG_PAGE(oldtargettag,
3101 : relation->rd_locator.dbOid,
3102 : relation->rd_id,
3103 : oldblkno);
3104 3 : SET_PREDICATELOCKTARGETTAG_PAGE(newtargettag,
3105 : relation->rd_locator.dbOid,
3106 : relation->rd_id,
3107 : newblkno);
3108 :
3109 3 : LWLockAcquire(SerializablePredicateListLock, LW_EXCLUSIVE);
3110 :
3111 : /*
3112 : * Try copying the locks over to the new page's tag, creating it if
3113 : * necessary.
3114 : */
3115 3 : success = TransferPredicateLocksToNewTarget(oldtargettag,
3116 : newtargettag,
3117 : false);
3118 :
3119 3 : if (!success)
3120 : {
3121 : /*
3122 : * No more predicate lock entries are available. Failure isn't an
3123 : * option here, so promote the page lock to a relation lock.
3124 : */
3125 :
3126 : /* Get the parent relation lock's lock tag */
3127 0 : success = GetParentPredicateLockTag(&oldtargettag,
3128 : &newtargettag);
3129 : Assert(success);
3130 :
3131 : /*
3132 : * Move the locks to the parent. This shouldn't fail.
3133 : *
3134 : * Note that here we are removing locks held by other backends,
3135 : * leading to a possible inconsistency in their local lock hash table.
3136 : * This is OK because we're replacing it with a lock that covers the
3137 : * old one.
3138 : */
3139 0 : success = TransferPredicateLocksToNewTarget(oldtargettag,
3140 : newtargettag,
3141 : true);
3142 : Assert(success);
3143 : }
3144 :
3145 3 : LWLockRelease(SerializablePredicateListLock);
3146 : }
3147 :
3148 : /*
3149 : * PredicateLockPageCombine
3150 : *
3151 : * Combines predicate locks for two existing pages.
3152 : * Skip if this is a temporary table or toast table.
3153 : *
3154 : * NOTE: A page combine affects all serializable transactions, even if it
3155 : * occurs in the context of another transaction isolation level.
3156 : */
3157 : void
3158 3461 : PredicateLockPageCombine(Relation relation, BlockNumber oldblkno,
3159 : BlockNumber newblkno)
3160 : {
3161 : /*
3162 : * Page combines differ from page splits in that we ought to be able to
3163 : * remove the locks on the old page after transferring them to the new
3164 : * page, instead of duplicating them. However, because we can't edit other
3165 : * backends' local lock tables, removing the old lock would leave them
3166 : * with an entry in their LocalPredicateLockHash for a lock they're not
3167 : * holding, which isn't acceptable. So we wind up having to do the same
3168 : * work as a page split, acquiring a lock on the new page and keeping the
3169 : * old page locked too. That can lead to some false positives, but should
3170 : * be rare in practice.
3171 : */
3172 3461 : PredicateLockPageSplit(relation, oldblkno, newblkno);
3173 3461 : }
3174 :
3175 : /*
3176 : * Walk the list of in-progress serializable transactions and find the new
3177 : * xmin.
3178 : */
3179 : static void
3180 893 : SetNewSxactGlobalXmin(void)
3181 : {
3182 : dlist_iter iter;
3183 :
3184 : Assert(LWLockHeldByMe(SerializableXactHashLock));
3185 :
3186 893 : PredXact->SxactGlobalXmin = InvalidTransactionId;
3187 893 : PredXact->SxactGlobalXminCount = 0;
3188 :
3189 3378 : dlist_foreach(iter, &PredXact->activeList)
3190 : {
3191 2485 : SERIALIZABLEXACT *sxact =
3192 2485 : dlist_container(SERIALIZABLEXACT, xactLink, iter.cur);
3193 :
3194 2485 : if (!SxactIsRolledBack(sxact)
3195 2179 : && !SxactIsCommitted(sxact)
3196 19 : && sxact != OldCommittedSxact)
3197 : {
3198 : Assert(sxact->xmin != InvalidTransactionId);
3199 19 : if (!TransactionIdIsValid(PredXact->SxactGlobalXmin)
3200 0 : || TransactionIdPrecedes(sxact->xmin,
3201 0 : PredXact->SxactGlobalXmin))
3202 : {
3203 19 : PredXact->SxactGlobalXmin = sxact->xmin;
3204 19 : PredXact->SxactGlobalXminCount = 1;
3205 : }
3206 0 : else if (TransactionIdEquals(sxact->xmin,
3207 : PredXact->SxactGlobalXmin))
3208 0 : PredXact->SxactGlobalXminCount++;
3209 : }
3210 : }
3211 :
3212 893 : SerialSetActiveSerXmin(PredXact->SxactGlobalXmin);
3213 893 : }
3214 :
3215 : /*
3216 : * ReleasePredicateLocks
3217 : *
3218 : * Releases predicate locks based on completion of the current transaction,
3219 : * whether committed or rolled back. It can also be called for a read only
3220 : * transaction when it becomes impossible for the transaction to become
3221 : * part of a dangerous structure.
3222 : *
3223 : * We do nothing unless this is a serializable transaction.
3224 : *
3225 : * This method must ensure that shared memory hash tables are cleaned
3226 : * up in some relatively timely fashion.
3227 : *
3228 : * If this transaction is committing and is holding any predicate locks,
3229 : * it must be added to a list of completed serializable transactions still
3230 : * holding locks.
3231 : *
3232 : * If isReadOnlySafe is true, then predicate locks are being released before
3233 : * the end of the transaction because MySerializableXact has been determined
3234 : * to be RO_SAFE. In non-parallel mode we can release it completely, but it
3235 : * in parallel mode we partially release the SERIALIZABLEXACT and keep it
3236 : * around until the end of the transaction, allowing each backend to clear its
3237 : * MySerializableXact variable and benefit from the optimization in its own
3238 : * time.
3239 : */
3240 : void
3241 629209 : ReleasePredicateLocks(bool isCommit, bool isReadOnlySafe)
3242 : {
3243 629209 : bool partiallyReleasing = false;
3244 : bool needToClear;
3245 : SERIALIZABLEXACT *roXact;
3246 : dlist_mutable_iter iter;
3247 :
3248 : /*
3249 : * We can't trust XactReadOnly here, because a transaction which started
3250 : * as READ WRITE can show as READ ONLY later, e.g., within
3251 : * subtransactions. We want to flag a transaction as READ ONLY if it
3252 : * commits without writing so that de facto READ ONLY transactions get the
3253 : * benefit of some RO optimizations, so we will use this local variable to
3254 : * get some cleanup logic right which is based on whether the transaction
3255 : * was declared READ ONLY at the top level.
3256 : */
3257 : bool topLevelIsDeclaredReadOnly;
3258 :
3259 : /* We can't be both committing and releasing early due to RO_SAFE. */
3260 : Assert(!(isCommit && isReadOnlySafe));
3261 :
3262 : /* Are we at the end of a transaction, that is, a commit or abort? */
3263 629209 : if (!isReadOnlySafe)
3264 : {
3265 : /*
3266 : * Parallel workers mustn't release predicate locks at the end of
3267 : * their transaction. The leader will do that at the end of its
3268 : * transaction.
3269 : */
3270 629174 : if (IsParallelWorker())
3271 : {
3272 6030 : ReleasePredicateLocksLocal();
3273 627630 : return;
3274 : }
3275 :
3276 : /*
3277 : * By the time the leader in a parallel query reaches end of
3278 : * transaction, it has waited for all workers to exit.
3279 : */
3280 : Assert(!ParallelContextActive());
3281 :
3282 : /*
3283 : * If the leader in a parallel query earlier stashed a partially
3284 : * released SERIALIZABLEXACT for final clean-up at end of transaction
3285 : * (because workers might still have been accessing it), then it's
3286 : * time to restore it.
3287 : */
3288 623144 : if (SavedSerializableXact != InvalidSerializableXact)
3289 : {
3290 : Assert(MySerializableXact == InvalidSerializableXact);
3291 1 : MySerializableXact = SavedSerializableXact;
3292 1 : SavedSerializableXact = InvalidSerializableXact;
3293 : Assert(SxactIsPartiallyReleased(MySerializableXact));
3294 : }
3295 : }
3296 :
3297 623179 : if (MySerializableXact == InvalidSerializableXact)
3298 : {
3299 : Assert(LocalPredicateLockHash == NULL);
3300 621597 : return;
3301 : }
3302 :
3303 1582 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
3304 :
3305 : /*
3306 : * If the transaction is committing, but it has been partially released
3307 : * already, then treat this as a roll back. It was marked as rolled back.
3308 : */
3309 1582 : if (isCommit && SxactIsPartiallyReleased(MySerializableXact))
3310 2 : isCommit = false;
3311 :
3312 : /*
3313 : * If we're called in the middle of a transaction because we discovered
3314 : * that the SXACT_FLAG_RO_SAFE flag was set, then we'll partially release
3315 : * it (that is, release the predicate locks and conflicts, but not the
3316 : * SERIALIZABLEXACT itself) if we're the first backend to have noticed.
3317 : */
3318 1582 : if (isReadOnlySafe && IsInParallelMode())
3319 : {
3320 : /*
3321 : * The leader needs to stash a pointer to it, so that it can
3322 : * completely release it at end-of-transaction.
3323 : */
3324 5 : if (!IsParallelWorker())
3325 1 : SavedSerializableXact = MySerializableXact;
3326 :
3327 : /*
3328 : * The first backend to reach this condition will partially release
3329 : * the SERIALIZABLEXACT. All others will just clear their
3330 : * backend-local state so that they stop doing SSI checks for the rest
3331 : * of the transaction.
3332 : */
3333 5 : if (SxactIsPartiallyReleased(MySerializableXact))
3334 : {
3335 3 : LWLockRelease(SerializableXactHashLock);
3336 3 : ReleasePredicateLocksLocal();
3337 3 : return;
3338 : }
3339 : else
3340 : {
3341 2 : MySerializableXact->flags |= SXACT_FLAG_PARTIALLY_RELEASED;
3342 2 : partiallyReleasing = true;
3343 : /* ... and proceed to perform the partial release below. */
3344 : }
3345 : }
3346 : Assert(!isCommit || SxactIsPrepared(MySerializableXact));
3347 : Assert(!isCommit || !SxactIsDoomed(MySerializableXact));
3348 : Assert(!SxactIsCommitted(MySerializableXact));
3349 : Assert(SxactIsPartiallyReleased(MySerializableXact)
3350 : || !SxactIsRolledBack(MySerializableXact));
3351 :
3352 : /* may not be serializable during COMMIT/ROLLBACK PREPARED */
3353 : Assert(MySerializableXact->pid == 0 || IsolationIsSerializable());
3354 :
3355 : /* We'd better not already be on the cleanup list. */
3356 : Assert(!SxactIsOnFinishedList(MySerializableXact));
3357 :
3358 1579 : topLevelIsDeclaredReadOnly = SxactIsReadOnly(MySerializableXact);
3359 :
3360 : /*
3361 : * We don't hold XidGenLock lock here, assuming that TransactionId is
3362 : * atomic!
