Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * heapam.c
4 : * heap access method code
5 : *
6 : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/access/heap/heapam.c
12 : *
13 : *
14 : * INTERFACE ROUTINES
15 : * heap_beginscan - begin relation scan
16 : * heap_rescan - restart a relation scan
17 : * heap_endscan - end relation scan
18 : * heap_getnext - retrieve next tuple in scan
19 : * heap_fetch - retrieve tuple with given tid
20 : * heap_insert - insert tuple into a relation
21 : * heap_multi_insert - insert multiple tuples into a relation
22 : * heap_delete - delete a tuple from a relation
23 : * heap_update - replace a tuple in a relation with another tuple
24 : *
25 : * NOTES
26 : * This file contains the heap_ routines which implement
27 : * the POSTGRES heap access method used for all POSTGRES
28 : * relations.
29 : *
30 : *-------------------------------------------------------------------------
31 : */
32 : #include "postgres.h"
33 :
34 : #include "access/heapam.h"
35 : #include "access/heaptoast.h"
36 : #include "access/hio.h"
37 : #include "access/multixact.h"
38 : #include "access/subtrans.h"
39 : #include "access/syncscan.h"
40 : #include "access/valid.h"
41 : #include "access/visibilitymap.h"
42 : #include "access/xloginsert.h"
43 : #include "catalog/pg_database.h"
44 : #include "catalog/pg_database_d.h"
45 : #include "commands/vacuum.h"
46 : #include "pgstat.h"
47 : #include "port/pg_bitutils.h"
48 : #include "storage/lmgr.h"
49 : #include "storage/predicate.h"
50 : #include "storage/proc.h"
51 : #include "storage/procarray.h"
52 : #include "utils/datum.h"
53 : #include "utils/injection_point.h"
54 : #include "utils/inval.h"
55 : #include "utils/spccache.h"
56 : #include "utils/syscache.h"
57 :
58 :
59 : static HeapTuple heap_prepare_insert(Relation relation, HeapTuple tup,
60 : TransactionId xid, CommandId cid, int options);
61 : static XLogRecPtr log_heap_update(Relation reln, Buffer oldbuf,
62 : Buffer newbuf, HeapTuple oldtup,
63 : HeapTuple newtup, HeapTuple old_key_tuple,
64 : bool all_visible_cleared, bool new_all_visible_cleared);
65 : #ifdef USE_ASSERT_CHECKING
66 : static void check_lock_if_inplace_updateable_rel(Relation relation,
67 : const ItemPointerData *otid,
68 : HeapTuple newtup);
69 : static void check_inplace_rel_lock(HeapTuple oldtup);
70 : #endif
71 : static Bitmapset *HeapDetermineColumnsInfo(Relation relation,
72 : Bitmapset *interesting_cols,
73 : Bitmapset *external_cols,
74 : HeapTuple oldtup, HeapTuple newtup,
75 : bool *has_external);
76 : static bool heap_acquire_tuplock(Relation relation, const ItemPointerData *tid,
77 : LockTupleMode mode, LockWaitPolicy wait_policy,
78 : bool *have_tuple_lock);
79 : static inline BlockNumber heapgettup_advance_block(HeapScanDesc scan,
80 : BlockNumber block,
81 : ScanDirection dir);
82 : static pg_noinline BlockNumber heapgettup_initial_block(HeapScanDesc scan,
83 : ScanDirection dir);
84 : static void compute_new_xmax_infomask(TransactionId xmax, uint16 old_infomask,
85 : uint16 old_infomask2, TransactionId add_to_xmax,
86 : LockTupleMode mode, bool is_update,
87 : TransactionId *result_xmax, uint16 *result_infomask,
88 : uint16 *result_infomask2);
89 : static TM_Result heap_lock_updated_tuple(Relation rel,
90 : uint16 prior_infomask,
91 : TransactionId prior_raw_xmax,
92 : const ItemPointerData *prior_ctid,
93 : TransactionId xid,
94 : LockTupleMode mode);
95 : static void GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
96 : uint16 *new_infomask2);
97 : static TransactionId MultiXactIdGetUpdateXid(TransactionId xmax,
98 : uint16 t_infomask);
99 : static bool DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
100 : LockTupleMode lockmode, bool *current_is_member);
101 : static void MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
102 : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
103 : int *remaining);
104 : static bool ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status,
105 : uint16 infomask, Relation rel, int *remaining,
106 : bool logLockFailure);
107 : static void index_delete_sort(TM_IndexDeleteOp *delstate);
108 : static int bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate);
109 : static XLogRecPtr log_heap_new_cid(Relation relation, HeapTuple tup);
110 : static HeapTuple ExtractReplicaIdentity(Relation relation, HeapTuple tp, bool key_required,
111 : bool *copy);
112 :
113 :
114 : /*
115 : * This table lists the heavyweight lock mode that corresponds to each tuple
116 : * lock mode, as well as one or two corresponding MultiXactStatus values:
117 : * .lockstatus to merely lock tuples, and .updstatus to update them. The
118 : * latter is set to -1 if the corresponding tuple lock mode does not allow
119 : * updating tuples -- see get_mxact_status_for_lock().
120 : *
121 : * These interact with InplaceUpdateTupleLock, an alias for ExclusiveLock.
122 : *
123 : * Don't look at lockstatus/updstatus directly! Use get_mxact_status_for_lock
124 : * instead.
125 : */
126 : static const struct
127 : {
128 : LOCKMODE hwlock;
129 : int lockstatus;
130 : int updstatus;
131 : } tupleLockExtraInfo[] =
132 :
133 : {
134 : [LockTupleKeyShare] = {
135 : .hwlock = AccessShareLock,
136 : .lockstatus = MultiXactStatusForKeyShare,
137 : /* KeyShare does not allow updating tuples */
138 : .updstatus = -1
139 : },
140 : [LockTupleShare] = {
141 : .hwlock = RowShareLock,
142 : .lockstatus = MultiXactStatusForShare,
143 : /* Share does not allow updating tuples */
144 : .updstatus = -1
145 : },
146 : [LockTupleNoKeyExclusive] = {
147 : .hwlock = ExclusiveLock,
148 : .lockstatus = MultiXactStatusForNoKeyUpdate,
149 : .updstatus = MultiXactStatusNoKeyUpdate
150 : },
151 : [LockTupleExclusive] = {
152 : .hwlock = AccessExclusiveLock,
153 : .lockstatus = MultiXactStatusForUpdate,
154 : .updstatus = MultiXactStatusUpdate
155 : }
156 : };
157 :
158 : /* Get the LOCKMODE for a given MultiXactStatus */
159 : #define LOCKMODE_from_mxstatus(status) \
160 : (tupleLockExtraInfo[TUPLOCK_from_mxstatus((status))].hwlock)
161 :
162 : /*
163 : * Acquire heavyweight locks on tuples, using a LockTupleMode strength value.
164 : * This is more readable than having every caller translate it to lock.h's
165 : * LOCKMODE.
166 : */
167 : #define LockTupleTuplock(rel, tup, mode) \
168 : LockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
169 : #define UnlockTupleTuplock(rel, tup, mode) \
170 : UnlockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
171 : #define ConditionalLockTupleTuplock(rel, tup, mode, log) \
172 : ConditionalLockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock, (log))
173 :
174 : #ifdef USE_PREFETCH
175 : /*
176 : * heap_index_delete_tuples and index_delete_prefetch_buffer use this
177 : * structure to coordinate prefetching activity
178 : */
179 : typedef struct
180 : {
181 : BlockNumber cur_hblkno;
182 : int next_item;
183 : int ndeltids;
184 : TM_IndexDelete *deltids;
185 : } IndexDeletePrefetchState;
186 : #endif
187 :
188 : /* heap_index_delete_tuples bottom-up index deletion costing constants */
189 : #define BOTTOMUP_MAX_NBLOCKS 6
190 : #define BOTTOMUP_TOLERANCE_NBLOCKS 3
191 :
192 : /*
193 : * heap_index_delete_tuples uses this when determining which heap blocks it
194 : * must visit to help its bottom-up index deletion caller
195 : */
196 : typedef struct IndexDeleteCounts
197 : {
198 : int16 npromisingtids; /* Number of "promising" TIDs in group */
199 : int16 ntids; /* Number of TIDs in group */
200 : int16 ifirsttid; /* Offset to group's first deltid */
201 : } IndexDeleteCounts;
202 :
203 : /*
204 : * This table maps tuple lock strength values for each particular
205 : * MultiXactStatus value.
206 : */
207 : static const int MultiXactStatusLock[MaxMultiXactStatus + 1] =
208 : {
209 : LockTupleKeyShare, /* ForKeyShare */
210 : LockTupleShare, /* ForShare */
211 : LockTupleNoKeyExclusive, /* ForNoKeyUpdate */
212 : LockTupleExclusive, /* ForUpdate */
213 : LockTupleNoKeyExclusive, /* NoKeyUpdate */
214 : LockTupleExclusive /* Update */
215 : };
216 :
217 : /* Get the LockTupleMode for a given MultiXactStatus */
218 : #define TUPLOCK_from_mxstatus(status) \
219 : (MultiXactStatusLock[(status)])
220 :
221 : /*
222 : * Check that we have a valid snapshot if we might need TOAST access.
223 : */
224 : static inline void
225 12746958 : AssertHasSnapshotForToast(Relation rel)
226 : {
227 : #ifdef USE_ASSERT_CHECKING
228 :
229 : /* bootstrap mode in particular breaks this rule */
230 : if (!IsNormalProcessingMode())
231 : return;
232 :
233 : /* if the relation doesn't have a TOAST table, we are good */
234 : if (!OidIsValid(rel->rd_rel->reltoastrelid))
235 : return;
236 :
237 : Assert(HaveRegisteredOrActiveSnapshot());
238 :
239 : #endif /* USE_ASSERT_CHECKING */
240 12746958 : }
241 :
242 : /* ----------------------------------------------------------------
243 : * heap support routines
244 : * ----------------------------------------------------------------
245 : */
246 :
247 : /*
248 : * Streaming read API callback for parallel sequential scans. Returns the next
249 : * block the caller wants from the read stream or InvalidBlockNumber when done.
250 : */
251 : static BlockNumber
252 132439 : heap_scan_stream_read_next_parallel(ReadStream *stream,
253 : void *callback_private_data,
254 : void *per_buffer_data)
255 : {
256 132439 : HeapScanDesc scan = (HeapScanDesc) callback_private_data;
257 :
258 : Assert(ScanDirectionIsForward(scan->rs_dir));
259 : Assert(scan->rs_base.rs_parallel);
260 :
261 132439 : if (unlikely(!scan->rs_inited))
262 : {
263 : /* parallel scan */
264 2237 : table_block_parallelscan_startblock_init(scan->rs_base.rs_rd,
265 2237 : scan->rs_parallelworkerdata,
266 2237 : (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel,
267 : scan->rs_startblock,
268 : scan->rs_numblocks);
269 :
270 : /* may return InvalidBlockNumber if there are no more blocks */
271 4474 : scan->rs_prefetch_block = table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
272 2237 : scan->rs_parallelworkerdata,
273 2237 : (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel);
274 2237 : scan->rs_inited = true;
275 : }
276 : else
277 : {
278 130202 : scan->rs_prefetch_block = table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
279 130202 : scan->rs_parallelworkerdata, (ParallelBlockTableScanDesc)
280 130202 : scan->rs_base.rs_parallel);
281 : }
282 :
283 132439 : return scan->rs_prefetch_block;
284 : }
285 :
286 : /*
287 : * Streaming read API callback for serial sequential and TID range scans.
288 : * Returns the next block the caller wants from the read stream or
289 : * InvalidBlockNumber when done.
290 : */
291 : static BlockNumber
292 5497957 : heap_scan_stream_read_next_serial(ReadStream *stream,
293 : void *callback_private_data,
294 : void *per_buffer_data)
295 : {
296 5497957 : HeapScanDesc scan = (HeapScanDesc) callback_private_data;
297 :
298 5497957 : if (unlikely(!scan->rs_inited))
299 : {
300 1601696 : scan->rs_prefetch_block = heapgettup_initial_block(scan, scan->rs_dir);
301 1601696 : scan->rs_inited = true;
302 : }
303 : else
304 3896261 : scan->rs_prefetch_block = heapgettup_advance_block(scan,
305 : scan->rs_prefetch_block,
306 : scan->rs_dir);
307 :
308 5497957 : return scan->rs_prefetch_block;
309 : }
310 :
311 : /*
312 : * Read stream API callback for bitmap heap scans.
313 : * Returns the next block the caller wants from the read stream or
314 : * InvalidBlockNumber when done.
315 : */
316 : static BlockNumber
317 263978 : bitmapheap_stream_read_next(ReadStream *pgsr, void *private_data,
318 : void *per_buffer_data)
319 : {
320 263978 : TBMIterateResult *tbmres = per_buffer_data;
321 263978 : BitmapHeapScanDesc bscan = (BitmapHeapScanDesc) private_data;
322 263978 : HeapScanDesc hscan = (HeapScanDesc) bscan;
323 263978 : TableScanDesc sscan = &hscan->rs_base;
324 :
325 : for (;;)
326 : {
327 263978 : CHECK_FOR_INTERRUPTS();
328 :
329 : /* no more entries in the bitmap */
330 263978 : if (!tbm_iterate(&sscan->st.rs_tbmiterator, tbmres))
331 15631 : return InvalidBlockNumber;
332 :
333 : /*
334 : * Ignore any claimed entries past what we think is the end of the
335 : * relation. It may have been extended after the start of our scan (we
336 : * only hold an AccessShareLock, and it could be inserts from this
337 : * backend). We don't take this optimization in SERIALIZABLE
338 : * isolation though, as we need to examine all invisible tuples
339 : * reachable by the index.
340 : */
341 248347 : if (!IsolationIsSerializable() &&
342 248238 : tbmres->blockno >= hscan->rs_nblocks)
343 0 : continue;
344 :
345 248347 : return tbmres->blockno;
346 : }
347 :
348 : /* not reachable */
349 : Assert(false);
350 : }
351 :
352 : /* ----------------
353 : * initscan - scan code common to heap_beginscan and heap_rescan
354 : * ----------------
355 : */
356 : static void
357 1631579 : initscan(HeapScanDesc scan, ScanKey key, bool keep_startblock)
358 : {
359 1631579 : ParallelBlockTableScanDesc bpscan = NULL;
360 : bool allow_strat;
361 : bool allow_sync;
362 :
363 : /*
364 : * Determine the number of blocks we have to scan.
365 : *
366 : * It is sufficient to do this once at scan start, since any tuples added
367 : * while the scan is in progress will be invisible to my snapshot anyway.
368 : * (That is not true when using a non-MVCC snapshot. However, we couldn't
369 : * guarantee to return tuples added after scan start anyway, since they
370 : * might go into pages we already scanned. To guarantee consistent
371 : * results for a non-MVCC snapshot, the caller must hold some higher-level
372 : * lock that ensures the interesting tuple(s) won't change.)
373 : */
374 1631579 : if (scan->rs_base.rs_parallel != NULL)
375 : {
376 3017 : bpscan = (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel;
377 3017 : scan->rs_nblocks = bpscan->phs_nblocks;
378 : }
379 : else
380 1628562 : scan->rs_nblocks = RelationGetNumberOfBlocks(scan->rs_base.rs_rd);
381 :
382 : /*
383 : * If the table is large relative to NBuffers, use a bulk-read access
384 : * strategy and enable synchronized scanning (see syncscan.c). Although
385 : * the thresholds for these features could be different, we make them the
386 : * same so that there are only two behaviors to tune rather than four.
387 : * (However, some callers need to be able to disable one or both of these
388 : * behaviors, independently of the size of the table; also there is a GUC
389 : * variable that can disable synchronized scanning.)
390 : *
391 : * Note that table_block_parallelscan_initialize has a very similar test;
392 : * if you change this, consider changing that one, too.
393 : */
394 1631577 : if (!RelationUsesLocalBuffers(scan->rs_base.rs_rd) &&
395 1621796 : scan->rs_nblocks > NBuffers / 4)
396 : {
397 14305 : allow_strat = (scan->rs_base.rs_flags & SO_ALLOW_STRAT) != 0;
398 14305 : allow_sync = (scan->rs_base.rs_flags & SO_ALLOW_SYNC) != 0;
399 : }
400 : else
401 1617272 : allow_strat = allow_sync = false;
402 :
403 1631577 : if (allow_strat)
404 : {
405 : /* During a rescan, keep the previous strategy object. */
406 12966 : if (scan->rs_strategy == NULL)
407 12771 : scan->rs_strategy = GetAccessStrategy(BAS_BULKREAD);
408 : }
409 : else
410 : {
411 1618611 : if (scan->rs_strategy != NULL)
412 0 : FreeAccessStrategy(scan->rs_strategy);
413 1618611 : scan->rs_strategy = NULL;
414 : }
415 :
416 1631577 : if (scan->rs_base.rs_parallel != NULL)
417 : {
418 : /* For parallel scan, believe whatever ParallelTableScanDesc says. */
419 3017 : if (scan->rs_base.rs_parallel->phs_syncscan)
420 2 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
421 : else
422 3015 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
423 :
424 : /*
425 : * If not rescanning, initialize the startblock. Finding the actual
426 : * start location is done in table_block_parallelscan_startblock_init,
427 : * based on whether an alternative start location has been set with
428 : * heap_setscanlimits, or using the syncscan location, when syncscan
429 : * is enabled.
430 : */
431 3017 : if (!keep_startblock)
432 2865 : scan->rs_startblock = InvalidBlockNumber;
433 : }
434 : else
435 : {
436 1628560 : if (keep_startblock)
437 : {
438 : /*
439 : * When rescanning, we want to keep the previous startblock
440 : * setting, so that rewinding a cursor doesn't generate surprising
441 : * results. Reset the active syncscan setting, though.
442 : */
443 864463 : if (allow_sync && synchronize_seqscans)
444 50 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
445 : else
446 864413 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
447 : }
448 764097 : else if (allow_sync && synchronize_seqscans)
449 : {
450 73 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
451 73 : scan->rs_startblock = ss_get_location(scan->rs_base.rs_rd, scan->rs_nblocks);
452 : }
453 : else
454 : {
455 764024 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
456 764024 : scan->rs_startblock = 0;
457 : }
458 : }
459 :
460 1631577 : scan->rs_numblocks = InvalidBlockNumber;
461 1631577 : scan->rs_inited = false;
462 1631577 : scan->rs_ctup.t_data = NULL;
463 1631577 : ItemPointerSetInvalid(&scan->rs_ctup.t_self);
464 1631577 : scan->rs_cbuf = InvalidBuffer;
465 1631577 : scan->rs_cblock = InvalidBlockNumber;
466 1631577 : scan->rs_ntuples = 0;
467 1631577 : scan->rs_cindex = 0;
468 :
469 : /*
470 : * Initialize to ForwardScanDirection because it is most common and
471 : * because heap scans go forward before going backward (e.g. CURSORs).
472 : */
473 1631577 : scan->rs_dir = ForwardScanDirection;
474 1631577 : scan->rs_prefetch_block = InvalidBlockNumber;
475 :
476 : /* page-at-a-time fields are always invalid when not rs_inited */
477 :
478 : /*
479 : * copy the scan key, if appropriate
480 : */
481 1631577 : if (key != NULL && scan->rs_base.rs_nkeys > 0)
482 252249 : memcpy(scan->rs_base.rs_key, key, scan->rs_base.rs_nkeys * sizeof(ScanKeyData));
483 :
484 : /*
485 : * Currently, we only have a stats counter for sequential heap scans (but
486 : * e.g for bitmap scans the underlying bitmap index scans will be counted,
487 : * and for sample scans we update stats for tuple fetches).
488 : */
489 1631577 : if (scan->rs_base.rs_flags & SO_TYPE_SEQSCAN)
490 1603374 : pgstat_count_heap_scan(scan->rs_base.rs_rd);
491 1631577 : }
492 :
493 : /*
494 : * heap_setscanlimits - restrict range of a heapscan
495 : *
496 : * startBlk is the page to start at
497 : * numBlks is number of pages to scan (InvalidBlockNumber means "all")
498 : */
499 : void
500 3274 : heap_setscanlimits(TableScanDesc sscan, BlockNumber startBlk, BlockNumber numBlks)
501 : {
502 3274 : HeapScanDesc scan = (HeapScanDesc) sscan;
503 :
504 : Assert(!scan->rs_inited); /* else too late to change */
505 : /* else rs_startblock is significant */
506 : Assert(!(scan->rs_base.rs_flags & SO_ALLOW_SYNC));
507 :
508 : /* Check startBlk is valid (but allow case of zero blocks...) */
509 : Assert(startBlk == 0 || startBlk < scan->rs_nblocks);
510 :
511 3274 : scan->rs_startblock = startBlk;
512 3274 : scan->rs_numblocks = numBlks;
513 3274 : }
514 :
515 : /*
516 : * Per-tuple loop for heap_prepare_pagescan(). Pulled out so it can be called
517 : * multiple times, with constant arguments for all_visible,
518 : * check_serializable.
519 : */
520 : pg_attribute_always_inline
521 : static int
522 3891536 : page_collect_tuples(HeapScanDesc scan, Snapshot snapshot,
523 : Page page, Buffer buffer,
524 : BlockNumber block, int lines,
525 : bool all_visible, bool check_serializable)
526 : {
527 3891536 : Oid relid = RelationGetRelid(scan->rs_base.rs_rd);
528 3891536 : int ntup = 0;
529 3891536 : int nvis = 0;
530 : BatchMVCCState batchmvcc;
531 :
532 : /* page at a time should have been disabled otherwise */
533 : Assert(IsMVCCSnapshot(snapshot));
534 :
535 : /* first find all tuples on the page */
536 213691236 : for (OffsetNumber lineoff = FirstOffsetNumber; lineoff <= lines; lineoff++)
537 : {
538 209799700 : ItemId lpp = PageGetItemId(page, lineoff);
539 : HeapTuple tup;
540 :
541 209799700 : if (unlikely(!ItemIdIsNormal(lpp)))
542 42499806 : continue;
543 :
544 : /*
545 : * If the page is not all-visible or we need to check serializability,
546 : * maintain enough state to be able to refind the tuple efficiently,
547 : * without again first needing to fetch the item and then via that the
548 : * tuple.
549 : */
550 167299894 : if (!all_visible || check_serializable)
551 : {
552 83370084 : tup = &batchmvcc.tuples[ntup];
553 :
554 83370084 : tup->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
555 83370084 : tup->t_len = ItemIdGetLength(lpp);
556 83370084 : tup->t_tableOid = relid;
557 83370084 : ItemPointerSet(&(tup->t_self), block, lineoff);
558 : }
559 :
560 : /*
561 : * If the page is all visible, these fields otherwise won't be
562 : * populated in loop below.
563 : */
564 167299894 : if (all_visible)
565 : {
566 83929810 : if (check_serializable)
567 : {
568 0 : batchmvcc.visible[ntup] = true;
569 : }
570 83929810 : scan->rs_vistuples[ntup] = lineoff;
571 : }
572 :
573 167299894 : ntup++;
574 : }
575 :
576 : Assert(ntup <= MaxHeapTuplesPerPage);
577 :
578 : /*
579 : * Unless the page is all visible, test visibility for all tuples one go.
580 : * That is considerably more efficient than calling
581 : * HeapTupleSatisfiesMVCC() one-by-one.
582 : */
583 3891536 : if (all_visible)
584 1582810 : nvis = ntup;
585 : else
586 2308726 : nvis = HeapTupleSatisfiesMVCCBatch(snapshot, buffer,
587 : ntup,
588 : &batchmvcc,
589 2308726 : scan->rs_vistuples);
590 :
591 : /*
592 : * So far we don't have batch API for testing serializabilty, so do so
593 : * one-by-one.
594 : */
595 3891536 : if (check_serializable)
596 : {
597 2076 : for (int i = 0; i < ntup; i++)
598 : {
599 1454 : HeapCheckForSerializableConflictOut(batchmvcc.visible[i],
600 : scan->rs_base.rs_rd,
601 : &batchmvcc.tuples[i],
602 : buffer, snapshot);
603 : }
604 : }
605 :
606 3891528 : return nvis;
607 : }
608 :
609 : /*
610 : * heap_prepare_pagescan - Prepare current scan page to be scanned in pagemode
611 : *
612 : * Preparation currently consists of 1. prune the scan's rs_cbuf page, and 2.
613 : * fill the rs_vistuples[] array with the OffsetNumbers of visible tuples.
614 : */
615 : void
616 3891536 : heap_prepare_pagescan(TableScanDesc sscan)
617 : {
618 3891536 : HeapScanDesc scan = (HeapScanDesc) sscan;
619 3891536 : Buffer buffer = scan->rs_cbuf;
620 3891536 : BlockNumber block = scan->rs_cblock;
621 : Snapshot snapshot;
622 : Page page;
623 : int lines;
624 : bool all_visible;
625 : bool check_serializable;
626 :
627 : Assert(BufferGetBlockNumber(buffer) == block);
628 :
629 : /* ensure we're not accidentally being used when not in pagemode */
630 : Assert(scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE);
631 3891536 : snapshot = scan->rs_base.rs_snapshot;
632 :
633 : /*
634 : * Prune and repair fragmentation for the whole page, if possible.
635 : */
636 3891536 : heap_page_prune_opt(scan->rs_base.rs_rd, buffer, &scan->rs_vmbuffer);
637 :
638 : /*
639 : * We must hold share lock on the buffer content while examining tuple
640 : * visibility. Afterwards, however, the tuples we have found to be
641 : * visible are guaranteed good as long as we hold the buffer pin.
642 : */
643 3891536 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
644 :
645 3891536 : page = BufferGetPage(buffer);
646 3891536 : lines = PageGetMaxOffsetNumber(page);
647 :
648 : /*
649 : * If the all-visible flag indicates that all tuples on the page are
650 : * visible to everyone, we can skip the per-tuple visibility tests.
651 : *
652 : * Note: In hot standby, a tuple that's already visible to all
653 : * transactions on the primary might still be invisible to a read-only
654 : * transaction in the standby. We partly handle this problem by tracking
655 : * the minimum xmin of visible tuples as the cut-off XID while marking a
656 : * page all-visible on the primary and WAL log that along with the
657 : * visibility map SET operation. In hot standby, we wait for (or abort)
658 : * all transactions that can potentially may not see one or more tuples on
659 : * the page. That's how index-only scans work fine in hot standby. A
660 : * crucial difference between index-only scans and heap scans is that the
661 : * index-only scan completely relies on the visibility map where as heap
662 : * scan looks at the page-level PD_ALL_VISIBLE flag. We are not sure if
663 : * the page-level flag can be trusted in the same way, because it might
664 : * get propagated somehow without being explicitly WAL-logged, e.g. via a
665 : * full page write. Until we can prove that beyond doubt, let's check each
666 : * tuple for visibility the hard way.
667 : */
668 3891536 : all_visible = PageIsAllVisible(page) && !snapshot->takenDuringRecovery;
669 : check_serializable =
670 3891536 : CheckForSerializableConflictOutNeeded(scan->rs_base.rs_rd, snapshot);
671 :
672 : /*
673 : * We call page_collect_tuples() with constant arguments, to get the
674 : * compiler to constant fold the constant arguments. Separate calls with
675 : * constant arguments, rather than variables, are needed on several
676 : * compilers to actually perform constant folding.
677 : */
678 3891536 : if (likely(all_visible))
679 : {
680 1582810 : if (likely(!check_serializable))
681 1582810 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
682 : block, lines, true, false);
683 : else
684 0 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
685 : block, lines, true, true);
686 : }
687 : else
688 : {
689 2308726 : if (likely(!check_serializable))
690 2308096 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
691 : block, lines, false, false);
692 : else
693 630 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
694 : block, lines, false, true);
695 : }
696 :
697 3891528 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
698 3891528 : }
699 :
700 : /*
701 : * heap_fetch_next_buffer - read and pin the next block from MAIN_FORKNUM.
702 : *
703 : * Read the next block of the scan relation from the read stream and save it
704 : * in the scan descriptor. It is already pinned.
705 : */
706 : static inline void
707 5119213 : heap_fetch_next_buffer(HeapScanDesc scan, ScanDirection dir)
708 : {
709 : Assert(scan->rs_read_stream);
710 :
711 : /* release previous scan buffer, if any */
712 5119213 : if (BufferIsValid(scan->rs_cbuf))
713 : {
714 3515276 : ReleaseBuffer(scan->rs_cbuf);
715 3515276 : scan->rs_cbuf = InvalidBuffer;
716 : }
717 :
718 : /*
719 : * Be sure to check for interrupts at least once per page. Checks at
720 : * higher code levels won't be able to stop a seqscan that encounters many
721 : * pages' worth of consecutive dead tuples.
722 : */
723 5119213 : CHECK_FOR_INTERRUPTS();
724 :
725 : /*
726 : * If the scan direction is changing, reset the prefetch block to the
727 : * current block. Otherwise, we will incorrectly prefetch the blocks
728 : * between the prefetch block and the current block again before
729 : * prefetching blocks in the new, correct scan direction.
730 : */
731 5119210 : if (unlikely(scan->rs_dir != dir))
732 : {
733 102 : scan->rs_prefetch_block = scan->rs_cblock;
734 102 : read_stream_reset(scan->rs_read_stream);
735 : }
736 :
737 5119210 : scan->rs_dir = dir;
738 :
739 5119210 : scan->rs_cbuf = read_stream_next_buffer(scan->rs_read_stream, NULL);
740 5119183 : if (BufferIsValid(scan->rs_cbuf))
741 4001813 : scan->rs_cblock = BufferGetBlockNumber(scan->rs_cbuf);
742 5119183 : }
743 :
744 : /*
745 : * heapgettup_initial_block - return the first BlockNumber to scan
746 : *
747 : * Returns InvalidBlockNumber when there are no blocks to scan. This can
748 : * occur with empty tables and in parallel scans when parallel workers get all
749 : * of the pages before we can get a chance to get our first page.
750 : */
751 : static pg_noinline BlockNumber
752 1601698 : heapgettup_initial_block(HeapScanDesc scan, ScanDirection dir)
753 : {
754 : Assert(!scan->rs_inited);
755 : Assert(scan->rs_base.rs_parallel == NULL);
756 :
757 : /* When there are no pages to scan, return InvalidBlockNumber */
758 1601698 : if (scan->rs_nblocks == 0 || scan->rs_numblocks == 0)
759 757573 : return InvalidBlockNumber;
760 :
761 844125 : if (ScanDirectionIsForward(dir))
762 : {
763 844083 : return scan->rs_startblock;
764 : }
765 : else
766 : {
767 : /*
768 : * Disable reporting to syncscan logic in a backwards scan; it's not
769 : * very likely anyone else is doing the same thing at the same time,
770 : * and much more likely that we'll just bollix things for forward
771 : * scanners.
772 : */
773 42 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
774 :
775 : /*
776 : * Start from last page of the scan. Ensure we take into account
777 : * rs_numblocks if it's been adjusted by heap_setscanlimits().
778 : */
779 42 : if (scan->rs_numblocks != InvalidBlockNumber)
780 4 : return (scan->rs_startblock + scan->rs_numblocks - 1) % scan->rs_nblocks;
781 :
782 38 : if (scan->rs_startblock > 0)
783 0 : return scan->rs_startblock - 1;
784 :
785 38 : return scan->rs_nblocks - 1;
786 : }
787 : }
788 :
789 :
790 : /*
791 : * heapgettup_start_page - helper function for heapgettup()
792 : *
793 : * Return the next page to scan based on the scan->rs_cbuf and set *linesleft
794 : * to the number of tuples on this page. Also set *lineoff to the first
795 : * offset to scan with forward scans getting the first offset and backward
796 : * getting the final offset on the page.
797 : */
798 : static Page
799 115963 : heapgettup_start_page(HeapScanDesc scan, ScanDirection dir, int *linesleft,
800 : OffsetNumber *lineoff)
801 : {
802 : Page page;
803 :
804 : Assert(scan->rs_inited);
805 : Assert(BufferIsValid(scan->rs_cbuf));
806 :
807 : /* Caller is responsible for ensuring buffer is locked if needed */
808 115963 : page = BufferGetPage(scan->rs_cbuf);
809 :
810 115963 : *linesleft = PageGetMaxOffsetNumber(page) - FirstOffsetNumber + 1;
811 :
812 115963 : if (ScanDirectionIsForward(dir))
813 115963 : *lineoff = FirstOffsetNumber;
814 : else
815 0 : *lineoff = (OffsetNumber) (*linesleft);
816 :
817 : /* lineoff now references the physically previous or next tid */
818 115963 : return page;
819 : }
820 :
821 :
822 : /*
823 : * heapgettup_continue_page - helper function for heapgettup()
824 : *
825 : * Return the next page to scan based on the scan->rs_cbuf and set *linesleft
826 : * to the number of tuples left to scan on this page. Also set *lineoff to
827 : * the next offset to scan according to the ScanDirection in 'dir'.
828 : */
829 : static inline Page
830 9235099 : heapgettup_continue_page(HeapScanDesc scan, ScanDirection dir, int *linesleft,
831 : OffsetNumber *lineoff)
832 : {
833 : Page page;
834 :
835 : Assert(scan->rs_inited);
836 : Assert(BufferIsValid(scan->rs_cbuf));
837 :
838 : /* Caller is responsible for ensuring buffer is locked if needed */
839 9235099 : page = BufferGetPage(scan->rs_cbuf);
840 :
841 9235099 : if (ScanDirectionIsForward(dir))
842 : {
843 9235099 : *lineoff = OffsetNumberNext(scan->rs_coffset);
844 9235099 : *linesleft = PageGetMaxOffsetNumber(page) - (*lineoff) + 1;
845 : }
846 : else
847 : {
848 : /*
849 : * The previous returned tuple may have been vacuumed since the
850 : * previous scan when we use a non-MVCC snapshot, so we must
851 : * re-establish the lineoff <= PageGetMaxOffsetNumber(page) invariant
852 : */
853 0 : *lineoff = Min(PageGetMaxOffsetNumber(page), OffsetNumberPrev(scan->rs_coffset));
854 0 : *linesleft = *lineoff;
855 : }
856 :
857 : /* lineoff now references the physically previous or next tid */
858 9235099 : return page;
859 : }
860 :
861 : /*
862 : * heapgettup_advance_block - helper for heap_fetch_next_buffer()
863 : *
864 : * Given the current block number, the scan direction, and various information
865 : * contained in the scan descriptor, calculate the BlockNumber to scan next
866 : * and return it. If there are no further blocks to scan, return
867 : * InvalidBlockNumber to indicate this fact to the caller.
868 : *
869 : * This should not be called to determine the initial block number -- only for
870 : * subsequent blocks.
871 : *
872 : * This also adjusts rs_numblocks when a limit has been imposed by
873 : * heap_setscanlimits().
874 : */
875 : static inline BlockNumber
876 3896263 : heapgettup_advance_block(HeapScanDesc scan, BlockNumber block, ScanDirection dir)
877 : {
878 : Assert(scan->rs_base.rs_parallel == NULL);
879 :
880 3896263 : if (likely(ScanDirectionIsForward(dir)))
881 : {
882 3896185 : block++;
883 :
884 : /* wrap back to the start of the heap */
885 3896185 : if (block >= scan->rs_nblocks)
886 729956 : block = 0;
887 :
888 : /*
889 : * Report our new scan position for synchronization purposes. We don't
890 : * do that when moving backwards, however. That would just mess up any
891 : * other forward-moving scanners.
892 : *
893 : * Note: we do this before checking for end of scan so that the final
894 : * state of the position hint is back at the start of the rel. That's
895 : * not strictly necessary, but otherwise when you run the same query
896 : * multiple times the starting position would shift a little bit
897 : * backwards on every invocation, which is confusing. We don't
898 : * guarantee any specific ordering in general, though.
899 : */
900 3896185 : if (scan->rs_base.rs_flags & SO_ALLOW_SYNC)
901 11629 : ss_report_location(scan->rs_base.rs_rd, block);
902 :
903 : /* we're done if we're back at where we started */
904 3896185 : if (block == scan->rs_startblock)
905 729906 : return InvalidBlockNumber;
906 :
907 : /* check if the limit imposed by heap_setscanlimits() is met */
908 3166279 : if (scan->rs_numblocks != InvalidBlockNumber)
909 : {
910 2822 : if (--scan->rs_numblocks == 0)
911 1590 : return InvalidBlockNumber;
912 : }
913 :
914 3164689 : return block;
915 : }
916 : else
917 : {
918 : /* we're done if the last block is the start position */
919 78 : if (block == scan->rs_startblock)
920 78 : return InvalidBlockNumber;
921 :
922 : /* check if the limit imposed by heap_setscanlimits() is met */
923 0 : if (scan->rs_numblocks != InvalidBlockNumber)
924 : {
925 0 : if (--scan->rs_numblocks == 0)
926 0 : return InvalidBlockNumber;
927 : }
928 :
929 : /* wrap to the end of the heap when the last page was page 0 */
930 0 : if (block == 0)
931 0 : block = scan->rs_nblocks;
932 :
933 0 : block--;
934 :
935 0 : return block;
936 : }
937 : }
938 :
939 : /* ----------------
940 : * heapgettup - fetch next heap tuple
941 : *
942 : * Initialize the scan if not already done; then advance to the next
943 : * tuple as indicated by "dir"; return the next tuple in scan->rs_ctup,
944 : * or set scan->rs_ctup.t_data = NULL if no more tuples.