3363 : *
3364 : * If this value is changing, we don't care that much whether we get the
3365 : * old or new value -- it is just used to determine how far
3366 : * SxactGlobalXmin must advance before this transaction can be fully
3367 : * cleaned up. The worst that could happen is we wait for one more
3368 : * transaction to complete before freeing some RAM; correctness of visible
3369 : * behavior is not affected.
3370 : */
3371 1579 : MySerializableXact->finishedBefore = XidFromFullTransactionId(TransamVariables->nextXid);
3372 :
3373 : /*
3374 : * If it's not a commit it's either a rollback or a read-only transaction
3375 : * flagged SXACT_FLAG_RO_SAFE, and we can clear our locks immediately.
3376 : */
3377 1579 : if (isCommit)
3378 : {
3379 1248 : MySerializableXact->flags |= SXACT_FLAG_COMMITTED;
3380 1248 : MySerializableXact->commitSeqNo = ++(PredXact->LastSxactCommitSeqNo);
3381 : /* Recognize implicit read-only transaction (commit without write). */
3382 1248 : if (!MyXactDidWrite)
3383 239 : MySerializableXact->flags |= SXACT_FLAG_READ_ONLY;
3384 : }
3385 : else
3386 : {
3387 : /*
3388 : * The DOOMED flag indicates that we intend to roll back this
3389 : * transaction and so it should not cause serialization failures for
3390 : * other transactions that conflict with it. Note that this flag might
3391 : * already be set, if another backend marked this transaction for
3392 : * abort.
3393 : *
3394 : * The ROLLED_BACK flag further indicates that ReleasePredicateLocks
3395 : * has been called, and so the SerializableXact is eligible for
3396 : * cleanup. This means it should not be considered when calculating
3397 : * SxactGlobalXmin.
3398 : */
3399 331 : MySerializableXact->flags |= SXACT_FLAG_DOOMED;
3400 331 : MySerializableXact->flags |= SXACT_FLAG_ROLLED_BACK;
3401 :
3402 : /*
3403 : * If the transaction was previously prepared, but is now failing due
3404 : * to a ROLLBACK PREPARED or (hopefully very rare) error after the
3405 : * prepare, clear the prepared flag. This simplifies conflict
3406 : * checking.
3407 : */
3408 331 : MySerializableXact->flags &= ~SXACT_FLAG_PREPARED;
3409 : }
3410 :
3411 1579 : if (!topLevelIsDeclaredReadOnly)
3412 : {
3413 : Assert(PredXact->WritableSxactCount > 0);
3414 1469 : if (--(PredXact->WritableSxactCount) == 0)
3415 : {
3416 : /*
3417 : * Release predicate locks and rw-conflicts in for all committed
3418 : * transactions. There are no longer any transactions which might
3419 : * conflict with the locks and no chance for new transactions to
3420 : * overlap. Similarly, existing conflicts in can't cause pivots,
3421 : * and any conflicts in which could have completed a dangerous
3422 : * structure would already have caused a rollback, so any
3423 : * remaining ones must be benign.
3424 : */
3425 884 : PredXact->CanPartialClearThrough = PredXact->LastSxactCommitSeqNo;
3426 : }
3427 : }
3428 : else
3429 : {
3430 : /*
3431 : * Read-only transactions: clear the list of transactions that might
3432 : * make us unsafe. Note that we use 'inLink' for the iteration as
3433 : * opposed to 'outLink' for the r/w xacts.
3434 : */
3435 152 : dlist_foreach_modify(iter, &MySerializableXact->possibleUnsafeConflicts)
3436 : {
3437 42 : RWConflict possibleUnsafeConflict =
3438 42 : dlist_container(RWConflictData, inLink, iter.cur);
3439 :
3440 : Assert(!SxactIsReadOnly(possibleUnsafeConflict->sxactOut));
3441 : Assert(MySerializableXact == possibleUnsafeConflict->sxactIn);
3442 :
3443 42 : ReleaseRWConflict(possibleUnsafeConflict);
3444 : }
3445 : }
3446 :
3447 : /* Check for conflict out to old committed transactions. */
3448 1579 : if (isCommit
3449 1248 : && !SxactIsReadOnly(MySerializableXact)
3450 1009 : && SxactHasSummaryConflictOut(MySerializableXact))
3451 : {
3452 : /*
3453 : * we don't know which old committed transaction we conflicted with,
3454 : * so be conservative and use FirstNormalSerCommitSeqNo here
3455 : */
3456 0 : MySerializableXact->SeqNo.earliestOutConflictCommit =
3457 : FirstNormalSerCommitSeqNo;
3458 0 : MySerializableXact->flags |= SXACT_FLAG_CONFLICT_OUT;
3459 : }
3460 :
3461 : /*
3462 : * Release all outConflicts to committed transactions. If we're rolling
3463 : * back clear them all. Set SXACT_FLAG_CONFLICT_OUT if any point to
3464 : * previously committed transactions.
3465 : */
3466 2266 : dlist_foreach_modify(iter, &MySerializableXact->outConflicts)
3467 : {
3468 687 : RWConflict conflict =
3469 : dlist_container(RWConflictData, outLink, iter.cur);
3470 :
3471 687 : if (isCommit
3472 455 : && !SxactIsReadOnly(MySerializableXact)
3473 347 : && SxactIsCommitted(conflict->sxactIn))
3474 : {
3475 96 : if ((MySerializableXact->flags & SXACT_FLAG_CONFLICT_OUT) == 0
3476 0 : || conflict->sxactIn->prepareSeqNo < MySerializableXact->SeqNo.earliestOutConflictCommit)
3477 96 : MySerializableXact->SeqNo.earliestOutConflictCommit = conflict->sxactIn->prepareSeqNo;
3478 96 : MySerializableXact->flags |= SXACT_FLAG_CONFLICT_OUT;
3479 : }
3480 :
3481 687 : if (!isCommit
3482 455 : || SxactIsCommitted(conflict->sxactIn)
3483 337 : || (conflict->sxactIn->SeqNo.lastCommitBeforeSnapshot >= PredXact->LastSxactCommitSeqNo))
3484 350 : ReleaseRWConflict(conflict);
3485 : }
3486 :
3487 : /*
3488 : * Release all inConflicts from committed and read-only transactions. If
3489 : * we're rolling back, clear them all.
3490 : */
3491 2360 : dlist_foreach_modify(iter, &MySerializableXact->inConflicts)
3492 : {
3493 781 : RWConflict conflict =
3494 781 : dlist_container(RWConflictData, inLink, iter.cur);
3495 :
3496 781 : if (!isCommit
3497 604 : || SxactIsCommitted(conflict->sxactOut)
3498 419 : || SxactIsReadOnly(conflict->sxactOut))
3499 442 : ReleaseRWConflict(conflict);
3500 : }
3501 :
3502 1579 : if (!topLevelIsDeclaredReadOnly)
3503 : {
3504 : /*
3505 : * Remove ourselves from the list of possible conflicts for concurrent
3506 : * READ ONLY transactions, flagging them as unsafe if we have a
3507 : * conflict out. If any are waiting DEFERRABLE transactions, wake them
3508 : * up if they are known safe or known unsafe.
3509 : */
3510 1561 : dlist_foreach_modify(iter, &MySerializableXact->possibleUnsafeConflicts)
3511 : {
3512 92 : RWConflict possibleUnsafeConflict =
3513 : dlist_container(RWConflictData, outLink, iter.cur);
3514 :
3515 92 : roXact = possibleUnsafeConflict->sxactIn;
3516 : Assert(MySerializableXact == possibleUnsafeConflict->sxactOut);
3517 : Assert(SxactIsReadOnly(roXact));
3518 :
3519 : /* Mark conflicted if necessary. */
3520 92 : if (isCommit
3521 89 : && MyXactDidWrite
3522 84 : && SxactHasConflictOut(MySerializableXact)
3523 13 : && (MySerializableXact->SeqNo.earliestOutConflictCommit
3524 13 : <= roXact->SeqNo.lastCommitBeforeSnapshot))
3525 : {
3526 : /*
3527 : * This releases possibleUnsafeConflict (as well as all other
3528 : * possible conflicts for roXact)
3529 : */
3530 3 : FlagSxactUnsafe(roXact);
3531 : }
3532 : else
3533 : {
3534 89 : ReleaseRWConflict(possibleUnsafeConflict);
3535 :
3536 : /*
3537 : * If we were the last possible conflict, flag it safe. The
3538 : * transaction can now safely release its predicate locks (but
3539 : * that transaction's backend has to do that itself).
3540 : */
3541 89 : if (dlist_is_empty(&roXact->possibleUnsafeConflicts))
3542 67 : roXact->flags |= SXACT_FLAG_RO_SAFE;
3543 : }
3544 :
3545 : /*
3546 : * Wake up the process for a waiting DEFERRABLE transaction if we
3547 : * now know it's either safe or conflicted.
3548 : */
3549 92 : if (SxactIsDeferrableWaiting(roXact) &&
3550 3 : (SxactIsROUnsafe(roXact) || SxactIsROSafe(roXact)))
3551 3 : ProcSendSignal(roXact->pgprocno);
3552 : }
3553 : }
3554 :
3555 : /*
3556 : * Check whether it's time to clean up old transactions. This can only be
3557 : * done when the last serializable transaction with the oldest xmin among
3558 : * serializable transactions completes. We then find the "new oldest"
3559 : * xmin and purge any transactions which finished before this transaction
3560 : * was launched.
3561 : *
3562 : * For parallel queries in read-only transactions, it might run twice. We
3563 : * only release the reference on the first call.
3564 : */
3565 1579 : needToClear = false;
3566 1579 : if ((partiallyReleasing ||
3567 1577 : !SxactIsPartiallyReleased(MySerializableXact)) &&
3568 1577 : TransactionIdEquals(MySerializableXact->xmin,
3569 : PredXact->SxactGlobalXmin))
3570 : {
3571 : Assert(PredXact->SxactGlobalXminCount > 0);
3572 1559 : if (--(PredXact->SxactGlobalXminCount) == 0)
3573 : {
3574 893 : SetNewSxactGlobalXmin();
3575 893 : needToClear = true;
3576 : }
3577 : }
3578 :
3579 1579 : LWLockRelease(SerializableXactHashLock);
3580 :
3581 1579 : LWLockAcquire(SerializableFinishedListLock, LW_EXCLUSIVE);
3582 :
3583 : /* Add this to the list of transactions to check for later cleanup. */
3584 1579 : if (isCommit)
3585 1248 : dlist_push_tail(FinishedSerializableTransactions,
3586 1248 : &MySerializableXact->finishedLink);
3587 :
3588 : /*
3589 : * If we're releasing a RO_SAFE transaction in parallel mode, we'll only
3590 : * partially release it. That's necessary because other backends may have
3591 : * a reference to it. The leader will release the SERIALIZABLEXACT itself
3592 : * at the end of the transaction after workers have stopped running.
3593 : */
3594 1579 : if (!isCommit)
3595 331 : ReleaseOneSerializableXact(MySerializableXact,
3596 331 : isReadOnlySafe && IsInParallelMode(),
3597 331 : false);
3598 :
3599 1579 : LWLockRelease(SerializableFinishedListLock);
3600 :
3601 1579 : if (needToClear)
3602 893 : ClearOldPredicateLocks();
3603 :
3604 1579 : ReleasePredicateLocksLocal();
3605 : }
3606 :
3607 : static void
3608 7612 : ReleasePredicateLocksLocal(void)
3609 : {
3610 7612 : MySerializableXact = InvalidSerializableXact;
3611 7612 : MyXactDidWrite = false;
3612 :
3613 : /* Delete per-transaction lock table */
3614 7612 : if (LocalPredicateLockHash != NULL)
3615 : {
3616 1578 : hash_destroy(LocalPredicateLockHash);
3617 1578 : LocalPredicateLockHash = NULL;
3618 : }
3619 7612 : }
3620 :
3621 : /*
3622 : * Clear old predicate locks, belonging to committed transactions that are no
3623 : * longer interesting to any in-progress transaction.
3624 : */
3625 : static void
3626 893 : ClearOldPredicateLocks(void)
3627 : {
3628 : dlist_mutable_iter iter;
3629 :
3630 : /*
3631 : * Loop through finished transactions. They are in commit order, so we can
3632 : * stop as soon as we find one that's still interesting.
3633 : */
3634 893 : LWLockAcquire(SerializableFinishedListLock, LW_EXCLUSIVE);
3635 893 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
3636 2150 : dlist_foreach_modify(iter, FinishedSerializableTransactions)
3637 : {
3638 1266 : SERIALIZABLEXACT *finishedSxact =
3639 1266 : dlist_container(SERIALIZABLEXACT, finishedLink, iter.cur);
3640 :
3641 1266 : if (!TransactionIdIsValid(PredXact->SxactGlobalXmin)
3642 29 : || TransactionIdPrecedesOrEquals(finishedSxact->finishedBefore,
3643 29 : PredXact->SxactGlobalXmin))
3644 : {
3645 : /*
3646 : * This transaction committed before any in-progress transaction
3647 : * took its snapshot. It's no longer interesting.
3648 : */
3649 1248 : LWLockRelease(SerializableXactHashLock);
3650 1248 : dlist_delete_thoroughly(&finishedSxact->finishedLink);
3651 1248 : ReleaseOneSerializableXact(finishedSxact, false, false);
3652 1248 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
3653 : }
3654 18 : else if (finishedSxact->commitSeqNo > PredXact->HavePartialClearedThrough
3655 18 : && finishedSxact->commitSeqNo <= PredXact->CanPartialClearThrough)
3656 : {
3657 : /*
3658 : * Any active transactions that took their snapshot before this
3659 : * transaction committed are read-only, so we can clear part of
3660 : * its state.
3661 : */
3662 9 : LWLockRelease(SerializableXactHashLock);
3663 :
3664 9 : if (SxactIsReadOnly(finishedSxact))
3665 : {
3666 : /* A read-only transaction can be removed entirely */
3667 0 : dlist_delete_thoroughly(&(finishedSxact->finishedLink));
3668 0 : ReleaseOneSerializableXact(finishedSxact, false, false);
3669 : }
3670 : else
3671 : {
3672 : /*
3673 : * A read-write transaction can only be partially cleared. We
3674 : * need to keep the SERIALIZABLEXACT but can release the
3675 : * SIREAD locks and conflicts in.
3676 : */
3677 9 : ReleaseOneSerializableXact(finishedSxact, true, false);
3678 : }
3679 :
3680 9 : PredXact->HavePartialClearedThrough = finishedSxact->commitSeqNo;
3681 9 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
3682 : }
3683 : else
3684 : {
3685 : /* Still interesting. */
3686 : break;
3687 : }
3688 : }
3689 893 : LWLockRelease(SerializableXactHashLock);
3690 :
3691 : /*
3692 : * Loop through predicate locks on dummy transaction for summarized data.
3693 : */
3694 893 : LWLockAcquire(SerializablePredicateListLock, LW_SHARED);
3695 893 : dlist_foreach_modify(iter, &OldCommittedSxact->predicateLocks)
3696 : {
3697 0 : PREDICATELOCK *predlock =
3698 0 : dlist_container(PREDICATELOCK, xactLink, iter.cur);
3699 : bool canDoPartialCleanup;
3700 :
3701 0 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
3702 : Assert(predlock->commitSeqNo != 0);
3703 : Assert(predlock->commitSeqNo != InvalidSerCommitSeqNo);
3704 0 : canDoPartialCleanup = (predlock->commitSeqNo <= PredXact->CanPartialClearThrough);
3705 0 : LWLockRelease(SerializableXactHashLock);
3706 :
3707 : /*
3708 : * If this lock originally belonged to an old enough transaction, we
3709 : * can release it.
3710 : */
3711 0 : if (canDoPartialCleanup)
3712 : {
3713 : PREDICATELOCKTAG tag;
3714 : PREDICATELOCKTARGET *target;
3715 : PREDICATELOCKTARGETTAG targettag;
3716 : uint32 targettaghash;
3717 : LWLock *partitionLock;
3718 :
3719 0 : tag = predlock->tag;
3720 0 : target = tag.myTarget;
3721 0 : targettag = target->tag;
3722 0 : targettaghash = PredicateLockTargetTagHashCode(&targettag);
3723 0 : partitionLock = PredicateLockHashPartitionLock(targettaghash);
3724 :
3725 0 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
3726 :
3727 0 : dlist_delete(&(predlock->targetLink));
3728 0 : dlist_delete(&(predlock->xactLink));
3729 :
3730 0 : hash_search_with_hash_value(PredicateLockHash, &tag,
3731 0 : PredicateLockHashCodeFromTargetHashCode(&tag,
3732 : targettaghash),
3733 : HASH_REMOVE, NULL);
3734 0 : RemoveTargetIfNoLongerUsed(target, targettaghash);
3735 :
3736 0 : LWLockRelease(partitionLock);
3737 : }
3738 : }
3739 :
3740 893 : LWLockRelease(SerializablePredicateListLock);
3741 893 : LWLockRelease(SerializableFinishedListLock);
3742 893 : }
3743 :
3744 : /*
3745 : * This is the normal way to delete anything from any of the predicate
3746 : * locking hash tables. Given a transaction which we know can be deleted:
3747 : * delete all predicate locks held by that transaction and any predicate
3748 : * lock targets which are now unreferenced by a lock; delete all conflicts
3749 : * for the transaction; delete all xid values for the transaction; then
3750 : * delete the transaction.
3751 : *
3752 : * When the partial flag is set, we can release all predicate locks and
3753 : * in-conflict information -- we've established that there are no longer
3754 : * any overlapping read write transactions for which this transaction could
3755 : * matter -- but keep the transaction entry itself and any outConflicts.
3756 : *
3757 : * When the summarize flag is set, we've run short of room for sxact data
3758 : * and must summarize to the SLRU. Predicate locks are transferred to a
3759 : * dummy "old" transaction, with duplicate locks on a single target
3760 : * collapsing to a single lock with the "latest" commitSeqNo from among
3761 : * the conflicting locks..
3762 : */
3763 : static void
3764 1588 : ReleaseOneSerializableXact(SERIALIZABLEXACT *sxact, bool partial,
3765 : bool summarize)
3766 : {
3767 : SERIALIZABLEXIDTAG sxidtag;
3768 : dlist_mutable_iter iter;
3769 :
3770 : Assert(sxact != NULL);
3771 : Assert(SxactIsRolledBack(sxact) || SxactIsCommitted(sxact));
3772 : Assert(partial || !SxactIsOnFinishedList(sxact));
3773 : Assert(LWLockHeldByMe(SerializableFinishedListLock));
3774 :
3775 : /*
3776 : * First release all the predicate locks held by this xact (or transfer
3777 : * them to OldCommittedSxact if summarize is true)
3778 : */
3779 1588 : LWLockAcquire(SerializablePredicateListLock, LW_SHARED);
3780 1588 : if (IsInParallelMode())
3781 3 : LWLockAcquire(&sxact->perXactPredicateListLock, LW_EXCLUSIVE);
3782 4456 : dlist_foreach_modify(iter, &sxact->predicateLocks)
3783 : {
3784 2868 : PREDICATELOCK *predlock =
3785 2868 : dlist_container(PREDICATELOCK, xactLink, iter.cur);
3786 : PREDICATELOCKTAG tag;
3787 : PREDICATELOCKTARGET *target;
3788 : PREDICATELOCKTARGETTAG targettag;
3789 : uint32 targettaghash;
3790 : LWLock *partitionLock;
3791 :
3792 2868 : tag = predlock->tag;
3793 2868 : target = tag.myTarget;
3794 2868 : targettag = target->tag;
3795 2868 : targettaghash = PredicateLockTargetTagHashCode(&targettag);
3796 2868 : partitionLock = PredicateLockHashPartitionLock(targettaghash);
3797 :
3798 2868 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
3799 :
3800 2868 : dlist_delete(&predlock->targetLink);
3801 :
3802 2868 : hash_search_with_hash_value(PredicateLockHash, &tag,
3803 2868 : PredicateLockHashCodeFromTargetHashCode(&tag,
3804 : targettaghash),
3805 : HASH_REMOVE, NULL);
3806 2868 : if (summarize)
3807 : {
3808 : bool found;
3809 :
3810 : /* Fold into dummy transaction list. */
3811 0 : tag.myXact = OldCommittedSxact;
3812 0 : predlock = hash_search_with_hash_value(PredicateLockHash, &tag,
3813 0 : PredicateLockHashCodeFromTargetHashCode(&tag,
3814 : targettaghash),
3815 : HASH_ENTER_NULL, &found);
3816 0 : if (!predlock)
3817 0 : ereport(ERROR,
3818 : (errcode(ERRCODE_OUT_OF_MEMORY),
3819 : errmsg("out of shared memory"),
3820 : errhint("You might need to increase \"%s\".", "max_pred_locks_per_transaction")));
3821 0 : if (found)
3822 : {
3823 : Assert(predlock->commitSeqNo != 0);
3824 : Assert(predlock->commitSeqNo != InvalidSerCommitSeqNo);
3825 0 : if (predlock->commitSeqNo < sxact->commitSeqNo)
3826 0 : predlock->commitSeqNo = sxact->commitSeqNo;
3827 : }
3828 : else
3829 : {
3830 0 : dlist_push_tail(&target->predicateLocks,
3831 : &predlock->targetLink);
3832 0 : dlist_push_tail(&OldCommittedSxact->predicateLocks,
3833 : &predlock->xactLink);
3834 0 : predlock->commitSeqNo = sxact->commitSeqNo;
3835 : }
3836 : }
3837 : else
3838 2868 : RemoveTargetIfNoLongerUsed(target, targettaghash);
3839 :
3840 2868 : LWLockRelease(partitionLock);
3841 : }
3842 :
3843 : /*
3844 : * Rather than retail removal, just re-init the head after we've run
3845 : * through the list.