945 : *
946 : * Note: the reason nkeys/key are passed separately, even though they are
947 : * kept in the scan descriptor, is that the caller may not want us to check
948 : * the scankeys.
949 : *
950 : * Note: when we fall off the end of the scan in either direction, we
951 : * reset rs_inited. This means that a further request with the same
952 : * scan direction will restart the scan, which is a bit odd, but a
953 : * request with the opposite scan direction will start a fresh scan
954 : * in the proper direction. The latter is required behavior for cursors,
955 : * while the former case is generally undefined behavior in Postgres
956 : * so we don't care too much.
957 : * ----------------
958 : */
959 : static void
960 9261371 : heapgettup(HeapScanDesc scan,
961 : ScanDirection dir,
962 : int nkeys,
963 : ScanKey key)
964 : {
965 9261371 : HeapTuple tuple = &(scan->rs_ctup);
966 : Page page;
967 : OffsetNumber lineoff;
968 : int linesleft;
969 :
970 9261371 : if (likely(scan->rs_inited))
971 : {
972 : /* continue from previously returned page/tuple */
973 9235099 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
974 9235099 : page = heapgettup_continue_page(scan, dir, &linesleft, &lineoff);
975 9235099 : goto continue_page;
976 : }
977 :
978 : /*
979 : * advance the scan until we find a qualifying tuple or run out of stuff
980 : * to scan
981 : */
982 : while (true)
983 : {
984 141368 : heap_fetch_next_buffer(scan, dir);
985 :
986 : /* did we run out of blocks to scan? */
987 141368 : if (!BufferIsValid(scan->rs_cbuf))
988 25405 : break;
989 :
990 : Assert(BufferGetBlockNumber(scan->rs_cbuf) == scan->rs_cblock);
991 :
992 115963 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
993 115963 : page = heapgettup_start_page(scan, dir, &linesleft, &lineoff);
994 9351062 : continue_page:
995 :
996 : /*
997 : * Only continue scanning the page while we have lines left.
998 : *
999 : * Note that this protects us from accessing line pointers past
1000 : * PageGetMaxOffsetNumber(); both for forward scans when we resume the
1001 : * table scan, and for when we start scanning a new page.
1002 : */
1003 9420535 : for (; linesleft > 0; linesleft--, lineoff += dir)
1004 : {
1005 : bool visible;
1006 9305439 : ItemId lpp = PageGetItemId(page, lineoff);
1007 :
1008 9305439 : if (!ItemIdIsNormal(lpp))
1009 47437 : continue;
1010 :
1011 9258002 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
1012 9258002 : tuple->t_len = ItemIdGetLength(lpp);
1013 9258002 : ItemPointerSet(&(tuple->t_self), scan->rs_cblock, lineoff);
1014 :
1015 9258002 : visible = HeapTupleSatisfiesVisibility(tuple,
1016 : scan->rs_base.rs_snapshot,
1017 : scan->rs_cbuf);
1018 :
1019 9258002 : HeapCheckForSerializableConflictOut(visible, scan->rs_base.rs_rd,
1020 : tuple, scan->rs_cbuf,
1021 : scan->rs_base.rs_snapshot);
1022 :
1023 : /* skip tuples not visible to this snapshot */
1024 9258002 : if (!visible)
1025 7713 : continue;
1026 :
1027 : /* skip any tuples that don't match the scan key */
1028 9250289 : if (key != NULL &&
1029 15039 : !HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
1030 : nkeys, key))
1031 14323 : continue;
1032 :
1033 9235966 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1034 9235966 : scan->rs_coffset = lineoff;
1035 9235966 : return;
1036 : }
1037 :
1038 : /*
1039 : * if we get here, it means we've exhausted the items on this page and
1040 : * it's time to move to the next.
1041 : */
1042 115096 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1043 : }
1044 :
1045 : /* end of scan */
1046 25405 : if (BufferIsValid(scan->rs_cbuf))
1047 0 : ReleaseBuffer(scan->rs_cbuf);
1048 :
1049 25405 : scan->rs_cbuf = InvalidBuffer;
1050 25405 : scan->rs_cblock = InvalidBlockNumber;
1051 25405 : scan->rs_prefetch_block = InvalidBlockNumber;
1052 25405 : tuple->t_data = NULL;
1053 25405 : scan->rs_inited = false;
1054 : }
1055 :
1056 : /* ----------------
1057 : * heapgettup_pagemode - fetch next heap tuple in page-at-a-time mode
1058 : *
1059 : * Same API as heapgettup, but used in page-at-a-time mode
1060 : *
1061 : * The internal logic is much the same as heapgettup's too, but there are some
1062 : * differences: we do not take the buffer content lock (that only needs to
1063 : * happen inside heap_prepare_pagescan), and we iterate through just the
1064 : * tuples listed in rs_vistuples[] rather than all tuples on the page. Notice
1065 : * that lineindex is 0-based, where the corresponding loop variable lineoff in
1066 : * heapgettup is 1-based.
1067 : * ----------------
1068 : */
1069 : static void
1070 70726326 : heapgettup_pagemode(HeapScanDesc scan,
1071 : ScanDirection dir,
1072 : int nkeys,
1073 : ScanKey key)
1074 : {
1075 70726326 : HeapTuple tuple = &(scan->rs_ctup);
1076 : Page page;
1077 : uint32 lineindex;
1078 : uint32 linesleft;
1079 :
1080 70726326 : if (likely(scan->rs_inited))
1081 : {
1082 : /* continue from previously returned page/tuple */
1083 69148661 : page = BufferGetPage(scan->rs_cbuf);
1084 :
1085 69148661 : lineindex = scan->rs_cindex + dir;
1086 69148661 : if (ScanDirectionIsForward(dir))
1087 69148223 : linesleft = scan->rs_ntuples - lineindex;
1088 : else
1089 438 : linesleft = scan->rs_cindex;
1090 : /* lineindex now references the next or previous visible tid */
1091 :
1092 69148661 : goto continue_page;
1093 : }
1094 :
1095 : /*
1096 : * advance the scan until we find a qualifying tuple or run out of stuff
1097 : * to scan
1098 : */
1099 : while (true)
1100 : {
1101 4977845 : heap_fetch_next_buffer(scan, dir);
1102 :
1103 : /* did we run out of blocks to scan? */
1104 4977815 : if (!BufferIsValid(scan->rs_cbuf))
1105 1091965 : break;
1106 :
1107 : Assert(BufferGetBlockNumber(scan->rs_cbuf) == scan->rs_cblock);
1108 :
1109 : /* prune the page and determine visible tuple offsets */
1110 3885850 : heap_prepare_pagescan((TableScanDesc) scan);
1111 3885842 : page = BufferGetPage(scan->rs_cbuf);
1112 3885842 : linesleft = scan->rs_ntuples;
1113 3885842 : lineindex = ScanDirectionIsForward(dir) ? 0 : linesleft - 1;
1114 :
1115 : /* block is the same for all tuples, set it once outside the loop */
1116 3885842 : ItemPointerSetBlockNumber(&tuple->t_self, scan->rs_cblock);
1117 :
1118 : /* lineindex now references the next or previous visible tid */
1119 73034503 : continue_page:
1120 :
1121 130714363 : for (; linesleft > 0; linesleft--, lineindex += dir)
1122 : {
1123 : ItemId lpp;
1124 : OffsetNumber lineoff;
1125 :
1126 : Assert(lineindex < scan->rs_ntuples);
1127 127314183 : lineoff = scan->rs_vistuples[lineindex];
1128 127314183 : lpp = PageGetItemId(page, lineoff);
1129 : Assert(ItemIdIsNormal(lpp));
1130 :
1131 127314183 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
1132 127314183 : tuple->t_len = ItemIdGetLength(lpp);
1133 127314183 : ItemPointerSetOffsetNumber(&tuple->t_self, lineoff);
1134 :
1135 : /* skip any tuples that don't match the scan key */
1136 127314183 : if (key != NULL &&
1137 58112217 : !HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
1138 : nkeys, key))
1139 57679860 : continue;
1140 :
1141 69634323 : scan->rs_cindex = lineindex;
1142 69634323 : return;
1143 : }
1144 : }
1145 :
1146 : /* end of scan */
1147 1091965 : if (BufferIsValid(scan->rs_cbuf))
1148 0 : ReleaseBuffer(scan->rs_cbuf);
1149 1091965 : scan->rs_cbuf = InvalidBuffer;
1150 1091965 : scan->rs_cblock = InvalidBlockNumber;
1151 1091965 : scan->rs_prefetch_block = InvalidBlockNumber;
1152 1091965 : tuple->t_data = NULL;
1153 1091965 : scan->rs_inited = false;
1154 : }
1155 :
1156 :
1157 : /* ----------------------------------------------------------------
1158 : * heap access method interface
1159 : * ----------------------------------------------------------------
1160 : */
1161 :
1162 :
1163 : TableScanDesc
1164 766964 : heap_beginscan(Relation relation, Snapshot snapshot,
1165 : int nkeys, ScanKey key,
1166 : ParallelTableScanDesc parallel_scan,
1167 : uint32 flags)
1168 : {
1169 : HeapScanDesc scan;
1170 :
1171 : /*
1172 : * increment relation ref count while scanning relation
1173 : *
1174 : * This is just to make really sure the relcache entry won't go away while
1175 : * the scan has a pointer to it. Caller should be holding the rel open
1176 : * anyway, so this is redundant in all normal scenarios...
1177 : */
1178 766964 : RelationIncrementReferenceCount(relation);
1179 :
1180 : /*
1181 : * allocate and initialize scan descriptor
1182 : */
1183 766964 : if (flags & SO_TYPE_BITMAPSCAN)
1184 : {
1185 13346 : BitmapHeapScanDesc bscan = palloc_object(BitmapHeapScanDescData);
1186 :
1187 : /*
1188 : * Bitmap Heap scans do not have any fields that a normal Heap Scan
1189 : * does not have, so no special initializations required here.
1190 : */
1191 13346 : scan = (HeapScanDesc) bscan;
1192 : }
1193 : else
1194 753618 : scan = (HeapScanDesc) palloc_object(HeapScanDescData);
1195 :
1196 766964 : scan->rs_base.rs_rd = relation;
1197 766964 : scan->rs_base.rs_snapshot = snapshot;
1198 766964 : scan->rs_base.rs_nkeys = nkeys;
1199 766964 : scan->rs_base.rs_flags = flags;
1200 766964 : scan->rs_base.rs_parallel = parallel_scan;
1201 766964 : scan->rs_strategy = NULL; /* set in initscan */
1202 766964 : scan->rs_cbuf = InvalidBuffer;
1203 :
1204 : /*
1205 : * Disable page-at-a-time mode if it's not a MVCC-safe snapshot.
1206 : */
1207 766964 : if (!(snapshot && IsMVCCSnapshot(snapshot)))
1208 36776 : scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
1209 :
1210 : /* Check that a historic snapshot is not used for non-catalog tables */
1211 766964 : if (snapshot &&
1212 756460 : IsHistoricMVCCSnapshot(snapshot) &&
1213 716 : !RelationIsAccessibleInLogicalDecoding(relation))
1214 : {
1215 0 : ereport(ERROR,
1216 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
1217 : errmsg("cannot query non-catalog table \"%s\" during logical decoding",
1218 : RelationGetRelationName(relation))));
1219 : }
1220 :
1221 : /*
1222 : * For seqscan and sample scans in a serializable transaction, acquire a
1223 : * predicate lock on the entire relation. This is required not only to
1224 : * lock all the matching tuples, but also to conflict with new insertions
1225 : * into the table. In an indexscan, we take page locks on the index pages
1226 : * covering the range specified in the scan qual, but in a heap scan there
1227 : * is nothing more fine-grained to lock. A bitmap scan is a different
1228 : * story, there we have already scanned the index and locked the index
1229 : * pages covering the predicate. But in that case we still have to lock
1230 : * any matching heap tuples. For sample scan we could optimize the locking
1231 : * to be at least page-level granularity, but we'd need to add per-tuple
1232 : * locking for that.
1233 : */
1234 766964 : if (scan->rs_base.rs_flags & (SO_TYPE_SEQSCAN | SO_TYPE_SAMPLESCAN))
1235 : {
1236 : /*
1237 : * Ensure a missing snapshot is noticed reliably, even if the
1238 : * isolation mode means predicate locking isn't performed (and
1239 : * therefore the snapshot isn't used here).
1240 : */
1241 : Assert(snapshot);
1242 741322 : PredicateLockRelation(relation, snapshot);
1243 : }
1244 :
1245 : /* we only need to set this up once */
1246 766964 : scan->rs_ctup.t_tableOid = RelationGetRelid(relation);
1247 :
1248 : /*
1249 : * Allocate memory to keep track of page allocation for parallel workers
1250 : * when doing a parallel scan.
1251 : */
1252 766964 : if (parallel_scan != NULL)
1253 2865 : scan->rs_parallelworkerdata = palloc_object(ParallelBlockTableScanWorkerData);
1254 : else
1255 764099 : scan->rs_parallelworkerdata = NULL;
1256 :
1257 : /*
1258 : * we do this here instead of in initscan() because heap_rescan also calls
1259 : * initscan() and we don't want to allocate memory again
1260 : */
1261 766964 : if (nkeys > 0)
1262 252249 : scan->rs_base.rs_key = palloc_array(ScanKeyData, nkeys);
1263 : else
1264 514715 : scan->rs_base.rs_key = NULL;
1265 :
1266 766964 : initscan(scan, key, false);
1267 :
1268 766962 : scan->rs_read_stream = NULL;
1269 :
1270 : /*
1271 : * Set up a read stream for sequential scans and TID range scans. This
1272 : * should be done after initscan() because initscan() allocates the
1273 : * BufferAccessStrategy object passed to the read stream API.
1274 : */
1275 766962 : if (scan->rs_base.rs_flags & SO_TYPE_SEQSCAN ||
1276 25736 : scan->rs_base.rs_flags & SO_TYPE_TIDRANGESCAN)
1277 742544 : {
1278 : ReadStreamBlockNumberCB cb;
1279 :
1280 742544 : if (scan->rs_base.rs_parallel)
1281 2865 : cb = heap_scan_stream_read_next_parallel;
1282 : else
1283 739679 : cb = heap_scan_stream_read_next_serial;
1284 :
1285 : /* ---
1286 : * It is safe to use batchmode as the only locks taken by `cb`
1287 : * are never taken while waiting for IO:
1288 : * - SyncScanLock is used in the non-parallel case
1289 : * - in the parallel case, only spinlocks and atomics are used
1290 : * ---
1291 : */
1292 742544 : scan->rs_read_stream = read_stream_begin_relation(READ_STREAM_SEQUENTIAL |
1293 : READ_STREAM_USE_BATCHING,
1294 : scan->rs_strategy,
1295 : scan->rs_base.rs_rd,
1296 : MAIN_FORKNUM,
1297 : cb,
1298 : scan,
1299 : 0);
1300 : }
1301 24418 : else if (scan->rs_base.rs_flags & SO_TYPE_BITMAPSCAN)
1302 : {
1303 13346 : scan->rs_read_stream = read_stream_begin_relation(READ_STREAM_DEFAULT |
1304 : READ_STREAM_USE_BATCHING,
1305 : scan->rs_strategy,
1306 : scan->rs_base.rs_rd,
1307 : MAIN_FORKNUM,
1308 : bitmapheap_stream_read_next,
1309 : scan,
1310 : sizeof(TBMIterateResult));
1311 : }
1312 :
1313 766962 : scan->rs_vmbuffer = InvalidBuffer;
1314 :
1315 766962 : return (TableScanDesc) scan;
1316 : }
1317 :
1318 : void
1319 864615 : heap_rescan(TableScanDesc sscan, ScanKey key, bool set_params,
1320 : bool allow_strat, bool allow_sync, bool allow_pagemode)
1321 : {
1322 864615 : HeapScanDesc scan = (HeapScanDesc) sscan;
1323 :
1324 864615 : if (set_params)
1325 : {
1326 19 : if (allow_strat)
1327 19 : scan->rs_base.rs_flags |= SO_ALLOW_STRAT;
1328 : else
1329 0 : scan->rs_base.rs_flags &= ~SO_ALLOW_STRAT;
1330 :
1331 19 : if (allow_sync)
1332 8 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
1333 : else
1334 11 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
1335 :
1336 19 : if (allow_pagemode && scan->rs_base.rs_snapshot &&
1337 19 : IsMVCCSnapshot(scan->rs_base.rs_snapshot))
1338 19 : scan->rs_base.rs_flags |= SO_ALLOW_PAGEMODE;
1339 : else
1340 0 : scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
1341 : }
1342 :
1343 : /*
1344 : * unpin scan buffers
1345 : */
1346 864615 : if (BufferIsValid(scan->rs_cbuf))
1347 : {
1348 3045 : ReleaseBuffer(scan->rs_cbuf);
1349 3045 : scan->rs_cbuf = InvalidBuffer;
1350 : }
1351 :
1352 864615 : if (BufferIsValid(scan->rs_vmbuffer))
1353 : {
1354 12 : ReleaseBuffer(scan->rs_vmbuffer);
1355 12 : scan->rs_vmbuffer = InvalidBuffer;
1356 : }
1357 :
1358 : /*
1359 : * SO_TYPE_BITMAPSCAN would be cleaned up here, but it does not hold any
1360 : * additional data vs a normal HeapScan
1361 : */
1362 :
1363 : /*
1364 : * The read stream is reset on rescan. This must be done before
1365 : * initscan(), as some state referred to by read_stream_reset() is reset
1366 : * in initscan().
1367 : */
1368 864615 : if (scan->rs_read_stream)
1369 864592 : read_stream_reset(scan->rs_read_stream);
1370 :
1371 : /*
1372 : * reinitialize scan descriptor
1373 : */
1374 864615 : initscan(scan, key, true);
1375 864615 : }
1376 :
1377 : void
1378 763757 : heap_endscan(TableScanDesc sscan)
1379 : {
1380 763757 : HeapScanDesc scan = (HeapScanDesc) sscan;
1381 :
1382 : /* Note: no locking manipulations needed */
1383 :
1384 : /*
1385 : * unpin scan buffers
1386 : */
1387 763757 : if (BufferIsValid(scan->rs_cbuf))
1388 481320 : ReleaseBuffer(scan->rs_cbuf);
1389 :
1390 763757 : if (BufferIsValid(scan->rs_vmbuffer))
1391 1473 : ReleaseBuffer(scan->rs_vmbuffer);
1392 :
1393 : /*
1394 : * Must free the read stream before freeing the BufferAccessStrategy.
1395 : */
1396 763757 : if (scan->rs_read_stream)
1397 752752 : read_stream_end(scan->rs_read_stream);
1398 :
1399 : /*
1400 : * decrement relation reference count and free scan descriptor storage
1401 : */
1402 763757 : RelationDecrementReferenceCount(scan->rs_base.rs_rd);
1403 :
1404 763757 : if (scan->rs_base.rs_key)
1405 252212 : pfree(scan->rs_base.rs_key);
1406 :
1407 763757 : if (scan->rs_strategy != NULL)
1408 12761 : FreeAccessStrategy(scan->rs_strategy);
1409 :
1410 763757 : if (scan->rs_parallelworkerdata != NULL)
1411 2865 : pfree(scan->rs_parallelworkerdata);
1412 :
1413 763757 : if (scan->rs_base.rs_flags & SO_TEMP_SNAPSHOT)
1414 47653 : UnregisterSnapshot(scan->rs_base.rs_snapshot);
1415 :
1416 763757 : pfree(scan);
1417 763757 : }
1418 :
1419 : HeapTuple
1420 11583927 : heap_getnext(TableScanDesc sscan, ScanDirection direction)
1421 : {
1422 11583927 : HeapScanDesc scan = (HeapScanDesc) sscan;
1423 :
1424 : /*
1425 : * This is still widely used directly, without going through table AM, so
1426 : * add a safety check. It's possible we should, at a later point,
1427 : * downgrade this to an assert. The reason for checking the AM routine,
1428 : * rather than the AM oid, is that this allows to write regression tests
1429 : * that create another AM reusing the heap handler.
1430 : */
1431 11583927 : if (unlikely(sscan->rs_rd->rd_tableam != GetHeapamTableAmRoutine()))
1432 0 : ereport(ERROR,
1433 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1434 : errmsg_internal("only heap AM is supported")));
1435 :
1436 : /* Note: no locking manipulations needed */
1437 :
1438 11583927 : if (scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE)
1439 2913349 : heapgettup_pagemode(scan, direction,
1440 2913349 : scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1441 : else
1442 8670578 : heapgettup(scan, direction,
1443 8670578 : scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1444 :
1445 11583926 : if (scan->rs_ctup.t_data == NULL)
1446 78849 : return NULL;
1447 :
1448 : /*
1449 : * if we get here it means we have a new current scan tuple, so point to
1450 : * the proper return buffer and return the tuple.
1451 : */
1452 :
1453 11505077 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1454 :
1455 11505077 : return &scan->rs_ctup;
1456 : }
1457 :
1458 : bool
1459 68397130 : heap_getnextslot(TableScanDesc sscan, ScanDirection direction, TupleTableSlot *slot)
1460 : {
1461 68397130 : HeapScanDesc scan = (HeapScanDesc) sscan;
1462 :
1463 : /* Note: no locking manipulations needed */
1464 :
1465 68397130 : if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1466 67806337 : heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1467 : else
1468 590793 : heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1469 :
1470 68397103 : if (scan->rs_ctup.t_data == NULL)
1471 : {
1472 1038383 : ExecClearTuple(slot);
1473 1038383 : return false;
1474 : }
1475 :
1476 : /*
1477 : * if we get here it means we have a new current scan tuple, so point to
1478 : * the proper return buffer and return the tuple.
1479 : */
1480 :
1481 67358720 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1482 :
1483 67358720 : ExecStoreBufferHeapTuple(&scan->rs_ctup, slot,
1484 : scan->rs_cbuf);
1485 67358720 : return true;
1486 : }
1487 :
1488 : void
1489 1378 : heap_set_tidrange(TableScanDesc sscan, ItemPointer mintid,
1490 : ItemPointer maxtid)
1491 : {
1492 1378 : HeapScanDesc scan = (HeapScanDesc) sscan;
1493 : BlockNumber startBlk;
1494 : BlockNumber numBlks;
1495 : ItemPointerData highestItem;
1496 : ItemPointerData lowestItem;
1497 :
1498 : /*
1499 : * For relations without any pages, we can simply leave the TID range
1500 : * unset. There will be no tuples to scan, therefore no tuples outside
1501 : * the given TID range.
1502 : */
1503 1378 : if (scan->rs_nblocks == 0)
1504 32 : return;
1505 :
1506 : /*
1507 : * Set up some ItemPointers which point to the first and last possible
1508 : * tuples in the heap.
1509 : */
1510 1370 : ItemPointerSet(&highestItem, scan->rs_nblocks - 1, MaxOffsetNumber);
1511 1370 : ItemPointerSet(&lowestItem, 0, FirstOffsetNumber);
1512 :
1513 : /*
1514 : * If the given maximum TID is below the highest possible TID in the
1515 : * relation, then restrict the range to that, otherwise we scan to the end
1516 : * of the relation.
1517 : */
1518 1370 : if (ItemPointerCompare(maxtid, &highestItem) < 0)
1519 172 : ItemPointerCopy(maxtid, &highestItem);
1520 :
1521 : /*
1522 : * If the given minimum TID is above the lowest possible TID in the
1523 : * relation, then restrict the range to only scan for TIDs above that.
1524 : */
1525 1370 : if (ItemPointerCompare(mintid, &lowestItem) > 0)
1526 1214 : ItemPointerCopy(mintid, &lowestItem);
1527 :
1528 : /*
1529 : * Check for an empty range and protect from would be negative results
1530 : * from the numBlks calculation below.
1531 : */
1532 1370 : if (ItemPointerCompare(&highestItem, &lowestItem) < 0)
1533 : {
1534 : /* Set an empty range of blocks to scan */
1535 24 : heap_setscanlimits(sscan, 0, 0);
1536 24 : return;
1537 : }
1538 :
1539 : /*
1540 : * Calculate the first block and the number of blocks we must scan. We
1541 : * could be more aggressive here and perform some more validation to try
1542 : * and further narrow the scope of blocks to scan by checking if the
1543 : * lowestItem has an offset above MaxOffsetNumber. In this case, we could
1544 : * advance startBlk by one. Likewise, if highestItem has an offset of 0
1545 : * we could scan one fewer blocks. However, such an optimization does not
1546 : * seem worth troubling over, currently.
1547 : */
1548 1346 : startBlk = ItemPointerGetBlockNumberNoCheck(&lowestItem);
1549 :
1550 1346 : numBlks = ItemPointerGetBlockNumberNoCheck(&highestItem) -
1551 1346 : ItemPointerGetBlockNumberNoCheck(&lowestItem) + 1;
1552 :
1553 : /* Set the start block and number of blocks to scan */
1554 1346 : heap_setscanlimits(sscan, startBlk, numBlks);
1555 :
1556 : /* Finally, set the TID range in sscan */
1557 1346 : ItemPointerCopy(&lowestItem, &sscan->st.tidrange.rs_mintid);
1558 1346 : ItemPointerCopy(&highestItem, &sscan->st.tidrange.rs_maxtid);
1559 : }
1560 :
1561 : bool
1562 6516 : heap_getnextslot_tidrange(TableScanDesc sscan, ScanDirection direction,
1563 : TupleTableSlot *slot)
1564 : {
1565 6516 : HeapScanDesc scan = (HeapScanDesc) sscan;
1566 6516 : ItemPointer mintid = &sscan->st.tidrange.rs_mintid;
1567 6516 : ItemPointer maxtid = &sscan->st.tidrange.rs_maxtid;
1568 :
1569 : /* Note: no locking manipulations needed */
1570 : for (;;)
1571 : {
1572 6640 : if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1573 6640 : heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1574 : else
1575 0 : heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1576 :
1577 6630 : if (scan->rs_ctup.t_data == NULL)
1578 : {
1579 138 : ExecClearTuple(slot);
1580 138 : return false;
1581 : }
1582 :
1583 : /*
1584 : * heap_set_tidrange will have used heap_setscanlimits to limit the
1585 : * range of pages we scan to only ones that can contain the TID range
1586 : * we're scanning for. Here we must filter out any tuples from these
1587 : * pages that are outside of that range.
1588 : */
1589 6492 : if (ItemPointerCompare(&scan->rs_ctup.t_self, mintid) < 0)
1590 : {
1591 124 : ExecClearTuple(slot);
1592 :
1593 : /*
1594 : * When scanning backwards, the TIDs will be in descending order.
1595 : * Future tuples in this direction will be lower still, so we can
1596 : * just return false to indicate there will be no more tuples.
1597 : */
1598 124 : if (ScanDirectionIsBackward(direction))
1599 0 : return false;
1600 :
1601 124 : continue;
1602 : }
1603 :
1604 : /*
1605 : * Likewise for the final page, we must filter out TIDs greater than
1606 : * maxtid.
1607 : */
1608 6368 : if (ItemPointerCompare(&scan->rs_ctup.t_self, maxtid) > 0)
1609 : {
1610 74 : ExecClearTuple(slot);
1611 :
1612 : /*
1613 : * When scanning forward, the TIDs will be in ascending order.
1614 : * Future tuples in this direction will be higher still, so we can
1615 : * just return false to indicate there will be no more tuples.
1616 : */
1617 74 : if (ScanDirectionIsForward(direction))
1618 74 : return false;
1619 0 : continue;
1620 : }
1621 :
1622 6294 : break;
1623 : }
1624 :
1625 : /*
1626 : * if we get here it means we have a new current scan tuple, so point to
1627 : * the proper return buffer and return the tuple.
1628 : */
1629 6294 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1630 :
1631 6294 : ExecStoreBufferHeapTuple(&scan->rs_ctup, slot, scan->rs_cbuf);
1632 6294 : return true;
1633 : }
1634 :
1635 : /*
1636 : * heap_fetch - retrieve tuple with given tid
1637 : *
1638 : * On entry, tuple->t_self is the TID to fetch. We pin the buffer holding
1639 : * the tuple, fill in the remaining fields of *tuple, and check the tuple
1640 : * against the specified snapshot.
1641 : *
1642 : * If successful (tuple found and passes snapshot time qual), then *userbuf
1643 : * is set to the buffer holding the tuple and true is returned. The caller
1644 : * must unpin the buffer when done with the tuple.
1645 : *
1646 : * If the tuple is not found (ie, item number references a deleted slot),
1647 : * then tuple->t_data is set to NULL, *userbuf is set to InvalidBuffer,
1648 : * and false is returned.
1649 : *
1650 : * If the tuple is found but fails the time qual check, then the behavior
1651 : * depends on the keep_buf parameter. If keep_buf is false, the results
1652 : * are the same as for the tuple-not-found case. If keep_buf is true,
1653 : * then tuple->t_data and *userbuf are returned as for the success case,
1654 : * and again the caller must unpin the buffer; but false is returned.
1655 : *
1656 : * heap_fetch does not follow HOT chains: only the exact TID requested will
1657 : * be fetched.
1658 : *
1659 : * It is somewhat inconsistent that we ereport() on invalid block number but
1660 : * return false on invalid item number. There are a couple of reasons though.
1661 : * One is that the caller can relatively easily check the block number for
1662 : * validity, but cannot check the item number without reading the page
1663 : * himself. Another is that when we are following a t_ctid link, we can be
1664 : * reasonably confident that the page number is valid (since VACUUM shouldn't
1665 : * truncate off the destination page without having killed the referencing
1666 : * tuple first), but the item number might well not be good.
1667 : */
1668 : bool
1669 595186 : heap_fetch(Relation relation,
1670 : Snapshot snapshot,
1671 : HeapTuple tuple,
1672 : Buffer *userbuf,
1673 : bool keep_buf)
1674 : {
1675 595186 : ItemPointer tid = &(tuple->t_self);
1676 : ItemId lp;
1677 : Buffer buffer;
1678 : Page page;
1679 : OffsetNumber offnum;
1680 : bool valid;
1681 :
1682 : /*
1683 : * Fetch and pin the appropriate page of the relation.
1684 : */
1685 595186 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
1686 :
1687 : /*
1688 : * Need share lock on buffer to examine tuple commit status.
1689 : */
1690 595178 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
1691 595178 : page = BufferGetPage(buffer);
1692 :
1693 : /*
1694 : * We'd better check for out-of-range offnum in case of VACUUM since the
1695 : * TID was obtained.
1696 : */
1697 595178 : offnum = ItemPointerGetOffsetNumber(tid);
1698 595178 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1699 : {
1700 4 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1701 4 : ReleaseBuffer(buffer);
1702 4 : *userbuf = InvalidBuffer;
1703 4 : tuple->t_data = NULL;
1704 4 : return false;
1705 : }
1706 :
1707 : /*
1708 : * get the item line pointer corresponding to the requested tid
1709 : */
1710 595174 : lp = PageGetItemId(page, offnum);
1711 :
1712 : /*
1713 : * Must check for deleted tuple.
1714 : */
1715 595174 : if (!ItemIdIsNormal(lp))
1716 : {
1717 299 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1718 299 : ReleaseBuffer(buffer);
1719 299 : *userbuf = InvalidBuffer;
1720 299 : tuple->t_data = NULL;
1721 299 : return false;
1722 : }
1723 :
1724 : /*
1725 : * fill in *tuple fields
1726 : */
1727 594875 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
1728 594875 : tuple->t_len = ItemIdGetLength(lp);
1729 594875 : tuple->t_tableOid = RelationGetRelid(relation);
1730 :
1731 : /*
1732 : * check tuple visibility, then release lock
1733 : */
1734 594875 : valid = HeapTupleSatisfiesVisibility(tuple, snapshot, buffer);
1735 :
1736 594875 : if (valid)
1737 594815 : PredicateLockTID(relation, &(tuple->t_self), snapshot,
1738 594815 : HeapTupleHeaderGetXmin(tuple->t_data));
1739 :
1740 594875 : HeapCheckForSerializableConflictOut(valid, relation, tuple, buffer, snapshot);
1741 :
1742 594875 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1743 :
1744 594875 : if (valid)
1745 : {
1746 : /*
1747 : * All checks passed, so return the tuple as valid. Caller is now
1748 : * responsible for releasing the buffer.
1749 : */
1750 594815 : *userbuf = buffer;
1751 :
1752 594815 : return true;
1753 : }
1754 :
1755 : /* Tuple failed time qual, but maybe caller wants to see it anyway. */
1756 60 : if (keep_buf)
1757 34 : *userbuf = buffer;
1758 : else
1759 : {
1760 26 : ReleaseBuffer(buffer);
1761 26 : *userbuf = InvalidBuffer;
1762 26 : tuple->t_data = NULL;
1763 : }
1764 :
1765 60 : return false;
1766 : }
1767 :
1768 : /*
1769 : * heap_hot_search_buffer - search HOT chain for tuple satisfying snapshot
1770 : *
1771 : * On entry, *tid is the TID of a tuple (either a simple tuple, or the root
1772 : * of a HOT chain), and buffer is the buffer holding this tuple. We search
1773 : * for the first chain member satisfying the given snapshot. If one is
1774 : * found, we update *tid to reference that tuple's offset number, and
1775 : * return true. If no match, return false without modifying *tid.
1776 : *
1777 : * heapTuple is a caller-supplied buffer. When a match is found, we return
1778 : * the tuple here, in addition to updating *tid. If no match is found, the
1779 : * contents of this buffer on return are undefined.
1780 : *
1781 : * If all_dead is not NULL, we check non-visible tuples to see if they are
1782 : * globally dead; *all_dead is set true if all members of the HOT chain
1783 : * are vacuumable, false if not.
1784 : *
1785 : * Unlike heap_fetch, the caller must already have pin and (at least) share
1786 : * lock on the buffer; it is still pinned/locked at exit.
1787 : */
1788 : bool
1789 28561158 : heap_hot_search_buffer(ItemPointer tid, Relation relation, Buffer buffer,
1790 : Snapshot snapshot, HeapTuple heapTuple,
1791 : bool *all_dead, bool first_call)
1792 : {
1793 28561158 : Page page = BufferGetPage(buffer);
1794 28561158 : TransactionId prev_xmax = InvalidTransactionId;
1795 : BlockNumber blkno;
1796 : OffsetNumber offnum;
1797 : bool at_chain_start;
1798 : bool valid;
1799 : bool skip;
1800 28561158 : GlobalVisState *vistest = NULL;
1801 :
1802 : /* If this is not the first call, previous call returned a (live!) tuple */
1803 28561158 : if (all_dead)
1804 24556593 : *all_dead = first_call;
1805 :
1806 28561158 : blkno = ItemPointerGetBlockNumber(tid);
1807 28561158 : offnum = ItemPointerGetOffsetNumber(tid);
1808 28561158 : at_chain_start = first_call;
1809 28561158 : skip = !first_call;
1810 :
1811 : /* XXX: we should assert that a snapshot is pushed or registered */
1812 : Assert(TransactionIdIsValid(RecentXmin));
1813 : Assert(BufferGetBlockNumber(buffer) == blkno);
1814 :
1815 : /* Scan through possible multiple members of HOT-chain */
1816 : for (;;)
1817 1763719 : {
1818 : ItemId lp;
1819 :
1820 : /* check for bogus TID */
1821 30324877 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1822 : break;
1823 :
1824 30324877 : lp = PageGetItemId(page, offnum);
1825 :
1826 : /* check for unused, dead, or redirected items */
1827 30324877 : if (!ItemIdIsNormal(lp))
1828 : {
1829 : /* We should only see a redirect at start of chain */
1830 1092568 : if (ItemIdIsRedirected(lp) && at_chain_start)
1831 : {
1832 : /* Follow the redirect */
1833 621660 : offnum = ItemIdGetRedirect(lp);
1834 621660 : at_chain_start = false;
1835 621660 : continue;
1836 : }
1837 : /* else must be end of chain */
1838 470908 : break;
1839 : }
1840 :
1841 : /*
1842 : * Update heapTuple to point to the element of the HOT chain we're
1843 : * currently investigating. Having t_self set correctly is important
1844 : * because the SSI checks and the *Satisfies routine for historical
1845 : * MVCC snapshots need the correct tid to decide about the visibility.
1846 : */
1847 29232309 : heapTuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
1848 29232309 : heapTuple->t_len = ItemIdGetLength(lp);
1849 29232309 : heapTuple->t_tableOid = RelationGetRelid(relation);
1850 29232309 : ItemPointerSet(&heapTuple->t_self, blkno, offnum);
1851 :
1852 : /*
1853 : * Shouldn't see a HEAP_ONLY tuple at chain start.
1854 : */
1855 29232309 : if (at_chain_start && HeapTupleIsHeapOnly(heapTuple))
1856 0 : break;
1857 :
1858 : /*
1859 : * The xmin should match the previous xmax value, else chain is
1860 : * broken.
1861 : */
1862 30374368 : if (TransactionIdIsValid(prev_xmax) &&
1863 1142059 : !TransactionIdEquals(prev_xmax,
1864 : HeapTupleHeaderGetXmin(heapTuple->t_data)))
1865 0 : break;
1866 :
1867 : /*
1868 : * When first_call is true (and thus, skip is initially false) we'll
1869 : * return the first tuple we find. But on later passes, heapTuple
1870 : * will initially be pointing to the tuple we returned last time.
1871 : * Returning it again would be incorrect (and would loop forever), so
1872 : * we skip it and return the next match we find.
1873 : */
1874 29232309 : if (!skip)
1875 : {
1876 : /* If it's visible per the snapshot, we must return it */
1877 29138488 : valid = HeapTupleSatisfiesVisibility(heapTuple, snapshot, buffer);
1878 29138488 : HeapCheckForSerializableConflictOut(valid, relation, heapTuple,
1879 : buffer, snapshot);
1880 :
1881 29138483 : if (valid)
1882 : {
1883 19899274 : ItemPointerSetOffsetNumber(tid, offnum);
1884 19899274 : PredicateLockTID(relation, &heapTuple->t_self, snapshot,
1885 19899274 : HeapTupleHeaderGetXmin(heapTuple->t_data));
1886 19899274 : if (all_dead)
1887 16275931 : *all_dead = false;
1888 19899274 : return true;
1889 : }
1890 : }
1891 9333030 : skip = false;
1892 :
1893 : /*
1894 : * If we can't see it, maybe no one else can either. At caller
1895 : * request, check whether all chain members are dead to all
1896 : * transactions.
1897 : *
1898 : * Note: if you change the criterion here for what is "dead", fix the
1899 : * planner's get_actual_variable_range() function to match.
1900 : */
1901 9333030 : if (all_dead && *all_dead)
1902 : {
1903 8504761 : if (!vistest)
1904 8338714 : vistest = GlobalVisTestFor(relation);
1905 :
1906 8504761 : if (!HeapTupleIsSurelyDead(heapTuple, vistest))
1907 8038830 : *all_dead = false;
1908 : }
1909 :
1910 : /*
1911 : * Check to see if HOT chain continues past this tuple; if so fetch
1912 : * the next offnum and loop around.
1913 : */
1914 9333030 : if (HeapTupleIsHotUpdated(heapTuple))
1915 : {
1916 : Assert(ItemPointerGetBlockNumber(&heapTuple->t_data->t_ctid) ==
1917 : blkno);
1918 1142059 : offnum = ItemPointerGetOffsetNumber(&heapTuple->t_data->t_ctid);
1919 1142059 : at_chain_start = false;
1920 1142059 : prev_xmax = HeapTupleHeaderGetUpdateXid(heapTuple->t_data);
1921 : }
1922 : else
1923 8190971 : break; /* end of chain */
1924 : }
1925 :
1926 8661879 : return false;
1927 : }
1928 :
1929 : /*
1930 : * heap_get_latest_tid - get the latest tid of a specified tuple
1931 : *
1932 : * Actually, this gets the latest version that is visible according to the
1933 : * scan's snapshot. Create a scan using SnapshotDirty to get the very latest,
1934 : * possibly uncommitted version.
1935 : *
1936 : * *tid is both an input and an output parameter: it is updated to
1937 : * show the latest version of the row. Note that it will not be changed
1938 : * if no version of the row passes the snapshot test.
1939 : */
1940 : void
1941 199 : heap_get_latest_tid(TableScanDesc sscan,
1942 : ItemPointer tid)
1943 : {
1944 199 : Relation relation = sscan->rs_rd;
1945 199 : Snapshot snapshot = sscan->rs_snapshot;
1946 : ItemPointerData ctid;
1947 : TransactionId priorXmax;
1948 :
1949 : /*
1950 : * table_tuple_get_latest_tid() verified that the passed in tid is valid.
1951 : * Assume that t_ctid links are valid however - there shouldn't be invalid
1952 : * ones in the table.
1953 : */
1954 : Assert(ItemPointerIsValid(tid));
1955 :
1956 : /*
1957 : * Loop to chase down t_ctid links. At top of loop, ctid is the tuple we
1958 : * need to examine, and *tid is the TID we will return if ctid turns out
1959 : * to be bogus.
1960 : *
1961 : * Note that we will loop until we reach the end of the t_ctid chain.
1962 : * Depending on the snapshot passed, there might be at most one visible
1963 : * version of the row, but we don't try to optimize for that.
1964 : */
1965 199 : ctid = *tid;
1966 199 : priorXmax = InvalidTransactionId; /* cannot check first XMIN */
1967 : for (;;)
1968 60 : {
1969 : Buffer buffer;
1970 : Page page;
1971 : OffsetNumber offnum;
1972 : ItemId lp;
1973 : HeapTupleData tp;
1974 : bool valid;
1975 :
1976 : /*
1977 : * Read, pin, and lock the page.
1978 : */
1979 259 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(&ctid));
1980 259 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
1981 259 : page = BufferGetPage(buffer);
1982 :
1983 : /*
1984 : * Check for bogus item number. This is not treated as an error
1985 : * condition because it can happen while following a t_ctid link. We
1986 : * just assume that the prior tid is OK and return it unchanged.
1987 : */
1988 259 : offnum = ItemPointerGetOffsetNumber(&ctid);
1989 259 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1990 : {
1991 0 : UnlockReleaseBuffer(buffer);
1992 0 : break;
1993 : }
1994 259 : lp = PageGetItemId(page, offnum);
1995 259 : if (!ItemIdIsNormal(lp))
1996 : {
1997 0 : UnlockReleaseBuffer(buffer);
1998 0 : break;
1999 : }
2000 :
2001 : /* OK to access the tuple */
2002 259 : tp.t_self = ctid;
2003 259 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
2004 259 : tp.t_len = ItemIdGetLength(lp);
2005 259 : tp.t_tableOid = RelationGetRelid(relation);
2006 :
2007 : /*
2008 : * After following a t_ctid link, we might arrive at an unrelated
2009 : * tuple. Check for XMIN match.
2010 : */
2011 319 : if (TransactionIdIsValid(priorXmax) &&
2012 60 : !TransactionIdEquals(priorXmax, HeapTupleHeaderGetXmin(tp.t_data)))
2013 : {
2014 0 : UnlockReleaseBuffer(buffer);
2015 0 : break;
2016 : }
2017 :
2018 : /*
2019 : * Check tuple visibility; if visible, set it as the new result
2020 : * candidate.
2021 : */
2022 259 : valid = HeapTupleSatisfiesVisibility(&tp, snapshot, buffer);
2023 259 : HeapCheckForSerializableConflictOut(valid, relation, &tp, buffer, snapshot);
2024 259 : if (valid)
2025 183 : *tid = ctid;
2026 :
2027 : /*
2028 : * If there's a valid t_ctid link, follow it, else we're done.
2029 : */
2030 367 : if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
2031 184 : HeapTupleHeaderIsOnlyLocked(tp.t_data) ||
2032 152 : HeapTupleHeaderIndicatesMovedPartitions(tp.t_data) ||
2033 76 : ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
2034 : {
2035 199 : UnlockReleaseBuffer(buffer);
2036 199 : break;
2037 : }
2038 :
2039 60 : ctid = tp.t_data->t_ctid;
2040 60 : priorXmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
2041 60 : UnlockReleaseBuffer(buffer);
2042 : } /* end of loop */
2043 199 : }
2044 :
2045 :
2046 : /*
2047 : * UpdateXmaxHintBits - update tuple hint bits after xmax transaction ends
2048 : *
2049 : * This is called after we have waited for the XMAX transaction to terminate.
2050 : * If the transaction aborted, we guarantee the XMAX_INVALID hint bit will
2051 : * be set on exit. If the transaction committed, we set the XMAX_COMMITTED
2052 : * hint bit if possible --- but beware that that may not yet be possible,
2053 : * if the transaction committed asynchronously.
2054 : *
2055 : * Note that if the transaction was a locker only, we set HEAP_XMAX_INVALID
2056 : * even if it commits.
2057 : *
2058 : * Hence callers should look only at XMAX_INVALID.
2059 : *
2060 : * Note this is not allowed for tuples whose xmax is a multixact.
2061 : */
2062 : static void
2063 229 : UpdateXmaxHintBits(HeapTupleHeader tuple, Buffer buffer, TransactionId xid)
2064 : {
2065 : Assert(TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple), xid));
2066 : Assert(!(tuple->t_infomask & HEAP_XMAX_IS_MULTI));
2067 :
2068 229 : if (!(tuple->t_infomask & (HEAP_XMAX_COMMITTED | HEAP_XMAX_INVALID)))
2069 : {
2070 406 : if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask) &&
2071 178 : TransactionIdDidCommit(xid))
2072 151 : HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_COMMITTED,
2073 : xid);
2074 : else
2075 77 : HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_INVALID,
2076 : InvalidTransactionId);
2077 : }
2078 229 : }
2079 :
2080 :
2081 : /*
2082 : * GetBulkInsertState - prepare status object for a bulk insert
2083 : */
2084 : BulkInsertState
2085 3476 : GetBulkInsertState(void)
2086 : {
2087 : BulkInsertState bistate;
2088 :
2089 3476 : bistate = (BulkInsertState) palloc_object(BulkInsertStateData);
2090 3476 : bistate->strategy = GetAccessStrategy(BAS_BULKWRITE);
2091 3476 : bistate->current_buf = InvalidBuffer;
2092 3476 : bistate->next_free = InvalidBlockNumber;
2093 3476 : bistate->last_free = InvalidBlockNumber;
2094 3476 : bistate->already_extended_by = 0;
2095 3476 : return bistate;
2096 : }
2097 :
2098 : /*
2099 : * FreeBulkInsertState - clean up after finishing a bulk insert
2100 : */
2101 : void
2102 3255 : FreeBulkInsertState(BulkInsertState bistate)
2103 : {
2104 3255 : if (bistate->current_buf != InvalidBuffer)
2105 2563 : ReleaseBuffer(bistate->current_buf);
2106 3255 : FreeAccessStrategy(bistate->strategy);
2107 3255 : pfree(bistate);
2108 3255 : }
2109 :
2110 : /*
2111 : * ReleaseBulkInsertStatePin - release a buffer currently held in bistate
2112 : */
2113 : void
2114 90779 : ReleaseBulkInsertStatePin(BulkInsertState bistate)
2115 : {
2116 90779 : if (bistate->current_buf != InvalidBuffer)
2117 40028 : ReleaseBuffer(bistate->current_buf);
2118 90779 : bistate->current_buf = InvalidBuffer;
2119 :
2120 : /*
2121 : * Despite the name, we also reset bulk relation extension state.
2122 : * Otherwise we can end up erroring out due to looking for free space in
2123 : * ->next_free of one partition, even though ->next_free was set when
2124 : * extending another partition. It could obviously also be bad for
2125 : * efficiency to look at existing blocks at offsets from another
2126 : * partition, even if we don't error out.
2127 : */
2128 90779 : bistate->next_free = InvalidBlockNumber;
2129 90779 : bistate->last_free = InvalidBlockNumber;
2130 90779 : }
2131 :
2132 :
2133 : /*
2134 : * heap_insert - insert tuple into a heap
2135 : *
2136 : * The new tuple is stamped with current transaction ID and the specified
2137 : * command ID.
2138 : *
2139 : * See table_tuple_insert for comments about most of the input flags, except
2140 : * that this routine directly takes a tuple rather than a slot.
2141 : *
2142 : * There's corresponding HEAP_INSERT_ options to all the TABLE_INSERT_
2143 : * options, and there additionally is HEAP_INSERT_SPECULATIVE which is used to
2144 : * implement table_tuple_insert_speculative().
2145 : *
2146 : * On return the header fields of *tup are updated to match the stored tuple;
2147 : * in particular tup->t_self receives the actual TID where the tuple was
2148 : * stored. But note that any toasting of fields within the tuple data is NOT
2149 : * reflected into *tup.
2150 : */
2151 : void
2152 10108893 : heap_insert(Relation relation, HeapTuple tup, CommandId cid,
2153 : int options, BulkInsertState bistate)
2154 : {
2155 10108893 : TransactionId xid = GetCurrentTransactionId();
2156 : HeapTuple heaptup;
2157 : Buffer buffer;
2158 10108886 : Buffer vmbuffer = InvalidBuffer;
2159 10108886 : bool all_visible_cleared = false;
2160 :
2161 : /* Cheap, simplistic check that the tuple matches the rel's rowtype. */
2162 : Assert(HeapTupleHeaderGetNatts(tup->t_data) <=
2163 : RelationGetNumberOfAttributes(relation));
2164 :
2165 10108886 : AssertHasSnapshotForToast(relation);
2166 :
2167 : /*
2168 : * Fill in tuple header fields and toast the tuple if necessary.
2169 : *
2170 : * Note: below this point, heaptup is the data we actually intend to store
2171 : * into the relation; tup is the caller's original untoasted data.
2172 : */
2173 10108886 : heaptup = heap_prepare_insert(relation, tup, xid, cid, options);
2174 :
2175 : /*
2176 : * Find buffer to insert this tuple into. If the page is all visible,
2177 : * this will also pin the requisite visibility map page.
2178 : */
2179 10108886 : buffer = RelationGetBufferForTuple(relation, heaptup->t_len,
2180 : InvalidBuffer, options, bistate,
2181 : &vmbuffer, NULL,
2182 : 0);
2183 :
2184 : /*
2185 : * We're about to do the actual insert -- but check for conflict first, to
2186 : * avoid possibly having to roll back work we've just done.
2187 : *
2188 : * This is safe without a recheck as long as there is no possibility of
2189 : * another process scanning the page between this check and the insert
2190 : * being visible to the scan (i.e., an exclusive buffer content lock is
2191 : * continuously held from this point until the tuple insert is visible).
2192 : *
2193 : * For a heap insert, we only need to check for table-level SSI locks. Our
2194 : * new tuple can't possibly conflict with existing tuple locks, and heap
2195 : * page locks are only consolidated versions of tuple locks; they do not
2196 : * lock "gaps" as index page locks do. So we don't need to specify a
2197 : * buffer when making the call, which makes for a faster check.
2198 : */
2199 10108886 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2200 :
2201 : /* NO EREPORT(ERROR) from here till changes are logged */
2202 10108874 : START_CRIT_SECTION();
2203 :
2204 10108874 : RelationPutHeapTuple(relation, buffer, heaptup,
2205 10108874 : (options & HEAP_INSERT_SPECULATIVE) != 0);
2206 :
2207 10108874 : if (PageIsAllVisible(BufferGetPage(buffer)))
2208 : {
2209 7296 : all_visible_cleared = true;
2210 7296 : PageClearAllVisible(BufferGetPage(buffer));
2211 7296 : visibilitymap_clear(relation,
2212 7296 : ItemPointerGetBlockNumber(&(heaptup->t_self)),
2213 : vmbuffer, VISIBILITYMAP_VALID_BITS);
2214 : }
2215 :
2216 : /*
2217 : * XXX Should we set PageSetPrunable on this page ?
2218 : *
2219 : * The inserting transaction may eventually abort thus making this tuple
2220 : * DEAD and hence available for pruning. Though we don't want to optimize
2221 : * for aborts, if no other tuple in this page is UPDATEd/DELETEd, the
2222 : * aborted tuple will never be pruned until next vacuum is triggered.
2223 : *
2224 : * If you do add PageSetPrunable here, add it in heap_xlog_insert too.
2225 : */
2226 :
2227 10108874 : MarkBufferDirty(buffer);
2228 :
2229 : /* XLOG stuff */
2230 10108874 : if (RelationNeedsWAL(relation))
2231 : {
2232 : xl_heap_insert xlrec;
2233 : xl_heap_header xlhdr;
2234 : XLogRecPtr recptr;
2235 8671048 : Page page = BufferGetPage(buffer);
2236 8671048 : uint8 info = XLOG_HEAP_INSERT;
2237 8671048 : int bufflags = 0;
2238 :
2239 : /*
2240 : * If this is a catalog, we need to transmit combo CIDs to properly
2241 : * decode, so log that as well.
2242 : */
2243 8671048 : if (RelationIsAccessibleInLogicalDecoding(relation))
2244 3498 : log_heap_new_cid(relation, heaptup);
2245 :
2246 : /*
2247 : * If this is the single and first tuple on page, we can reinit the
2248 : * page instead of restoring the whole thing. Set flag, and hide
2249 : * buffer references from XLogInsert.
2250 : */
2251 8780679 : if (ItemPointerGetOffsetNumber(&(heaptup->t_self)) == FirstOffsetNumber &&
2252 109631 : PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
2253 : {
2254 108755 : info |= XLOG_HEAP_INIT_PAGE;
2255 108755 : bufflags |= REGBUF_WILL_INIT;
2256 : }
2257 :
2258 8671048 : xlrec.offnum = ItemPointerGetOffsetNumber(&heaptup->t_self);
2259 8671048 : xlrec.flags = 0;
2260 8671048 : if (all_visible_cleared)
2261 7292 : xlrec.flags |= XLH_INSERT_ALL_VISIBLE_CLEARED;
2262 8671048 : if (options & HEAP_INSERT_SPECULATIVE)
2263 2206 : xlrec.flags |= XLH_INSERT_IS_SPECULATIVE;
2264 : Assert(ItemPointerGetBlockNumber(&heaptup->t_self) == BufferGetBlockNumber(buffer));
2265 :
2266 : /*
2267 : * For logical decoding, we need the tuple even if we're doing a full
2268 : * page write, so make sure it's included even if we take a full-page
2269 : * image. (XXX We could alternatively store a pointer into the FPW).
2270 : */
2271 8671048 : if (RelationIsLogicallyLogged(relation) &&
2272 252700 : !(options & HEAP_INSERT_NO_LOGICAL))
2273 : {
2274 252661 : xlrec.flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
2275 252661 : bufflags |= REGBUF_KEEP_DATA;
2276 :
2277 252661 : if (IsToastRelation(relation))
2278 1781 : xlrec.flags |= XLH_INSERT_ON_TOAST_RELATION;
2279 : }
2280 :
2281 8671048 : XLogBeginInsert();
2282 8671048 : XLogRegisterData(&xlrec, SizeOfHeapInsert);
2283 :
2284 8671048 : xlhdr.t_infomask2 = heaptup->t_data->t_infomask2;
2285 8671048 : xlhdr.t_infomask = heaptup->t_data->t_infomask;
2286 8671048 : xlhdr.t_hoff = heaptup->t_data->t_hoff;
2287 :
2288 : /*
2289 : * note we mark xlhdr as belonging to buffer; if XLogInsert decides to
2290 : * write the whole page to the xlog, we don't need to store
2291 : * xl_heap_header in the xlog.
2292 : */
2293 8671048 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD | bufflags);
2294 8671048 : XLogRegisterBufData(0, &xlhdr, SizeOfHeapHeader);
2295 : /* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
2296 8671048 : XLogRegisterBufData(0,
2297 8671048 : (char *) heaptup->t_data + SizeofHeapTupleHeader,
2298 8671048 : heaptup->t_len - SizeofHeapTupleHeader);
2299 :
2300 : /* filtering by origin on a row level is much more efficient */
2301 8671048 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
2302 :
2303 8671048 : recptr = XLogInsert(RM_HEAP_ID, info);
2304 :
2305 8671048 : PageSetLSN(page, recptr);
2306 : }
2307 :
2308 10108874 : END_CRIT_SECTION();
2309 :
2310 10108874 : UnlockReleaseBuffer(buffer);
2311 10108874 : if (vmbuffer != InvalidBuffer)
2312 7665 : ReleaseBuffer(vmbuffer);
2313 :
2314 : /*
2315 : * If tuple is cacheable, mark it for invalidation from the caches in case
2316 : * we abort. Note it is OK to do this after releasing the buffer, because
2317 : * the heaptup data structure is all in local memory, not in the shared
2318 : * buffer.
2319 : */
2320 10108874 : CacheInvalidateHeapTuple(relation, heaptup, NULL);
2321 :
2322 : /* Note: speculative insertions are counted too, even if aborted later */
2323 10108874 : pgstat_count_heap_insert(relation, 1);
2324 :
2325 : /*
2326 : * If heaptup is a private copy, release it. Don't forget to copy t_self
2327 : * back to the caller's image, too.
2328 : */
2329 10108874 : if (heaptup != tup)
2330 : {
2331 21108 : tup->t_self = heaptup->t_self;
2332 21108 : heap_freetuple(heaptup);
2333 : }
2334 10108874 : }
2335 :
2336 : /*
2337 : * Subroutine for heap_insert(). Prepares a tuple for insertion. This sets the
2338 : * tuple header fields and toasts the tuple if necessary. Returns a toasted
2339 : * version of the tuple if it was toasted, or the original tuple if not. Note
2340 : * that in any case, the header fields are also set in the original tuple.
2341 : */
2342 : static HeapTuple
2343 11879116 : heap_prepare_insert(Relation relation, HeapTuple tup, TransactionId xid,
2344 : CommandId cid, int options)
2345 : {
2346 : /*
2347 : * To allow parallel inserts, we need to ensure that they are safe to be
2348 : * performed in workers. We have the infrastructure to allow parallel
2349 : * inserts in general except for the cases where inserts generate a new
2350 : * CommandId (eg. inserts into a table having a foreign key column).
2351 : */
2352 11879116 : if (IsParallelWorker())
2353 0 : ereport(ERROR,
2354 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
2355 : errmsg("cannot insert tuples in a parallel worker")));
2356 :
2357 11879116 : tup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
2358 11879116 : tup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
2359 11879116 : tup->t_data->t_infomask |= HEAP_XMAX_INVALID;
2360 11879116 : HeapTupleHeaderSetXmin(tup->t_data, xid);
2361 11879116 : if (options & HEAP_INSERT_FROZEN)
2362 102651 : HeapTupleHeaderSetXminFrozen(tup->t_data);
2363 :
2364 11879116 : HeapTupleHeaderSetCmin(tup->t_data, cid);
2365 11879116 : HeapTupleHeaderSetXmax(tup->t_data, 0); /* for cleanliness */
2366 11879116 : tup->t_tableOid = RelationGetRelid(relation);
2367 :
2368 : /*
2369 : * If the new tuple is too big for storage or contains already toasted
2370 : * out-of-line attributes from some other relation, invoke the toaster.
2371 : */
2372 11879116 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
2373 33825 : relation->rd_rel->relkind != RELKIND_MATVIEW)
2374 : {
2375 : /* toast table entries should never be recursively toasted */
2376 : Assert(!HeapTupleHasExternal(tup));
2377 33764 : return tup;
2378 : }
2379 11845352 : else if (HeapTupleHasExternal(tup) || tup->t_len > TOAST_TUPLE_THRESHOLD)
2380 21167 : return heap_toast_insert_or_update(relation, tup, NULL, options);
2381 : else
2382 11824185 : return tup;
2383 : }
2384 :
2385 : /*
2386 : * Helper for heap_multi_insert() that computes the number of entire pages
2387 : * that inserting the remaining heaptuples requires. Used to determine how
2388 : * much the relation needs to be extended by.
2389 : */
2390 : static int
2391 464720 : heap_multi_insert_pages(HeapTuple *heaptuples, int done, int ntuples, Size saveFreeSpace)
2392 : {
2393 464720 : size_t page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
2394 464720 : int npages = 1;
2395 :
2396 2910530 : for (int i = done; i < ntuples; i++)
2397 : {
2398 2445810 : size_t tup_sz = sizeof(ItemIdData) + MAXALIGN(heaptuples[i]->t_len);
2399 :
2400 2445810 : if (page_avail < tup_sz)
2401 : {
2402 17503 : npages++;
2403 17503 : page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
2404 : }
2405 2445810 : page_avail -= tup_sz;
2406 : }
2407 :
2408 464720 : return npages;
2409 : }
2410 :
2411 : /*
2412 : * heap_multi_insert - insert multiple tuples into a heap
2413 : *
2414 : * This is like heap_insert(), but inserts multiple tuples in one operation.
2415 : * That's faster than calling heap_insert() in a loop, because when multiple
2416 : * tuples can be inserted on a single page, we can write just a single WAL
2417 : * record covering all of them, and only need to lock/unlock the page once.
2418 : *
2419 : * Note: this leaks memory into the current memory context. You can create a
2420 : * temporary context before calling this, if that's a problem.
2421 : */
2422 : void
2423 456790 : heap_multi_insert(Relation relation, TupleTableSlot **slots, int ntuples,
2424 : CommandId cid, int options, BulkInsertState bistate)
2425 : {
2426 456790 : TransactionId xid = GetCurrentTransactionId();
2427 : HeapTuple *heaptuples;
2428 : int i;
2429 : int ndone;
2430 : PGAlignedBlock scratch;
2431 : Page page;
2432 456790 : Buffer vmbuffer = InvalidBuffer;
2433 : bool needwal;
2434 : Size saveFreeSpace;
2435 456790 : bool need_tuple_data = RelationIsLogicallyLogged(relation);
2436 456790 : bool need_cids = RelationIsAccessibleInLogicalDecoding(relation);
2437 456790 : bool starting_with_empty_page = false;
2438 456790 : int npages = 0;
2439 456790 : int npages_used = 0;
2440 :
2441 : /* currently not needed (thus unsupported) for heap_multi_insert() */
2442 : Assert(!(options & HEAP_INSERT_NO_LOGICAL));
2443 :
2444 456790 : AssertHasSnapshotForToast(relation);
2445 :
2446 456790 : needwal = RelationNeedsWAL(relation);
2447 456790 : saveFreeSpace = RelationGetTargetPageFreeSpace(relation,
2448 : HEAP_DEFAULT_FILLFACTOR);
2449 :
2450 : /* Toast and set header data in all the slots */
2451 456790 : heaptuples = palloc(ntuples * sizeof(HeapTuple));
2452 2227020 : for (i = 0; i < ntuples; i++)
2453 : {
2454 : HeapTuple tuple;
2455 :
2456 1770230 : tuple = ExecFetchSlotHeapTuple(slots[i], true, NULL);
2457 1770230 : slots[i]->tts_tableOid = RelationGetRelid(relation);
2458 1770230 : tuple->t_tableOid = slots[i]->tts_tableOid;
2459 1770230 : heaptuples[i] = heap_prepare_insert(relation, tuple, xid, cid,
2460 : options);
2461 : }
2462 :
2463 : /*
2464 : * We're about to do the actual inserts -- but check for conflict first,
2465 : * to minimize the possibility of having to roll back work we've just
2466 : * done.
2467 : *
2468 : * A check here does not definitively prevent a serialization anomaly;
2469 : * that check MUST be done at least past the point of acquiring an
2470 : * exclusive buffer content lock on every buffer that will be affected,
2471 : * and MAY be done after all inserts are reflected in the buffers and
2472 : * those locks are released; otherwise there is a race condition. Since
2473 : * multiple buffers can be locked and unlocked in the loop below, and it
2474 : * would not be feasible to identify and lock all of those buffers before
2475 : * the loop, we must do a final check at the end.
2476 : *
2477 : * The check here could be omitted with no loss of correctness; it is
2478 : * present strictly as an optimization.
2479 : *
2480 : * For heap inserts, we only need to check for table-level SSI locks. Our
2481 : * new tuples can't possibly conflict with existing tuple locks, and heap
2482 : * page locks are only consolidated versions of tuple locks; they do not
2483 : * lock "gaps" as index page locks do. So we don't need to specify a
2484 : * buffer when making the call, which makes for a faster check.
2485 : */
2486 456790 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2487 :
2488 456790 : ndone = 0;
2489 930751 : while (ndone < ntuples)
2490 : {
2491 : Buffer buffer;
2492 473961 : bool all_visible_cleared = false;
2493 473961 : bool all_frozen_set = false;
2494 : int nthispage;
2495 :
2496 473961 : CHECK_FOR_INTERRUPTS();
2497 :
2498 : /*
2499 : * Compute number of pages needed to fit the to-be-inserted tuples in
2500 : * the worst case. This will be used to determine how much to extend
2501 : * the relation by in RelationGetBufferForTuple(), if needed. If we
2502 : * filled a prior page from scratch, we can just update our last
2503 : * computation, but if we started with a partially filled page,
2504 : * recompute from scratch, the number of potentially required pages
2505 : * can vary due to tuples needing to fit onto the page, page headers
2506 : * etc.
2507 : */
2508 473961 : if (ndone == 0 || !starting_with_empty_page)
2509 : {
2510 464720 : npages = heap_multi_insert_pages(heaptuples, ndone, ntuples,
2511 : saveFreeSpace);
2512 464720 : npages_used = 0;
2513 : }
2514 : else
2515 9241 : npages_used++;
2516 :
2517 : /*
2518 : * Find buffer where at least the next tuple will fit. If the page is
2519 : * all-visible, this will also pin the requisite visibility map page.
2520 : *
2521 : * Also pin visibility map page if COPY FREEZE inserts tuples into an
2522 : * empty page. See all_frozen_set below.
2523 : */
2524 473961 : buffer = RelationGetBufferForTuple(relation, heaptuples[ndone]->t_len,
2525 : InvalidBuffer, options, bistate,
2526 : &vmbuffer, NULL,
2527 : npages - npages_used);
2528 473961 : page = BufferGetPage(buffer);
2529 :
2530 473961 : starting_with_empty_page = PageGetMaxOffsetNumber(page) == 0;
2531 :
2532 473961 : if (starting_with_empty_page && (options & HEAP_INSERT_FROZEN))
2533 : {
2534 1665 : all_frozen_set = true;
2535 : /* Lock the vmbuffer before entering the critical section */
2536 1665 : LockBuffer(vmbuffer, BUFFER_LOCK_EXCLUSIVE);
2537 : }
2538 :
2539 : /* NO EREPORT(ERROR) from here till changes are logged */
2540 473961 : START_CRIT_SECTION();
2541 :
2542 : /*
2543 : * RelationGetBufferForTuple has ensured that the first tuple fits.
2544 : * Put that on the page, and then as many other tuples as fit.
2545 : */
2546 473961 : RelationPutHeapTuple(relation, buffer, heaptuples[ndone], false);
2547 :
2548 : /*
2549 : * For logical decoding we need combo CIDs to properly decode the
2550 : * catalog.
2551 : */
2552 473961 : if (needwal && need_cids)
2553 5139 : log_heap_new_cid(relation, heaptuples[ndone]);
2554 :
2555 1770230 : for (nthispage = 1; ndone + nthispage < ntuples; nthispage++)
2556 : {
2557 1313440 : HeapTuple heaptup = heaptuples[ndone + nthispage];
2558 :
2559 1313440 : if (PageGetHeapFreeSpace(page) < MAXALIGN(heaptup->t_len) + saveFreeSpace)
2560 17171 : break;
2561 :
2562 1296269 : RelationPutHeapTuple(relation, buffer, heaptup, false);
2563 :
2564 : /*
2565 : * For logical decoding we need combo CIDs to properly decode the
2566 : * catalog.
2567 : */
2568 1296269 : if (needwal && need_cids)
2569 4802 : log_heap_new_cid(relation, heaptup);
2570 : }
2571 :
2572 : /*
2573 : * If the page is all visible, need to clear that, unless we're only
2574 : * going to add further frozen rows to it.
2575 : *
2576 : * If we're only adding already frozen rows to a previously empty
2577 : * page, mark it as all-frozen and update the visibility map. We're
2578 : * already holding a pin on the vmbuffer.
2579 : */
2580 473961 : if (PageIsAllVisible(page) && !(options & HEAP_INSERT_FROZEN))
2581 : {
2582 3763 : all_visible_cleared = true;
2583 3763 : PageClearAllVisible(page);
2584 3763 : visibilitymap_clear(relation,
2585 : BufferGetBlockNumber(buffer),
2586 : vmbuffer, VISIBILITYMAP_VALID_BITS);
2587 : }
2588 470198 : else if (all_frozen_set)
2589 : {
2590 1665 : PageSetAllVisible(page);
2591 1665 : PageClearPrunable(page);
2592 1665 : visibilitymap_set_vmbits(BufferGetBlockNumber(buffer),
2593 : vmbuffer,
2594 : VISIBILITYMAP_ALL_VISIBLE |
2595 : VISIBILITYMAP_ALL_FROZEN,
2596 : relation->rd_locator);
2597 : }
2598 :
2599 : /*
2600 : * XXX Should we set PageSetPrunable on this page ? See heap_insert()
2601 : */
2602 :
2603 473961 : MarkBufferDirty(buffer);
2604 :
2605 : /* XLOG stuff */
2606 473961 : if (needwal)
2607 : {
2608 : XLogRecPtr recptr;
2609 : xl_heap_multi_insert *xlrec;
2610 470061 : uint8 info = XLOG_HEAP2_MULTI_INSERT;
2611 : char *tupledata;
2612 : int totaldatalen;
2613 470061 : char *scratchptr = scratch.data;
2614 : bool init;
2615 470061 : int bufflags = 0;
2616 :
2617 : /*
2618 : * If the page was previously empty, we can reinit the page
2619 : * instead of restoring the whole thing.
2620 : */
2621 470061 : init = starting_with_empty_page;
2622 :
2623 : /* allocate xl_heap_multi_insert struct from the scratch area */
2624 470061 : xlrec = (xl_heap_multi_insert *) scratchptr;
2625 470061 : scratchptr += SizeOfHeapMultiInsert;
2626 :
2627 : /*
2628 : * Allocate offsets array. Unless we're reinitializing the page,
2629 : * in that case the tuples are stored in order starting at
2630 : * FirstOffsetNumber and we don't need to store the offsets
2631 : * explicitly.
2632 : */
2633 470061 : if (!init)
2634 454235 : scratchptr += nthispage * sizeof(OffsetNumber);
2635 :
2636 : /* the rest of the scratch space is used for tuple data */
2637 470061 : tupledata = scratchptr;
2638 :
2639 : /* check that the mutually exclusive flags are not both set */
2640 : Assert(!(all_visible_cleared && all_frozen_set));
2641 :
2642 470061 : xlrec->flags = 0;
2643 470061 : if (all_visible_cleared)
2644 3763 : xlrec->flags = XLH_INSERT_ALL_VISIBLE_CLEARED;
2645 :
2646 : /*
2647 : * We don't have to worry about including a conflict xid in the
2648 : * WAL record, as HEAP_INSERT_FROZEN intentionally violates
2649 : * visibility rules.
2650 : */
2651 470061 : if (all_frozen_set)
2652 17 : xlrec->flags = XLH_INSERT_ALL_FROZEN_SET;
2653 :
2654 470061 : xlrec->ntuples = nthispage;
2655 :
2656 : /*
2657 : * Write out an xl_multi_insert_tuple and the tuple data itself
2658 : * for each tuple.
2659 : */
2660 2033699 : for (i = 0; i < nthispage; i++)
2661 : {
2662 1563638 : HeapTuple heaptup = heaptuples[ndone + i];
2663 : xl_multi_insert_tuple *tuphdr;
2664 : int datalen;
2665 :
2666 1563638 : if (!init)
2667 934013 : xlrec->offsets[i] = ItemPointerGetOffsetNumber(&heaptup->t_self);
2668 : /* xl_multi_insert_tuple needs two-byte alignment. */
2669 1563638 : tuphdr = (xl_multi_insert_tuple *) SHORTALIGN(scratchptr);
2670 1563638 : scratchptr = ((char *) tuphdr) + SizeOfMultiInsertTuple;
2671 :
2672 1563638 : tuphdr->t_infomask2 = heaptup->t_data->t_infomask2;
2673 1563638 : tuphdr->t_infomask = heaptup->t_data->t_infomask;
2674 1563638 : tuphdr->t_hoff = heaptup->t_data->t_hoff;
2675 :
2676 : /* write bitmap [+ padding] [+ oid] + data */
2677 1563638 : datalen = heaptup->t_len - SizeofHeapTupleHeader;
2678 1563638 : memcpy(scratchptr,
2679 1563638 : (char *) heaptup->t_data + SizeofHeapTupleHeader,
2680 : datalen);
2681 1563638 : tuphdr->datalen = datalen;
2682 1563638 : scratchptr += datalen;
2683 : }
2684 470061 : totaldatalen = scratchptr - tupledata;
2685 : Assert((scratchptr - scratch.data) < BLCKSZ);
2686 :
2687 470061 : if (need_tuple_data)
2688 72 : xlrec->flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
2689 :
2690 : /*
2691 : * Signal that this is the last xl_heap_multi_insert record
2692 : * emitted by this call to heap_multi_insert(). Needed for logical
2693 : * decoding so it knows when to cleanup temporary data.
2694 : */
2695 470061 : if (ndone + nthispage == ntuples)
2696 456333 : xlrec->flags |= XLH_INSERT_LAST_IN_MULTI;
2697 :
2698 470061 : if (init)
2699 : {
2700 15826 : info |= XLOG_HEAP_INIT_PAGE;
2701 15826 : bufflags |= REGBUF_WILL_INIT;
2702 : }
2703 :
2704 : /*
2705 : * If we're doing logical decoding, include the new tuple data
2706 : * even if we take a full-page image of the page.