3846 : */
3847 1588 : dlist_init(&sxact->predicateLocks);
3848 :
3849 1588 : if (IsInParallelMode())
3850 3 : LWLockRelease(&sxact->perXactPredicateListLock);
3851 1588 : LWLockRelease(SerializablePredicateListLock);
3852 :
3853 1588 : sxidtag.xid = sxact->topXid;
3854 1588 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
3855 :
3856 : /* Release all outConflicts (unless 'partial' is true) */
3857 1588 : if (!partial)
3858 : {
3859 1577 : dlist_foreach_modify(iter, &sxact->outConflicts)
3860 : {
3861 0 : RWConflict conflict =
3862 : dlist_container(RWConflictData, outLink, iter.cur);
3863 :
3864 0 : if (summarize)
3865 0 : conflict->sxactIn->flags |= SXACT_FLAG_SUMMARY_CONFLICT_IN;
3866 0 : ReleaseRWConflict(conflict);
3867 : }
3868 : }
3869 :
3870 : /* Release all inConflicts. */
3871 1588 : dlist_foreach_modify(iter, &sxact->inConflicts)
3872 : {
3873 0 : RWConflict conflict =
3874 0 : dlist_container(RWConflictData, inLink, iter.cur);
3875 :
3876 0 : if (summarize)
3877 0 : conflict->sxactOut->flags |= SXACT_FLAG_SUMMARY_CONFLICT_OUT;
3878 0 : ReleaseRWConflict(conflict);
3879 : }
3880 :
3881 : /* Finally, get rid of the xid and the record of the transaction itself. */
3882 1588 : if (!partial)
3883 : {
3884 1577 : if (sxidtag.xid != InvalidTransactionId)
3885 1302 : hash_search(SerializableXidHash, &sxidtag, HASH_REMOVE, NULL);
3886 1577 : ReleasePredXact(sxact);
3887 : }
3888 :
3889 1588 : LWLockRelease(SerializableXactHashLock);
3890 1588 : }
3891 :
3892 : /*
3893 : * Tests whether the given top level transaction is concurrent with
3894 : * (overlaps) our current transaction.
3895 : *
3896 : * We need to identify the top level transaction for SSI, anyway, so pass
3897 : * that to this function to save the overhead of checking the snapshot's
3898 : * subxip array.
3899 : */
3900 : static bool
3901 536 : XidIsConcurrent(TransactionId xid)
3902 : {
3903 : Snapshot snap;
3904 :
3905 : Assert(TransactionIdIsValid(xid));
3906 : Assert(!TransactionIdEquals(xid, GetTopTransactionIdIfAny()));
3907 :
3908 536 : snap = GetTransactionSnapshot();
3909 :
3910 536 : if (TransactionIdPrecedes(xid, snap->xmin))
3911 0 : return false;
3912 :
3913 536 : if (TransactionIdFollowsOrEquals(xid, snap->xmax))
3914 524 : return true;
3915 :
3916 12 : return pg_lfind32(xid, snap->xip, snap->xcnt);
3917 : }
3918 :
3919 : bool
3920 45881407 : CheckForSerializableConflictOutNeeded(Relation relation, Snapshot snapshot)
3921 : {
3922 45881407 : if (!SerializationNeededForRead(relation, snapshot))
3923 45872831 : return false;
3924 :
3925 : /* Check if someone else has already decided that we need to die */
3926 8576 : if (SxactIsDoomed(MySerializableXact))
3927 : {
3928 0 : ereport(ERROR,
3929 : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
3930 : errmsg("could not serialize access due to read/write dependencies among transactions"),
3931 : errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict out checking."),
3932 : errhint("The transaction might succeed if retried.")));
3933 : }
3934 :
3935 8576 : return true;
3936 : }
3937 :
3938 : /*
3939 : * CheckForSerializableConflictOut
3940 : * A table AM is reading a tuple that has been modified. If it determines
3941 : * that the tuple version it is reading is not visible to us, it should
3942 : * pass in the top level xid of the transaction that created it.
3943 : * Otherwise, if it determines that it is visible to us but it has been
3944 : * deleted or there is a newer version available due to an update, it
3945 : * should pass in the top level xid of the modifying transaction.
3946 : *
3947 : * This function will check for overlap with our own transaction. If the given
3948 : * xid is also serializable and the transactions overlap (i.e., they cannot see
3949 : * each other's writes), then we have a conflict out.
3950 : */
3951 : void
3952 567 : CheckForSerializableConflictOut(Relation relation, TransactionId xid, Snapshot snapshot)
3953 : {
3954 : SERIALIZABLEXIDTAG sxidtag;
3955 : SERIALIZABLEXID *sxid;
3956 : SERIALIZABLEXACT *sxact;
3957 :
3958 567 : if (!SerializationNeededForRead(relation, snapshot))
3959 200 : return;
3960 :
3961 : /* Check if someone else has already decided that we need to die */
3962 567 : if (SxactIsDoomed(MySerializableXact))
3963 : {
3964 0 : ereport(ERROR,
3965 : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
3966 : errmsg("could not serialize access due to read/write dependencies among transactions"),
3967 : errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict out checking."),
3968 : errhint("The transaction might succeed if retried.")));
3969 : }
3970 : Assert(TransactionIdIsValid(xid));
3971 :
3972 567 : if (TransactionIdEquals(xid, GetTopTransactionIdIfAny()))
3973 0 : return;
3974 :
3975 : /*
3976 : * Find sxact or summarized info for the top level xid.
3977 : */
3978 567 : sxidtag.xid = xid;
3979 567 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
3980 : sxid = (SERIALIZABLEXID *)
3981 567 : hash_search(SerializableXidHash, &sxidtag, HASH_FIND, NULL);
3982 567 : if (!sxid)
3983 : {
3984 : /*
3985 : * Transaction not found in "normal" SSI structures. Check whether it
3986 : * got pushed out to SLRU storage for "old committed" transactions.
3987 : */
3988 : SerCommitSeqNo conflictCommitSeqNo;
3989 :
3990 21 : conflictCommitSeqNo = SerialGetMinConflictCommitSeqNo(xid);
3991 21 : if (conflictCommitSeqNo != 0)
3992 : {
3993 0 : if (conflictCommitSeqNo != InvalidSerCommitSeqNo
3994 0 : && (!SxactIsReadOnly(MySerializableXact)
3995 0 : || conflictCommitSeqNo
3996 0 : <= MySerializableXact->SeqNo.lastCommitBeforeSnapshot))
3997 0 : ereport(ERROR,
3998 : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
3999 : errmsg("could not serialize access due to read/write dependencies among transactions"),
4000 : errdetail_internal("Reason code: Canceled on conflict out to old pivot %u.", xid),
4001 : errhint("The transaction might succeed if retried.")));
4002 :
4003 0 : if (SxactHasSummaryConflictIn(MySerializableXact)
4004 0 : || !dlist_is_empty(&MySerializableXact->inConflicts))
4005 0 : ereport(ERROR,
4006 : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4007 : errmsg("could not serialize access due to read/write dependencies among transactions"),
4008 : errdetail_internal("Reason code: Canceled on identification as a pivot, with conflict out to old committed transaction %u.", xid),
4009 : errhint("The transaction might succeed if retried.")));
4010 :
4011 0 : MySerializableXact->flags |= SXACT_FLAG_SUMMARY_CONFLICT_OUT;
4012 : }
4013 :
4014 : /* It's not serializable or otherwise not important. */
4015 21 : LWLockRelease(SerializableXactHashLock);
4016 21 : return;
4017 : }
4018 546 : sxact = sxid->myXact;
4019 : Assert(TransactionIdEquals(sxact->topXid, xid));
4020 546 : if (sxact == MySerializableXact || SxactIsDoomed(sxact))
4021 : {
4022 : /* Can't conflict with ourself or a transaction that will roll back. */
4023 4 : LWLockRelease(SerializableXactHashLock);
4024 4 : return;
4025 : }
4026 :
4027 : /*
4028 : * We have a conflict out to a transaction which has a conflict out to a
4029 : * summarized transaction. That summarized transaction must have
4030 : * committed first, and we can't tell when it committed in relation to our
4031 : * snapshot acquisition, so something needs to be canceled.