2707 : */
2708 470061 : if (need_tuple_data)
2709 72 : bufflags |= REGBUF_KEEP_DATA;
2710 :
2711 470061 : XLogBeginInsert();
2712 470061 : XLogRegisterData(xlrec, tupledata - scratch.data);
2713 470061 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD | bufflags);
2714 470061 : if (all_frozen_set)
2715 17 : XLogRegisterBuffer(1, vmbuffer, 0);
2716 :
2717 470061 : XLogRegisterBufData(0, tupledata, totaldatalen);
2718 :
2719 : /* filtering by origin on a row level is much more efficient */
2720 470061 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
2721 :
2722 470061 : recptr = XLogInsert(RM_HEAP2_ID, info);
2723 :
2724 470061 : PageSetLSN(page, recptr);
2725 470061 : if (all_frozen_set)
2726 : {
2727 : Assert(BufferIsDirty(vmbuffer));
2728 17 : PageSetLSN(BufferGetPage(vmbuffer), recptr);
2729 : }
2730 : }
2731 :
2732 473961 : END_CRIT_SECTION();
2733 :
2734 473961 : if (all_frozen_set)
2735 1665 : LockBuffer(vmbuffer, BUFFER_LOCK_UNLOCK);
2736 :
2737 473961 : UnlockReleaseBuffer(buffer);
2738 473961 : ndone += nthispage;
2739 :
2740 : /*
2741 : * NB: Only release vmbuffer after inserting all tuples - it's fairly
2742 : * likely that we'll insert into subsequent heap pages that are likely
2743 : * to use the same vm page.
2744 : */
2745 : }
2746 :
2747 : /* We're done with inserting all tuples, so release the last vmbuffer. */
2748 456790 : if (vmbuffer != InvalidBuffer)
2749 3847 : ReleaseBuffer(vmbuffer);
2750 :
2751 : /*
2752 : * We're done with the actual inserts. Check for conflicts again, to
2753 : * ensure that all rw-conflicts in to these inserts are detected. Without
2754 : * this final check, a sequential scan of the heap may have locked the
2755 : * table after the "before" check, missing one opportunity to detect the
2756 : * conflict, and then scanned the table before the new tuples were there,
2757 : * missing the other chance to detect the conflict.
2758 : *
2759 : * For heap inserts, we only need to check for table-level SSI locks. Our
2760 : * new tuples can't possibly conflict with existing tuple locks, and heap
2761 : * page locks are only consolidated versions of tuple locks; they do not
2762 : * lock "gaps" as index page locks do. So we don't need to specify a
2763 : * buffer when making the call.
2764 : */
2765 456790 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2766 :
2767 : /*
2768 : * If tuples are cacheable, mark them for invalidation from the caches in
2769 : * case we abort. Note it is OK to do this after releasing the buffer,
2770 : * because the heaptuples data structure is all in local memory, not in
2771 : * the shared buffer.
2772 : */
2773 456790 : if (IsCatalogRelation(relation))
2774 : {
2775 1512578 : for (i = 0; i < ntuples; i++)
2776 1057248 : CacheInvalidateHeapTuple(relation, heaptuples[i], NULL);
2777 : }
2778 :
2779 : /* copy t_self fields back to the caller's slots */
2780 2227020 : for (i = 0; i < ntuples; i++)
2781 1770230 : slots[i]->tts_tid = heaptuples[i]->t_self;
2782 :
2783 456790 : pgstat_count_heap_insert(relation, ntuples);
2784 456790 : }
2785 :
2786 : /*
2787 : * simple_heap_insert - insert a tuple
2788 : *
2789 : * Currently, this routine differs from heap_insert only in supplying
2790 : * a default command ID and not allowing access to the speedup options.
2791 : *
2792 : * This should be used rather than using heap_insert directly in most places
2793 : * where we are modifying system catalogs.
2794 : */
2795 : void
2796 992104 : simple_heap_insert(Relation relation, HeapTuple tup)
2797 : {
2798 992104 : heap_insert(relation, tup, GetCurrentCommandId(true), 0, NULL);
2799 992104 : }
2800 :
2801 : /*
2802 : * Given infomask/infomask2, compute the bits that must be saved in the
2803 : * "infobits" field of xl_heap_delete, xl_heap_update, xl_heap_lock,
2804 : * xl_heap_lock_updated WAL records.
2805 : *
2806 : * See fix_infomask_from_infobits.
2807 : */
2808 : static uint8
2809 2392921 : compute_infobits(uint16 infomask, uint16 infomask2)
2810 : {
2811 : return
2812 2392921 : ((infomask & HEAP_XMAX_IS_MULTI) != 0 ? XLHL_XMAX_IS_MULTI : 0) |
2813 2392921 : ((infomask & HEAP_XMAX_LOCK_ONLY) != 0 ? XLHL_XMAX_LOCK_ONLY : 0) |
2814 2392921 : ((infomask & HEAP_XMAX_EXCL_LOCK) != 0 ? XLHL_XMAX_EXCL_LOCK : 0) |
2815 : /* note we ignore HEAP_XMAX_SHR_LOCK here */
2816 4785842 : ((infomask & HEAP_XMAX_KEYSHR_LOCK) != 0 ? XLHL_XMAX_KEYSHR_LOCK : 0) |
2817 : ((infomask2 & HEAP_KEYS_UPDATED) != 0 ?
2818 2392921 : XLHL_KEYS_UPDATED : 0);
2819 : }
2820 :
2821 : /*
2822 : * Given two versions of the same t_infomask for a tuple, compare them and
2823 : * return whether the relevant status for a tuple Xmax has changed. This is
2824 : * used after a buffer lock has been released and reacquired: we want to ensure
2825 : * that the tuple state continues to be the same it was when we previously
2826 : * examined it.
2827 : *
2828 : * Note the Xmax field itself must be compared separately.
2829 : */
2830 : static inline bool
2831 5398 : xmax_infomask_changed(uint16 new_infomask, uint16 old_infomask)
2832 : {
2833 5398 : const uint16 interesting =
2834 : HEAP_XMAX_IS_MULTI | HEAP_XMAX_LOCK_ONLY | HEAP_LOCK_MASK;
2835 :
2836 5398 : if ((new_infomask & interesting) != (old_infomask & interesting))
2837 17 : return true;
2838 :
2839 5381 : return false;
2840 : }
2841 :
2842 : /*
2843 : * heap_delete - delete a tuple
2844 : *
2845 : * See table_tuple_delete() for an explanation of the parameters, except that
2846 : * this routine directly takes a tuple rather than a slot.
2847 : *
2848 : * In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
2849 : * t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
2850 : * only for TM_SelfModified, since we cannot obtain cmax from a combo CID
2851 : * generated by another transaction).
2852 : */
2853 : TM_Result
2854 1844687 : heap_delete(Relation relation, const ItemPointerData *tid,
2855 : CommandId cid, Snapshot crosscheck, bool wait,
2856 : TM_FailureData *tmfd, bool changingPart)
2857 : {
2858 : TM_Result result;
2859 1844687 : TransactionId xid = GetCurrentTransactionId();
2860 : ItemId lp;
2861 : HeapTupleData tp;
2862 : Page page;
2863 : BlockNumber block;
2864 : Buffer buffer;
2865 1844687 : Buffer vmbuffer = InvalidBuffer;
2866 : TransactionId new_xmax;
2867 : uint16 new_infomask,
2868 : new_infomask2;
2869 1844687 : bool have_tuple_lock = false;
2870 : bool iscombo;
2871 1844687 : bool all_visible_cleared = false;
2872 1844687 : HeapTuple old_key_tuple = NULL; /* replica identity of the tuple */
2873 1844687 : bool old_key_copied = false;
2874 :
2875 : Assert(ItemPointerIsValid(tid));
2876 :
2877 1844687 : AssertHasSnapshotForToast(relation);
2878 :
2879 : /*
2880 : * Forbid this during a parallel operation, lest it allocate a combo CID.
2881 : * Other workers might need that combo CID for visibility checks, and we
2882 : * have no provision for broadcasting it to them.
2883 : */
2884 1844687 : if (IsInParallelMode())
2885 0 : ereport(ERROR,
2886 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
2887 : errmsg("cannot delete tuples during a parallel operation")));
2888 :
2889 1844687 : block = ItemPointerGetBlockNumber(tid);
2890 1844687 : buffer = ReadBuffer(relation, block);
2891 1844687 : page = BufferGetPage(buffer);
2892 :
2893 : /*
2894 : * Before locking the buffer, pin the visibility map page if it appears to
2895 : * be necessary. Since we haven't got the lock yet, someone else might be
2896 : * in the middle of changing this, so we'll need to recheck after we have
2897 : * the lock.
2898 : */
2899 1844687 : if (PageIsAllVisible(page))
2900 259 : visibilitymap_pin(relation, block, &vmbuffer);
2901 :
2902 1844687 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2903 :
2904 1844687 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
2905 : Assert(ItemIdIsNormal(lp));
2906 :
2907 1844687 : tp.t_tableOid = RelationGetRelid(relation);
2908 1844687 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
2909 1844687 : tp.t_len = ItemIdGetLength(lp);
2910 1844687 : tp.t_self = *tid;
2911 :
2912 1 : l1:
2913 :
2914 : /*
2915 : * If we didn't pin the visibility map page and the page has become all
2916 : * visible while we were busy locking the buffer, we'll have to unlock and
2917 : * re-lock, to avoid holding the buffer lock across an I/O. That's a bit
2918 : * unfortunate, but hopefully shouldn't happen often.
2919 : */
2920 1844688 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
2921 : {
2922 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2923 0 : visibilitymap_pin(relation, block, &vmbuffer);
2924 0 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2925 : }
2926 :
2927 1844688 : result = HeapTupleSatisfiesUpdate(&tp, cid, buffer);
2928 :
2929 1844688 : if (result == TM_Invisible)
2930 : {
2931 0 : UnlockReleaseBuffer(buffer);
2932 0 : ereport(ERROR,
2933 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2934 : errmsg("attempted to delete invisible tuple")));
2935 : }
2936 1844688 : else if (result == TM_BeingModified && wait)
2937 : {
2938 : TransactionId xwait;
2939 : uint16 infomask;
2940 :
2941 : /* must copy state data before unlocking buffer */
2942 40594 : xwait = HeapTupleHeaderGetRawXmax(tp.t_data);
2943 40594 : infomask = tp.t_data->t_infomask;
2944 :
2945 : /*
2946 : * Sleep until concurrent transaction ends -- except when there's a
2947 : * single locker and it's our own transaction. Note we don't care
2948 : * which lock mode the locker has, because we need the strongest one.
2949 : *
2950 : * Before sleeping, we need to acquire tuple lock to establish our
2951 : * priority for the tuple (see heap_lock_tuple). LockTuple will
2952 : * release us when we are next-in-line for the tuple.
2953 : *
2954 : * If we are forced to "start over" below, we keep the tuple lock;
2955 : * this arranges that we stay at the head of the line while rechecking
2956 : * tuple state.
2957 : */
2958 40594 : if (infomask & HEAP_XMAX_IS_MULTI)
2959 : {
2960 8 : bool current_is_member = false;
2961 :
2962 8 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
2963 : LockTupleExclusive, ¤t_is_member))
2964 : {
2965 8 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2966 :
2967 : /*
2968 : * Acquire the lock, if necessary (but skip it when we're
2969 : * requesting a lock and already have one; avoids deadlock).
2970 : */
2971 8 : if (!current_is_member)
2972 6 : heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
2973 : LockWaitBlock, &have_tuple_lock);
2974 :
2975 : /* wait for multixact */
2976 8 : MultiXactIdWait((MultiXactId) xwait, MultiXactStatusUpdate, infomask,
2977 : relation, &(tp.t_self), XLTW_Delete,
2978 : NULL);
2979 8 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2980 :
2981 : /*
2982 : * If xwait had just locked the tuple then some other xact
2983 : * could update this tuple before we get to this point. Check
2984 : * for xmax change, and start over if so.
2985 : *
2986 : * We also must start over if we didn't pin the VM page, and
2987 : * the page has become all visible.
2988 : */
2989 16 : if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
2990 16 : xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
2991 8 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
2992 : xwait))
2993 0 : goto l1;
2994 : }
2995 :
2996 : /*
2997 : * You might think the multixact is necessarily done here, but not
2998 : * so: it could have surviving members, namely our own xact or
2999 : * other subxacts of this backend. It is legal for us to delete
3000 : * the tuple in either case, however (the latter case is
3001 : * essentially a situation of upgrading our former shared lock to
3002 : * exclusive). We don't bother changing the on-disk hint bits
3003 : * since we are about to overwrite the xmax altogether.
3004 : */
3005 : }
3006 40586 : else if (!TransactionIdIsCurrentTransactionId(xwait))
3007 : {
3008 : /*
3009 : * Wait for regular transaction to end; but first, acquire tuple
3010 : * lock.
3011 : */
3012 52 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3013 52 : heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
3014 : LockWaitBlock, &have_tuple_lock);
3015 52 : XactLockTableWait(xwait, relation, &(tp.t_self), XLTW_Delete);
3016 48 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3017 :
3018 : /*
3019 : * xwait is done, but if xwait had just locked the tuple then some
3020 : * other xact could update this tuple before we get to this point.
3021 : * Check for xmax change, and start over if so.
3022 : *
3023 : * We also must start over if we didn't pin the VM page, and the
3024 : * page has become all visible.
3025 : */
3026 96 : if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
3027 95 : xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
3028 47 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
3029 : xwait))
3030 1 : goto l1;
3031 :
3032 : /* Otherwise check if it committed or aborted */
3033 47 : UpdateXmaxHintBits(tp.t_data, buffer, xwait);
3034 : }
3035 :
3036 : /*
3037 : * We may overwrite if previous xmax aborted, or if it committed but
3038 : * only locked the tuple without updating it.
3039 : */
3040 81158 : if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
3041 40600 : HEAP_XMAX_IS_LOCKED_ONLY(tp.t_data->t_infomask) ||
3042 31 : HeapTupleHeaderIsOnlyLocked(tp.t_data))
3043 40562 : result = TM_Ok;
3044 27 : else if (!ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
3045 23 : result = TM_Updated;
3046 : else
3047 4 : result = TM_Deleted;
3048 : }
3049 :
3050 : /* sanity check the result HeapTupleSatisfiesUpdate() and the logic above */
3051 : if (result != TM_Ok)
3052 : {
3053 : Assert(result == TM_SelfModified ||
3054 : result == TM_Updated ||
3055 : result == TM_Deleted ||
3056 : result == TM_BeingModified);
3057 : Assert(!(tp.t_data->t_infomask & HEAP_XMAX_INVALID));
3058 : Assert(result != TM_Updated ||
3059 : !ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid));
3060 : }
3061 :
3062 1844683 : if (crosscheck != InvalidSnapshot && result == TM_Ok)
3063 : {
3064 : /* Perform additional check for transaction-snapshot mode RI updates */
3065 1 : if (!HeapTupleSatisfiesVisibility(&tp, crosscheck, buffer))
3066 1 : result = TM_Updated;
3067 : }
3068 :
3069 1844683 : if (result != TM_Ok)
3070 : {
3071 69 : tmfd->ctid = tp.t_data->t_ctid;
3072 69 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
3073 69 : if (result == TM_SelfModified)
3074 28 : tmfd->cmax = HeapTupleHeaderGetCmax(tp.t_data);
3075 : else
3076 41 : tmfd->cmax = InvalidCommandId;
3077 69 : UnlockReleaseBuffer(buffer);
3078 69 : if (have_tuple_lock)
3079 27 : UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
3080 69 : if (vmbuffer != InvalidBuffer)
3081 0 : ReleaseBuffer(vmbuffer);
3082 69 : return result;
3083 : }
3084 :
3085 : /*
3086 : * We're about to do the actual delete -- check for conflict first, to
3087 : * avoid possibly having to roll back work we've just done.
3088 : *
3089 : * This is safe without a recheck as long as there is no possibility of
3090 : * another process scanning the page between this check and the delete
3091 : * being visible to the scan (i.e., an exclusive buffer content lock is
3092 : * continuously held from this point until the tuple delete is visible).
3093 : */
3094 1844614 : CheckForSerializableConflictIn(relation, tid, BufferGetBlockNumber(buffer));
3095 :
3096 : /* replace cid with a combo CID if necessary */
3097 1844600 : HeapTupleHeaderAdjustCmax(tp.t_data, &cid, &iscombo);
3098 :
3099 : /*
3100 : * Compute replica identity tuple before entering the critical section so
3101 : * we don't PANIC upon a memory allocation failure.
3102 : */
3103 1844600 : old_key_tuple = ExtractReplicaIdentity(relation, &tp, true, &old_key_copied);
3104 :
3105 : /*
3106 : * If this is the first possibly-multixact-able operation in the current
3107 : * transaction, set my per-backend OldestMemberMXactId setting. We can be
3108 : * certain that the transaction will never become a member of any older
3109 : * MultiXactIds than that. (We have to do this even if we end up just
3110 : * using our own TransactionId below, since some other backend could
3111 : * incorporate our XID into a MultiXact immediately afterwards.)
3112 : */
3113 1844600 : MultiXactIdSetOldestMember();
3114 :
3115 1844600 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(tp.t_data),
3116 1844600 : tp.t_data->t_infomask, tp.t_data->t_infomask2,
3117 : xid, LockTupleExclusive, true,
3118 : &new_xmax, &new_infomask, &new_infomask2);
3119 :
3120 1844600 : START_CRIT_SECTION();
3121 :
3122 : /*
3123 : * If this transaction commits, the tuple will become DEAD sooner or
3124 : * later. Set flag that this page is a candidate for pruning once our xid
3125 : * falls below the OldestXmin horizon. If the transaction finally aborts,
3126 : * the subsequent page pruning will be a no-op and the hint will be
3127 : * cleared.
3128 : */
3129 1844600 : PageSetPrunable(page, xid);
3130 :
3131 1844600 : if (PageIsAllVisible(page))
3132 : {
3133 259 : all_visible_cleared = true;
3134 259 : PageClearAllVisible(page);
3135 259 : visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
3136 : vmbuffer, VISIBILITYMAP_VALID_BITS);
3137 : }
3138 :
3139 : /* store transaction information of xact deleting the tuple */
3140 1844600 : tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
3141 1844600 : tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
3142 1844600 : tp.t_data->t_infomask |= new_infomask;
3143 1844600 : tp.t_data->t_infomask2 |= new_infomask2;
3144 1844600 : HeapTupleHeaderClearHotUpdated(tp.t_data);
3145 1844600 : HeapTupleHeaderSetXmax(tp.t_data, new_xmax);
3146 1844600 : HeapTupleHeaderSetCmax(tp.t_data, cid, iscombo);
3147 : /* Make sure there is no forward chain link in t_ctid */
3148 1844600 : tp.t_data->t_ctid = tp.t_self;
3149 :
3150 : /* Signal that this is actually a move into another partition */
3151 1844600 : if (changingPart)
3152 638 : HeapTupleHeaderSetMovedPartitions(tp.t_data);
3153 :
3154 1844600 : MarkBufferDirty(buffer);
3155 :
3156 : /*
3157 : * XLOG stuff
3158 : *
3159 : * NB: heap_abort_speculative() uses the same xlog record and replay
3160 : * routines.
3161 : */
3162 1844600 : if (RelationNeedsWAL(relation))
3163 : {
3164 : xl_heap_delete xlrec;
3165 : xl_heap_header xlhdr;
3166 : XLogRecPtr recptr;
3167 :
3168 : /*
3169 : * For logical decode we need combo CIDs to properly decode the
3170 : * catalog
3171 : */
3172 1761340 : if (RelationIsAccessibleInLogicalDecoding(relation))
3173 6403 : log_heap_new_cid(relation, &tp);
3174 :
3175 1761340 : xlrec.flags = 0;
3176 1761340 : if (all_visible_cleared)
3177 259 : xlrec.flags |= XLH_DELETE_ALL_VISIBLE_CLEARED;
3178 1761340 : if (changingPart)
3179 638 : xlrec.flags |= XLH_DELETE_IS_PARTITION_MOVE;
3180 3522680 : xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
3181 1761340 : tp.t_data->t_infomask2);
3182 1761340 : xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
3183 1761340 : xlrec.xmax = new_xmax;
3184 :
3185 1761340 : if (old_key_tuple != NULL)
3186 : {
3187 47019 : if (relation->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
3188 132 : xlrec.flags |= XLH_DELETE_CONTAINS_OLD_TUPLE;
3189 : else
3190 46887 : xlrec.flags |= XLH_DELETE_CONTAINS_OLD_KEY;
3191 : }
3192 :
3193 1761340 : XLogBeginInsert();
3194 1761340 : XLogRegisterData(&xlrec, SizeOfHeapDelete);
3195 :
3196 1761340 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
3197 :
3198 : /*
3199 : * Log replica identity of the deleted tuple if there is one
3200 : */
3201 1761340 : if (old_key_tuple != NULL)
3202 : {
3203 47019 : xlhdr.t_infomask2 = old_key_tuple->t_data->t_infomask2;
3204 47019 : xlhdr.t_infomask = old_key_tuple->t_data->t_infomask;
3205 47019 : xlhdr.t_hoff = old_key_tuple->t_data->t_hoff;
3206 :
3207 47019 : XLogRegisterData(&xlhdr, SizeOfHeapHeader);
3208 47019 : XLogRegisterData((char *) old_key_tuple->t_data
3209 : + SizeofHeapTupleHeader,
3210 47019 : old_key_tuple->t_len
3211 : - SizeofHeapTupleHeader);
3212 : }
3213 :
3214 : /* filtering by origin on a row level is much more efficient */
3215 1761340 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
3216 :
3217 1761340 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
3218 :
3219 1761340 : PageSetLSN(page, recptr);
3220 : }
3221 :
3222 1844600 : END_CRIT_SECTION();
3223 :
3224 1844600 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3225 :
3226 1844600 : if (vmbuffer != InvalidBuffer)
3227 259 : ReleaseBuffer(vmbuffer);
3228 :
3229 : /*
3230 : * If the tuple has toasted out-of-line attributes, we need to delete
3231 : * those items too. We have to do this before releasing the buffer
3232 : * because we need to look at the contents of the tuple, but it's OK to
3233 : * release the content lock on the buffer first.
3234 : */
3235 1848329 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
3236 3742 : relation->rd_rel->relkind != RELKIND_MATVIEW)
3237 : {
3238 : /* toast table entries should never be recursively toasted */
3239 : Assert(!HeapTupleHasExternal(&tp));
3240 : }
3241 1840871 : else if (HeapTupleHasExternal(&tp))
3242 504 : heap_toast_delete(relation, &tp, false);
3243 :
3244 : /*
3245 : * Mark tuple for invalidation from system caches at next command
3246 : * boundary. We have to do this before releasing the buffer because we
3247 : * need to look at the contents of the tuple.
3248 : */
3249 1844600 : CacheInvalidateHeapTuple(relation, &tp, NULL);
3250 :
3251 : /* Now we can release the buffer */
3252 1844600 : ReleaseBuffer(buffer);
3253 :
3254 : /*
3255 : * Release the lmgr tuple lock, if we had it.
3256 : */
3257 1844600 : if (have_tuple_lock)
3258 26 : UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
3259 :
3260 1844600 : pgstat_count_heap_delete(relation);
3261 :
3262 1844600 : if (old_key_tuple != NULL && old_key_copied)
3263 46888 : heap_freetuple(old_key_tuple);
3264 :
3265 1844600 : return TM_Ok;
3266 : }
3267 :
3268 : /*
3269 : * simple_heap_delete - delete a tuple
3270 : *
3271 : * This routine may be used to delete a tuple when concurrent updates of
3272 : * the target tuple are not expected (for example, because we have a lock
3273 : * on the relation associated with the tuple). Any failure is reported
3274 : * via ereport().
3275 : */
3276 : void
3277 793293 : simple_heap_delete(Relation relation, const ItemPointerData *tid)
3278 : {
3279 : TM_Result result;
3280 : TM_FailureData tmfd;
3281 :
3282 793293 : result = heap_delete(relation, tid,
3283 : GetCurrentCommandId(true), InvalidSnapshot,
3284 : true /* wait for commit */ ,
3285 : &tmfd, false /* changingPart */ );
3286 793293 : switch (result)
3287 : {
3288 0 : case TM_SelfModified:
3289 : /* Tuple was already updated in current command? */
3290 0 : elog(ERROR, "tuple already updated by self");
3291 : break;
3292 :
3293 793293 : case TM_Ok:
3294 : /* done successfully */
3295 793293 : break;
3296 :
3297 0 : case TM_Updated:
3298 0 : elog(ERROR, "tuple concurrently updated");
3299 : break;
3300 :
3301 0 : case TM_Deleted:
3302 0 : elog(ERROR, "tuple concurrently deleted");
3303 : break;
3304 :
3305 0 : default:
3306 0 : elog(ERROR, "unrecognized heap_delete status: %u", result);
3307 : break;
3308 : }
3309 793293 : }
3310 :
3311 : /*
3312 : * heap_update - replace a tuple
3313 : *
3314 : * See table_tuple_update() for an explanation of the parameters, except that
3315 : * this routine directly takes a tuple rather than a slot.
3316 : *
3317 : * In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
3318 : * t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
3319 : * only for TM_SelfModified, since we cannot obtain cmax from a combo CID
3320 : * generated by another transaction).
3321 : */
3322 : TM_Result
3323 336595 : heap_update(Relation relation, const ItemPointerData *otid, HeapTuple newtup,
3324 : CommandId cid, Snapshot crosscheck, bool wait,
3325 : TM_FailureData *tmfd, LockTupleMode *lockmode,
3326 : TU_UpdateIndexes *update_indexes)
3327 : {
3328 : TM_Result result;
3329 336595 : TransactionId xid = GetCurrentTransactionId();
3330 : Bitmapset *hot_attrs;
3331 : Bitmapset *sum_attrs;
3332 : Bitmapset *key_attrs;
3333 : Bitmapset *id_attrs;
3334 : Bitmapset *interesting_attrs;
3335 : Bitmapset *modified_attrs;
3336 : ItemId lp;
3337 : HeapTupleData oldtup;
3338 : HeapTuple heaptup;
3339 336595 : HeapTuple old_key_tuple = NULL;
3340 336595 : bool old_key_copied = false;
3341 : Page page,
3342 : newpage;
3343 : BlockNumber block;
3344 : MultiXactStatus mxact_status;
3345 : Buffer buffer,
3346 : newbuf,
3347 336595 : vmbuffer = InvalidBuffer,
3348 336595 : vmbuffer_new = InvalidBuffer;
3349 : bool need_toast;
3350 : Size newtupsize,
3351 : pagefree;
3352 336595 : bool have_tuple_lock = false;
3353 : bool iscombo;
3354 336595 : bool use_hot_update = false;
3355 336595 : bool summarized_update = false;
3356 : bool key_intact;
3357 336595 : bool all_visible_cleared = false;
3358 336595 : bool all_visible_cleared_new = false;
3359 : bool checked_lockers;
3360 : bool locker_remains;
3361 336595 : bool id_has_external = false;
3362 : TransactionId xmax_new_tuple,
3363 : xmax_old_tuple;
3364 : uint16 infomask_old_tuple,
3365 : infomask2_old_tuple,
3366 : infomask_new_tuple,
3367 : infomask2_new_tuple;
3368 :
3369 : Assert(ItemPointerIsValid(otid));
3370 :
3371 : /* Cheap, simplistic check that the tuple matches the rel's rowtype. */
3372 : Assert(HeapTupleHeaderGetNatts(newtup->t_data) <=
3373 : RelationGetNumberOfAttributes(relation));
3374 :
3375 336595 : AssertHasSnapshotForToast(relation);
3376 :
3377 : /*
3378 : * Forbid this during a parallel operation, lest it allocate a combo CID.
3379 : * Other workers might need that combo CID for visibility checks, and we
3380 : * have no provision for broadcasting it to them.
3381 : */
3382 336595 : if (IsInParallelMode())
3383 0 : ereport(ERROR,
3384 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
3385 : errmsg("cannot update tuples during a parallel operation")));
3386 :
3387 : #ifdef USE_ASSERT_CHECKING
3388 : check_lock_if_inplace_updateable_rel(relation, otid, newtup);
3389 : #endif
3390 :
3391 : /*
3392 : * Fetch the list of attributes to be checked for various operations.
3393 : *
3394 : * For HOT considerations, this is wasted effort if we fail to update or
3395 : * have to put the new tuple on a different page. But we must compute the
3396 : * list before obtaining buffer lock --- in the worst case, if we are
3397 : * doing an update on one of the relevant system catalogs, we could
3398 : * deadlock if we try to fetch the list later. In any case, the relcache
3399 : * caches the data so this is usually pretty cheap.
3400 : *
3401 : * We also need columns used by the replica identity and columns that are
3402 : * considered the "key" of rows in the table.
3403 : *
3404 : * Note that we get copies of each bitmap, so we need not worry about
3405 : * relcache flush happening midway through.
3406 : */
3407 336595 : hot_attrs = RelationGetIndexAttrBitmap(relation,
3408 : INDEX_ATTR_BITMAP_HOT_BLOCKING);
3409 336595 : sum_attrs = RelationGetIndexAttrBitmap(relation,
3410 : INDEX_ATTR_BITMAP_SUMMARIZED);
3411 336595 : key_attrs = RelationGetIndexAttrBitmap(relation, INDEX_ATTR_BITMAP_KEY);
3412 336595 : id_attrs = RelationGetIndexAttrBitmap(relation,
3413 : INDEX_ATTR_BITMAP_IDENTITY_KEY);
3414 336595 : interesting_attrs = NULL;
3415 336595 : interesting_attrs = bms_add_members(interesting_attrs, hot_attrs);
3416 336595 : interesting_attrs = bms_add_members(interesting_attrs, sum_attrs);
3417 336595 : interesting_attrs = bms_add_members(interesting_attrs, key_attrs);
3418 336595 : interesting_attrs = bms_add_members(interesting_attrs, id_attrs);
3419 :
3420 336595 : block = ItemPointerGetBlockNumber(otid);
3421 336595 : INJECTION_POINT("heap_update-before-pin", NULL);
3422 336595 : buffer = ReadBuffer(relation, block);
3423 336595 : page = BufferGetPage(buffer);
3424 :
3425 : /*
3426 : * Before locking the buffer, pin the visibility map page if it appears to
3427 : * be necessary. Since we haven't got the lock yet, someone else might be
3428 : * in the middle of changing this, so we'll need to recheck after we have
3429 : * the lock.
3430 : */
3431 336595 : if (PageIsAllVisible(page))
3432 2116 : visibilitymap_pin(relation, block, &vmbuffer);
3433 :
3434 336595 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3435 :
3436 336595 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(otid));
3437 :
3438 : /*
3439 : * Usually, a buffer pin and/or snapshot blocks pruning of otid, ensuring
3440 : * we see LP_NORMAL here. When the otid origin is a syscache, we may have
3441 : * neither a pin nor a snapshot. Hence, we may see other LP_ states, each
3442 : * of which indicates concurrent pruning.
3443 : *
3444 : * Failing with TM_Updated would be most accurate. However, unlike other
3445 : * TM_Updated scenarios, we don't know the successor ctid in LP_UNUSED and
3446 : * LP_DEAD cases. While the distinction between TM_Updated and TM_Deleted
3447 : * does matter to SQL statements UPDATE and MERGE, those SQL statements
3448 : * hold a snapshot that ensures LP_NORMAL. Hence, the choice between
3449 : * TM_Updated and TM_Deleted affects only the wording of error messages.
3450 : * Settle on TM_Deleted, for two reasons. First, it avoids complicating
3451 : * the specification of when tmfd->ctid is valid. Second, it creates
3452 : * error log evidence that we took this branch.
3453 : *
3454 : * Since it's possible to see LP_UNUSED at otid, it's also possible to see
3455 : * LP_NORMAL for a tuple that replaced LP_UNUSED. If it's a tuple for an
3456 : * unrelated row, we'll fail with "duplicate key value violates unique".
3457 : * XXX if otid is the live, newer version of the newtup row, we'll discard
3458 : * changes originating in versions of this catalog row after the version
3459 : * the caller got from syscache. See syscache-update-pruned.spec.
3460 : */
3461 336595 : if (!ItemIdIsNormal(lp))
3462 : {
3463 : Assert(RelationSupportsSysCache(RelationGetRelid(relation)));
3464 :
3465 1 : UnlockReleaseBuffer(buffer);
3466 : Assert(!have_tuple_lock);
3467 1 : if (vmbuffer != InvalidBuffer)
3468 1 : ReleaseBuffer(vmbuffer);
3469 1 : tmfd->ctid = *otid;
3470 1 : tmfd->xmax = InvalidTransactionId;
3471 1 : tmfd->cmax = InvalidCommandId;
3472 1 : *update_indexes = TU_None;
3473 :
3474 1 : bms_free(hot_attrs);
3475 1 : bms_free(sum_attrs);
3476 1 : bms_free(key_attrs);
3477 1 : bms_free(id_attrs);
3478 : /* modified_attrs not yet initialized */
3479 1 : bms_free(interesting_attrs);
3480 1 : return TM_Deleted;
3481 : }
3482 :
3483 : /*
3484 : * Fill in enough data in oldtup for HeapDetermineColumnsInfo to work
3485 : * properly.
3486 : */
3487 336594 : oldtup.t_tableOid = RelationGetRelid(relation);
3488 336594 : oldtup.t_data = (HeapTupleHeader) PageGetItem(page, lp);
3489 336594 : oldtup.t_len = ItemIdGetLength(lp);
3490 336594 : oldtup.t_self = *otid;
3491 :
3492 : /* the new tuple is ready, except for this: */
3493 336594 : newtup->t_tableOid = RelationGetRelid(relation);
3494 :
3495 : /*
3496 : * Determine columns modified by the update. Additionally, identify
3497 : * whether any of the unmodified replica identity key attributes in the
3498 : * old tuple is externally stored or not. This is required because for
3499 : * such attributes the flattened value won't be WAL logged as part of the
3500 : * new tuple so we must include it as part of the old_key_tuple. See
3501 : * ExtractReplicaIdentity.
3502 : */
3503 336594 : modified_attrs = HeapDetermineColumnsInfo(relation, interesting_attrs,
3504 : id_attrs, &oldtup,
3505 : newtup, &id_has_external);
3506 :
3507 : /*
3508 : * If we're not updating any "key" column, we can grab a weaker lock type.
3509 : * This allows for more concurrency when we are running simultaneously
3510 : * with foreign key checks.
3511 : *
3512 : * Note that if a column gets detoasted while executing the update, but
3513 : * the value ends up being the same, this test will fail and we will use
3514 : * the stronger lock. This is acceptable; the important case to optimize
3515 : * is updates that don't manipulate key columns, not those that
3516 : * serendipitously arrive at the same key values.
3517 : */
3518 336594 : if (!bms_overlap(modified_attrs, key_attrs))
3519 : {
3520 331037 : *lockmode = LockTupleNoKeyExclusive;
3521 331037 : mxact_status = MultiXactStatusNoKeyUpdate;
3522 331037 : key_intact = true;
3523 :
3524 : /*
3525 : * If this is the first possibly-multixact-able operation in the
3526 : * current transaction, set my per-backend OldestMemberMXactId
3527 : * setting. We can be certain that the transaction will never become a
3528 : * member of any older MultiXactIds than that. (We have to do this
3529 : * even if we end up just using our own TransactionId below, since
3530 : * some other backend could incorporate our XID into a MultiXact
3531 : * immediately afterwards.)
3532 : */
3533 331037 : MultiXactIdSetOldestMember();
3534 : }
3535 : else
3536 : {
3537 5557 : *lockmode = LockTupleExclusive;
3538 5557 : mxact_status = MultiXactStatusUpdate;
3539 5557 : key_intact = false;
3540 : }
3541 :
3542 : /*
3543 : * Note: beyond this point, use oldtup not otid to refer to old tuple.
3544 : * otid may very well point at newtup->t_self, which we will overwrite
3545 : * with the new tuple's location, so there's great risk of confusion if we
3546 : * use otid anymore.
3547 : */
3548 :
3549 1 : l2:
3550 336595 : checked_lockers = false;
3551 336595 : locker_remains = false;
3552 336595 : result = HeapTupleSatisfiesUpdate(&oldtup, cid, buffer);
3553 :
3554 : /* see below about the "no wait" case */
3555 : Assert(result != TM_BeingModified || wait);
3556 :
3557 336595 : if (result == TM_Invisible)
3558 : {
3559 0 : UnlockReleaseBuffer(buffer);
3560 0 : ereport(ERROR,
3561 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3562 : errmsg("attempted to update invisible tuple")));
3563 : }
3564 336595 : else if (result == TM_BeingModified && wait)
3565 : {
3566 : TransactionId xwait;
3567 : uint16 infomask;
3568 36443 : bool can_continue = false;
3569 :
3570 : /*
3571 : * XXX note that we don't consider the "no wait" case here. This
3572 : * isn't a problem currently because no caller uses that case, but it
3573 : * should be fixed if such a caller is introduced. It wasn't a
3574 : * problem previously because this code would always wait, but now
3575 : * that some tuple locks do not conflict with one of the lock modes we
3576 : * use, it is possible that this case is interesting to handle
3577 : * specially.
3578 : *
3579 : * This may cause failures with third-party code that calls
3580 : * heap_update directly.
3581 : */
3582 :
3583 : /* must copy state data before unlocking buffer */
3584 36443 : xwait = HeapTupleHeaderGetRawXmax(oldtup.t_data);
3585 36443 : infomask = oldtup.t_data->t_infomask;
3586 :
3587 : /*
3588 : * Now we have to do something about the existing locker. If it's a
3589 : * multi, sleep on it; we might be awakened before it is completely
3590 : * gone (or even not sleep at all in some cases); we need to preserve
3591 : * it as locker, unless it is gone completely.
3592 : *
3593 : * If it's not a multi, we need to check for sleeping conditions
3594 : * before actually going to sleep. If the update doesn't conflict
3595 : * with the locks, we just continue without sleeping (but making sure
3596 : * it is preserved).
3597 : *
3598 : * Before sleeping, we need to acquire tuple lock to establish our
3599 : * priority for the tuple (see heap_lock_tuple). LockTuple will
3600 : * release us when we are next-in-line for the tuple. Note we must
3601 : * not acquire the tuple lock until we're sure we're going to sleep;
3602 : * otherwise we're open for race conditions with other transactions
3603 : * holding the tuple lock which sleep on us.
3604 : *
3605 : * If we are forced to "start over" below, we keep the tuple lock;
3606 : * this arranges that we stay at the head of the line while rechecking
3607 : * tuple state.
3608 : */
3609 36443 : if (infomask & HEAP_XMAX_IS_MULTI)
3610 : {
3611 : TransactionId update_xact;
3612 : int remain;
3613 179 : bool current_is_member = false;
3614 :
3615 179 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
3616 : *lockmode, ¤t_is_member))
3617 : {
3618 8 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3619 :
3620 : /*
3621 : * Acquire the lock, if necessary (but skip it when we're
3622 : * requesting a lock and already have one; avoids deadlock).
3623 : */
3624 8 : if (!current_is_member)
3625 0 : heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3626 : LockWaitBlock, &have_tuple_lock);
3627 :
3628 : /* wait for multixact */
3629 8 : MultiXactIdWait((MultiXactId) xwait, mxact_status, infomask,
3630 : relation, &oldtup.t_self, XLTW_Update,
3631 : &remain);
3632 8 : checked_lockers = true;
3633 8 : locker_remains = remain != 0;
3634 8 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3635 :
3636 : /*
3637 : * If xwait had just locked the tuple then some other xact
3638 : * could update this tuple before we get to this point. Check
3639 : * for xmax change, and start over if so.
3640 : */
3641 8 : if (xmax_infomask_changed(oldtup.t_data->t_infomask,
3642 8 : infomask) ||
3643 8 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3644 : xwait))
3645 0 : goto l2;
3646 : }
3647 :
3648 : /*
3649 : * Note that the multixact may not be done by now. It could have
3650 : * surviving members; our own xact or other subxacts of this
3651 : * backend, and also any other concurrent transaction that locked
3652 : * the tuple with LockTupleKeyShare if we only got
3653 : * LockTupleNoKeyExclusive. If this is the case, we have to be
3654 : * careful to mark the updated tuple with the surviving members in
3655 : * Xmax.
3656 : *
3657 : * Note that there could have been another update in the
3658 : * MultiXact. In that case, we need to check whether it committed
3659 : * or aborted. If it aborted we are safe to update it again;
3660 : * otherwise there is an update conflict, and we have to return
3661 : * TableTuple{Deleted, Updated} below.
3662 : *
3663 : * In the LockTupleExclusive case, we still need to preserve the
3664 : * surviving members: those would include the tuple locks we had
3665 : * before this one, which are important to keep in case this
3666 : * subxact aborts.
3667 : */
3668 179 : if (!HEAP_XMAX_IS_LOCKED_ONLY(oldtup.t_data->t_infomask))
3669 8 : update_xact = HeapTupleGetUpdateXid(oldtup.t_data);
3670 : else
3671 171 : update_xact = InvalidTransactionId;
3672 :
3673 : /*
3674 : * There was no UPDATE in the MultiXact; or it aborted. No
3675 : * TransactionIdIsInProgress() call needed here, since we called
3676 : * MultiXactIdWait() above.
3677 : */
3678 187 : if (!TransactionIdIsValid(update_xact) ||
3679 8 : TransactionIdDidAbort(update_xact))
3680 172 : can_continue = true;
3681 : }
3682 36264 : else if (TransactionIdIsCurrentTransactionId(xwait))
3683 : {
3684 : /*
3685 : * The only locker is ourselves; we can avoid grabbing the tuple
3686 : * lock here, but must preserve our locking information.
3687 : */
3688 36153 : checked_lockers = true;
3689 36153 : locker_remains = true;
3690 36153 : can_continue = true;
3691 : }
3692 111 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) && key_intact)
3693 : {
3694 : /*
3695 : * If it's just a key-share locker, and we're not changing the key
3696 : * columns, we don't need to wait for it to end; but we need to
3697 : * preserve it as locker.
3698 : */
3699 29 : checked_lockers = true;
3700 29 : locker_remains = true;
3701 29 : can_continue = true;
3702 : }
3703 : else
3704 : {
3705 : /*
3706 : * Wait for regular transaction to end; but first, acquire tuple
3707 : * lock.
3708 : */
3709 82 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3710 82 : heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3711 : LockWaitBlock, &have_tuple_lock);
3712 82 : XactLockTableWait(xwait, relation, &oldtup.t_self,
3713 : XLTW_Update);
3714 82 : checked_lockers = true;
3715 82 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3716 :
3717 : /*
3718 : * xwait is done, but if xwait had just locked the tuple then some
3719 : * other xact could update this tuple before we get to this point.
3720 : * Check for xmax change, and start over if so.
3721 : */
3722 163 : if (xmax_infomask_changed(oldtup.t_data->t_infomask, infomask) ||
3723 81 : !TransactionIdEquals(xwait,
3724 : HeapTupleHeaderGetRawXmax(oldtup.t_data)))
3725 1 : goto l2;
3726 :
3727 : /* Otherwise check if it committed or aborted */
3728 81 : UpdateXmaxHintBits(oldtup.t_data, buffer, xwait);
3729 81 : if (oldtup.t_data->t_infomask & HEAP_XMAX_INVALID)
3730 22 : can_continue = true;
3731 : }
3732 :
3733 36442 : if (can_continue)
3734 36376 : result = TM_Ok;
3735 66 : else if (!ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid))
3736 61 : result = TM_Updated;
3737 : else
3738 5 : result = TM_Deleted;
3739 : }
3740 :
3741 : /* Sanity check the result HeapTupleSatisfiesUpdate() and the logic above */
3742 : if (result != TM_Ok)
3743 : {
3744 : Assert(result == TM_SelfModified ||
3745 : result == TM_Updated ||
3746 : result == TM_Deleted ||
3747 : result == TM_BeingModified);
3748 : Assert(!(oldtup.t_data->t_infomask & HEAP_XMAX_INVALID));
3749 : Assert(result != TM_Updated ||
3750 : !ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid));
3751 : }
3752 :
3753 336594 : if (crosscheck != InvalidSnapshot && result == TM_Ok)
3754 : {
3755 : /* Perform additional check for transaction-snapshot mode RI updates */
3756 1 : if (!HeapTupleSatisfiesVisibility(&oldtup, crosscheck, buffer))
3757 1 : result = TM_Updated;
3758 : }
3759 :
3760 336594 : if (result != TM_Ok)
3761 : {
3762 181 : tmfd->ctid = oldtup.t_data->t_ctid;
3763 181 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(oldtup.t_data);
3764 181 : if (result == TM_SelfModified)
3765 69 : tmfd->cmax = HeapTupleHeaderGetCmax(oldtup.t_data);
3766 : else
3767 112 : tmfd->cmax = InvalidCommandId;
3768 181 : UnlockReleaseBuffer(buffer);
3769 181 : if (have_tuple_lock)
3770 59 : UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
3771 181 : if (vmbuffer != InvalidBuffer)
3772 0 : ReleaseBuffer(vmbuffer);
3773 181 : *update_indexes = TU_None;
3774 :
3775 181 : bms_free(hot_attrs);
3776 181 : bms_free(sum_attrs);
3777 181 : bms_free(key_attrs);
3778 181 : bms_free(id_attrs);
3779 181 : bms_free(modified_attrs);
3780 181 : bms_free(interesting_attrs);
3781 181 : return result;
3782 : }
3783 :
3784 : /*
3785 : * If we didn't pin the visibility map page and the page has become all
3786 : * visible while we were busy locking the buffer, or during some
3787 : * subsequent window during which we had it unlocked, we'll have to unlock
3788 : * and re-lock, to avoid holding the buffer lock across an I/O. That's a
3789 : * bit unfortunate, especially since we'll now have to recheck whether the
3790 : * tuple has been locked or updated under us, but hopefully it won't
3791 : * happen very often.
3792 : */
3793 336413 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
3794 : {
3795 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3796 0 : visibilitymap_pin(relation, block, &vmbuffer);
3797 0 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3798 0 : goto l2;
3799 : }
3800 :
3801 : /* Fill in transaction status data */
3802 :
3803 : /*
3804 : * If the tuple we're updating is locked, we need to preserve the locking
3805 : * info in the old tuple's Xmax. Prepare a new Xmax value for this.
3806 : */
3807 336413 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3808 336413 : oldtup.t_data->t_infomask,
3809 336413 : oldtup.t_data->t_infomask2,
3810 : xid, *lockmode, true,
3811 : &xmax_old_tuple, &infomask_old_tuple,
3812 : &infomask2_old_tuple);
3813 :
3814 : /*
3815 : * And also prepare an Xmax value for the new copy of the tuple. If there
3816 : * was no xmax previously, or there was one but all lockers are now gone,
3817 : * then use InvalidTransactionId; otherwise, get the xmax from the old
3818 : * tuple. (In rare cases that might also be InvalidTransactionId and yet
3819 : * not have the HEAP_XMAX_INVALID bit set; that's fine.)
3820 : */
3821 372767 : if ((oldtup.t_data->t_infomask & HEAP_XMAX_INVALID) ||
3822 72708 : HEAP_LOCKED_UPGRADED(oldtup.t_data->t_infomask) ||
3823 36183 : (checked_lockers && !locker_remains))
3824 300059 : xmax_new_tuple = InvalidTransactionId;
3825 : else
3826 36354 : xmax_new_tuple = HeapTupleHeaderGetRawXmax(oldtup.t_data);
3827 :
3828 336413 : if (!TransactionIdIsValid(xmax_new_tuple))
3829 : {
3830 300059 : infomask_new_tuple = HEAP_XMAX_INVALID;
3831 300059 : infomask2_new_tuple = 0;
3832 : }
3833 : else
3834 : {
3835 : /*
3836 : * If we found a valid Xmax for the new tuple, then the infomask bits
3837 : * to use on the new tuple depend on what was there on the old one.
3838 : * Note that since we're doing an update, the only possibility is that
3839 : * the lockers had FOR KEY SHARE lock.
3840 : */
3841 36354 : if (oldtup.t_data->t_infomask & HEAP_XMAX_IS_MULTI)
3842 : {
3843 172 : GetMultiXactIdHintBits(xmax_new_tuple, &infomask_new_tuple,
3844 : &infomask2_new_tuple);
3845 : }
3846 : else
3847 : {
3848 36182 : infomask_new_tuple = HEAP_XMAX_KEYSHR_LOCK | HEAP_XMAX_LOCK_ONLY;
3849 36182 : infomask2_new_tuple = 0;
3850 : }
3851 : }
3852 :
3853 : /*
3854 : * Prepare the new tuple with the appropriate initial values of Xmin and
3855 : * Xmax, as well as initial infomask bits as computed above.
3856 : */
3857 336413 : newtup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
3858 336413 : newtup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
3859 336413 : HeapTupleHeaderSetXmin(newtup->t_data, xid);
3860 336413 : HeapTupleHeaderSetCmin(newtup->t_data, cid);
3861 336413 : newtup->t_data->t_infomask |= HEAP_UPDATED | infomask_new_tuple;
3862 336413 : newtup->t_data->t_infomask2 |= infomask2_new_tuple;
3863 336413 : HeapTupleHeaderSetXmax(newtup->t_data, xmax_new_tuple);
3864 :
3865 : /*
3866 : * Replace cid with a combo CID if necessary. Note that we already put
3867 : * the plain cid into the new tuple.
3868 : */
3869 336413 : HeapTupleHeaderAdjustCmax(oldtup.t_data, &cid, &iscombo);
3870 :
3871 : /*
3872 : * If the toaster needs to be activated, OR if the new tuple will not fit
3873 : * on the same page as the old, then we need to release the content lock
3874 : * (but not the pin!) on the old tuple's buffer while we are off doing
3875 : * TOAST and/or table-file-extension work. We must mark the old tuple to
3876 : * show that it's locked, else other processes may try to update it
3877 : * themselves.
3878 : *
3879 : * We need to invoke the toaster if there are already any out-of-line
3880 : * toasted values present, or if the new tuple is over-threshold.
3881 : */
3882 336413 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
3883 0 : relation->rd_rel->relkind != RELKIND_MATVIEW)
3884 : {
3885 : /* toast table entries should never be recursively toasted */
3886 : Assert(!HeapTupleHasExternal(&oldtup));
3887 : Assert(!HeapTupleHasExternal(newtup));
3888 0 : need_toast = false;
3889 : }
3890 : else
3891 1008736 : need_toast = (HeapTupleHasExternal(&oldtup) ||
3892 672323 : HeapTupleHasExternal(newtup) ||
3893 335878 : newtup->t_len > TOAST_TUPLE_THRESHOLD);
3894 :
3895 336413 : pagefree = PageGetHeapFreeSpace(page);
3896 :
3897 336413 : newtupsize = MAXALIGN(newtup->t_len);
3898 :
3899 336413 : if (need_toast || newtupsize > pagefree)
3900 157320 : {
3901 : TransactionId xmax_lock_old_tuple;
3902 : uint16 infomask_lock_old_tuple,
3903 : infomask2_lock_old_tuple;
3904 157320 : bool cleared_all_frozen = false;
3905 :
3906 : /*
3907 : * To prevent concurrent sessions from updating the tuple, we have to
3908 : * temporarily mark it locked, while we release the page-level lock.
3909 : *
3910 : * To satisfy the rule that any xid potentially appearing in a buffer
3911 : * written out to disk, we unfortunately have to WAL log this
3912 : * temporary modification. We can reuse xl_heap_lock for this
3913 : * purpose. If we crash/error before following through with the
3914 : * actual update, xmax will be of an aborted transaction, allowing
3915 : * other sessions to proceed.
3916 : */
3917 :
3918 : /*
3919 : * Compute xmax / infomask appropriate for locking the tuple. This has
3920 : * to be done separately from the combo that's going to be used for
3921 : * updating, because the potentially created multixact would otherwise
3922 : * be wrong.
3923 : */
3924 157320 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3925 157320 : oldtup.t_data->t_infomask,
3926 157320 : oldtup.t_data->t_infomask2,
3927 : xid, *lockmode, false,
3928 : &xmax_lock_old_tuple, &infomask_lock_old_tuple,
3929 : &infomask2_lock_old_tuple);
3930 :
3931 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(infomask_lock_old_tuple));
3932 :
3933 157320 : START_CRIT_SECTION();
3934 :
3935 : /* Clear obsolete visibility flags ... */
3936 157320 : oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
3937 157320 : oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
3938 157320 : HeapTupleClearHotUpdated(&oldtup);
3939 : /* ... and store info about transaction updating this tuple */
3940 : Assert(TransactionIdIsValid(xmax_lock_old_tuple));
3941 157320 : HeapTupleHeaderSetXmax(oldtup.t_data, xmax_lock_old_tuple);
3942 157320 : oldtup.t_data->t_infomask |= infomask_lock_old_tuple;
3943 157320 : oldtup.t_data->t_infomask2 |= infomask2_lock_old_tuple;
3944 157320 : HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
3945 :
3946 : /* temporarily make it look not-updated, but locked */
3947 157320 : oldtup.t_data->t_ctid = oldtup.t_self;
3948 :
3949 : /*
3950 : * Clear all-frozen bit on visibility map if needed. We could
3951 : * immediately reset ALL_VISIBLE, but given that the WAL logging
3952 : * overhead would be unchanged, that doesn't seem necessarily
3953 : * worthwhile.
3954 : */
3955 158442 : if (PageIsAllVisible(page) &&
3956 1122 : visibilitymap_clear(relation, block, vmbuffer,
3957 : VISIBILITYMAP_ALL_FROZEN))
3958 883 : cleared_all_frozen = true;
3959 :
3960 157320 : MarkBufferDirty(buffer);
3961 :
3962 157320 : if (RelationNeedsWAL(relation))
3963 : {
3964 : xl_heap_lock xlrec;
3965 : XLogRecPtr recptr;
3966 :
3967 147181 : XLogBeginInsert();
3968 147181 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
3969 :
3970 147181 : xlrec.offnum = ItemPointerGetOffsetNumber(&oldtup.t_self);
3971 147181 : xlrec.xmax = xmax_lock_old_tuple;
3972 294362 : xlrec.infobits_set = compute_infobits(oldtup.t_data->t_infomask,
3973 147181 : oldtup.t_data->t_infomask2);
3974 147181 : xlrec.flags =
3975 147181 : cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
3976 147181 : XLogRegisterData(&xlrec, SizeOfHeapLock);
3977 147181 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
3978 147181 : PageSetLSN(page, recptr);
3979 : }
3980 :
3981 157320 : END_CRIT_SECTION();
3982 :
3983 157320 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3984 :
3985 : /*
3986 : * Let the toaster do its thing, if needed.
3987 : *
3988 : * Note: below this point, heaptup is the data we actually intend to
3989 : * store into the relation; newtup is the caller's original untoasted
3990 : * data.
3991 : */
3992 157320 : if (need_toast)
3993 : {
3994 : /* Note we always use WAL and FSM during updates */
3995 1834 : heaptup = heap_toast_insert_or_update(relation, newtup, &oldtup, 0);
3996 1834 : newtupsize = MAXALIGN(heaptup->t_len);
3997 : }
3998 : else
3999 155486 : heaptup = newtup;
4000 :
4001 : /*
4002 : * Now, do we need a new page for the tuple, or not? This is a bit
4003 : * tricky since someone else could have added tuples to the page while
4004 : * we weren't looking. We have to recheck the available space after
4005 : * reacquiring the buffer lock. But don't bother to do that if the
4006 : * former amount of free space is still not enough; it's unlikely
4007 : * there's more free now than before.
4008 : *
4009 : * What's more, if we need to get a new page, we will need to acquire
4010 : * buffer locks on both old and new pages. To avoid deadlock against
4011 : * some other backend trying to get the same two locks in the other
4012 : * order, we must be consistent about the order we get the locks in.
4013 : * We use the rule "lock the lower-numbered page of the relation
4014 : * first". To implement this, we must do RelationGetBufferForTuple
4015 : * while not holding the lock on the old page, and we must rely on it
4016 : * to get the locks on both pages in the correct order.
4017 : *
4018 : * Another consideration is that we need visibility map page pin(s) if
4019 : * we will have to clear the all-visible flag on either page. If we
4020 : * call RelationGetBufferForTuple, we rely on it to acquire any such
4021 : * pins; but if we don't, we have to handle that here. Hence we need
4022 : * a loop.
4023 : */
4024 : for (;;)
4025 : {
4026 157320 : if (newtupsize > pagefree)
4027 : {
4028 : /* It doesn't fit, must use RelationGetBufferForTuple. */
4029 156643 : newbuf = RelationGetBufferForTuple(relation, heaptup->t_len,
4030 : buffer, 0, NULL,
4031 : &vmbuffer_new, &vmbuffer,
4032 : 0);
4033 : /* We're all done. */
4034 156643 : break;
4035 : }
4036 : /* Acquire VM page pin if needed and we don't have it. */
4037 677 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
4038 0 : visibilitymap_pin(relation, block, &vmbuffer);
4039 : /* Re-acquire the lock on the old tuple's page. */
4040 677 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
4041 : /* Re-check using the up-to-date free space */
4042 677 : pagefree = PageGetHeapFreeSpace(page);
4043 677 : if (newtupsize > pagefree ||
4044 677 : (vmbuffer == InvalidBuffer && PageIsAllVisible(page)))
4045 : {
4046 : /*
4047 : * Rats, it doesn't fit anymore, or somebody just now set the
4048 : * all-visible flag. We must now unlock and loop to avoid
4049 : * deadlock. Fortunately, this path should seldom be taken.
4050 : */
4051 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
4052 : }
4053 : else
4054 : {
4055 : /* We're all done. */
4056 677 : newbuf = buffer;
4057 677 : break;
4058 : }
4059 : }
4060 : }
4061 : else
4062 : {
4063 : /* No TOAST work needed, and it'll fit on same page */
4064 179093 : newbuf = buffer;
4065 179093 : heaptup = newtup;
4066 : }
4067 :
4068 336413 : newpage = BufferGetPage(newbuf);
4069 :
4070 : /*
4071 : * We're about to do the actual update -- check for conflict first, to
4072 : * avoid possibly having to roll back work we've just done.
4073 : *
4074 : * This is safe without a recheck as long as there is no possibility of
4075 : * another process scanning the pages between this check and the update
4076 : * being visible to the scan (i.e., exclusive buffer content lock(s) are
4077 : * continuously held from this point until the tuple update is visible).
4078 : *
4079 : * For the new tuple the only check needed is at the relation level, but
4080 : * since both tuples are in the same relation and the check for oldtup
4081 : * will include checking the relation level, there is no benefit to a
4082 : * separate check for the new tuple.
4083 : */
4084 336413 : CheckForSerializableConflictIn(relation, &oldtup.t_self,
4085 : BufferGetBlockNumber(buffer));
4086 :
4087 : /*
4088 : * At this point newbuf and buffer are both pinned and locked, and newbuf
4089 : * has enough space for the new tuple. If they are the same buffer, only
4090 : * one pin is held.
4091 : */
4092 :
4093 336401 : if (newbuf == buffer)
4094 : {
4095 : /*
4096 : * Since the new tuple is going into the same page, we might be able
4097 : * to do a HOT update. Check if any of the index columns have been
4098 : * changed.
4099 : */
4100 179758 : if (!bms_overlap(modified_attrs, hot_attrs))
4101 : {
4102 163843 : use_hot_update = true;
4103 :
4104 : /*
4105 : * If none of the columns that are used in hot-blocking indexes
4106 : * were updated, we can apply HOT, but we do still need to check
4107 : * if we need to update the summarizing indexes, and update those
4108 : * indexes if the columns were updated, or we may fail to detect
4109 : * e.g. value bound changes in BRIN minmax indexes.
4110 : */
4111 163843 : if (bms_overlap(modified_attrs, sum_attrs))
4112 2188 : summarized_update = true;
4113 : }
4114 : }
4115 : else
4116 : {
4117 : /* Set a hint that the old page could use prune/defrag */
4118 156643 : PageSetFull(page);
4119 : }
4120 :
4121 : /*
4122 : * Compute replica identity tuple before entering the critical section so
4123 : * we don't PANIC upon a memory allocation failure.
4124 : * ExtractReplicaIdentity() will return NULL if nothing needs to be
4125 : * logged. Pass old key required as true only if the replica identity key
4126 : * columns are modified or it has external data.
4127 : */
4128 336401 : old_key_tuple = ExtractReplicaIdentity(relation, &oldtup,
4129 336401 : bms_overlap(modified_attrs, id_attrs) ||
4130 : id_has_external,
4131 336401 : &old_key_copied);
4132 :
4133 : /* NO EREPORT(ERROR) from here till changes are logged */
4134 336401 : START_CRIT_SECTION();
4135 :
4136 : /*
4137 : * If this transaction commits, the old tuple will become DEAD sooner or
4138 : * later. Set flag that this page is a candidate for pruning once our xid
4139 : * falls below the OldestXmin horizon. If the transaction finally aborts,
4140 : * the subsequent page pruning will be a no-op and the hint will be
4141 : * cleared.
4142 : *
4143 : * XXX Should we set hint on newbuf as well? If the transaction aborts,
4144 : * there would be a prunable tuple in the newbuf; but for now we choose
4145 : * not to optimize for aborts. Note that heap_xlog_update must be kept in
4146 : * sync if this decision changes.
4147 : */
4148 336401 : PageSetPrunable(page, xid);
4149 :
4150 336401 : if (use_hot_update)
4151 : {
4152 : /* Mark the old tuple as HOT-updated */
4153 163843 : HeapTupleSetHotUpdated(&oldtup);
4154 : /* And mark the new tuple as heap-only */
4155 163843 : HeapTupleSetHeapOnly(heaptup);
4156 : /* Mark the caller's copy too, in case different from heaptup */
4157 163843 : HeapTupleSetHeapOnly(newtup);
4158 : }
4159 : else
4160 : {
4161 : /* Make sure tuples are correctly marked as not-HOT */
4162 172558 : HeapTupleClearHotUpdated(&oldtup);
4163 172558 : HeapTupleClearHeapOnly(heaptup);
4164 172558 : HeapTupleClearHeapOnly(newtup);
4165 : }
4166 :
4167 336401 : RelationPutHeapTuple(relation, newbuf, heaptup, false); /* insert new tuple */
4168 :
4169 :
4170 : /* Clear obsolete visibility flags, possibly set by ourselves above... */
4171 336401 : oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
4172 336401 : oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
4173 : /* ... and store info about transaction updating this tuple */
4174 : Assert(TransactionIdIsValid(xmax_old_tuple));
4175 336401 : HeapTupleHeaderSetXmax(oldtup.t_data, xmax_old_tuple);
4176 336401 : oldtup.t_data->t_infomask |= infomask_old_tuple;
4177 336401 : oldtup.t_data->t_infomask2 |= infomask2_old_tuple;
4178 336401 : HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
4179 :
4180 : /* record address of new tuple in t_ctid of old one */
4181 336401 : oldtup.t_data->t_ctid = heaptup->t_self;
4182 :
4183 : /* clear PD_ALL_VISIBLE flags, reset all visibilitymap bits */
4184 336401 : if (PageIsAllVisible(page))
4185 : {
4186 2115 : all_visible_cleared = true;
4187 2115 : PageClearAllVisible(page);
4188 2115 : visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
4189 : vmbuffer, VISIBILITYMAP_VALID_BITS);
4190 : }
4191 336401 : if (newbuf != buffer && PageIsAllVisible(newpage))
4192 : {
4193 787 : all_visible_cleared_new = true;
4194 787 : PageClearAllVisible(newpage);
4195 787 : visibilitymap_clear(relation, BufferGetBlockNumber(newbuf),
4196 : vmbuffer_new, VISIBILITYMAP_VALID_BITS);
4197 : }
4198 :
4199 336401 : if (newbuf != buffer)
4200 156643 : MarkBufferDirty(newbuf);
4201 336401 : MarkBufferDirty(buffer);
4202 :
4203 : /* XLOG stuff */
4204 336401 : if (RelationNeedsWAL(relation))
4205 : {
4206 : XLogRecPtr recptr;
4207 :
4208 : /*
4209 : * For logical decoding we need combo CIDs to properly decode the
4210 : * catalog.
4211 : */
4212 324625 : if (RelationIsAccessibleInLogicalDecoding(relation))
4213 : {
4214 2623 : log_heap_new_cid(relation, &oldtup);
4215 2623 : log_heap_new_cid(relation, heaptup);
4216 : }
4217 :
4218 324625 : recptr = log_heap_update(relation, buffer,
4219 : newbuf, &oldtup, heaptup,
4220 : old_key_tuple,
4221 : all_visible_cleared,
4222 : all_visible_cleared_new);
4223 324625 : if (newbuf != buffer)
4224 : {
4225 146512 : PageSetLSN(newpage, recptr);
4226 : }
4227 324625 : PageSetLSN(page, recptr);
4228 : }
4229 :
4230 336401 : END_CRIT_SECTION();
4231 :
4232 336401 : if (newbuf != buffer)
4233 156643 : LockBuffer(newbuf, BUFFER_LOCK_UNLOCK);
4234 336401 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
4235 :
4236 : /*
4237 : * Mark old tuple for invalidation from system caches at next command
4238 : * boundary, and mark the new tuple for invalidation in case we abort. We
4239 : * have to do this before releasing the buffer because oldtup is in the
4240 : * buffer. (heaptup is all in local memory, but it's necessary to process
4241 : * both tuple versions in one call to inval.c so we can avoid redundant
4242 : * sinval messages.)
4243 : */
4244 336401 : CacheInvalidateHeapTuple(relation, &oldtup, heaptup);
4245 :
4246 : /* Now we can release the buffer(s) */
4247 336401 : if (newbuf != buffer)
4248 156643 : ReleaseBuffer(newbuf);
4249 336401 : ReleaseBuffer(buffer);
4250 336401 : if (BufferIsValid(vmbuffer_new))
4251 787 : ReleaseBuffer(vmbuffer_new);
4252 336401 : if (BufferIsValid(vmbuffer))
4253 2115 : ReleaseBuffer(vmbuffer);
4254 :
4255 : /*
4256 : * Release the lmgr tuple lock, if we had it.
4257 : */
4258 336401 : if (have_tuple_lock)
4259 22 : UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
4260 :
4261 336401 : pgstat_count_heap_update(relation, use_hot_update, newbuf != buffer);
4262 :
4263 : /*
4264 : * If heaptup is a private copy, release it. Don't forget to copy t_self
4265 : * back to the caller's image, too.
4266 : */
4267 336401 : if (heaptup != newtup)
4268 : {
4269 1780 : newtup->t_self = heaptup->t_self;
4270 1780 : heap_freetuple(heaptup);
4271 : }
4272 :
4273 : /*
4274 : * If it is a HOT update, the update may still need to update summarized
4275 : * indexes, lest we fail to update those summaries and get incorrect
4276 : * results (for example, minmax bounds of the block may change with this
4277 : * update).
4278 : */
4279 336401 : if (use_hot_update)
4280 : {
4281 163843 : if (summarized_update)
4282 2188 : *update_indexes = TU_Summarizing;
4283 : else
4284 161655 : *update_indexes = TU_None;
4285 : }
4286 : else
4287 172558 : *update_indexes = TU_All;
4288 :
4289 336401 : if (old_key_tuple != NULL && old_key_copied)
4290 85 : heap_freetuple(old_key_tuple);
4291 :
4292 336401 : bms_free(hot_attrs);
4293 336401 : bms_free(sum_attrs);
4294 336401 : bms_free(key_attrs);
4295 336401 : bms_free(id_attrs);
4296 336401 : bms_free(modified_attrs);
4297 336401 : bms_free(interesting_attrs);
4298 :
4299 336401 : return TM_Ok;
4300 : }
4301 :
4302 : #ifdef USE_ASSERT_CHECKING
4303 : /*
4304 : * Confirm adequate lock held during heap_update(), per rules from
4305 : * README.tuplock section "Locking to write inplace-updated tables".
4306 : */
4307 : static void
4308 : check_lock_if_inplace_updateable_rel(Relation relation,
4309 : const ItemPointerData *otid,
4310 : HeapTuple newtup)
4311 : {
4312 : /* LOCKTAG_TUPLE acceptable for any catalog */
4313 : switch (RelationGetRelid(relation))
4314 : {
4315 : case RelationRelationId:
4316 : case DatabaseRelationId:
4317 : {
4318 : LOCKTAG tuptag;
4319 :
4320 : SET_LOCKTAG_TUPLE(tuptag,
4321 : relation->rd_lockInfo.lockRelId.dbId,
4322 : relation->rd_lockInfo.lockRelId.relId,
4323 : ItemPointerGetBlockNumber(otid),
4324 : ItemPointerGetOffsetNumber(otid));
4325 : if (LockHeldByMe(&tuptag, InplaceUpdateTupleLock, false))
4326 : return;
4327 : }
4328 : break;
4329 : default:
4330 : Assert(!IsInplaceUpdateRelation(relation));
4331 : return;
4332 : }
4333 :
4334 : switch (RelationGetRelid(relation))
4335 : {
4336 : case RelationRelationId:
4337 : {
4338 : /* LOCKTAG_TUPLE or LOCKTAG_RELATION ok */
4339 : Form_pg_class classForm = (Form_pg_class) GETSTRUCT(newtup);
4340 : Oid relid = classForm->oid;
4341 : Oid dbid;
4342 : LOCKTAG tag;
4343 :
4344 : if (IsSharedRelation(relid))
4345 : dbid = InvalidOid;
4346 : else
4347 : dbid = MyDatabaseId;
4348 :
4349 : if (classForm->relkind == RELKIND_INDEX)
4350 : {
4351 : Relation irel = index_open(relid, AccessShareLock);
4352 :
4353 : SET_LOCKTAG_RELATION(tag, dbid, irel->rd_index->indrelid);
4354 : index_close(irel, AccessShareLock);
4355 : }
4356 : else
4357 : SET_LOCKTAG_RELATION(tag, dbid, relid);
4358 :
4359 : if (!LockHeldByMe(&tag, ShareUpdateExclusiveLock, false) &&
4360 : !LockHeldByMe(&tag, ShareRowExclusiveLock, true))
4361 : elog(WARNING,
4362 : "missing lock for relation \"%s\" (OID %u, relkind %c) @ TID (%u,%u)",
4363 : NameStr(classForm->relname),
4364 : relid,
4365 : classForm->relkind,
4366 : ItemPointerGetBlockNumber(otid),
4367 : ItemPointerGetOffsetNumber(otid));
4368 : }
4369 : break;
4370 : case DatabaseRelationId:
4371 : {
4372 : /* LOCKTAG_TUPLE required */
4373 : Form_pg_database dbForm = (Form_pg_database) GETSTRUCT(newtup);
4374 :
4375 : elog(WARNING,
4376 : "missing lock on database \"%s\" (OID %u) @ TID (%u,%u)",
4377 : NameStr(dbForm->datname),
4378 : dbForm->oid,
4379 : ItemPointerGetBlockNumber(otid),
4380 : ItemPointerGetOffsetNumber(otid));
4381 : }
4382 : break;
4383 : }
4384 : }
4385 :
4386 : /*
4387 : * Confirm adequate relation lock held, per rules from README.tuplock section
4388 : * "Locking to write inplace-updated tables".
4389 : */
4390 : static void
4391 : check_inplace_rel_lock(HeapTuple oldtup)
4392 : {
4393 : Form_pg_class classForm = (Form_pg_class) GETSTRUCT(oldtup);
4394 : Oid relid = classForm->oid;
4395 : Oid dbid;
4396 : LOCKTAG tag;
4397 :
4398 : if (IsSharedRelation(relid))
4399 : dbid = InvalidOid;
4400 : else
4401 : dbid = MyDatabaseId;
4402 :
4403 : if (classForm->relkind == RELKIND_INDEX)
4404 : {
4405 : Relation irel = index_open(relid, AccessShareLock);
4406 :
4407 : SET_LOCKTAG_RELATION(tag, dbid, irel->rd_index->indrelid);
4408 : index_close(irel, AccessShareLock);
4409 : }
4410 : else
4411 : SET_LOCKTAG_RELATION(tag, dbid, relid);
4412 :
4413 : if (!LockHeldByMe(&tag, ShareUpdateExclusiveLock, true))
4414 : elog(WARNING,
4415 : "missing lock for relation \"%s\" (OID %u, relkind %c) @ TID (%u,%u)",
4416 : NameStr(classForm->relname),
4417 : relid,
4418 : classForm->relkind,
4419 : ItemPointerGetBlockNumber(&oldtup->t_self),
4420 : ItemPointerGetOffsetNumber(&oldtup->t_self));
4421 : }
4422 : #endif
4423 :
4424 : /*
4425 : * Check if the specified attribute's values are the same. Subroutine for
4426 : * HeapDetermineColumnsInfo.
4427 : */
4428 : static bool
4429 873466 : heap_attr_equals(TupleDesc tupdesc, int attrnum, Datum value1, Datum value2,
4430 : bool isnull1, bool isnull2)
4431 : {
4432 : /*
4433 : * If one value is NULL and other is not, then they are certainly not
4434 : * equal
4435 : */
4436 873466 : if (isnull1 != isnull2)
4437 60 : return false;
4438 :
4439 : /*
4440 : * If both are NULL, they can be considered equal.
4441 : */
4442 873406 : if (isnull1)
4443 6641 : return true;
4444 :
4445 : /*
4446 : * We do simple binary comparison of the two datums. This may be overly
4447 : * strict because there can be multiple binary representations for the
4448 : * same logical value. But we should be OK as long as there are no false
4449 : * positives. Using a type-specific equality operator is messy because
4450 : * there could be multiple notions of equality in different operator
4451 : * classes; furthermore, we cannot safely invoke user-defined functions
4452 : * while holding exclusive buffer lock.
4453 : */
4454 866765 : if (attrnum <= 0)
4455 : {
4456 : /* The only allowed system columns are OIDs, so do this */
4457 0 : return (DatumGetObjectId(value1) == DatumGetObjectId(value2));
4458 : }
4459 : else
4460 : {
4461 : CompactAttribute *att;
4462 :
4463 : Assert(attrnum <= tupdesc->natts);
4464 866765 : att = TupleDescCompactAttr(tupdesc, attrnum - 1);
4465 866765 : return datumIsEqual(value1, value2, att->attbyval, att->attlen);
4466 : }
4467 : }
4468 :
4469 : /*
4470 : * Check which columns are being updated.