4032 : */
4033 542 : if (SxactHasSummaryConflictOut(sxact))
4034 : {
4035 0 : if (!SxactIsPrepared(sxact))
4036 : {
4037 0 : sxact->flags |= SXACT_FLAG_DOOMED;
4038 0 : LWLockRelease(SerializableXactHashLock);
4039 0 : return;
4040 : }
4041 : else
4042 : {
4043 0 : LWLockRelease(SerializableXactHashLock);
4044 0 : ereport(ERROR,
4045 : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4046 : errmsg("could not serialize access due to read/write dependencies among transactions"),
4047 : errdetail_internal("Reason code: Canceled on conflict out to old pivot."),
4048 : errhint("The transaction might succeed if retried.")));
4049 : }
4050 : }
4051 :
4052 : /*
4053 : * If this is a read-only transaction and the writing transaction has
4054 : * committed, and it doesn't have a rw-conflict to a transaction which
4055 : * committed before it, no conflict.
4056 : */
4057 542 : if (SxactIsReadOnly(MySerializableXact)
4058 119 : && SxactIsCommitted(sxact)
4059 8 : && !SxactHasSummaryConflictOut(sxact)
4060 8 : && (!SxactHasConflictOut(sxact)
4061 2 : || MySerializableXact->SeqNo.lastCommitBeforeSnapshot < sxact->SeqNo.earliestOutConflictCommit))
4062 : {
4063 : /* Read-only transaction will appear to run first. No conflict. */
4064 6 : LWLockRelease(SerializableXactHashLock);
4065 6 : return;
4066 : }
4067 :
4068 536 : if (!XidIsConcurrent(xid))
4069 : {
4070 : /* This write was already in our snapshot; no conflict. */
4071 0 : LWLockRelease(SerializableXactHashLock);
4072 0 : return;
4073 : }
4074 :
4075 536 : if (RWConflictExists(MySerializableXact, sxact))
4076 : {
4077 : /* We don't want duplicate conflict records in the list. */
4078 169 : LWLockRelease(SerializableXactHashLock);
4079 169 : return;
4080 : }
4081 :
4082 : /*
4083 : * Flag the conflict. But first, if this conflict creates a dangerous
4084 : * structure, ereport an error.
4085 : */
4086 367 : FlagRWConflict(MySerializableXact, sxact);
4087 354 : LWLockRelease(SerializableXactHashLock);
4088 : }
4089 :
4090 : /*
4091 : * Check a particular target for rw-dependency conflict in. A subroutine of
4092 : * CheckForSerializableConflictIn().
4093 : */
4094 : static void
4095 7616 : CheckTargetForConflictsIn(PREDICATELOCKTARGETTAG *targettag)
4096 : {
4097 : uint32 targettaghash;
4098 : LWLock *partitionLock;
4099 : PREDICATELOCKTARGET *target;
4100 7616 : PREDICATELOCK *mypredlock = NULL;
4101 : PREDICATELOCKTAG mypredlocktag;
4102 : dlist_mutable_iter iter;
4103 :
4104 : Assert(MySerializableXact != InvalidSerializableXact);
4105 :
4106 : /*
4107 : * The same hash and LW lock apply to the lock target and the lock itself.
4108 : */
4109 7616 : targettaghash = PredicateLockTargetTagHashCode(targettag);
4110 7616 : partitionLock = PredicateLockHashPartitionLock(targettaghash);
4111 7616 : LWLockAcquire(partitionLock, LW_SHARED);
4112 : target = (PREDICATELOCKTARGET *)
4113 7616 : hash_search_with_hash_value(PredicateLockTargetHash,
4114 : targettag, targettaghash,
4115 : HASH_FIND, NULL);
4116 7616 : if (!target)
4117 : {
4118 : /* Nothing has this target locked; we're done here. */
4119 5711 : LWLockRelease(partitionLock);
4120 5711 : return;
4121 : }
4122 :
4123 : /*
4124 : * Each lock for an overlapping transaction represents a conflict: a
4125 : * rw-dependency in to this transaction.
4126 : */
4127 1905 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
4128 :
4129 4290 : dlist_foreach_modify(iter, &target->predicateLocks)
4130 : {
4131 2452 : PREDICATELOCK *predlock =
4132 2452 : dlist_container(PREDICATELOCK, targetLink, iter.cur);
4133 2452 : SERIALIZABLEXACT *sxact = predlock->tag.myXact;
4134 :
4135 2452 : if (sxact == MySerializableXact)
4136 : {
4137 : /*
4138 : * If we're getting a write lock on a tuple, we don't need a
4139 : * predicate (SIREAD) lock on the same tuple. We can safely remove
4140 : * our SIREAD lock, but we'll defer doing so until after the loop
4141 : * because that requires upgrading to an exclusive partition lock.
4142 : *
4143 : * We can't use this optimization within a subtransaction because
4144 : * the subtransaction could roll back, and we would be left
4145 : * without any lock at the top level.
4146 : */
4147 1594 : if (!IsSubTransaction()
4148 1594 : && GET_PREDICATELOCKTARGETTAG_OFFSET(*targettag))
4149 : {
4150 401 : mypredlock = predlock;
4151 401 : mypredlocktag = predlock->tag;
4152 : }
4153 : }
4154 858 : else if (!SxactIsDoomed(sxact)
4155 858 : && (!SxactIsCommitted(sxact)
4156 88 : || TransactionIdPrecedes(GetTransactionSnapshot()->xmin,
4157 : sxact->finishedBefore))
4158 849 : && !RWConflictExists(sxact, MySerializableXact))
4159 : {
4160 505 : LWLockRelease(SerializableXactHashLock);
4161 505 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
4162 :
4163 : /*
4164 : * Re-check after getting exclusive lock because the other
4165 : * transaction may have flagged a conflict.
4166 : */
4167 505 : if (!SxactIsDoomed(sxact)
4168 505 : && (!SxactIsCommitted(sxact)
4169 77 : || TransactionIdPrecedes(GetTransactionSnapshot()->xmin,
4170 : sxact->finishedBefore))
4171 505 : && !RWConflictExists(sxact, MySerializableXact))
4172 : {
4173 505 : FlagRWConflict(sxact, MySerializableXact);
4174 : }
4175 :
4176 438 : LWLockRelease(SerializableXactHashLock);
4177 438 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
4178 : }
4179 : }
4180 1838 : LWLockRelease(SerializableXactHashLock);
4181 1838 : LWLockRelease(partitionLock);
4182 :
4183 : /*
4184 : * If we found one of our own SIREAD locks to remove, remove it now.
4185 : *
4186 : * At this point our transaction already has a RowExclusiveLock on the
4187 : * relation, so we are OK to drop the predicate lock on the tuple, if
4188 : * found, without fearing that another write against the tuple will occur
4189 : * before the MVCC information makes it to the buffer.
4190 : */
4191 1838 : if (mypredlock != NULL)
4192 : {
4193 : uint32 predlockhashcode;
4194 : PREDICATELOCK *rmpredlock;
4195 :
4196 394 : LWLockAcquire(SerializablePredicateListLock, LW_SHARED);
4197 394 : if (IsInParallelMode())
4198 0 : LWLockAcquire(&MySerializableXact->perXactPredicateListLock, LW_EXCLUSIVE);
4199 394 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
4200 394 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
4201 :
4202 : /*
4203 : * Remove the predicate lock from shared memory, if it wasn't removed
4204 : * while the locks were released. One way that could happen is from
4205 : * autovacuum cleaning up an index.
4206 : */
4207 394 : predlockhashcode = PredicateLockHashCodeFromTargetHashCode
4208 : (&mypredlocktag, targettaghash);
4209 : rmpredlock = (PREDICATELOCK *)
4210 394 : hash_search_with_hash_value(PredicateLockHash,
4211 : &mypredlocktag,
4212 : predlockhashcode,
4213 : HASH_FIND, NULL);
4214 394 : if (rmpredlock != NULL)
4215 : {
4216 : Assert(rmpredlock == mypredlock);
4217 :
4218 394 : dlist_delete(&(mypredlock->targetLink));
4219 394 : dlist_delete(&(mypredlock->xactLink));
4220 :
4221 : rmpredlock = (PREDICATELOCK *)
4222 394 : hash_search_with_hash_value(PredicateLockHash,
4223 : &mypredlocktag,
4224 : predlockhashcode,
4225 : HASH_REMOVE, NULL);
4226 : Assert(rmpredlock == mypredlock);
4227 :
4228 394 : RemoveTargetIfNoLongerUsed(target, targettaghash);
4229 : }
4230 :
4231 394 : LWLockRelease(SerializableXactHashLock);
4232 394 : LWLockRelease(partitionLock);
4233 394 : if (IsInParallelMode())
4234 0 : LWLockRelease(&MySerializableXact->perXactPredicateListLock);
4235 394 : LWLockRelease(SerializablePredicateListLock);
4236 :
4237 394 : if (rmpredlock != NULL)
4238 : {
4239 : /*
4240 : * Remove entry in local lock table if it exists. It's OK if it
4241 : * doesn't exist; that means the lock was transferred to a new
4242 : * target by a different backend.
4243 : */
4244 394 : hash_search_with_hash_value(LocalPredicateLockHash,
4245 : targettag, targettaghash,
4246 : HASH_REMOVE, NULL);
4247 :
4248 394 : DecrementParentLocks(targettag);
4249 : }
4250 : }
4251 : }
4252 :
4253 : /*
4254 : * CheckForSerializableConflictIn
4255 : * We are writing the given tuple. If that indicates a rw-conflict
4256 : * in from another serializable transaction, take appropriate action.
4257 : *
4258 : * Skip checking for any granularity for which a parameter is missing.
4259 : *
4260 : * A tuple update or delete is in conflict if we have a predicate lock
4261 : * against the relation or page in which the tuple exists, or against the
4262 : * tuple itself.
4263 : */
4264 : void
4265 24569114 : CheckForSerializableConflictIn(Relation relation, const ItemPointerData *tid, BlockNumber blkno)
4266 : {
4267 : PREDICATELOCKTARGETTAG targettag;
4268 :
4269 24569114 : if (!SerializationNeededForWrite(relation))
4270 24564608 : return;
4271 :
4272 : /* Check if someone else has already decided that we need to die */
4273 4506 : if (SxactIsDoomed(MySerializableXact))
4274 1 : ereport(ERROR,
4275 : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4276 : errmsg("could not serialize access due to read/write dependencies among transactions"),
4277 : errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict in checking."),
4278 : errhint("The transaction might succeed if retried.")));
4279 :
4280 : /*
4281 : * We're doing a write which might cause rw-conflicts now or later.
4282 : * Memorize that fact.