4471 : *
4472 : * Given an updated tuple, determine (and return into the output bitmapset),
4473 : * from those listed as interesting, the set of columns that changed.
4474 : *
4475 : * has_external indicates if any of the unmodified attributes (from those
4476 : * listed as interesting) of the old tuple is a member of external_cols and is
4477 : * stored externally.
4478 : */
4479 : static Bitmapset *
4480 336594 : HeapDetermineColumnsInfo(Relation relation,
4481 : Bitmapset *interesting_cols,
4482 : Bitmapset *external_cols,
4483 : HeapTuple oldtup, HeapTuple newtup,
4484 : bool *has_external)
4485 : {
4486 : int attidx;
4487 336594 : Bitmapset *modified = NULL;
4488 336594 : TupleDesc tupdesc = RelationGetDescr(relation);
4489 :
4490 336594 : attidx = -1;
4491 1210060 : while ((attidx = bms_next_member(interesting_cols, attidx)) >= 0)
4492 : {
4493 : /* attidx is zero-based, attrnum is the normal attribute number */
4494 873466 : AttrNumber attrnum = attidx + FirstLowInvalidHeapAttributeNumber;
4495 : Datum value1,
4496 : value2;
4497 : bool isnull1,
4498 : isnull2;
4499 :
4500 : /*
4501 : * If it's a whole-tuple reference, say "not equal". It's not really
4502 : * worth supporting this case, since it could only succeed after a
4503 : * no-op update, which is hardly a case worth optimizing for.
4504 : */
4505 873466 : if (attrnum == 0)
4506 : {
4507 0 : modified = bms_add_member(modified, attidx);
4508 841722 : continue;
4509 : }
4510 :
4511 : /*
4512 : * Likewise, automatically say "not equal" for any system attribute
4513 : * other than tableOID; we cannot expect these to be consistent in a
4514 : * HOT chain, or even to be set correctly yet in the new tuple.
4515 : */
4516 873466 : if (attrnum < 0)
4517 : {
4518 0 : if (attrnum != TableOidAttributeNumber)
4519 : {
4520 0 : modified = bms_add_member(modified, attidx);
4521 0 : continue;
4522 : }
4523 : }
4524 :
4525 : /*
4526 : * Extract the corresponding values. XXX this is pretty inefficient
4527 : * if there are many indexed columns. Should we do a single
4528 : * heap_deform_tuple call on each tuple, instead? But that doesn't
4529 : * work for system columns ...
4530 : */
4531 873466 : value1 = heap_getattr(oldtup, attrnum, tupdesc, &isnull1);
4532 873466 : value2 = heap_getattr(newtup, attrnum, tupdesc, &isnull2);
4533 :
4534 873466 : if (!heap_attr_equals(tupdesc, attrnum, value1,
4535 : value2, isnull1, isnull2))
4536 : {
4537 34013 : modified = bms_add_member(modified, attidx);
4538 34013 : continue;
4539 : }
4540 :
4541 : /*
4542 : * No need to check attributes that can't be stored externally. Note
4543 : * that system attributes can't be stored externally.
4544 : */
4545 839453 : if (attrnum < 0 || isnull1 ||
4546 832812 : TupleDescCompactAttr(tupdesc, attrnum - 1)->attlen != -1)
4547 807709 : continue;
4548 :
4549 : /*
4550 : * Check if the old tuple's attribute is stored externally and is a
4551 : * member of external_cols.
4552 : */
4553 31749 : if (VARATT_IS_EXTERNAL((varlena *) DatumGetPointer(value1)) &&
4554 5 : bms_is_member(attidx, external_cols))
4555 2 : *has_external = true;
4556 : }
4557 :
4558 336594 : return modified;
4559 : }
4560 :
4561 : /*
4562 : * simple_heap_update - replace a tuple
4563 : *
4564 : * This routine may be used to update a tuple when concurrent updates of
4565 : * the target tuple are not expected (for example, because we have a lock
4566 : * on the relation associated with the tuple). Any failure is reported
4567 : * via ereport().
4568 : */
4569 : void
4570 133850 : simple_heap_update(Relation relation, const ItemPointerData *otid, HeapTuple tup,
4571 : TU_UpdateIndexes *update_indexes)
4572 : {
4573 : TM_Result result;
4574 : TM_FailureData tmfd;
4575 : LockTupleMode lockmode;
4576 :
4577 133850 : result = heap_update(relation, otid, tup,
4578 : GetCurrentCommandId(true), InvalidSnapshot,
4579 : true /* wait for commit */ ,
4580 : &tmfd, &lockmode, update_indexes);
4581 133850 : switch (result)
4582 : {
4583 0 : case TM_SelfModified:
4584 : /* Tuple was already updated in current command? */
4585 0 : elog(ERROR, "tuple already updated by self");
4586 : break;
4587 :
4588 133849 : case TM_Ok:
4589 : /* done successfully */
4590 133849 : break;
4591 :
4592 0 : case TM_Updated:
4593 0 : elog(ERROR, "tuple concurrently updated");
4594 : break;
4595 :
4596 1 : case TM_Deleted:
4597 1 : elog(ERROR, "tuple concurrently deleted");
4598 : break;
4599 :
4600 0 : default:
4601 0 : elog(ERROR, "unrecognized heap_update status: %u", result);
4602 : break;
4603 : }
4604 133849 : }
4605 :
4606 :
4607 : /*
4608 : * Return the MultiXactStatus corresponding to the given tuple lock mode.
4609 : */
4610 : static MultiXactStatus
4611 115506 : get_mxact_status_for_lock(LockTupleMode mode, bool is_update)
4612 : {
4613 : int retval;
4614 :
4615 115506 : if (is_update)
4616 216 : retval = tupleLockExtraInfo[mode].updstatus;
4617 : else
4618 115290 : retval = tupleLockExtraInfo[mode].lockstatus;
4619 :
4620 115506 : if (retval == -1)
4621 0 : elog(ERROR, "invalid lock tuple mode %d/%s", mode,
4622 : is_update ? "true" : "false");
4623 :
4624 115506 : return (MultiXactStatus) retval;
4625 : }
4626 :
4627 : /*
4628 : * heap_lock_tuple - lock a tuple in shared or exclusive mode
4629 : *
4630 : * Note that this acquires a buffer pin, which the caller must release.
4631 : *
4632 : * Input parameters:
4633 : * relation: relation containing tuple (caller must hold suitable lock)
4634 : * cid: current command ID (used for visibility test, and stored into
4635 : * tuple's cmax if lock is successful)
4636 : * mode: indicates if shared or exclusive tuple lock is desired
4637 : * wait_policy: what to do if tuple lock is not available
4638 : * follow_updates: if true, follow the update chain to also lock descendant
4639 : * tuples.
4640 : *
4641 : * Output parameters:
4642 : * *tuple: all fields filled in
4643 : * *buffer: set to buffer holding tuple (pinned but not locked at exit)
4644 : * *tmfd: filled in failure cases (see below)
4645 : *
4646 : * Function results are the same as the ones for table_tuple_lock().
4647 : *
4648 : * In the failure cases other than TM_Invisible, the routine fills
4649 : * *tmfd with the tuple's t_ctid, t_xmax (resolving a possible MultiXact,
4650 : * if necessary), and t_cmax (the last only for TM_SelfModified,
4651 : * since we cannot obtain cmax from a combo CID generated by another
4652 : * transaction).
4653 : * See comments for struct TM_FailureData for additional info.
4654 : *
4655 : * See README.tuplock for a thorough explanation of this mechanism.
4656 : */
4657 : TM_Result
4658 559748 : heap_lock_tuple(Relation relation, HeapTuple tuple,
4659 : CommandId cid, LockTupleMode mode, LockWaitPolicy wait_policy,
4660 : bool follow_updates,
4661 : Buffer *buffer, TM_FailureData *tmfd)
4662 : {
4663 : TM_Result result;
4664 559748 : ItemPointer tid = &(tuple->t_self);
4665 : ItemId lp;
4666 : Page page;
4667 559748 : Buffer vmbuffer = InvalidBuffer;
4668 : BlockNumber block;
4669 : TransactionId xid,
4670 : xmax;
4671 : uint16 old_infomask,
4672 : new_infomask,
4673 : new_infomask2;
4674 559748 : bool first_time = true;
4675 559748 : bool skip_tuple_lock = false;
4676 559748 : bool have_tuple_lock = false;
4677 559748 : bool cleared_all_frozen = false;
4678 :
4679 559748 : *buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
4680 559748 : block = ItemPointerGetBlockNumber(tid);
4681 :
4682 : /*
4683 : * Before locking the buffer, pin the visibility map page if it appears to
4684 : * be necessary. Since we haven't got the lock yet, someone else might be
4685 : * in the middle of changing this, so we'll need to recheck after we have
4686 : * the lock.
4687 : */
4688 559748 : if (PageIsAllVisible(BufferGetPage(*buffer)))
4689 401796 : visibilitymap_pin(relation, block, &vmbuffer);
4690 :
4691 559748 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4692 :
4693 559748 : page = BufferGetPage(*buffer);
4694 559748 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
4695 : Assert(ItemIdIsNormal(lp));
4696 :
4697 559748 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
4698 559748 : tuple->t_len = ItemIdGetLength(lp);
4699 559748 : tuple->t_tableOid = RelationGetRelid(relation);
4700 :
4701 17 : l3:
4702 559765 : result = HeapTupleSatisfiesUpdate(tuple, cid, *buffer);
4703 :
4704 559765 : if (result == TM_Invisible)
4705 : {
4706 : /*
4707 : * This is possible, but only when locking a tuple for ON CONFLICT DO
4708 : * SELECT/UPDATE. We return this value here rather than throwing an
4709 : * error in order to give that case the opportunity to throw a more
4710 : * specific error.
4711 : */
4712 28 : result = TM_Invisible;
4713 28 : goto out_locked;
4714 : }
4715 559737 : else if (result == TM_BeingModified ||
4716 78551 : result == TM_Updated ||
4717 : result == TM_Deleted)
4718 : {
4719 : TransactionId xwait;
4720 : uint16 infomask;
4721 : uint16 infomask2;
4722 : bool require_sleep;
4723 : ItemPointerData t_ctid;
4724 :
4725 : /* must copy state data before unlocking buffer */
4726 481188 : xwait = HeapTupleHeaderGetRawXmax(tuple->t_data);
4727 481188 : infomask = tuple->t_data->t_infomask;
4728 481188 : infomask2 = tuple->t_data->t_infomask2;
4729 481188 : ItemPointerCopy(&tuple->t_data->t_ctid, &t_ctid);
4730 :
4731 481188 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
4732 :
4733 : /*
4734 : * If any subtransaction of the current top transaction already holds
4735 : * a lock as strong as or stronger than what we're requesting, we
4736 : * effectively hold the desired lock already. We *must* succeed
4737 : * without trying to take the tuple lock, else we will deadlock
4738 : * against anyone wanting to acquire a stronger lock.
4739 : *
4740 : * Note we only do this the first time we loop on the HTSU result;
4741 : * there is no point in testing in subsequent passes, because
4742 : * evidently our own transaction cannot have acquired a new lock after
4743 : * the first time we checked.
4744 : */
4745 481188 : if (first_time)
4746 : {
4747 481176 : first_time = false;
4748 :
4749 481176 : if (infomask & HEAP_XMAX_IS_MULTI)
4750 : {
4751 : int i;
4752 : int nmembers;
4753 : MultiXactMember *members;
4754 :
4755 : /*
4756 : * We don't need to allow old multixacts here; if that had
4757 : * been the case, HeapTupleSatisfiesUpdate would have returned
4758 : * MayBeUpdated and we wouldn't be here.
4759 : */
4760 : nmembers =
4761 73295 : GetMultiXactIdMembers(xwait, &members, false,
4762 73295 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
4763 :
4764 1422658 : for (i = 0; i < nmembers; i++)
4765 : {
4766 : /* only consider members of our own transaction */
4767 1349377 : if (!TransactionIdIsCurrentTransactionId(members[i].xid))
4768 1349327 : continue;
4769 :
4770 50 : if (TUPLOCK_from_mxstatus(members[i].status) >= mode)
4771 : {
4772 14 : pfree(members);
4773 14 : result = TM_Ok;
4774 14 : goto out_unlocked;
4775 : }
4776 : else
4777 : {
4778 : /*
4779 : * Disable acquisition of the heavyweight tuple lock.
4780 : * Otherwise, when promoting a weaker lock, we might
4781 : * deadlock with another locker that has acquired the
4782 : * heavyweight tuple lock and is waiting for our
4783 : * transaction to finish.
4784 : *
4785 : * Note that in this case we still need to wait for
4786 : * the multixact if required, to avoid acquiring
4787 : * conflicting locks.
4788 : */
4789 36 : skip_tuple_lock = true;
4790 : }
4791 : }
4792 :
4793 73281 : if (members)
4794 73281 : pfree(members);
4795 : }
4796 407881 : else if (TransactionIdIsCurrentTransactionId(xwait))
4797 : {
4798 406487 : switch (mode)
4799 : {
4800 400052 : case LockTupleKeyShare:
4801 : Assert(HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) ||
4802 : HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4803 : HEAP_XMAX_IS_EXCL_LOCKED(infomask));
4804 400052 : result = TM_Ok;
4805 400052 : goto out_unlocked;
4806 38 : case LockTupleShare:
4807 44 : if (HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4808 6 : HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4809 : {
4810 32 : result = TM_Ok;
4811 32 : goto out_unlocked;
4812 : }
4813 6 : break;
4814 82 : case LockTupleNoKeyExclusive:
4815 82 : if (HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4816 : {
4817 70 : result = TM_Ok;
4818 70 : goto out_unlocked;
4819 : }
4820 12 : break;
4821 6315 : case LockTupleExclusive:
4822 6315 : if (HEAP_XMAX_IS_EXCL_LOCKED(infomask) &&
4823 1273 : infomask2 & HEAP_KEYS_UPDATED)
4824 : {
4825 1245 : result = TM_Ok;
4826 1245 : goto out_unlocked;
4827 : }
4828 5070 : break;
4829 : }
4830 : }
4831 : }
4832 :
4833 : /*
4834 : * Initially assume that we will have to wait for the locking
4835 : * transaction(s) to finish. We check various cases below in which
4836 : * this can be turned off.
4837 : */
4838 79775 : require_sleep = true;
4839 79775 : if (mode == LockTupleKeyShare)
4840 : {
4841 : /*
4842 : * If we're requesting KeyShare, and there's no update present, we
4843 : * don't need to wait. Even if there is an update, we can still
4844 : * continue if the key hasn't been modified.
4845 : *
4846 : * However, if there are updates, we need to walk the update chain
4847 : * to mark future versions of the row as locked, too. That way,
4848 : * if somebody deletes that future version, we're protected
4849 : * against the key going away. This locking of future versions
4850 : * could block momentarily, if a concurrent transaction is
4851 : * deleting a key; or it could return a value to the effect that
4852 : * the transaction deleting the key has already committed. So we
4853 : * do this before re-locking the buffer; otherwise this would be
4854 : * prone to deadlocks.
4855 : *
4856 : * Note that the TID we're locking was grabbed before we unlocked
4857 : * the buffer. For it to change while we're not looking, the
4858 : * other properties we're testing for below after re-locking the
4859 : * buffer would also change, in which case we would restart this
4860 : * loop above.
4861 : */
4862 73888 : if (!(infomask2 & HEAP_KEYS_UPDATED))
4863 : {
4864 : bool updated;
4865 :
4866 73845 : updated = !HEAP_XMAX_IS_LOCKED_ONLY(infomask);
4867 :
4868 : /*
4869 : * If there are updates, follow the update chain; bail out if
4870 : * that cannot be done.
4871 : */
4872 73845 : if (follow_updates && updated &&
4873 2169 : !ItemPointerEquals(&tuple->t_self, &t_ctid))
4874 : {
4875 : TM_Result res;
4876 :
4877 2169 : res = heap_lock_updated_tuple(relation,
4878 : infomask, xwait, &t_ctid,
4879 : GetCurrentTransactionId(),
4880 : mode);
4881 2169 : if (res != TM_Ok)
4882 : {
4883 6 : result = res;
4884 : /* recovery code expects to have buffer lock held */
4885 6 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4886 200 : goto failed;
4887 : }
4888 : }
4889 :
4890 73839 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4891 :
4892 : /*
4893 : * Make sure it's still an appropriate lock, else start over.
4894 : * Also, if it wasn't updated before we released the lock, but
4895 : * is updated now, we start over too; the reason is that we
4896 : * now need to follow the update chain to lock the new
4897 : * versions.
4898 : */
4899 73839 : if (!HeapTupleHeaderIsOnlyLocked(tuple->t_data) &&
4900 2151 : ((tuple->t_data->t_infomask2 & HEAP_KEYS_UPDATED) ||
4901 2151 : !updated))
4902 17 : goto l3;
4903 :
4904 : /* Things look okay, so we can skip sleeping */
4905 73839 : require_sleep = false;
4906 :
4907 : /*
4908 : * Note we allow Xmax to change here; other updaters/lockers
4909 : * could have modified it before we grabbed the buffer lock.
4910 : * However, this is not a problem, because with the recheck we
4911 : * just did we ensure that they still don't conflict with the
4912 : * lock we want.
4913 : */
4914 : }
4915 : }
4916 5887 : else if (mode == LockTupleShare)
4917 : {
4918 : /*
4919 : * If we're requesting Share, we can similarly avoid sleeping if
4920 : * there's no update and no exclusive lock present.
4921 : */
4922 477 : if (HEAP_XMAX_IS_LOCKED_ONLY(infomask) &&
4923 477 : !HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4924 : {
4925 471 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4926 :
4927 : /*
4928 : * Make sure it's still an appropriate lock, else start over.
4929 : * See above about allowing xmax to change.
4930 : */
4931 942 : if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
4932 471 : HEAP_XMAX_IS_EXCL_LOCKED(tuple->t_data->t_infomask))
4933 0 : goto l3;
4934 471 : require_sleep = false;
4935 : }
4936 : }
4937 5410 : else if (mode == LockTupleNoKeyExclusive)
4938 : {
4939 : /*
4940 : * If we're requesting NoKeyExclusive, we might also be able to
4941 : * avoid sleeping; just ensure that there no conflicting lock
4942 : * already acquired.
4943 : */
4944 173 : if (infomask & HEAP_XMAX_IS_MULTI)
4945 : {
4946 26 : if (!DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
4947 : mode, NULL))
4948 : {
4949 : /*
4950 : * No conflict, but if the xmax changed under us in the
4951 : * meantime, start over.
4952 : */
4953 13 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4954 26 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4955 13 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4956 : xwait))
4957 0 : goto l3;
4958 :
4959 : /* otherwise, we're good */
4960 13 : require_sleep = false;
4961 : }
4962 : }
4963 147 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask))
4964 : {
4965 18 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4966 :
4967 : /* if the xmax changed in the meantime, start over */
4968 36 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4969 18 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4970 : xwait))
4971 0 : goto l3;
4972 : /* otherwise, we're good */
4973 18 : require_sleep = false;
4974 : }
4975 : }
4976 :
4977 : /*
4978 : * As a check independent from those above, we can also avoid sleeping
4979 : * if the current transaction is the sole locker of the tuple. Note
4980 : * that the strength of the lock already held is irrelevant; this is
4981 : * not about recording the lock in Xmax (which will be done regardless
4982 : * of this optimization, below). Also, note that the cases where we
4983 : * hold a lock stronger than we are requesting are already handled
4984 : * above by not doing anything.
4985 : *
4986 : * Note we only deal with the non-multixact case here; MultiXactIdWait
4987 : * is well equipped to deal with this situation on its own.
4988 : */
4989 85153 : if (require_sleep && !(infomask & HEAP_XMAX_IS_MULTI) &&
4990 5384 : TransactionIdIsCurrentTransactionId(xwait))
4991 : {
4992 : /* ... but if the xmax changed in the meantime, start over */
4993 5070 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4994 10140 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4995 5070 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4996 : xwait))
4997 0 : goto l3;
4998 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask));
4999 5070 : require_sleep = false;
5000 : }
5001 :
5002 : /*
5003 : * Time to sleep on the other transaction/multixact, if necessary.
5004 : *
5005 : * If the other transaction is an update/delete that's already
5006 : * committed, then sleeping cannot possibly do any good: if we're
5007 : * required to sleep, get out to raise an error instead.
5008 : *
5009 : * By here, we either have already acquired the buffer exclusive lock,
5010 : * or we must wait for the locking transaction or multixact; so below
5011 : * we ensure that we grab buffer lock after the sleep.
5012 : */
5013 79769 : if (require_sleep && (result == TM_Updated || result == TM_Deleted))
5014 : {
5015 156 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5016 156 : goto failed;
5017 : }
5018 79613 : else if (require_sleep)
5019 : {
5020 : /*
5021 : * Acquire tuple lock to establish our priority for the tuple, or
5022 : * die trying. LockTuple will release us when we are next-in-line
5023 : * for the tuple. We must do this even if we are share-locking,
5024 : * but not if we already have a weaker lock on the tuple.
5025 : *
5026 : * If we are forced to "start over" below, we keep the tuple lock;
5027 : * this arranges that we stay at the head of the line while
5028 : * rechecking tuple state.
5029 : */
5030 202 : if (!skip_tuple_lock &&
5031 185 : !heap_acquire_tuplock(relation, tid, mode, wait_policy,
5032 : &have_tuple_lock))
5033 : {
5034 : /*
5035 : * This can only happen if wait_policy is Skip and the lock
5036 : * couldn't be obtained.
5037 : */
5038 1 : result = TM_WouldBlock;
5039 : /* recovery code expects to have buffer lock held */
5040 1 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5041 1 : goto failed;
5042 : }
5043 :
5044 200 : if (infomask & HEAP_XMAX_IS_MULTI)
5045 : {
5046 43 : MultiXactStatus status = get_mxact_status_for_lock(mode, false);
5047 :
5048 : /* We only ever lock tuples, never update them */
5049 43 : if (status >= MultiXactStatusNoKeyUpdate)
5050 0 : elog(ERROR, "invalid lock mode in heap_lock_tuple");
5051 :
5052 : /* wait for multixact to end, or die trying */
5053 43 : switch (wait_policy)
5054 : {
5055 37 : case LockWaitBlock:
5056 37 : MultiXactIdWait((MultiXactId) xwait, status, infomask,
5057 37 : relation, &tuple->t_self, XLTW_Lock, NULL);
5058 37 : break;
5059 2 : case LockWaitSkip:
5060 2 : if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
5061 : status, infomask, relation,
5062 : NULL, false))
5063 : {
5064 2 : result = TM_WouldBlock;
5065 : /* recovery code expects to have buffer lock held */
5066 2 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5067 2 : goto failed;
5068 : }
5069 0 : break;
5070 4 : case LockWaitError:
5071 4 : if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
5072 : status, infomask, relation,
5073 : NULL, log_lock_failures))
5074 4 : ereport(ERROR,
5075 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5076 : errmsg("could not obtain lock on row in relation \"%s\"",
5077 : RelationGetRelationName(relation))));
5078 :
5079 0 : break;
5080 : }
5081 :
5082 : /*
5083 : * Of course, the multixact might not be done here: if we're
5084 : * requesting a light lock mode, other transactions with light
5085 : * locks could still be alive, as well as locks owned by our
5086 : * own xact or other subxacts of this backend. We need to
5087 : * preserve the surviving MultiXact members. Note that it
5088 : * isn't absolutely necessary in the latter case, but doing so
5089 : * is simpler.
5090 : */
5091 : }
5092 : else
5093 : {
5094 : /* wait for regular transaction to end, or die trying */
5095 157 : switch (wait_policy)
5096 : {
5097 116 : case LockWaitBlock:
5098 116 : XactLockTableWait(xwait, relation, &tuple->t_self,
5099 : XLTW_Lock);
5100 116 : break;
5101 33 : case LockWaitSkip:
5102 33 : if (!ConditionalXactLockTableWait(xwait, false))
5103 : {
5104 33 : result = TM_WouldBlock;
5105 : /* recovery code expects to have buffer lock held */
5106 33 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5107 33 : goto failed;
5108 : }
5109 0 : break;
5110 8 : case LockWaitError:
5111 8 : if (!ConditionalXactLockTableWait(xwait, log_lock_failures))
5112 8 : ereport(ERROR,
5113 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5114 : errmsg("could not obtain lock on row in relation \"%s\"",
5115 : RelationGetRelationName(relation))));
5116 0 : break;
5117 : }
5118 : }
5119 :
5120 : /* if there are updates, follow the update chain */
5121 153 : if (follow_updates && !HEAP_XMAX_IS_LOCKED_ONLY(infomask) &&
5122 59 : !ItemPointerEquals(&tuple->t_self, &t_ctid))
5123 : {
5124 : TM_Result res;
5125 :
5126 46 : res = heap_lock_updated_tuple(relation,
5127 : infomask, xwait, &t_ctid,
5128 : GetCurrentTransactionId(),
5129 : mode);
5130 46 : if (res != TM_Ok)
5131 : {
5132 2 : result = res;
5133 : /* recovery code expects to have buffer lock held */
5134 2 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5135 2 : goto failed;
5136 : }
5137 : }
5138 :
5139 151 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5140 :
5141 : /*
5142 : * xwait is done, but if xwait had just locked the tuple then some
5143 : * other xact could update this tuple before we get to this point.
5144 : * Check for xmax change, and start over if so.
5145 : */
5146 287 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
5147 136 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
5148 : xwait))
5149 17 : goto l3;
5150 :
5151 134 : if (!(infomask & HEAP_XMAX_IS_MULTI))
5152 : {
5153 : /*
5154 : * Otherwise check if it committed or aborted. Note we cannot
5155 : * be here if the tuple was only locked by somebody who didn't
5156 : * conflict with us; that would have been handled above. So
5157 : * that transaction must necessarily be gone by now. But
5158 : * don't check for this in the multixact case, because some
5159 : * locker transactions might still be running.
5160 : */
5161 101 : UpdateXmaxHintBits(tuple->t_data, *buffer, xwait);
5162 : }
5163 : }
5164 :
5165 : /* By here, we're certain that we hold buffer exclusive lock again */
5166 :
5167 : /*
5168 : * We may lock if previous xmax aborted, or if it committed but only
5169 : * locked the tuple without updating it; or if we didn't have to wait
5170 : * at all for whatever reason.
5171 : */
5172 79545 : if (!require_sleep ||
5173 232 : (tuple->t_data->t_infomask & HEAP_XMAX_INVALID) ||
5174 180 : HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
5175 82 : HeapTupleHeaderIsOnlyLocked(tuple->t_data))
5176 79470 : result = TM_Ok;
5177 75 : else if (!ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid))
5178 57 : result = TM_Updated;
5179 : else
5180 18 : result = TM_Deleted;
5181 : }
5182 :
5183 78549 : failed:
5184 158294 : if (result != TM_Ok)
5185 : {
5186 : Assert(result == TM_SelfModified || result == TM_Updated ||
5187 : result == TM_Deleted || result == TM_WouldBlock);
5188 :
5189 : /*
5190 : * When locking a tuple under LockWaitSkip semantics and we fail with
5191 : * TM_WouldBlock above, it's possible for concurrent transactions to
5192 : * release the lock and set HEAP_XMAX_INVALID in the meantime. So
5193 : * this assert is slightly different from the equivalent one in
5194 : * heap_delete and heap_update.
5195 : */
5196 : Assert((result == TM_WouldBlock) ||
5197 : !(tuple->t_data->t_infomask & HEAP_XMAX_INVALID));
5198 : Assert(result != TM_Updated ||
5199 : !ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid));
5200 283 : tmfd->ctid = tuple->t_data->t_ctid;
5201 283 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(tuple->t_data);
5202 283 : if (result == TM_SelfModified)
5203 8 : tmfd->cmax = HeapTupleHeaderGetCmax(tuple->t_data);
5204 : else
5205 275 : tmfd->cmax = InvalidCommandId;
5206 283 : goto out_locked;
5207 : }
5208 :
5209 : /*
5210 : * If we didn't pin the visibility map page and the page has become all
5211 : * visible while we were busy locking the buffer, or during some
5212 : * subsequent window during which we had it unlocked, we'll have to unlock
5213 : * and re-lock, to avoid holding the buffer lock across I/O. That's a bit
5214 : * unfortunate, especially since we'll now have to recheck whether the
5215 : * tuple has been locked or updated under us, but hopefully it won't
5216 : * happen very often.
5217 : */
5218 158011 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
5219 : {
5220 0 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
5221 0 : visibilitymap_pin(relation, block, &vmbuffer);
5222 0 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5223 0 : goto l3;
5224 : }
5225 :
5226 158011 : xmax = HeapTupleHeaderGetRawXmax(tuple->t_data);
5227 158011 : old_infomask = tuple->t_data->t_infomask;
5228 :
5229 : /*
5230 : * If this is the first possibly-multixact-able operation in the current
5231 : * transaction, set my per-backend OldestMemberMXactId setting. We can be
5232 : * certain that the transaction will never become a member of any older
5233 : * MultiXactIds than that. (We have to do this even if we end up just
5234 : * using our own TransactionId below, since some other backend could
5235 : * incorporate our XID into a MultiXact immediately afterwards.)
5236 : */
5237 158011 : MultiXactIdSetOldestMember();
5238 :
5239 : /*
5240 : * Compute the new xmax and infomask to store into the tuple. Note we do
5241 : * not modify the tuple just yet, because that would leave it in the wrong
5242 : * state if multixact.c elogs.
5243 : */
5244 158011 : compute_new_xmax_infomask(xmax, old_infomask, tuple->t_data->t_infomask2,
5245 : GetCurrentTransactionId(), mode, false,
5246 : &xid, &new_infomask, &new_infomask2);
5247 :
5248 158011 : START_CRIT_SECTION();
5249 :
5250 : /*
5251 : * Store transaction information of xact locking the tuple.
5252 : *
5253 : * Note: Cmax is meaningless in this context, so don't set it; this avoids
5254 : * possibly generating a useless combo CID. Moreover, if we're locking a
5255 : * previously updated tuple, it's important to preserve the Cmax.
5256 : *
5257 : * Also reset the HOT UPDATE bit, but only if there's no update; otherwise
5258 : * we would break the HOT chain.
5259 : */
5260 158011 : tuple->t_data->t_infomask &= ~HEAP_XMAX_BITS;
5261 158011 : tuple->t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
5262 158011 : tuple->t_data->t_infomask |= new_infomask;
5263 158011 : tuple->t_data->t_infomask2 |= new_infomask2;
5264 158011 : if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
5265 155864 : HeapTupleHeaderClearHotUpdated(tuple->t_data);
5266 158011 : HeapTupleHeaderSetXmax(tuple->t_data, xid);
5267 :
5268 : /*
5269 : * Make sure there is no forward chain link in t_ctid. Note that in the
5270 : * cases where the tuple has been updated, we must not overwrite t_ctid,
5271 : * because it was set by the updater. Moreover, if the tuple has been
5272 : * updated, we need to follow the update chain to lock the new versions of
5273 : * the tuple as well.
5274 : */
5275 158011 : if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
5276 155864 : tuple->t_data->t_ctid = *tid;
5277 :
5278 : /* Clear only the all-frozen bit on visibility map if needed */
5279 160013 : if (PageIsAllVisible(page) &&
5280 2002 : visibilitymap_clear(relation, block, vmbuffer,
5281 : VISIBILITYMAP_ALL_FROZEN))
5282 16 : cleared_all_frozen = true;
5283 :
5284 :
5285 158011 : MarkBufferDirty(*buffer);
5286 :
5287 : /*
5288 : * XLOG stuff. You might think that we don't need an XLOG record because
5289 : * there is no state change worth restoring after a crash. You would be
5290 : * wrong however: we have just written either a TransactionId or a
5291 : * MultiXactId that may never have been seen on disk before, and we need
5292 : * to make sure that there are XLOG entries covering those ID numbers.
5293 : * Else the same IDs might be re-used after a crash, which would be
5294 : * disastrous if this page made it to disk before the crash. Essentially
5295 : * we have to enforce the WAL log-before-data rule even in this case.
5296 : * (Also, in a PITR log-shipping or 2PC environment, we have to have XLOG
5297 : * entries for everything anyway.)
5298 : */
5299 158011 : if (RelationNeedsWAL(relation))
5300 : {
5301 : xl_heap_lock xlrec;
5302 : XLogRecPtr recptr;
5303 :
5304 157582 : XLogBeginInsert();
5305 157582 : XLogRegisterBuffer(0, *buffer, REGBUF_STANDARD);
5306 :
5307 157582 : xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
5308 157582 : xlrec.xmax = xid;
5309 315164 : xlrec.infobits_set = compute_infobits(new_infomask,
5310 157582 : tuple->t_data->t_infomask2);
5311 157582 : xlrec.flags = cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
5312 157582 : XLogRegisterData(&xlrec, SizeOfHeapLock);
5313 :
5314 : /* we don't decode row locks atm, so no need to log the origin */
5315 :
5316 157582 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
5317 :
5318 157582 : PageSetLSN(page, recptr);
5319 : }
5320 :
5321 158011 : END_CRIT_SECTION();
5322 :
5323 158011 : result = TM_Ok;
5324 :
5325 158322 : out_locked:
5326 158322 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
5327 :
5328 559735 : out_unlocked:
5329 559735 : if (BufferIsValid(vmbuffer))
5330 401796 : ReleaseBuffer(vmbuffer);
5331 :
5332 : /*
5333 : * Don't update the visibility map here. Locking a tuple doesn't change
5334 : * visibility info.
5335 : */
5336 :
5337 : /*
5338 : * Now that we have successfully marked the tuple as locked, we can
5339 : * release the lmgr tuple lock, if we had it.
5340 : */
5341 559735 : if (have_tuple_lock)
5342 166 : UnlockTupleTuplock(relation, tid, mode);
5343 :
5344 559735 : return result;
5345 : }
5346 :
5347 : /*
5348 : * Acquire heavyweight lock on the given tuple, in preparation for acquiring
5349 : * its normal, Xmax-based tuple lock.
5350 : *
5351 : * have_tuple_lock is an input and output parameter: on input, it indicates
5352 : * whether the lock has previously been acquired (and this function does
5353 : * nothing in that case). If this function returns success, have_tuple_lock
5354 : * has been flipped to true.
5355 : *
5356 : * Returns false if it was unable to obtain the lock; this can only happen if
5357 : * wait_policy is Skip.
5358 : */
5359 : static bool
5360 325 : heap_acquire_tuplock(Relation relation, const ItemPointerData *tid, LockTupleMode mode,
5361 : LockWaitPolicy wait_policy, bool *have_tuple_lock)
5362 : {
5363 325 : if (*have_tuple_lock)
5364 9 : return true;
5365 :
5366 316 : switch (wait_policy)
5367 : {
5368 271 : case LockWaitBlock:
5369 271 : LockTupleTuplock(relation, tid, mode);
5370 271 : break;
5371 :
5372 34 : case LockWaitSkip:
5373 34 : if (!ConditionalLockTupleTuplock(relation, tid, mode, false))
5374 1 : return false;
5375 33 : break;
5376 :
5377 11 : case LockWaitError:
5378 11 : if (!ConditionalLockTupleTuplock(relation, tid, mode, log_lock_failures))
5379 1 : ereport(ERROR,
5380 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5381 : errmsg("could not obtain lock on row in relation \"%s\"",
5382 : RelationGetRelationName(relation))));
5383 10 : break;
5384 : }
5385 314 : *have_tuple_lock = true;
5386 :
5387 314 : return true;
5388 : }
5389 :
5390 : /*
5391 : * Given an original set of Xmax and infomask, and a transaction (identified by
5392 : * add_to_xmax) acquiring a new lock of some mode, compute the new Xmax and
5393 : * corresponding infomasks to use on the tuple.
5394 : *
5395 : * Note that this might have side effects such as creating a new MultiXactId.
5396 : *
5397 : * Most callers will have called HeapTupleSatisfiesUpdate before this function;
5398 : * that will have set the HEAP_XMAX_INVALID bit if the xmax was a MultiXactId
5399 : * but it was not running anymore. There is a race condition, which is that the
5400 : * MultiXactId may have finished since then, but that uncommon case is handled
5401 : * either here, or within MultiXactIdExpand.
5402 : *
5403 : * There is a similar race condition possible when the old xmax was a regular
5404 : * TransactionId. We test TransactionIdIsInProgress again just to narrow the
5405 : * window, but it's still possible to end up creating an unnecessary
5406 : * MultiXactId. Fortunately this is harmless.
5407 : */
5408 : static void
5409 2498525 : compute_new_xmax_infomask(TransactionId xmax, uint16 old_infomask,
5410 : uint16 old_infomask2, TransactionId add_to_xmax,
5411 : LockTupleMode mode, bool is_update,
5412 : TransactionId *result_xmax, uint16 *result_infomask,
5413 : uint16 *result_infomask2)
5414 : {
5415 : TransactionId new_xmax;
5416 : uint16 new_infomask,
5417 : new_infomask2;
5418 :
5419 : Assert(TransactionIdIsCurrentTransactionId(add_to_xmax));
5420 :
5421 104353 : l5:
5422 2602878 : new_infomask = 0;
5423 2602878 : new_infomask2 = 0;
5424 2602878 : if (old_infomask & HEAP_XMAX_INVALID)
5425 : {
5426 : /*
5427 : * No previous locker; we just insert our own TransactionId.