4283 : */
4284 4505 : MyXactDidWrite = true;
4285 :
4286 : /*
4287 : * It is important that we check for locks from the finest granularity to
4288 : * the coarsest granularity, so that granularity promotion doesn't cause
4289 : * us to miss a lock. The new (coarser) lock will be acquired before the
4290 : * old (finer) locks are released.
4291 : *
4292 : * It is not possible to take and hold a lock across the checks for all
4293 : * granularities because each target could be in a separate partition.
4294 : */
4295 4505 : if (tid != NULL)
4296 : {
4297 657 : SET_PREDICATELOCKTARGETTAG_TUPLE(targettag,
4298 : relation->rd_locator.dbOid,
4299 : relation->rd_id,
4300 : ItemPointerGetBlockNumber(tid),
4301 : ItemPointerGetOffsetNumber(tid));
4302 657 : CheckTargetForConflictsIn(&targettag);
4303 : }
4304 :
4305 4482 : if (blkno != InvalidBlockNumber)
4306 : {
4307 2507 : SET_PREDICATELOCKTARGETTAG_PAGE(targettag,
4308 : relation->rd_locator.dbOid,
4309 : relation->rd_id,
4310 : blkno);
4311 2507 : CheckTargetForConflictsIn(&targettag);
4312 : }
4313 :
4314 4452 : SET_PREDICATELOCKTARGETTAG_RELATION(targettag,
4315 : relation->rd_locator.dbOid,
4316 : relation->rd_id);
4317 4452 : CheckTargetForConflictsIn(&targettag);
4318 : }
4319 :
4320 : /*
4321 : * CheckTableForSerializableConflictIn
4322 : * The entire table is going through a DDL-style logical mass delete
4323 : * like TRUNCATE or DROP TABLE. If that causes a rw-conflict in from
4324 : * another serializable transaction, take appropriate action.
4325 : *
4326 : * While these operations do not operate entirely within the bounds of
4327 : * snapshot isolation, they can occur inside a serializable transaction, and
4328 : * will logically occur after any reads which saw rows which were destroyed
4329 : * by these operations, so we do what we can to serialize properly under
4330 : * SSI.
4331 : *
4332 : * The relation passed in must be a heap relation. Any predicate lock of any
4333 : * granularity on the heap will cause a rw-conflict in to this transaction.
4334 : * Predicate locks on indexes do not matter because they only exist to guard
4335 : * against conflicting inserts into the index, and this is a mass *delete*.
4336 : * When a table is truncated or dropped, the index will also be truncated
4337 : * or dropped, and we'll deal with locks on the index when that happens.
4338 : *
4339 : * Dropping or truncating a table also needs to drop any existing predicate
4340 : * locks on heap tuples or pages, because they're about to go away. This
4341 : * should be done before altering the predicate locks because the transaction
4342 : * could be rolled back because of a conflict, in which case the lock changes
4343 : * are not needed. (At the moment, we don't actually bother to drop the
4344 : * existing locks on a dropped or truncated table at the moment. That might
4345 : * lead to some false positives, but it doesn't seem worth the trouble.)
4346 : */
4347 : void
4348 35333 : CheckTableForSerializableConflictIn(Relation relation)
4349 : {
4350 : HASH_SEQ_STATUS seqstat;
4351 : PREDICATELOCKTARGET *target;
4352 : Oid dbId;
4353 : Oid heapId;
4354 : int i;
4355 :
4356 : /*
4357 : * Bail out quickly if there are no serializable transactions running.
4358 : * It's safe to check this without taking locks because the caller is
4359 : * holding an ACCESS EXCLUSIVE lock on the relation. No new locks which
4360 : * would matter here can be acquired while that is held.
4361 : */
4362 35333 : if (!TransactionIdIsValid(PredXact->SxactGlobalXmin))
4363 35313 : return;
4364 :
4365 133 : if (!SerializationNeededForWrite(relation))
4366 113 : return;
4367 :
4368 : /*
4369 : * We're doing a write which might cause rw-conflicts now or later.
4370 : * Memorize that fact.
4371 : */
4372 20 : MyXactDidWrite = true;
4373 :
4374 : Assert(relation->rd_index == NULL); /* not an index relation */
4375 :
4376 20 : dbId = relation->rd_locator.dbOid;
4377 20 : heapId = relation->rd_id;
4378 :
4379 20 : LWLockAcquire(SerializablePredicateListLock, LW_EXCLUSIVE);
4380 340 : for (i = 0; i < NUM_PREDICATELOCK_PARTITIONS; i++)
4381 320 : LWLockAcquire(PredicateLockHashPartitionLockByIndex(i), LW_SHARED);
4382 20 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
4383 :
4384 : /* Scan through target list */
4385 20 : hash_seq_init(&seqstat, PredicateLockTargetHash);
4386 :
4387 70 : while ((target = (PREDICATELOCKTARGET *) hash_seq_search(&seqstat)))
4388 : {
4389 : dlist_mutable_iter iter;
4390 :
4391 : /*
4392 : * Check whether this is a target which needs attention.
4393 : */
4394 50 : if (GET_PREDICATELOCKTARGETTAG_RELATION(target->tag) != heapId)
4395 41 : continue; /* wrong relation id */
4396 9 : if (GET_PREDICATELOCKTARGETTAG_DB(target->tag) != dbId)
4397 0 : continue; /* wrong database id */
4398 :
4399 : /*
4400 : * Loop through locks for this target and flag conflicts.
4401 : */
4402 18 : dlist_foreach_modify(iter, &target->predicateLocks)
4403 : {
4404 9 : PREDICATELOCK *predlock =
4405 9 : dlist_container(PREDICATELOCK, targetLink, iter.cur);
4406 :
4407 9 : if (predlock->tag.myXact != MySerializableXact
4408 0 : && !RWConflictExists(predlock->tag.myXact, MySerializableXact))
4409 : {
4410 0 : FlagRWConflict(predlock->tag.myXact, MySerializableXact);
4411 : }
4412 : }
4413 : }
4414 :
4415 : /* Release locks in reverse order */
4416 20 : LWLockRelease(SerializableXactHashLock);
4417 340 : for (i = NUM_PREDICATELOCK_PARTITIONS - 1; i >= 0; i--)
4418 320 : LWLockRelease(PredicateLockHashPartitionLockByIndex(i));
4419 20 : LWLockRelease(SerializablePredicateListLock);
4420 : }
4421 :
4422 :
4423 : /*
4424 : * Flag a rw-dependency between two serializable transactions.
4425 : *
4426 : * The caller is responsible for ensuring that we have a LW lock on
4427 : * the transaction hash table.
4428 : */
4429 : static void
4430 872 : FlagRWConflict(SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer)
4431 : {
4432 : Assert(reader != writer);
4433 :
4434 : /* First, see if this conflict causes failure. */
4435 872 : OnConflict_CheckForSerializationFailure(reader, writer);
4436 :
4437 : /* Actually do the conflict flagging. */
4438 792 : if (reader == OldCommittedSxact)
4439 0 : writer->flags |= SXACT_FLAG_SUMMARY_CONFLICT_IN;
4440 792 : else if (writer == OldCommittedSxact)
4441 0 : reader->flags |= SXACT_FLAG_SUMMARY_CONFLICT_OUT;
4442 : else
4443 792 : SetRWConflict(reader, writer);
4444 792 : }
4445 :
4446 : /*----------------------------------------------------------------------------
4447 : * We are about to add a RW-edge to the dependency graph - check that we don't
4448 : * introduce a dangerous structure by doing so, and abort one of the
4449 : * transactions if so.
4450 : *
4451 : * A serialization failure can only occur if there is a dangerous structure
4452 : * in the dependency graph:
4453 : *
4454 : * Tin ------> Tpivot ------> Tout
4455 : * rw rw
4456 : *
4457 : * Furthermore, Tout must commit first.
4458 : *
4459 : * One more optimization is that if Tin is declared READ ONLY (or commits
4460 : * without writing), we can only have a problem if Tout committed before Tin
4461 : * acquired its snapshot.
4462 : *----------------------------------------------------------------------------
4463 : */
4464 : static void
4465 872 : OnConflict_CheckForSerializationFailure(const SERIALIZABLEXACT *reader,
4466 : SERIALIZABLEXACT *writer)
4467 : {
4468 : bool failure;
4469 :
4470 : Assert(LWLockHeldByMe(SerializableXactHashLock));
4471 :
4472 872 : failure = false;
4473 :
4474 : /*------------------------------------------------------------------------
4475 : * Check for already-committed writer with rw-conflict out flagged
4476 : * (conflict-flag on W means that T2 committed before W):
4477 : *
4478 : * R ------> W ------> T2
4479 : * rw rw
4480 : *
4481 : * That is a dangerous structure, so we must abort. (Since the writer
4482 : * has already committed, we must be the reader)
4483 : *------------------------------------------------------------------------
4484 : */
4485 872 : if (SxactIsCommitted(writer)
4486 18 : && (SxactHasConflictOut(writer) || SxactHasSummaryConflictOut(writer)))
4487 2 : failure = true;
4488 :
4489 : /*------------------------------------------------------------------------
4490 : * Check whether the writer has become a pivot with an out-conflict
4491 : * committed transaction (T2), and T2 committed first:
4492 : *
4493 : * R ------> W ------> T2
4494 : * rw rw
4495 : *
4496 : * Because T2 must've committed first, there is no anomaly if:
4497 : * - the reader committed before T2
4498 : * - the writer committed before T2
4499 : * - the reader is a READ ONLY transaction and the reader was concurrent
4500 : * with T2 (= reader acquired its snapshot before T2 committed)
4501 : *
4502 : * We also handle the case that T2 is prepared but not yet committed
4503 : * here. In that case T2 has already checked for conflicts, so if it
4504 : * commits first, making the above conflict real, it's too late for it
4505 : * to abort.