5428 : *
5429 : * Note that it's critical that this case be the first one checked,
5430 : * because there are several blocks below that come back to this one
5431 : * to implement certain optimizations; old_infomask might contain
5432 : * other dirty bits in those cases, but we don't really care.
5433 : */
5434 2421836 : if (is_update)
5435 : {
5436 2180797 : new_xmax = add_to_xmax;
5437 2180797 : if (mode == LockTupleExclusive)
5438 1882233 : new_infomask2 |= HEAP_KEYS_UPDATED;
5439 : }
5440 : else
5441 : {
5442 241039 : new_infomask |= HEAP_XMAX_LOCK_ONLY;
5443 241039 : switch (mode)
5444 : {
5445 3446 : case LockTupleKeyShare:
5446 3446 : new_xmax = add_to_xmax;
5447 3446 : new_infomask |= HEAP_XMAX_KEYSHR_LOCK;
5448 3446 : break;
5449 811 : case LockTupleShare:
5450 811 : new_xmax = add_to_xmax;
5451 811 : new_infomask |= HEAP_XMAX_SHR_LOCK;
5452 811 : break;
5453 140893 : case LockTupleNoKeyExclusive:
5454 140893 : new_xmax = add_to_xmax;
5455 140893 : new_infomask |= HEAP_XMAX_EXCL_LOCK;
5456 140893 : break;
5457 95889 : case LockTupleExclusive:
5458 95889 : new_xmax = add_to_xmax;
5459 95889 : new_infomask |= HEAP_XMAX_EXCL_LOCK;
5460 95889 : new_infomask2 |= HEAP_KEYS_UPDATED;
5461 95889 : break;
5462 0 : default:
5463 0 : new_xmax = InvalidTransactionId; /* silence compiler */
5464 0 : elog(ERROR, "invalid lock mode");
5465 : }
5466 : }
5467 : }
5468 181042 : else if (old_infomask & HEAP_XMAX_IS_MULTI)
5469 : {
5470 : MultiXactStatus new_status;
5471 :
5472 : /*
5473 : * Currently we don't allow XMAX_COMMITTED to be set for multis, so
5474 : * cross-check.
5475 : */
5476 : Assert(!(old_infomask & HEAP_XMAX_COMMITTED));
5477 :
5478 : /*
5479 : * A multixact together with LOCK_ONLY set but neither lock bit set
5480 : * (i.e. a pg_upgraded share locked tuple) cannot possibly be running
5481 : * anymore. This check is critical for databases upgraded by
5482 : * pg_upgrade; both MultiXactIdIsRunning and MultiXactIdExpand assume
5483 : * that such multis are never passed.
5484 : */
5485 75558 : if (HEAP_LOCKED_UPGRADED(old_infomask))
5486 : {
5487 0 : old_infomask &= ~HEAP_XMAX_IS_MULTI;
5488 0 : old_infomask |= HEAP_XMAX_INVALID;
5489 0 : goto l5;
5490 : }
5491 :
5492 : /*
5493 : * If the XMAX is already a MultiXactId, then we need to expand it to
5494 : * include add_to_xmax; but if all the members were lockers and are
5495 : * all gone, we can do away with the IS_MULTI bit and just set
5496 : * add_to_xmax as the only locker/updater. If all lockers are gone
5497 : * and we have an updater that aborted, we can also do without a
5498 : * multi.
5499 : *
5500 : * The cost of doing GetMultiXactIdMembers would be paid by
5501 : * MultiXactIdExpand if we weren't to do this, so this check is not
5502 : * incurring extra work anyhow.
5503 : */
5504 75558 : if (!MultiXactIdIsRunning(xmax, HEAP_XMAX_IS_LOCKED_ONLY(old_infomask)))
5505 : {
5506 24 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) ||
5507 8 : !TransactionIdDidCommit(MultiXactIdGetUpdateXid(xmax,
5508 : old_infomask)))
5509 : {
5510 : /*
5511 : * Reset these bits and restart; otherwise fall through to
5512 : * create a new multi below.
5513 : */
5514 24 : old_infomask &= ~HEAP_XMAX_IS_MULTI;
5515 24 : old_infomask |= HEAP_XMAX_INVALID;
5516 24 : goto l5;
5517 : }
5518 : }
5519 :
5520 75534 : new_status = get_mxact_status_for_lock(mode, is_update);
5521 :
5522 75534 : new_xmax = MultiXactIdExpand((MultiXactId) xmax, add_to_xmax,
5523 : new_status);
5524 75534 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5525 : }
5526 105484 : else if (old_infomask & HEAP_XMAX_COMMITTED)
5527 : {
5528 : /*
5529 : * It's a committed update, so we need to preserve him as updater of
5530 : * the tuple.
5531 : */
5532 : MultiXactStatus status;
5533 : MultiXactStatus new_status;
5534 :
5535 13 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5536 0 : status = MultiXactStatusUpdate;
5537 : else
5538 13 : status = MultiXactStatusNoKeyUpdate;
5539 :
5540 13 : new_status = get_mxact_status_for_lock(mode, is_update);
5541 :
5542 : /*
5543 : * since it's not running, it's obviously impossible for the old
5544 : * updater to be identical to the current one, so we need not check
5545 : * for that case as we do in the block above.
5546 : */
5547 13 : new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5548 13 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5549 : }
5550 105471 : else if (TransactionIdIsInProgress(xmax))
5551 : {
5552 : /*
5553 : * If the XMAX is a valid, in-progress TransactionId, then we need to
5554 : * create a new MultiXactId that includes both the old locker or
5555 : * updater and our own TransactionId.
5556 : */
5557 : MultiXactStatus new_status;
5558 : MultiXactStatus old_status;
5559 : LockTupleMode old_mode;
5560 :
5561 105462 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
5562 : {
5563 105436 : if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
5564 5717 : old_status = MultiXactStatusForKeyShare;
5565 99719 : else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
5566 468 : old_status = MultiXactStatusForShare;
5567 99251 : else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
5568 : {
5569 99251 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5570 92800 : old_status = MultiXactStatusForUpdate;
5571 : else
5572 6451 : old_status = MultiXactStatusForNoKeyUpdate;
5573 : }
5574 : else
5575 : {
5576 : /*
5577 : * LOCK_ONLY can be present alone only when a page has been
5578 : * upgraded by pg_upgrade. But in that case,
5579 : * TransactionIdIsInProgress() should have returned false. We
5580 : * assume it's no longer locked in this case.
5581 : */
5582 0 : elog(WARNING, "LOCK_ONLY found for Xid in progress %u", xmax);
5583 0 : old_infomask |= HEAP_XMAX_INVALID;
5584 0 : old_infomask &= ~HEAP_XMAX_LOCK_ONLY;
5585 0 : goto l5;
5586 : }
5587 : }
5588 : else
5589 : {
5590 : /* it's an update, but which kind? */
5591 26 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5592 0 : old_status = MultiXactStatusUpdate;
5593 : else
5594 26 : old_status = MultiXactStatusNoKeyUpdate;
5595 : }
5596 :
5597 105462 : old_mode = TUPLOCK_from_mxstatus(old_status);
5598 :
5599 : /*
5600 : * If the lock to be acquired is for the same TransactionId as the
5601 : * existing lock, there's an optimization possible: consider only the
5602 : * strongest of both locks as the only one present, and restart.
5603 : */
5604 105462 : if (xmax == add_to_xmax)
5605 : {
5606 : /*
5607 : * Note that it's not possible for the original tuple to be
5608 : * updated: we wouldn't be here because the tuple would have been
5609 : * invisible and we wouldn't try to update it. As a subtlety,
5610 : * this code can also run when traversing an update chain to lock
5611 : * future versions of a tuple. But we wouldn't be here either,
5612 : * because the add_to_xmax would be different from the original
5613 : * updater.
5614 : */
5615 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
5616 :
5617 : /* acquire the strongest of both */
5618 104321 : if (mode < old_mode)
5619 52198 : mode = old_mode;
5620 : /* mustn't touch is_update */
5621 :
5622 104321 : old_infomask |= HEAP_XMAX_INVALID;
5623 104321 : goto l5;
5624 : }
5625 :
5626 : /* otherwise, just fall back to creating a new multixact */
5627 1141 : new_status = get_mxact_status_for_lock(mode, is_update);
5628 1141 : new_xmax = MultiXactIdCreate(xmax, old_status,
5629 : add_to_xmax, new_status);
5630 1141 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5631 : }
5632 14 : else if (!HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) &&
5633 5 : TransactionIdDidCommit(xmax))
5634 1 : {
5635 : /*
5636 : * It's a committed update, so we gotta preserve him as updater of the
5637 : * tuple.
5638 : */
5639 : MultiXactStatus status;
5640 : MultiXactStatus new_status;
5641 :
5642 1 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5643 0 : status = MultiXactStatusUpdate;
5644 : else
5645 1 : status = MultiXactStatusNoKeyUpdate;
5646 :
5647 1 : new_status = get_mxact_status_for_lock(mode, is_update);
5648 :
5649 : /*
5650 : * since it's not running, it's obviously impossible for the old
5651 : * updater to be identical to the current one, so we need not check
5652 : * for that case as we do in the block above.
5653 : */
5654 1 : new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5655 1 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5656 : }
5657 : else
5658 : {
5659 : /*
5660 : * Can get here iff the locking/updating transaction was running when
5661 : * the infomask was extracted from the tuple, but finished before
5662 : * TransactionIdIsInProgress got to run. Deal with it as if there was
5663 : * no locker at all in the first place.
5664 : */
5665 8 : old_infomask |= HEAP_XMAX_INVALID;
5666 8 : goto l5;
5667 : }
5668 :
5669 2498525 : *result_infomask = new_infomask;
5670 2498525 : *result_infomask2 = new_infomask2;
5671 2498525 : *result_xmax = new_xmax;
5672 2498525 : }
5673 :
5674 : /*
5675 : * Subroutine for heap_lock_updated_tuple_rec.
5676 : *
5677 : * Given a hypothetical multixact status held by the transaction identified
5678 : * with the given xid, does the current transaction need to wait, fail, or can
5679 : * it continue if it wanted to acquire a lock of the given mode? "needwait"
5680 : * is set to true if waiting is necessary; if it can continue, then TM_Ok is
5681 : * returned. If the lock is already held by the current transaction, return
5682 : * TM_SelfModified. In case of a conflict with another transaction, a
5683 : * different HeapTupleSatisfiesUpdate return code is returned.
5684 : *
5685 : * The held status is said to be hypothetical because it might correspond to a
5686 : * lock held by a single Xid, i.e. not a real MultiXactId; we express it this
5687 : * way for simplicity of API.
5688 : */
5689 : static TM_Result
5690 38774 : test_lockmode_for_conflict(MultiXactStatus status, TransactionId xid,
5691 : LockTupleMode mode, HeapTuple tup,
5692 : bool *needwait)
5693 : {
5694 : MultiXactStatus wantedstatus;
5695 :
5696 38774 : *needwait = false;
5697 38774 : wantedstatus = get_mxact_status_for_lock(mode, false);
5698 :
5699 : /*
5700 : * Note: we *must* check TransactionIdIsInProgress before
5701 : * TransactionIdDidAbort/Commit; see comment at top of heapam_visibility.c
5702 : * for an explanation.
5703 : */
5704 38774 : if (TransactionIdIsCurrentTransactionId(xid))
5705 : {
5706 : /*
5707 : * The tuple has already been locked by our own transaction. This is
5708 : * very rare but can happen if multiple transactions are trying to
5709 : * lock an ancient version of the same tuple.
5710 : */
5711 0 : return TM_SelfModified;
5712 : }
5713 38774 : else if (TransactionIdIsInProgress(xid))
5714 : {
5715 : /*
5716 : * If the locking transaction is running, what we do depends on
5717 : * whether the lock modes conflict: if they do, then we must wait for
5718 : * it to finish; otherwise we can fall through to lock this tuple
5719 : * version without waiting.
5720 : */
5721 36539 : if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
5722 36539 : LOCKMODE_from_mxstatus(wantedstatus)))
5723 : {
5724 8 : *needwait = true;
5725 : }
5726 :
5727 : /*
5728 : * If we set needwait above, then this value doesn't matter;
5729 : * otherwise, this value signals to caller that it's okay to proceed.
5730 : */
5731 36539 : return TM_Ok;
5732 : }
5733 2235 : else if (TransactionIdDidAbort(xid))
5734 206 : return TM_Ok;
5735 2029 : else if (TransactionIdDidCommit(xid))
5736 : {
5737 : /*
5738 : * The other transaction committed. If it was only a locker, then the
5739 : * lock is completely gone now and we can return success; but if it
5740 : * was an update, then what we do depends on whether the two lock
5741 : * modes conflict. If they conflict, then we must report error to
5742 : * caller. But if they don't, we can fall through to allow the current
5743 : * transaction to lock the tuple.
5744 : *
5745 : * Note: the reason we worry about ISUPDATE here is because as soon as
5746 : * a transaction ends, all its locks are gone and meaningless, and
5747 : * thus we can ignore them; whereas its updates persist. In the
5748 : * TransactionIdIsInProgress case, above, we don't need to check
5749 : * because we know the lock is still "alive" and thus a conflict needs
5750 : * always be checked.
5751 : */
5752 2029 : if (!ISUPDATE_from_mxstatus(status))
5753 2020 : return TM_Ok;
5754 :
5755 9 : if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
5756 9 : LOCKMODE_from_mxstatus(wantedstatus)))
5757 : {
5758 : /* bummer */
5759 8 : if (!ItemPointerEquals(&tup->t_self, &tup->t_data->t_ctid))
5760 6 : return TM_Updated;
5761 : else
5762 2 : return TM_Deleted;
5763 : }
5764 :
5765 1 : return TM_Ok;
5766 : }
5767 :
5768 : /* Not in progress, not aborted, not committed -- must have crashed */
5769 0 : return TM_Ok;
5770 : }
5771 :
5772 :
5773 : /*
5774 : * Recursive part of heap_lock_updated_tuple
5775 : *
5776 : * Fetch the tuple pointed to by tid in rel, and mark it as locked by the given
5777 : * xid with the given mode; if this tuple is updated, recurse to lock the new
5778 : * version as well.
5779 : */
5780 : static TM_Result
5781 2213 : heap_lock_updated_tuple_rec(Relation rel, TransactionId priorXmax,
5782 : const ItemPointerData *tid, TransactionId xid,
5783 : LockTupleMode mode)
5784 : {
5785 : TM_Result result;
5786 : ItemPointerData tupid;
5787 : HeapTupleData mytup;
5788 : Buffer buf;
5789 : uint16 new_infomask,
5790 : new_infomask2,
5791 : old_infomask,
5792 : old_infomask2;
5793 : TransactionId xmax,
5794 : new_xmax;
5795 2213 : bool cleared_all_frozen = false;
5796 : bool pinned_desired_page;
5797 2213 : Buffer vmbuffer = InvalidBuffer;
5798 : BlockNumber block;
5799 :
5800 2213 : ItemPointerCopy(tid, &tupid);
5801 :
5802 : for (;;)
5803 : {
5804 2216 : new_infomask = 0;
5805 2216 : new_xmax = InvalidTransactionId;
5806 2216 : block = ItemPointerGetBlockNumber(&tupid);
5807 2216 : ItemPointerCopy(&tupid, &(mytup.t_self));
5808 :
5809 2216 : if (!heap_fetch(rel, SnapshotAny, &mytup, &buf, false))
5810 : {
5811 : /*
5812 : * if we fail to find the updated version of the tuple, it's
5813 : * because it was vacuumed/pruned away after its creator
5814 : * transaction aborted. So behave as if we got to the end of the
5815 : * chain, and there's no further tuple to lock: return success to
5816 : * caller.
5817 : */
5818 0 : result = TM_Ok;
5819 0 : goto out_unlocked;
5820 : }
5821 :
5822 2216 : l4:
5823 2224 : CHECK_FOR_INTERRUPTS();
5824 :
5825 : /*
5826 : * Before locking the buffer, pin the visibility map page if it
5827 : * appears to be necessary. Since we haven't got the lock yet,
5828 : * someone else might be in the middle of changing this, so we'll need
5829 : * to recheck after we have the lock.
5830 : */
5831 2224 : if (PageIsAllVisible(BufferGetPage(buf)))
5832 : {
5833 0 : visibilitymap_pin(rel, block, &vmbuffer);
5834 0 : pinned_desired_page = true;
5835 : }
5836 : else
5837 2224 : pinned_desired_page = false;
5838 :
5839 2224 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
5840 :
5841 : /*
5842 : * If we didn't pin the visibility map page and the page has become
5843 : * all visible while we were busy locking the buffer, we'll have to
5844 : * unlock and re-lock, to avoid holding the buffer lock across I/O.
5845 : * That's a bit unfortunate, but hopefully shouldn't happen often.
5846 : *
5847 : * Note: in some paths through this function, we will reach here
5848 : * holding a pin on a vm page that may or may not be the one matching
5849 : * this page. If this page isn't all-visible, we won't use the vm
5850 : * page, but we hold onto such a pin till the end of the function.
5851 : */
5852 2224 : if (!pinned_desired_page && PageIsAllVisible(BufferGetPage(buf)))
5853 : {
5854 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5855 0 : visibilitymap_pin(rel, block, &vmbuffer);
5856 0 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
5857 : }
5858 :
5859 : /*
5860 : * Check the tuple XMIN against prior XMAX, if any. If we reached the
5861 : * end of the chain, we're done, so return success.
5862 : */
5863 4448 : if (TransactionIdIsValid(priorXmax) &&
5864 2224 : !TransactionIdEquals(HeapTupleHeaderGetXmin(mytup.t_data),
5865 : priorXmax))
5866 : {
5867 2 : result = TM_Ok;
5868 2 : goto out_locked;
5869 : }
5870 :
5871 : /*
5872 : * Also check Xmin: if this tuple was created by an aborted
5873 : * (sub)transaction, then we already locked the last live one in the
5874 : * chain, thus we're done, so return success.
5875 : */
5876 2222 : if (TransactionIdDidAbort(HeapTupleHeaderGetXmin(mytup.t_data)))
5877 : {
5878 25 : result = TM_Ok;
5879 25 : goto out_locked;
5880 : }
5881 :
5882 2197 : old_infomask = mytup.t_data->t_infomask;
5883 2197 : old_infomask2 = mytup.t_data->t_infomask2;
5884 2197 : xmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
5885 :
5886 : /*
5887 : * If this tuple version has been updated or locked by some concurrent
5888 : * transaction(s), what we do depends on whether our lock mode
5889 : * conflicts with what those other transactions hold, and also on the
5890 : * status of them.
5891 : */
5892 2197 : if (!(old_infomask & HEAP_XMAX_INVALID))
5893 : {
5894 : TransactionId rawxmax;
5895 : bool needwait;
5896 :
5897 2138 : rawxmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
5898 2138 : if (old_infomask & HEAP_XMAX_IS_MULTI)
5899 : {
5900 : int nmembers;
5901 : int i;
5902 : MultiXactMember *members;
5903 :
5904 : /*
5905 : * We don't need a test for pg_upgrade'd tuples: this is only
5906 : * applied to tuples after the first in an update chain. Said
5907 : * first tuple in the chain may well be locked-in-9.2-and-
5908 : * pg_upgraded, but that one was already locked by our caller,
5909 : * not us; and any subsequent ones cannot be because our
5910 : * caller must necessarily have obtained a snapshot later than
5911 : * the pg_upgrade itself.
5912 : */
5913 : Assert(!HEAP_LOCKED_UPGRADED(mytup.t_data->t_infomask));
5914 :
5915 2109 : nmembers = GetMultiXactIdMembers(rawxmax, &members, false,
5916 2109 : HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
5917 40854 : for (i = 0; i < nmembers; i++)
5918 : {
5919 38745 : result = test_lockmode_for_conflict(members[i].status,
5920 38745 : members[i].xid,
5921 : mode,
5922 : &mytup,
5923 : &needwait);
5924 :
5925 : /*
5926 : * If the tuple was already locked by ourselves in a
5927 : * previous iteration of this (say heap_lock_tuple was
5928 : * forced to restart the locking loop because of a change
5929 : * in xmax), then we hold the lock already on this tuple
5930 : * version and we don't need to do anything; and this is
5931 : * not an error condition either. We just need to skip
5932 : * this tuple and continue locking the next version in the
5933 : * update chain.
5934 : */
5935 38745 : if (result == TM_SelfModified)
5936 : {
5937 0 : pfree(members);
5938 0 : goto next;
5939 : }
5940 :
5941 38745 : if (needwait)
5942 : {
5943 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5944 0 : XactLockTableWait(members[i].xid, rel,
5945 : &mytup.t_self,
5946 : XLTW_LockUpdated);
5947 0 : pfree(members);
5948 0 : goto l4;
5949 : }
5950 38745 : if (result != TM_Ok)
5951 : {
5952 0 : pfree(members);
5953 0 : goto out_locked;
5954 : }
5955 : }
5956 2109 : if (members)
5957 2109 : pfree(members);
5958 : }
5959 : else
5960 : {
5961 : MultiXactStatus status;
5962 :
5963 : /*
5964 : * For a non-multi Xmax, we first need to compute the
5965 : * corresponding MultiXactStatus by using the infomask bits.
5966 : */
5967 29 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
5968 : {
5969 10 : if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
5970 10 : status = MultiXactStatusForKeyShare;
5971 0 : else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
5972 0 : status = MultiXactStatusForShare;
5973 0 : else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
5974 : {
5975 0 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5976 0 : status = MultiXactStatusForUpdate;
5977 : else
5978 0 : status = MultiXactStatusForNoKeyUpdate;
5979 : }
5980 : else
5981 : {
5982 : /*
5983 : * LOCK_ONLY present alone (a pg_upgraded tuple marked
5984 : * as share-locked in the old cluster) shouldn't be
5985 : * seen in the middle of an update chain.
5986 : */
5987 0 : elog(ERROR, "invalid lock status in tuple");
5988 : }
5989 : }
5990 : else
5991 : {
5992 : /* it's an update, but which kind? */
5993 19 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5994 14 : status = MultiXactStatusUpdate;
5995 : else
5996 5 : status = MultiXactStatusNoKeyUpdate;
5997 : }
5998 :
5999 29 : result = test_lockmode_for_conflict(status, rawxmax, mode,
6000 : &mytup, &needwait);
6001 :
6002 : /*
6003 : * If the tuple was already locked by ourselves in a previous
6004 : * iteration of this (say heap_lock_tuple was forced to
6005 : * restart the locking loop because of a change in xmax), then
6006 : * we hold the lock already on this tuple version and we don't
6007 : * need to do anything; and this is not an error condition
6008 : * either. We just need to skip this tuple and continue
6009 : * locking the next version in the update chain.
6010 : */
6011 29 : if (result == TM_SelfModified)
6012 0 : goto next;
6013 :
6014 29 : if (needwait)
6015 : {
6016 8 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
6017 8 : XactLockTableWait(rawxmax, rel, &mytup.t_self,
6018 : XLTW_LockUpdated);
6019 8 : goto l4;
6020 : }
6021 21 : if (result != TM_Ok)
6022 : {
6023 8 : goto out_locked;
6024 : }
6025 : }
6026 : }
6027 :
6028 : /* compute the new Xmax and infomask values for the tuple ... */
6029 2181 : compute_new_xmax_infomask(xmax, old_infomask, mytup.t_data->t_infomask2,
6030 : xid, mode, false,
6031 : &new_xmax, &new_infomask, &new_infomask2);
6032 :
6033 2181 : if (PageIsAllVisible(BufferGetPage(buf)) &&
6034 0 : visibilitymap_clear(rel, block, vmbuffer,
6035 : VISIBILITYMAP_ALL_FROZEN))
6036 0 : cleared_all_frozen = true;
6037 :
6038 2181 : START_CRIT_SECTION();
6039 :
6040 : /* ... and set them */
6041 2181 : HeapTupleHeaderSetXmax(mytup.t_data, new_xmax);
6042 2181 : mytup.t_data->t_infomask &= ~HEAP_XMAX_BITS;
6043 2181 : mytup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
6044 2181 : mytup.t_data->t_infomask |= new_infomask;
6045 2181 : mytup.t_data->t_infomask2 |= new_infomask2;
6046 :
6047 2181 : MarkBufferDirty(buf);
6048 :
6049 : /* XLOG stuff */
6050 2181 : if (RelationNeedsWAL(rel))
6051 : {
6052 : xl_heap_lock_updated xlrec;
6053 : XLogRecPtr recptr;
6054 2181 : Page page = BufferGetPage(buf);
6055 :
6056 2181 : XLogBeginInsert();
6057 2181 : XLogRegisterBuffer(0, buf, REGBUF_STANDARD);
6058 :
6059 2181 : xlrec.offnum = ItemPointerGetOffsetNumber(&mytup.t_self);
6060 2181 : xlrec.xmax = new_xmax;
6061 2181 : xlrec.infobits_set = compute_infobits(new_infomask, new_infomask2);
6062 2181 : xlrec.flags =
6063 2181 : cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
6064 :
6065 2181 : XLogRegisterData(&xlrec, SizeOfHeapLockUpdated);
6066 :
6067 2181 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_LOCK_UPDATED);
6068 :
6069 2181 : PageSetLSN(page, recptr);
6070 : }
6071 :
6072 2181 : END_CRIT_SECTION();
6073 :
6074 2181 : next:
6075 : /* if we find the end of update chain, we're done. */
6076 4362 : if (mytup.t_data->t_infomask & HEAP_XMAX_INVALID ||
6077 4362 : HeapTupleHeaderIndicatesMovedPartitions(mytup.t_data) ||
6078 2185 : ItemPointerEquals(&mytup.t_self, &mytup.t_data->t_ctid) ||
6079 4 : HeapTupleHeaderIsOnlyLocked(mytup.t_data))
6080 : {
6081 2178 : result = TM_Ok;
6082 2178 : goto out_locked;
6083 : }
6084 :
6085 : /* tail recursion */
6086 3 : priorXmax = HeapTupleHeaderGetUpdateXid(mytup.t_data);
6087 3 : ItemPointerCopy(&(mytup.t_data->t_ctid), &tupid);
6088 3 : UnlockReleaseBuffer(buf);
6089 : }
6090 :
6091 : result = TM_Ok;
6092 :
6093 2213 : out_locked:
6094 2213 : UnlockReleaseBuffer(buf);
6095 :
6096 2213 : out_unlocked:
6097 2213 : if (vmbuffer != InvalidBuffer)
6098 0 : ReleaseBuffer(vmbuffer);
6099 :
6100 2213 : return result;
6101 : }
6102 :
6103 : /*
6104 : * heap_lock_updated_tuple
6105 : * Follow update chain when locking an updated tuple, acquiring locks (row
6106 : * marks) on the updated versions.
6107 : *
6108 : * 'prior_infomask', 'prior_raw_xmax' and 'prior_ctid' are the corresponding
6109 : * fields from the initial tuple. We will lock the tuples starting from the
6110 : * one that 'prior_ctid' points to. Note: This function does not lock the
6111 : * initial tuple itself.
6112 : *
6113 : * This function doesn't check visibility, it just unconditionally marks the
6114 : * tuple(s) as locked. If any tuple in the updated chain is being deleted
6115 : * concurrently (or updated with the key being modified), sleep until the
6116 : * transaction doing it is finished.
6117 : *
6118 : * Note that we don't acquire heavyweight tuple locks on the tuples we walk
6119 : * when we have to wait for other transactions to release them, as opposed to
6120 : * what heap_lock_tuple does. The reason is that having more than one
6121 : * transaction walking the chain is probably uncommon enough that risk of
6122 : * starvation is not likely: one of the preconditions for being here is that
6123 : * the snapshot in use predates the update that created this tuple (because we
6124 : * started at an earlier version of the tuple), but at the same time such a
6125 : * transaction cannot be using repeatable read or serializable isolation
6126 : * levels, because that would lead to a serializability failure.
6127 : */
6128 : static TM_Result
6129 2215 : heap_lock_updated_tuple(Relation rel,
6130 : uint16 prior_infomask,
6131 : TransactionId prior_raw_xmax,
6132 : const ItemPointerData *prior_ctid,
6133 : TransactionId xid, LockTupleMode mode)
6134 : {
6135 2215 : INJECTION_POINT("heap_lock_updated_tuple", NULL);
6136 :
6137 : /*
6138 : * If the tuple has moved into another partition (effectively a delete)
6139 : * stop here.
6140 : */
6141 2215 : if (!ItemPointerIndicatesMovedPartitions(prior_ctid))
6142 : {
6143 : TransactionId prior_xmax;
6144 :
6145 : /*
6146 : * If this is the first possibly-multixact-able operation in the
6147 : * current transaction, set my per-backend OldestMemberMXactId
6148 : * setting. We can be certain that the transaction will never become a
6149 : * member of any older MultiXactIds than that. (We have to do this
6150 : * even if we end up just using our own TransactionId below, since
6151 : * some other backend could incorporate our XID into a MultiXact
6152 : * immediately afterwards.)
6153 : */
6154 2213 : MultiXactIdSetOldestMember();
6155 :
6156 4426 : prior_xmax = (prior_infomask & HEAP_XMAX_IS_MULTI) ?
6157 2213 : MultiXactIdGetUpdateXid(prior_raw_xmax, prior_infomask) : prior_raw_xmax;
6158 2213 : return heap_lock_updated_tuple_rec(rel, prior_xmax, prior_ctid, xid, mode);
6159 : }
6160 :
6161 : /* nothing to lock */
6162 2 : return TM_Ok;
6163 : }
6164 :
6165 : /*
6166 : * heap_finish_speculative - mark speculative insertion as successful
6167 : *
6168 : * To successfully finish a speculative insertion we have to clear speculative
6169 : * token from tuple. To do so the t_ctid field, which will contain a
6170 : * speculative token value, is modified in place to point to the tuple itself,
6171 : * which is characteristic of a newly inserted ordinary tuple.
6172 : *
6173 : * NB: It is not ok to commit without either finishing or aborting a
6174 : * speculative insertion. We could treat speculative tuples of committed
6175 : * transactions implicitly as completed, but then we would have to be prepared
6176 : * to deal with speculative tokens on committed tuples. That wouldn't be
6177 : * difficult - no-one looks at the ctid field of a tuple with invalid xmax -
6178 : * but clearing the token at completion isn't very expensive either.
6179 : * An explicit confirmation WAL record also makes logical decoding simpler.
6180 : */
6181 : void
6182 2215 : heap_finish_speculative(Relation relation, const ItemPointerData *tid)
6183 : {
6184 : Buffer buffer;
6185 : Page page;
6186 : OffsetNumber offnum;
6187 : ItemId lp;
6188 : HeapTupleHeader htup;
6189 :
6190 2215 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
6191 2215 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6192 2215 : page = BufferGetPage(buffer);
6193 :
6194 2215 : offnum = ItemPointerGetOffsetNumber(tid);
6195 2215 : if (offnum < 1 || offnum > PageGetMaxOffsetNumber(page))
6196 0 : elog(ERROR, "offnum out of range");
6197 2215 : lp = PageGetItemId(page, offnum);
6198 2215 : if (!ItemIdIsNormal(lp))
6199 0 : elog(ERROR, "invalid lp");
6200 :
6201 2215 : htup = (HeapTupleHeader) PageGetItem(page, lp);
6202 :
6203 : /* NO EREPORT(ERROR) from here till changes are logged */
6204 2215 : START_CRIT_SECTION();
6205 :
6206 : Assert(HeapTupleHeaderIsSpeculative(htup));
6207 :
6208 2215 : MarkBufferDirty(buffer);
6209 :
6210 : /*
6211 : * Replace the speculative insertion token with a real t_ctid, pointing to
6212 : * itself like it does on regular tuples.
6213 : */
6214 2215 : htup->t_ctid = *tid;
6215 :
6216 : /* XLOG stuff */
6217 2215 : if (RelationNeedsWAL(relation))
6218 : {
6219 : xl_heap_confirm xlrec;
6220 : XLogRecPtr recptr;
6221 :
6222 2195 : xlrec.offnum = ItemPointerGetOffsetNumber(tid);
6223 :
6224 2195 : XLogBeginInsert();
6225 :
6226 : /* We want the same filtering on this as on a plain insert */
6227 2195 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
6228 :
6229 2195 : XLogRegisterData(&xlrec, SizeOfHeapConfirm);
6230 2195 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
6231 :
6232 2195 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_CONFIRM);
6233 :
6234 2195 : PageSetLSN(page, recptr);
6235 : }
6236 :
6237 2215 : END_CRIT_SECTION();
6238 :
6239 2215 : UnlockReleaseBuffer(buffer);
6240 2215 : }
6241 :
6242 : /*
6243 : * heap_abort_speculative - kill a speculatively inserted tuple
6244 : *
6245 : * Marks a tuple that was speculatively inserted in the same command as dead,
6246 : * by setting its xmin as invalid. That makes it immediately appear as dead
6247 : * to all transactions, including our own. In particular, it makes
6248 : * HeapTupleSatisfiesDirty() regard the tuple as dead, so that another backend
6249 : * inserting a duplicate key value won't unnecessarily wait for our whole
6250 : * transaction to finish (it'll just wait for our speculative insertion to
6251 : * finish).
6252 : *
6253 : * Killing the tuple prevents "unprincipled deadlocks", which are deadlocks
6254 : * that arise due to a mutual dependency that is not user visible. By
6255 : * definition, unprincipled deadlocks cannot be prevented by the user
6256 : * reordering lock acquisition in client code, because the implementation level
6257 : * lock acquisitions are not under the user's direct control. If speculative
6258 : * inserters did not take this precaution, then under high concurrency they
6259 : * could deadlock with each other, which would not be acceptable.
6260 : *
6261 : * This is somewhat redundant with heap_delete, but we prefer to have a
6262 : * dedicated routine with stripped down requirements. Note that this is also
6263 : * used to delete the TOAST tuples created during speculative insertion.
6264 : *
6265 : * This routine does not affect logical decoding as it only looks at
6266 : * confirmation records.
6267 : */
6268 : void
6269 16 : heap_abort_speculative(Relation relation, const ItemPointerData *tid)
6270 : {
6271 16 : TransactionId xid = GetCurrentTransactionId();
6272 : ItemId lp;
6273 : HeapTupleData tp;
6274 : Page page;
6275 : BlockNumber block;
6276 : Buffer buffer;
6277 :
6278 : Assert(ItemPointerIsValid(tid));
6279 :
6280 16 : block = ItemPointerGetBlockNumber(tid);
6281 16 : buffer = ReadBuffer(relation, block);
6282 16 : page = BufferGetPage(buffer);
6283 :
6284 16 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6285 :
6286 : /*
6287 : * Page can't be all visible, we just inserted into it, and are still
6288 : * running.
6289 : */
6290 : Assert(!PageIsAllVisible(page));
6291 :
6292 16 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
6293 : Assert(ItemIdIsNormal(lp));
6294 :
6295 16 : tp.t_tableOid = RelationGetRelid(relation);
6296 16 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
6297 16 : tp.t_len = ItemIdGetLength(lp);
6298 16 : tp.t_self = *tid;
6299 :
6300 : /*
6301 : * Sanity check that the tuple really is a speculatively inserted tuple,
6302 : * inserted by us.
6303 : */
6304 16 : if (tp.t_data->t_choice.t_heap.t_xmin != xid)
6305 0 : elog(ERROR, "attempted to kill a tuple inserted by another transaction");
6306 16 : if (!(IsToastRelation(relation) || HeapTupleHeaderIsSpeculative(tp.t_data)))
6307 0 : elog(ERROR, "attempted to kill a non-speculative tuple");
6308 : Assert(!HeapTupleHeaderIsHeapOnly(tp.t_data));
6309 :
6310 : /*
6311 : * No need to check for serializable conflicts here. There is never a
6312 : * need for a combo CID, either. No need to extract replica identity, or
6313 : * do anything special with infomask bits.
6314 : */
6315 :
6316 16 : START_CRIT_SECTION();
6317 :
6318 : /*
6319 : * The tuple will become DEAD immediately. Flag that this page is a
6320 : * candidate for pruning by setting xmin to TransactionXmin. While not
6321 : * immediately prunable, it is the oldest xid we can cheaply determine
6322 : * that's safe against wraparound / being older than the table's
6323 : * relfrozenxid. To defend against the unlikely case of a new relation
6324 : * having a newer relfrozenxid than our TransactionXmin, use relfrozenxid
6325 : * if so (vacuum can't subsequently move relfrozenxid to beyond
6326 : * TransactionXmin, so there's no race here).
6327 : */
6328 : Assert(TransactionIdIsValid(TransactionXmin));
6329 : {
6330 16 : TransactionId relfrozenxid = relation->rd_rel->relfrozenxid;
6331 : TransactionId prune_xid;
6332 :
6333 16 : if (TransactionIdPrecedes(TransactionXmin, relfrozenxid))
6334 0 : prune_xid = relfrozenxid;
6335 : else
6336 16 : prune_xid = TransactionXmin;
6337 16 : PageSetPrunable(page, prune_xid);
6338 : }
6339 :
6340 : /* store transaction information of xact deleting the tuple */
6341 16 : tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
6342 16 : tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
6343 :
6344 : /*
6345 : * Set the tuple header xmin to InvalidTransactionId. This makes the
6346 : * tuple immediately invisible everyone. (In particular, to any
6347 : * transactions waiting on the speculative token, woken up later.)