4506 : *------------------------------------------------------------------------
4507 : */
4508 872 : if (!failure && SxactHasSummaryConflictOut(writer))
4509 0 : failure = true;
4510 872 : else if (!failure)
4511 : {
4512 : dlist_iter iter;
4513 :
4514 1087 : dlist_foreach(iter, &writer->outConflicts)
4515 : {
4516 292 : RWConflict conflict =
4517 : dlist_container(RWConflictData, outLink, iter.cur);
4518 292 : SERIALIZABLEXACT *t2 = conflict->sxactIn;
4519 :
4520 292 : if (SxactIsPrepared(t2)
4521 81 : && (!SxactIsCommitted(reader)
4522 65 : || t2->prepareSeqNo <= reader->commitSeqNo)
4523 81 : && (!SxactIsCommitted(writer)
4524 0 : || t2->prepareSeqNo <= writer->commitSeqNo)
4525 81 : && (!SxactIsReadOnly(reader)
4526 12 : || t2->prepareSeqNo <= reader->SeqNo.lastCommitBeforeSnapshot))
4527 : {
4528 75 : failure = true;
4529 75 : break;
4530 : }
4531 : }
4532 : }
4533 :
4534 : /*------------------------------------------------------------------------
4535 : * Check whether the reader has become a pivot with a writer
4536 : * that's committed (or prepared):
4537 : *
4538 : * T0 ------> R ------> W
4539 : * rw rw
4540 : *
4541 : * Because W must've committed first for an anomaly to occur, there is no
4542 : * anomaly if:
4543 : * - T0 committed before the writer
4544 : * - T0 is READ ONLY, and overlaps the writer
4545 : *------------------------------------------------------------------------
4546 : */
4547 872 : if (!failure && SxactIsPrepared(writer) && !SxactIsReadOnly(reader))
4548 : {
4549 18 : if (SxactHasSummaryConflictIn(reader))
4550 : {
4551 0 : failure = true;
4552 : }
4553 : else
4554 : {
4555 : dlist_iter iter;
4556 :
4557 : /*
4558 : * The unconstify is needed as we have no const version of
4559 : * dlist_foreach().
4560 : */
4561 18 : dlist_foreach(iter, &unconstify(SERIALIZABLEXACT *, reader)->inConflicts)
4562 : {
4563 11 : const RWConflict conflict =
4564 11 : dlist_container(RWConflictData, inLink, iter.cur);
4565 11 : const SERIALIZABLEXACT *t0 = conflict->sxactOut;
4566 :
4567 11 : if (!SxactIsDoomed(t0)
4568 11 : && (!SxactIsCommitted(t0)
4569 11 : || t0->commitSeqNo >= writer->prepareSeqNo)
4570 11 : && (!SxactIsReadOnly(t0)
4571 0 : || t0->SeqNo.lastCommitBeforeSnapshot >= writer->prepareSeqNo))
4572 : {
4573 11 : failure = true;
4574 11 : break;
4575 : }
4576 : }
4577 : }
4578 : }
4579 :
4580 872 : if (failure)
4581 : {
4582 : /*
4583 : * We have to kill a transaction to avoid a possible anomaly from
4584 : * occurring. If the writer is us, we can just ereport() to cause a
4585 : * transaction abort. Otherwise we flag the writer for termination,
4586 : * causing it to abort when it tries to commit. However, if the writer
4587 : * is a prepared transaction, already prepared, we can't abort it
4588 : * anymore, so we have to kill the reader instead.
4589 : */
4590 88 : if (MySerializableXact == writer)
4591 : {
4592 67 : LWLockRelease(SerializableXactHashLock);
4593 67 : ereport(ERROR,
4594 : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4595 : errmsg("could not serialize access due to read/write dependencies among transactions"),
4596 : errdetail_internal("Reason code: Canceled on identification as a pivot, during write."),
4597 : errhint("The transaction might succeed if retried.")));
4598 : }
4599 21 : else if (SxactIsPrepared(writer))
4600 : {
4601 13 : LWLockRelease(SerializableXactHashLock);
4602 :
4603 : /* if we're not the writer, we have to be the reader */
4604 : Assert(MySerializableXact == reader);
4605 13 : ereport(ERROR,
4606 : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4607 : errmsg("could not serialize access due to read/write dependencies among transactions"),
4608 : errdetail_internal("Reason code: Canceled on conflict out to pivot %u, during read.", writer->topXid),
4609 : errhint("The transaction might succeed if retried.")));
4610 : }
4611 8 : writer->flags |= SXACT_FLAG_DOOMED;
4612 : }
4613 792 : }
4614 :
4615 : /*
4616 : * PreCommit_CheckForSerializationFailure
4617 : * Check for dangerous structures in a serializable transaction
4618 : * at commit.
4619 : *
4620 : * We're checking for a dangerous structure as each conflict is recorded.
4621 : * The only way we could have a problem at commit is if this is the "out"
4622 : * side of a pivot, and neither the "in" side nor the pivot has yet
4623 : * committed.
4624 : *
4625 : * If a dangerous structure is found, the pivot (the near conflict) is
4626 : * marked for death, because rolling back another transaction might mean
4627 : * that we fail without ever making progress. This transaction is
4628 : * committing writes, so letting it commit ensures progress. If we
4629 : * canceled the far conflict, it might immediately fail again on retry.
4630 : */
4631 : void
4632 592094 : PreCommit_CheckForSerializationFailure(void)
4633 : {
4634 : dlist_iter near_iter;
4635 :
4636 592094 : if (MySerializableXact == InvalidSerializableXact)
4637 590682 : return;
4638 :
4639 : Assert(IsolationIsSerializable());
4640 :
4641 1412 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
4642 :
4643 : /*
4644 : * Check if someone else has already decided that we need to die. Since
4645 : * we set our own DOOMED flag when partially releasing, ignore in that
4646 : * case.
4647 : */
4648 1412 : if (SxactIsDoomed(MySerializableXact) &&
4649 156 : !SxactIsPartiallyReleased(MySerializableXact))
4650 : {
4651 155 : LWLockRelease(SerializableXactHashLock);
4652 155 : ereport(ERROR,
4653 : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4654 : errmsg("could not serialize access due to read/write dependencies among transactions"),
4655 : errdetail_internal("Reason code: Canceled on identification as a pivot, during commit attempt."),
4656 : errhint("The transaction might succeed if retried.")));
4657 : }
4658 :
4659 1863 : dlist_foreach(near_iter, &MySerializableXact->inConflicts)
4660 : {
4661 606 : RWConflict nearConflict =
4662 606 : dlist_container(RWConflictData, inLink, near_iter.cur);
4663 :
4664 606 : if (!SxactIsCommitted(nearConflict->sxactOut)
4665 421 : && !SxactIsDoomed(nearConflict->sxactOut))
4666 : {
4667 : dlist_iter far_iter;
4668 :
4669 451 : dlist_foreach(far_iter, &nearConflict->sxactOut->inConflicts)
4670 : {
4671 182 : RWConflict farConflict =
4672 182 : dlist_container(RWConflictData, inLink, far_iter.cur);
4673 :
4674 182 : if (farConflict->sxactOut == MySerializableXact
4675 42 : || (!SxactIsCommitted(farConflict->sxactOut)
4676 24 : && !SxactIsReadOnly(farConflict->sxactOut)
4677 12 : && !SxactIsDoomed(farConflict->sxactOut)))
4678 : {
4679 : /*
4680 : * Normally, we kill the pivot transaction to make sure we
4681 : * make progress if the failing transaction is retried.
4682 : * However, we can't kill it if it's already prepared, so
4683 : * in that case we commit suicide instead.
4684 : */
4685 152 : if (SxactIsPrepared(nearConflict->sxactOut))
4686 : {
4687 0 : LWLockRelease(SerializableXactHashLock);
4688 0 : ereport(ERROR,
4689 : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4690 : errmsg("could not serialize access due to read/write dependencies among transactions"),
4691 : errdetail_internal("Reason code: Canceled on commit attempt with conflict in from prepared pivot."),
4692 : errhint("The transaction might succeed if retried.")));
4693 : }
4694 152 : nearConflict->sxactOut->flags |= SXACT_FLAG_DOOMED;
4695 152 : break;
4696 : }
4697 : }
4698 : }
4699 : }
4700 :
4701 1257 : MySerializableXact->prepareSeqNo = ++(PredXact->LastSxactCommitSeqNo);
4702 1257 : MySerializableXact->flags |= SXACT_FLAG_PREPARED;
4703 :
4704 1257 : LWLockRelease(SerializableXactHashLock);
4705 : }
4706 :
4707 : /*------------------------------------------------------------------------*/
4708 :
4709 : /*
4710 : * Two-phase commit support
4711 : */
4712 :
4713 : /*
4714 : * AtPrepare_Locks
4715 : * Do the preparatory work for a PREPARE: make 2PC state file
4716 : * records for all predicate locks currently held.
4717 : */
4718 : void
4719 320 : AtPrepare_PredicateLocks(void)
4720 : {
4721 : SERIALIZABLEXACT *sxact;
4722 : TwoPhasePredicateRecord record;
4723 : TwoPhasePredicateXactRecord *xactRecord;
4724 : TwoPhasePredicateLockRecord *lockRecord;
4725 : dlist_iter iter;
4726 :
4727 320 : sxact = MySerializableXact;
4728 320 : xactRecord = &(record.data.xactRecord);
4729 320 : lockRecord = &(record.data.lockRecord);
4730 :
4731 320 : if (MySerializableXact == InvalidSerializableXact)
4732 308 : return;
4733 :
4734 : /* Generate an xact record for our SERIALIZABLEXACT */
4735 12 : record.type = TWOPHASEPREDICATERECORD_XACT;
4736 12 : xactRecord->xmin = MySerializableXact->xmin;
4737 12 : xactRecord->flags = MySerializableXact->flags;
4738 :
4739 : /*
4740 : * Note that we don't include the list of conflicts in our out in the
4741 : * statefile, because new conflicts can be added even after the
4742 : * transaction prepares. We'll just make a conservative assumption during
4743 : * recovery instead.
4744 : */
4745 :
4746 12 : RegisterTwoPhaseRecord(TWOPHASE_RM_PREDICATELOCK_ID, 0,
4747 : &record, sizeof(record));
4748 :
4749 : /*
4750 : * Generate a lock record for each lock.
4751 : *
4752 : * To do this, we need to walk the predicate lock list in our sxact rather
4753 : * than using the local predicate lock table because the latter is not
4754 : * guaranteed to be accurate.
4755 : */
4756 12 : LWLockAcquire(SerializablePredicateListLock, LW_SHARED);
4757 :
4758 : /*
4759 : * No need to take sxact->perXactPredicateListLock in parallel mode
4760 : * because there cannot be any parallel workers running while we are
4761 : * preparing a transaction.
4762 : */
4763 : Assert(!IsParallelWorker() && !ParallelContextActive());
4764 :
4765 22 : dlist_foreach(iter, &sxact->predicateLocks)
4766 : {
4767 10 : PREDICATELOCK *predlock =
4768 10 : dlist_container(PREDICATELOCK, xactLink, iter.cur);
4769 :
4770 10 : record.type = TWOPHASEPREDICATERECORD_LOCK;
4771 10 : lockRecord->target = predlock->tag.myTarget->tag;
4772 :
4773 10 : RegisterTwoPhaseRecord(TWOPHASE_RM_PREDICATELOCK_ID, 0,
4774 : &record, sizeof(record));
4775 : }
4776 :
4777 12 : LWLockRelease(SerializablePredicateListLock);
4778 : }
4779 :
4780 : /*
4781 : * PostPrepare_Locks
4782 : * Clean up after successful PREPARE. Unlike the non-predicate
4783 : * lock manager, we do not need to transfer locks to a dummy
4784 : * PGPROC because our SERIALIZABLEXACT will stay around
4785 : * anyway. We only need to clean up our local state.
4786 : */
4787 : void
4788 320 : PostPrepare_PredicateLocks(FullTransactionId fxid)
4789 : {
4790 320 : if (MySerializableXact == InvalidSerializableXact)
4791 308 : return;
4792 :
4793 : Assert(SxactIsPrepared(MySerializableXact));
4794 :
4795 12 : MySerializableXact->pid = 0;
4796 12 : MySerializableXact->pgprocno = INVALID_PROC_NUMBER;
4797 :
4798 12 : hash_destroy(LocalPredicateLockHash);
4799 12 : LocalPredicateLockHash = NULL;
4800 :
4801 12 : MySerializableXact = InvalidSerializableXact;
4802 12 : MyXactDidWrite = false;
4803 : }
4804 :
4805 : /*
4806 : * PredicateLockTwoPhaseFinish
4807 : * Release a prepared transaction's predicate locks once it
4808 : * commits or aborts.
4809 : */
4810 : void
4811 327 : PredicateLockTwoPhaseFinish(FullTransactionId fxid, bool isCommit)
4812 : {
4813 : SERIALIZABLEXID *sxid;
4814 : SERIALIZABLEXIDTAG sxidtag;
4815 :
4816 327 : sxidtag.xid = XidFromFullTransactionId(fxid);
4817 :
4818 327 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
4819 : sxid = (SERIALIZABLEXID *)
4820 327 : hash_search(SerializableXidHash, &sxidtag, HASH_FIND, NULL);
4821 327 : LWLockRelease(SerializableXactHashLock);
4822 :
4823 : /* xid will not be found if it wasn't a serializable transaction */
4824 327 : if (sxid == NULL)
4825 315 : return;
4826 :
4827 : /* Release its locks */
4828 12 : MySerializableXact = sxid->myXact;
4829 12 : MyXactDidWrite = true; /* conservatively assume that we wrote
4830 : * something */
4831 12 : ReleasePredicateLocks(isCommit, false);
4832 : }
4833 :
4834 : /*
4835 : * Re-acquire a predicate lock belonging to a transaction that was prepared.
4836 : */
4837 : void
4838 0 : predicatelock_twophase_recover(FullTransactionId fxid, uint16 info,
4839 : void *recdata, uint32 len)
4840 : {
4841 : TwoPhasePredicateRecord *record;
4842 0 : TransactionId xid = XidFromFullTransactionId(fxid);
4843 :
4844 : Assert(len == sizeof(TwoPhasePredicateRecord));
4845 :
4846 0 : record = (TwoPhasePredicateRecord *) recdata;
4847 :
4848 : Assert((record->type == TWOPHASEPREDICATERECORD_XACT) ||
4849 : (record->type == TWOPHASEPREDICATERECORD_LOCK));
4850 :
4851 0 : if (record->type == TWOPHASEPREDICATERECORD_XACT)
4852 : {
4853 : /* Per-transaction record. Set up a SERIALIZABLEXACT. */
4854 : TwoPhasePredicateXactRecord *xactRecord;
4855 : SERIALIZABLEXACT *sxact;
4856 : SERIALIZABLEXID *sxid;
4857 : SERIALIZABLEXIDTAG sxidtag;
4858 : bool found;
4859 :
4860 0 : xactRecord = (TwoPhasePredicateXactRecord *) &record->data.xactRecord;
4861 :
4862 0 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
4863 0 : sxact = CreatePredXact();
4864 0 : if (!sxact)
4865 0 : ereport(ERROR,
4866 : (errcode(ERRCODE_OUT_OF_MEMORY),
4867 : errmsg("out of shared memory")));
4868 :
4869 : /* vxid for a prepared xact is INVALID_PROC_NUMBER/xid; no pid */
4870 0 : sxact->vxid.procNumber = INVALID_PROC_NUMBER;
4871 0 : sxact->vxid.localTransactionId = (LocalTransactionId) xid;
4872 0 : sxact->pid = 0;
4873 0 : sxact->pgprocno = INVALID_PROC_NUMBER;
4874 :
4875 : /* a prepared xact hasn't committed yet */
4876 0 : sxact->prepareSeqNo = RecoverySerCommitSeqNo;
4877 0 : sxact->commitSeqNo = InvalidSerCommitSeqNo;
4878 0 : sxact->finishedBefore = InvalidTransactionId;
4879 :
4880 0 : sxact->SeqNo.lastCommitBeforeSnapshot = RecoverySerCommitSeqNo;
4881 :
4882 : /*
4883 : * Don't need to track this; no transactions running at the time the
4884 : * recovered xact started are still active, except possibly other
4885 : * prepared xacts and we don't care whether those are RO_SAFE or not.
4886 : */
4887 0 : dlist_init(&(sxact->possibleUnsafeConflicts));
4888 :
4889 0 : dlist_init(&(sxact->predicateLocks));
4890 0 : dlist_node_init(&sxact->finishedLink);
4891 :
4892 0 : sxact->topXid = xid;
4893 0 : sxact->xmin = xactRecord->xmin;
4894 0 : sxact->flags = xactRecord->flags;
4895 : Assert(SxactIsPrepared(sxact));
4896 0 : if (!SxactIsReadOnly(sxact))
4897 : {
4898 0 : ++(PredXact->WritableSxactCount);
4899 : Assert(PredXact->WritableSxactCount <=
4900 : (MaxBackends + max_prepared_xacts));
4901 : }
4902 :
4903 : /*
4904 : * We don't know whether the transaction had any conflicts or not, so
4905 : * we'll conservatively assume that it had both a conflict in and a
4906 : * conflict out, and represent that with the summary conflict flags.
4907 : */
4908 0 : dlist_init(&(sxact->outConflicts));
4909 0 : dlist_init(&(sxact->inConflicts));
4910 0 : sxact->flags |= SXACT_FLAG_SUMMARY_CONFLICT_IN;
4911 0 : sxact->flags |= SXACT_FLAG_SUMMARY_CONFLICT_OUT;
4912 :
4913 : /* Register the transaction's xid */
4914 0 : sxidtag.xid = xid;
4915 0 : sxid = (SERIALIZABLEXID *) hash_search(SerializableXidHash,
4916 : &sxidtag,
4917 : HASH_ENTER, &found);
4918 : Assert(sxid != NULL);
4919 : Assert(!found);
4920 0 : sxid->myXact = sxact;
4921 :
4922 : /*
4923 : * Update global xmin. Note that this is a special case compared to
4924 : * registering a normal transaction, because the global xmin might go
4925 : * backwards. That's OK, because until recovery is over we're not
4926 : * going to complete any transactions or create any non-prepared
4927 : * transactions, so there's no danger of throwing away.
4928 : */
4929 0 : if ((!TransactionIdIsValid(PredXact->SxactGlobalXmin)) ||
4930 0 : (TransactionIdFollows(PredXact->SxactGlobalXmin, sxact->xmin)))
4931 : {
4932 0 : PredXact->SxactGlobalXmin = sxact->xmin;
4933 0 : PredXact->SxactGlobalXminCount = 1;
4934 0 : SerialSetActiveSerXmin(sxact->xmin);
4935 : }
4936 0 : else if (TransactionIdEquals(sxact->xmin, PredXact->SxactGlobalXmin))
4937 : {
4938 : Assert(PredXact->SxactGlobalXminCount > 0);
4939 0 : PredXact->SxactGlobalXminCount++;
4940 : }
4941 :
4942 0 : LWLockRelease(SerializableXactHashLock);
4943 : }
4944 0 : else if (record->type == TWOPHASEPREDICATERECORD_LOCK)
4945 : {
4946 : /* Lock record. Recreate the PREDICATELOCK */
4947 : TwoPhasePredicateLockRecord *lockRecord;
4948 : SERIALIZABLEXID *sxid;
4949 : SERIALIZABLEXACT *sxact;
4950 : SERIALIZABLEXIDTAG sxidtag;
4951 : uint32 targettaghash;
4952 :
4953 0 : lockRecord = (TwoPhasePredicateLockRecord *) &record->data.lockRecord;
4954 0 : targettaghash = PredicateLockTargetTagHashCode(&lockRecord->target);
4955 :
4956 0 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
4957 0 : sxidtag.xid = xid;
4958 : sxid = (SERIALIZABLEXID *)
4959 0 : hash_search(SerializableXidHash, &sxidtag, HASH_FIND, NULL);
4960 0 : LWLockRelease(SerializableXactHashLock);
4961 :
4962 : Assert(sxid != NULL);
4963 0 : sxact = sxid->myXact;
4964 : Assert(sxact != InvalidSerializableXact);
4965 :
4966 0 : CreatePredicateLock(&lockRecord->target, targettaghash, sxact);
4967 : }
4968 0 : }
4969 :
4970 : /*
4971 : * Prepare to share the current SERIALIZABLEXACT with parallel workers.
4972 : * Return a handle object that can be used by AttachSerializableXact() in a
4973 : * parallel worker.
4974 : */
4975 : SerializableXactHandle
4976 680 : ShareSerializableXact(void)
4977 : {
4978 680 : return MySerializableXact;
4979 : }
4980 :
4981 : /*
4982 : * Allow parallel workers to import the leader's SERIALIZABLEXACT.
4983 : */
4984 : void
4985 2010 : AttachSerializableXact(SerializableXactHandle handle)
4986 : {
4987 :
4988 : Assert(MySerializableXact == InvalidSerializableXact);
4989 :
4990 2010 : MySerializableXact = (SERIALIZABLEXACT *) handle;
4991 2010 : if (MySerializableXact != InvalidSerializableXact)
4992 13 : CreateLocalPredicateLockHash();
4993 2010 : }
|