6348 : */
6349 16 : HeapTupleHeaderSetXmin(tp.t_data, InvalidTransactionId);
6350 :
6351 : /* Clear the speculative insertion token too */
6352 16 : tp.t_data->t_ctid = tp.t_self;
6353 :
6354 16 : MarkBufferDirty(buffer);
6355 :
6356 : /*
6357 : * XLOG stuff
6358 : *
6359 : * The WAL records generated here match heap_delete(). The same recovery
6360 : * routines are used.
6361 : */
6362 16 : if (RelationNeedsWAL(relation))
6363 : {
6364 : xl_heap_delete xlrec;
6365 : XLogRecPtr recptr;
6366 :
6367 12 : xlrec.flags = XLH_DELETE_IS_SUPER;
6368 24 : xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
6369 12 : tp.t_data->t_infomask2);
6370 12 : xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
6371 12 : xlrec.xmax = xid;
6372 :
6373 12 : XLogBeginInsert();
6374 12 : XLogRegisterData(&xlrec, SizeOfHeapDelete);
6375 12 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
6376 :
6377 : /* No replica identity & replication origin logged */
6378 :
6379 12 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
6380 :
6381 12 : PageSetLSN(page, recptr);
6382 : }
6383 :
6384 16 : END_CRIT_SECTION();
6385 :
6386 16 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6387 :
6388 16 : if (HeapTupleHasExternal(&tp))
6389 : {
6390 : Assert(!IsToastRelation(relation));
6391 1 : heap_toast_delete(relation, &tp, true);
6392 : }
6393 :
6394 : /*
6395 : * Never need to mark tuple for invalidation, since catalogs don't support
6396 : * speculative insertion
6397 : */
6398 :
6399 : /* Now we can release the buffer */
6400 16 : ReleaseBuffer(buffer);
6401 :
6402 : /* count deletion, as we counted the insertion too */
6403 16 : pgstat_count_heap_delete(relation);
6404 16 : }
6405 :
6406 : /*
6407 : * heap_inplace_lock - protect inplace update from concurrent heap_update()
6408 : *
6409 : * Evaluate whether the tuple's state is compatible with a no-key update.
6410 : * Current transaction rowmarks are fine, as is KEY SHARE from any
6411 : * transaction. If compatible, return true with the buffer exclusive-locked,
6412 : * and the caller must release that by calling
6413 : * heap_inplace_update_and_unlock(), calling heap_inplace_unlock(), or raising
6414 : * an error. Otherwise, call release_callback(arg), wait for blocking
6415 : * transactions to end, and return false.
6416 : *
6417 : * Since this is intended for system catalogs and SERIALIZABLE doesn't cover
6418 : * DDL, this doesn't guarantee any particular predicate locking.
6419 : *
6420 : * heap_delete() is a rarer source of blocking transactions (xwait). We'll
6421 : * wait for such a transaction just like for the normal heap_update() case.
6422 : * Normal concurrent DROP commands won't cause that, because all inplace
6423 : * updaters take some lock that conflicts with DROP. An explicit SQL "DELETE
6424 : * FROM pg_class" can cause it. By waiting, if the concurrent transaction
6425 : * executed both "DELETE FROM pg_class" and "INSERT INTO pg_class", our caller
6426 : * can find the successor tuple.
6427 : *
6428 : * Readers of inplace-updated fields expect changes to those fields are
6429 : * durable. For example, vac_truncate_clog() reads datfrozenxid from
6430 : * pg_database tuples via catalog snapshots. A future snapshot must not
6431 : * return a lower datfrozenxid for the same database OID (lower in the
6432 : * FullTransactionIdPrecedes() sense). We achieve that since no update of a
6433 : * tuple can start while we hold a lock on its buffer. In cases like
6434 : * BEGIN;GRANT;CREATE INDEX;COMMIT we're inplace-updating a tuple visible only
6435 : * to this transaction. ROLLBACK then is one case where it's okay to lose
6436 : * inplace updates. (Restoring relhasindex=false on ROLLBACK is fine, since
6437 : * any concurrent CREATE INDEX would have blocked, then inplace-updated the
6438 : * committed tuple.)
6439 : *
6440 : * In principle, we could avoid waiting by overwriting every tuple in the
6441 : * updated tuple chain. Reader expectations permit updating a tuple only if
6442 : * it's aborted, is the tail of the chain, or we already updated the tuple
6443 : * referenced in its t_ctid. Hence, we would need to overwrite the tuples in
6444 : * order from tail to head. That would imply either (a) mutating all tuples
6445 : * in one critical section or (b) accepting a chance of partial completion.
6446 : * Partial completion of a relfrozenxid update would have the weird
6447 : * consequence that the table's next VACUUM could see the table's relfrozenxid
6448 : * move forward between vacuum_get_cutoffs() and finishing.
6449 : */
6450 : bool
6451 214226 : heap_inplace_lock(Relation relation,
6452 : HeapTuple oldtup_ptr, Buffer buffer,
6453 : void (*release_callback) (void *), void *arg)
6454 : {
6455 214226 : HeapTupleData oldtup = *oldtup_ptr; /* minimize diff vs. heap_update() */
6456 : TM_Result result;
6457 : bool ret;
6458 :
6459 : #ifdef USE_ASSERT_CHECKING
6460 : if (RelationGetRelid(relation) == RelationRelationId)
6461 : check_inplace_rel_lock(oldtup_ptr);
6462 : #endif
6463 :
6464 : Assert(BufferIsValid(buffer));
6465 :
6466 : /*
6467 : * Register shared cache invals if necessary. Other sessions may finish
6468 : * inplace updates of this tuple between this step and LockTuple(). Since
6469 : * inplace updates don't change cache keys, that's harmless.
6470 : *
6471 : * While it's tempting to register invals only after confirming we can
6472 : * return true, the following obstacle precludes reordering steps that
6473 : * way. Registering invals might reach a CatalogCacheInitializeCache()
6474 : * that locks "buffer". That would hang indefinitely if running after our
6475 : * own LockBuffer(). Hence, we must register invals before LockBuffer().
6476 : */
6477 214226 : CacheInvalidateHeapTupleInplace(relation, oldtup_ptr);
6478 :
6479 214226 : LockTuple(relation, &oldtup.t_self, InplaceUpdateTupleLock);
6480 214226 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6481 :
6482 : /*----------
6483 : * Interpret HeapTupleSatisfiesUpdate() like heap_update() does, except:
6484 : *
6485 : * - wait unconditionally
6486 : * - already locked tuple above, since inplace needs that unconditionally
6487 : * - don't recheck header after wait: simpler to defer to next iteration
6488 : * - don't try to continue even if the updater aborts: likewise
6489 : * - no crosscheck
6490 : */
6491 214226 : result = HeapTupleSatisfiesUpdate(&oldtup, GetCurrentCommandId(false),
6492 : buffer);
6493 :
6494 214226 : if (result == TM_Invisible)
6495 : {
6496 : /* no known way this can happen */
6497 0 : ereport(ERROR,
6498 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6499 : errmsg_internal("attempted to overwrite invisible tuple")));
6500 : }
6501 214226 : else if (result == TM_SelfModified)
6502 : {
6503 : /*
6504 : * CREATE INDEX might reach this if an expression is silly enough to
6505 : * call e.g. SELECT ... FROM pg_class FOR SHARE. C code of other SQL
6506 : * statements might get here after a heap_update() of the same row, in
6507 : * the absence of an intervening CommandCounterIncrement().
6508 : */
6509 0 : ereport(ERROR,
6510 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6511 : errmsg("tuple to be updated was already modified by an operation triggered by the current command")));
6512 : }
6513 214226 : else if (result == TM_BeingModified)
6514 : {
6515 : TransactionId xwait;
6516 : uint16 infomask;
6517 :
6518 30 : xwait = HeapTupleHeaderGetRawXmax(oldtup.t_data);
6519 30 : infomask = oldtup.t_data->t_infomask;
6520 :
6521 30 : if (infomask & HEAP_XMAX_IS_MULTI)
6522 : {
6523 5 : LockTupleMode lockmode = LockTupleNoKeyExclusive;
6524 5 : MultiXactStatus mxact_status = MultiXactStatusNoKeyUpdate;
6525 : int remain;
6526 :
6527 5 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
6528 : lockmode, NULL))
6529 : {
6530 2 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6531 2 : release_callback(arg);
6532 2 : ret = false;
6533 2 : MultiXactIdWait((MultiXactId) xwait, mxact_status, infomask,
6534 : relation, &oldtup.t_self, XLTW_Update,
6535 : &remain);
6536 : }
6537 : else
6538 3 : ret = true;
6539 : }
6540 25 : else if (TransactionIdIsCurrentTransactionId(xwait))
6541 1 : ret = true;
6542 24 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask))
6543 1 : ret = true;
6544 : else
6545 : {
6546 23 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6547 23 : release_callback(arg);
6548 23 : ret = false;
6549 23 : XactLockTableWait(xwait, relation, &oldtup.t_self,
6550 : XLTW_Update);
6551 : }
6552 : }
6553 : else
6554 : {
6555 214196 : ret = (result == TM_Ok);
6556 214196 : if (!ret)
6557 : {
6558 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6559 0 : release_callback(arg);
6560 : }
6561 : }
6562 :
6563 : /*
6564 : * GetCatalogSnapshot() relies on invalidation messages to know when to
6565 : * take a new snapshot. COMMIT of xwait is responsible for sending the
6566 : * invalidation. We're not acquiring heavyweight locks sufficient to
6567 : * block if not yet sent, so we must take a new snapshot to ensure a later
6568 : * attempt has a fair chance. While we don't need this if xwait aborted,
6569 : * don't bother optimizing that.
6570 : */
6571 214226 : if (!ret)
6572 : {
6573 25 : UnlockTuple(relation, &oldtup.t_self, InplaceUpdateTupleLock);
6574 25 : ForgetInplace_Inval();
6575 25 : InvalidateCatalogSnapshot();
6576 : }
6577 214226 : return ret;
6578 : }
6579 :
6580 : /*
6581 : * heap_inplace_update_and_unlock - core of systable_inplace_update_finish
6582 : *
6583 : * The tuple cannot change size, and therefore its header fields and null
6584 : * bitmap (if any) don't change either.
6585 : *
6586 : * Since we hold LOCKTAG_TUPLE, no updater has a local copy of this tuple.
6587 : */
6588 : void
6589 95748 : heap_inplace_update_and_unlock(Relation relation,
6590 : HeapTuple oldtup, HeapTuple tuple,
6591 : Buffer buffer)
6592 : {
6593 95748 : HeapTupleHeader htup = oldtup->t_data;
6594 : uint32 oldlen;
6595 : uint32 newlen;
6596 : char *dst;
6597 : char *src;
6598 95748 : int nmsgs = 0;
6599 95748 : SharedInvalidationMessage *invalMessages = NULL;
6600 95748 : bool RelcacheInitFileInval = false;
6601 :
6602 : Assert(ItemPointerEquals(&oldtup->t_self, &tuple->t_self));
6603 95748 : oldlen = oldtup->t_len - htup->t_hoff;
6604 95748 : newlen = tuple->t_len - tuple->t_data->t_hoff;
6605 95748 : if (oldlen != newlen || htup->t_hoff != tuple->t_data->t_hoff)
6606 0 : elog(ERROR, "wrong tuple length");
6607 :
6608 95748 : dst = (char *) htup + htup->t_hoff;
6609 95748 : src = (char *) tuple->t_data + tuple->t_data->t_hoff;
6610 :
6611 : /* Like RecordTransactionCommit(), log only if needed */
6612 95748 : if (XLogStandbyInfoActive())
6613 60727 : nmsgs = inplaceGetInvalidationMessages(&invalMessages,
6614 : &RelcacheInitFileInval);
6615 :
6616 : /*
6617 : * Unlink relcache init files as needed. If unlinking, acquire
6618 : * RelCacheInitLock until after associated invalidations. By doing this
6619 : * in advance, if we checkpoint and then crash between inplace
6620 : * XLogInsert() and inval, we don't rely on StartupXLOG() ->
6621 : * RelationCacheInitFileRemove(). That uses elevel==LOG, so replay would
6622 : * neglect to PANIC on EIO.
6623 : */
6624 95748 : PreInplace_Inval();
6625 :
6626 : /*----------
6627 : * NO EREPORT(ERROR) from here till changes are complete
6628 : *
6629 : * Our exclusive buffer lock won't stop a reader having already pinned and
6630 : * checked visibility for this tuple. With the usual order of changes
6631 : * (i.e. updating the buffer contents before WAL logging), a reader could
6632 : * observe our not-yet-persistent update to relfrozenxid and update
6633 : * datfrozenxid based on that. A crash in that moment could allow
6634 : * datfrozenxid to overtake relfrozenxid:
6635 : *
6636 : * ["D" is a VACUUM (ONLY_DATABASE_STATS)]
6637 : * ["R" is a VACUUM tbl]
6638 : * D: vac_update_datfrozenxid() -> systable_beginscan(pg_class)
6639 : * D: systable_getnext() returns pg_class tuple of tbl
6640 : * R: memcpy() into pg_class tuple of tbl
6641 : * D: raise pg_database.datfrozenxid, XLogInsert(), finish
6642 : * [crash]
6643 : * [recovery restores datfrozenxid w/o relfrozenxid]
6644 : *
6645 : * We avoid that by using a temporary copy of the buffer to hide our
6646 : * change from other backends until the change has been WAL-logged. We
6647 : * apply our change to the temporary copy and WAL-log it, before modifying
6648 : * the real page. That way any action a reader of the in-place-updated
6649 : * value takes will be WAL logged after this change.
6650 : */
6651 95748 : START_CRIT_SECTION();
6652 :
6653 95748 : MarkBufferDirty(buffer);
6654 :
6655 : /* XLOG stuff */
6656 95748 : if (RelationNeedsWAL(relation))
6657 : {
6658 : xl_heap_inplace xlrec;
6659 : PGAlignedBlock copied_buffer;
6660 95740 : char *origdata = (char *) BufferGetBlock(buffer);
6661 95740 : Page page = BufferGetPage(buffer);
6662 95740 : uint16 lower = ((PageHeader) page)->pd_lower;
6663 95740 : uint16 upper = ((PageHeader) page)->pd_upper;
6664 : uintptr_t dst_offset_in_block;
6665 : RelFileLocator rlocator;
6666 : ForkNumber forkno;
6667 : BlockNumber blkno;
6668 : XLogRecPtr recptr;
6669 :
6670 95740 : xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
6671 95740 : xlrec.dbId = MyDatabaseId;
6672 95740 : xlrec.tsId = MyDatabaseTableSpace;
6673 95740 : xlrec.relcacheInitFileInval = RelcacheInitFileInval;
6674 95740 : xlrec.nmsgs = nmsgs;
6675 :
6676 95740 : XLogBeginInsert();
6677 95740 : XLogRegisterData(&xlrec, MinSizeOfHeapInplace);
6678 95740 : if (nmsgs != 0)
6679 44662 : XLogRegisterData(invalMessages,
6680 : nmsgs * sizeof(SharedInvalidationMessage));
6681 :
6682 : /* register block matching what buffer will look like after changes */
6683 95740 : memcpy(copied_buffer.data, origdata, lower);
6684 95740 : memcpy(copied_buffer.data + upper, origdata + upper, BLCKSZ - upper);
6685 95740 : dst_offset_in_block = dst - origdata;
6686 95740 : memcpy(copied_buffer.data + dst_offset_in_block, src, newlen);
6687 95740 : BufferGetTag(buffer, &rlocator, &forkno, &blkno);
6688 : Assert(forkno == MAIN_FORKNUM);
6689 95740 : XLogRegisterBlock(0, &rlocator, forkno, blkno, copied_buffer.data,
6690 : REGBUF_STANDARD);
6691 95740 : XLogRegisterBufData(0, src, newlen);
6692 :
6693 : /* inplace updates aren't decoded atm, don't log the origin */
6694 :
6695 95740 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_INPLACE);
6696 :
6697 95740 : PageSetLSN(page, recptr);
6698 : }
6699 :
6700 95748 : memcpy(dst, src, newlen);
6701 :
6702 95748 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6703 :
6704 : /*
6705 : * Send invalidations to shared queue. SearchSysCacheLocked1() assumes we
6706 : * do this before UnlockTuple().
6707 : */
6708 95748 : AtInplace_Inval();
6709 :
6710 95748 : END_CRIT_SECTION();
6711 95748 : UnlockTuple(relation, &tuple->t_self, InplaceUpdateTupleLock);
6712 :
6713 95748 : AcceptInvalidationMessages(); /* local processing of just-sent inval */
6714 :
6715 : /*
6716 : * Queue a transactional inval, for logical decoding and for third-party
6717 : * code that might have been relying on it since long before inplace
6718 : * update adopted immediate invalidation. See README.tuplock section
6719 : * "Reading inplace-updated columns" for logical decoding details.
6720 : */
6721 95748 : if (!IsBootstrapProcessingMode())
6722 79683 : CacheInvalidateHeapTuple(relation, tuple, NULL);
6723 95748 : }
6724 :
6725 : /*
6726 : * heap_inplace_unlock - reverse of heap_inplace_lock
6727 : */
6728 : void
6729 118453 : heap_inplace_unlock(Relation relation,
6730 : HeapTuple oldtup, Buffer buffer)
6731 : {
6732 118453 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6733 118453 : UnlockTuple(relation, &oldtup->t_self, InplaceUpdateTupleLock);
6734 118453 : ForgetInplace_Inval();
6735 118453 : }
6736 :
6737 : #define FRM_NOOP 0x0001
6738 : #define FRM_INVALIDATE_XMAX 0x0002
6739 : #define FRM_RETURN_IS_XID 0x0004
6740 : #define FRM_RETURN_IS_MULTI 0x0008
6741 : #define FRM_MARK_COMMITTED 0x0010
6742 :
6743 : /*
6744 : * FreezeMultiXactId
6745 : * Determine what to do during freezing when a tuple is marked by a
6746 : * MultiXactId.
6747 : *
6748 : * "flags" is an output value; it's used to tell caller what to do on return.
6749 : * "pagefrz" is an input/output value, used to manage page level freezing.
6750 : *
6751 : * Possible values that we can set in "flags":
6752 : * FRM_NOOP
6753 : * don't do anything -- keep existing Xmax
6754 : * FRM_INVALIDATE_XMAX
6755 : * mark Xmax as InvalidTransactionId and set XMAX_INVALID flag.
6756 : * FRM_RETURN_IS_XID
6757 : * The Xid return value is a single update Xid to set as xmax.
6758 : * FRM_MARK_COMMITTED
6759 : * Xmax can be marked as HEAP_XMAX_COMMITTED
6760 : * FRM_RETURN_IS_MULTI
6761 : * The return value is a new MultiXactId to set as new Xmax.
6762 : * (caller must obtain proper infomask bits using GetMultiXactIdHintBits)
6763 : *
6764 : * Caller delegates control of page freezing to us. In practice we always
6765 : * force freezing of caller's page unless FRM_NOOP processing is indicated.
6766 : * We help caller ensure that XIDs < FreezeLimit and MXIDs < MultiXactCutoff
6767 : * can never be left behind. We freely choose when and how to process each
6768 : * Multi, without ever violating the cutoff postconditions for freezing.
6769 : *
6770 : * It's useful to remove Multis on a proactive timeline (relative to freezing
6771 : * XIDs) to keep MultiXact member SLRU buffer misses to a minimum. It can also
6772 : * be cheaper in the short run, for us, since we too can avoid SLRU buffer
6773 : * misses through eager processing.
6774 : *
6775 : * NB: Creates a _new_ MultiXactId when FRM_RETURN_IS_MULTI is set, though only
6776 : * when FreezeLimit and/or MultiXactCutoff cutoffs leave us with no choice.
6777 : * This can usually be put off, which is usually enough to avoid it altogether.
6778 : * Allocating new multis during VACUUM should be avoided on general principle;
6779 : * only VACUUM can advance relminmxid, so allocating new Multis here comes with
6780 : * its own special risks.
6781 : *
6782 : * NB: Caller must maintain "no freeze" NewRelfrozenXid/NewRelminMxid trackers
6783 : * using heap_tuple_should_freeze when we haven't forced page-level freezing.
6784 : *
6785 : * NB: Caller should avoid needlessly calling heap_tuple_should_freeze when we
6786 : * have already forced page-level freezing, since that might incur the same
6787 : * SLRU buffer misses that we specifically intended to avoid by freezing.
6788 : */
6789 : static TransactionId
6790 6 : FreezeMultiXactId(MultiXactId multi, uint16 t_infomask,
6791 : const struct VacuumCutoffs *cutoffs, uint16 *flags,
6792 : HeapPageFreeze *pagefrz)
6793 : {
6794 : TransactionId newxmax;
6795 : MultiXactMember *members;
6796 : int nmembers;
6797 : bool need_replace;
6798 : int nnewmembers;
6799 : MultiXactMember *newmembers;
6800 : bool has_lockers;
6801 : TransactionId update_xid;
6802 : bool update_committed;
6803 : TransactionId FreezePageRelfrozenXid;
6804 :
6805 6 : *flags = 0;
6806 :
6807 : /* We should only be called in Multis */
6808 : Assert(t_infomask & HEAP_XMAX_IS_MULTI);
6809 :
6810 12 : if (!MultiXactIdIsValid(multi) ||
6811 6 : HEAP_LOCKED_UPGRADED(t_infomask))
6812 : {
6813 0 : *flags |= FRM_INVALIDATE_XMAX;
6814 0 : pagefrz->freeze_required = true;
6815 0 : return InvalidTransactionId;
6816 : }
6817 6 : else if (MultiXactIdPrecedes(multi, cutoffs->relminmxid))
6818 0 : ereport(ERROR,
6819 : (errcode(ERRCODE_DATA_CORRUPTED),
6820 : errmsg_internal("found multixact %u from before relminmxid %u",
6821 : multi, cutoffs->relminmxid)));
6822 6 : else if (MultiXactIdPrecedes(multi, cutoffs->OldestMxact))
6823 : {
6824 : TransactionId update_xact;
6825 :
6826 : /*
6827 : * This old multi cannot possibly have members still running, but
6828 : * verify just in case. If it was a locker only, it can be removed
6829 : * without any further consideration; but if it contained an update,
6830 : * we might need to preserve it.
6831 : */
6832 4 : if (MultiXactIdIsRunning(multi,
6833 4 : HEAP_XMAX_IS_LOCKED_ONLY(t_infomask)))
6834 0 : ereport(ERROR,
6835 : (errcode(ERRCODE_DATA_CORRUPTED),
6836 : errmsg_internal("multixact %u from before multi freeze cutoff %u found to be still running",
6837 : multi, cutoffs->OldestMxact)));
6838 :
6839 4 : if (HEAP_XMAX_IS_LOCKED_ONLY(t_infomask))
6840 : {
6841 4 : *flags |= FRM_INVALIDATE_XMAX;
6842 4 : pagefrz->freeze_required = true;
6843 4 : return InvalidTransactionId;
6844 : }
6845 :
6846 : /* replace multi with single XID for its updater? */
6847 0 : update_xact = MultiXactIdGetUpdateXid(multi, t_infomask);
6848 0 : if (TransactionIdPrecedes(update_xact, cutoffs->relfrozenxid))
6849 0 : ereport(ERROR,
6850 : (errcode(ERRCODE_DATA_CORRUPTED),
6851 : errmsg_internal("multixact %u contains update XID %u from before relfrozenxid %u",
6852 : multi, update_xact,
6853 : cutoffs->relfrozenxid)));
6854 0 : else if (TransactionIdPrecedes(update_xact, cutoffs->OldestXmin))
6855 : {
6856 : /*
6857 : * Updater XID has to have aborted (otherwise the tuple would have
6858 : * been pruned away instead, since updater XID is < OldestXmin).
6859 : * Just remove xmax.
6860 : */
6861 0 : if (TransactionIdDidCommit(update_xact))
6862 0 : ereport(ERROR,
6863 : (errcode(ERRCODE_DATA_CORRUPTED),
6864 : errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
6865 : multi, update_xact,
6866 : cutoffs->OldestXmin)));
6867 0 : *flags |= FRM_INVALIDATE_XMAX;
6868 0 : pagefrz->freeze_required = true;
6869 0 : return InvalidTransactionId;
6870 : }
6871 :
6872 : /* Have to keep updater XID as new xmax */
6873 0 : *flags |= FRM_RETURN_IS_XID;
6874 0 : pagefrz->freeze_required = true;
6875 0 : return update_xact;
6876 : }
6877 :
6878 : /*
6879 : * Some member(s) of this Multi may be below FreezeLimit xid cutoff, so we
6880 : * need to walk the whole members array to figure out what to do, if
6881 : * anything.
6882 : */
6883 : nmembers =
6884 2 : GetMultiXactIdMembers(multi, &members, false,
6885 2 : HEAP_XMAX_IS_LOCKED_ONLY(t_infomask));
6886 2 : if (nmembers <= 0)
6887 : {
6888 : /* Nothing worth keeping */
6889 0 : *flags |= FRM_INVALIDATE_XMAX;
6890 0 : pagefrz->freeze_required = true;
6891 0 : return InvalidTransactionId;
6892 : }
6893 :
6894 : /*
6895 : * The FRM_NOOP case is the only case where we might need to ratchet back
6896 : * FreezePageRelfrozenXid or FreezePageRelminMxid. It is also the only
6897 : * case where our caller might ratchet back its NoFreezePageRelfrozenXid
6898 : * or NoFreezePageRelminMxid "no freeze" trackers to deal with a multi.
6899 : * FRM_NOOP handling should result in the NewRelfrozenXid/NewRelminMxid
6900 : * trackers managed by VACUUM being ratcheting back by xmax to the degree
6901 : * required to make it safe to leave xmax undisturbed, independent of
6902 : * whether or not page freezing is triggered somewhere else.
6903 : *
6904 : * Our policy is to force freezing in every case other than FRM_NOOP,
6905 : * which obviates the need to maintain either set of trackers, anywhere.
6906 : * Every other case will reliably execute a freeze plan for xmax that
6907 : * either replaces xmax with an XID/MXID >= OldestXmin/OldestMxact, or
6908 : * sets xmax to an InvalidTransactionId XID, rendering xmax fully frozen.
6909 : * (VACUUM's NewRelfrozenXid/NewRelminMxid trackers are initialized with
6910 : * OldestXmin/OldestMxact, so later values never need to be tracked here.)
6911 : */
6912 2 : need_replace = false;
6913 2 : FreezePageRelfrozenXid = pagefrz->FreezePageRelfrozenXid;
6914 4 : for (int i = 0; i < nmembers; i++)
6915 : {
6916 3 : TransactionId xid = members[i].xid;
6917 :
6918 : Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
6919 :
6920 3 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
6921 : {
6922 : /* Can't violate the FreezeLimit postcondition */
6923 1 : need_replace = true;
6924 1 : break;
6925 : }
6926 2 : if (TransactionIdPrecedes(xid, FreezePageRelfrozenXid))
6927 0 : FreezePageRelfrozenXid = xid;
6928 : }
6929 :
6930 : /* Can't violate the MultiXactCutoff postcondition, either */
6931 2 : if (!need_replace)
6932 1 : need_replace = MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff);
6933 :
6934 2 : if (!need_replace)
6935 : {
6936 : /*
6937 : * vacuumlazy.c might ratchet back NewRelminMxid, NewRelfrozenXid, or
6938 : * both together to make it safe to retain this particular multi after
6939 : * freezing its page
6940 : */
6941 1 : *flags |= FRM_NOOP;
6942 1 : pagefrz->FreezePageRelfrozenXid = FreezePageRelfrozenXid;
6943 1 : if (MultiXactIdPrecedes(multi, pagefrz->FreezePageRelminMxid))
6944 0 : pagefrz->FreezePageRelminMxid = multi;
6945 1 : pfree(members);
6946 1 : return multi;
6947 : }
6948 :
6949 : /*
6950 : * Do a more thorough second pass over the multi to figure out which
6951 : * member XIDs actually need to be kept. Checking the precise status of
6952 : * individual members might even show that we don't need to keep anything.
6953 : * That is quite possible even though the Multi must be >= OldestMxact,
6954 : * since our second pass only keeps member XIDs when it's truly necessary;
6955 : * even member XIDs >= OldestXmin often won't be kept by second pass.
6956 : */
6957 1 : nnewmembers = 0;
6958 1 : newmembers = palloc_array(MultiXactMember, nmembers);
6959 1 : has_lockers = false;
6960 1 : update_xid = InvalidTransactionId;
6961 1 : update_committed = false;
6962 :
6963 : /*
6964 : * Determine whether to keep each member xid, or to ignore it instead
6965 : */
6966 3 : for (int i = 0; i < nmembers; i++)
6967 : {
6968 2 : TransactionId xid = members[i].xid;
6969 2 : MultiXactStatus mstatus = members[i].status;
6970 :
6971 : Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
6972 :
6973 2 : if (!ISUPDATE_from_mxstatus(mstatus))
6974 : {
6975 : /*
6976 : * Locker XID (not updater XID). We only keep lockers that are
6977 : * still running.
6978 : */
6979 4 : if (TransactionIdIsCurrentTransactionId(xid) ||
6980 2 : TransactionIdIsInProgress(xid))
6981 : {
6982 1 : if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
6983 0 : ereport(ERROR,
6984 : (errcode(ERRCODE_DATA_CORRUPTED),
6985 : errmsg_internal("multixact %u contains running locker XID %u from before removable cutoff %u",
6986 : multi, xid,
6987 : cutoffs->OldestXmin)));
6988 1 : newmembers[nnewmembers++] = members[i];
6989 1 : has_lockers = true;
6990 : }
6991 :
6992 2 : continue;
6993 : }
6994 :
6995 : /*
6996 : * Updater XID (not locker XID). Should we keep it?
6997 : *
6998 : * Since the tuple wasn't totally removed when vacuum pruned, the
6999 : * update Xid cannot possibly be older than OldestXmin cutoff unless
7000 : * the updater XID aborted. If the updater transaction is known
7001 : * aborted or crashed then it's okay to ignore it, otherwise not.
7002 : *
7003 : * In any case the Multi should never contain two updaters, whatever
7004 : * their individual commit status. Check for that first, in passing.
7005 : */
7006 0 : if (TransactionIdIsValid(update_xid))
7007 0 : ereport(ERROR,
7008 : (errcode(ERRCODE_DATA_CORRUPTED),
7009 : errmsg_internal("multixact %u has two or more updating members",
7010 : multi),
7011 : errdetail_internal("First updater XID=%u second updater XID=%u.",
7012 : update_xid, xid)));
7013 :
7014 : /*
7015 : * As with all tuple visibility routines, it's critical to test
7016 : * TransactionIdIsInProgress before TransactionIdDidCommit, because of
7017 : * race conditions explained in detail in heapam_visibility.c.
7018 : */
7019 0 : if (TransactionIdIsCurrentTransactionId(xid) ||
7020 0 : TransactionIdIsInProgress(xid))
7021 0 : update_xid = xid;
7022 0 : else if (TransactionIdDidCommit(xid))
7023 : {
7024 : /*
7025 : * The transaction committed, so we can tell caller to set
7026 : * HEAP_XMAX_COMMITTED. (We can only do this because we know the
7027 : * transaction is not running.)
7028 : */
7029 0 : update_committed = true;
7030 0 : update_xid = xid;
7031 : }
7032 : else
7033 : {
7034 : /*
7035 : * Not in progress, not committed -- must be aborted or crashed;
7036 : * we can ignore it.
7037 : */
7038 0 : continue;
7039 : }
7040 :
7041 : /*
7042 : * We determined that updater must be kept -- add it to pending new
7043 : * members list
7044 : */
7045 0 : if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
7046 0 : ereport(ERROR,
7047 : (errcode(ERRCODE_DATA_CORRUPTED),
7048 : errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
7049 : multi, xid, cutoffs->OldestXmin)));
7050 0 : newmembers[nnewmembers++] = members[i];
7051 : }
7052 :
7053 1 : pfree(members);
7054 :
7055 : /*
7056 : * Determine what to do with caller's multi based on information gathered
7057 : * during our second pass
7058 : */
7059 1 : if (nnewmembers == 0)
7060 : {
7061 : /* Nothing worth keeping */
7062 0 : *flags |= FRM_INVALIDATE_XMAX;
7063 0 : newxmax = InvalidTransactionId;
7064 : }
7065 1 : else if (TransactionIdIsValid(update_xid) && !has_lockers)
7066 : {
7067 : /*
7068 : * If there's a single member and it's an update, pass it back alone
7069 : * without creating a new Multi. (XXX we could do this when there's a
7070 : * single remaining locker, too, but that would complicate the API too
7071 : * much; moreover, the case with the single updater is more
7072 : * interesting, because those are longer-lived.)
7073 : */
7074 : Assert(nnewmembers == 1);
7075 0 : *flags |= FRM_RETURN_IS_XID;
7076 0 : if (update_committed)
7077 0 : *flags |= FRM_MARK_COMMITTED;
7078 0 : newxmax = update_xid;
7079 : }
7080 : else
7081 : {
7082 : /*
7083 : * Create a new multixact with the surviving members of the previous
7084 : * one, to set as new Xmax in the tuple
7085 : */
7086 1 : newxmax = MultiXactIdCreateFromMembers(nnewmembers, newmembers);
7087 1 : *flags |= FRM_RETURN_IS_MULTI;
7088 : }
7089 :
7090 1 : pfree(newmembers);
7091 :
7092 1 : pagefrz->freeze_required = true;
7093 1 : return newxmax;
7094 : }
7095 :
7096 : /*
7097 : * heap_prepare_freeze_tuple
7098 : *
7099 : * Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
7100 : * are older than the OldestXmin and/or OldestMxact freeze cutoffs. If so,
7101 : * setup enough state (in the *frz output argument) to enable caller to
7102 : * process this tuple as part of freezing its page, and return true. Return
7103 : * false if nothing can be changed about the tuple right now.
7104 : *
7105 : * FreezePageConflictXid is advanced only for xmin/xvac freezing, not for xmax
7106 : * changes. We only remove xmax state here when it is lock-only, or when the
7107 : * updater XID (including an updater member of a MultiXact) must be aborted;
7108 : * otherwise, the tuple would already be removable. Neither case affects
7109 : * visibility on a standby.
7110 : *
7111 : * Also sets *totally_frozen to true if the tuple will be totally frozen once
7112 : * caller executes returned freeze plan (or if the tuple was already totally
7113 : * frozen by an earlier VACUUM). This indicates that there are no remaining
7114 : * XIDs or MultiXactIds that will need to be processed by a future VACUUM.
7115 : *
7116 : * VACUUM caller must assemble HeapTupleFreeze freeze plan entries for every
7117 : * tuple that we returned true for, and then execute freezing. Caller must
7118 : * initialize pagefrz fields for page as a whole before first call here for
7119 : * each heap page.
7120 : *
7121 : * VACUUM caller decides on whether or not to freeze the page as a whole.
7122 : * We'll often prepare freeze plans for a page that caller just discards.
7123 : * However, VACUUM doesn't always get to make a choice; it must freeze when
7124 : * pagefrz.freeze_required is set, to ensure that any XIDs < FreezeLimit (and
7125 : * MXIDs < MultiXactCutoff) can never be left behind. We help to make sure
7126 : * that VACUUM always follows that rule.
7127 : *
7128 : * We sometimes force freezing of xmax MultiXactId values long before it is
7129 : * strictly necessary to do so just to ensure the FreezeLimit postcondition.
7130 : * It's worth processing MultiXactIds proactively when it is cheap to do so,
7131 : * and it's convenient to make that happen by piggy-backing it on the "force
7132 : * freezing" mechanism. Conversely, we sometimes delay freezing MultiXactIds
7133 : * because it is expensive right now (though only when it's still possible to
7134 : * do so without violating the FreezeLimit/MultiXactCutoff postcondition).
7135 : *
7136 : * It is assumed that the caller has checked the tuple with
7137 : * HeapTupleSatisfiesVacuum() and determined that it is not HEAPTUPLE_DEAD
7138 : * (else we should be removing the tuple, not freezing it).
7139 : *
7140 : * NB: This function has side effects: it might allocate a new MultiXactId.
7141 : * It will be set as tuple's new xmax when our *frz output is processed within
7142 : * heap_execute_freeze_tuple later on. If the tuple is in a shared buffer
7143 : * then caller had better have an exclusive lock on it already.
7144 : */
7145 : bool
7146 10354946 : heap_prepare_freeze_tuple(HeapTupleHeader tuple,
7147 : const struct VacuumCutoffs *cutoffs,
7148 : HeapPageFreeze *pagefrz,
7149 : HeapTupleFreeze *frz, bool *totally_frozen)
7150 : {
7151 10354946 : bool xmin_already_frozen = false,
7152 10354946 : xmax_already_frozen = false;
7153 10354946 : bool freeze_xmin = false,
7154 10354946 : replace_xvac = false,
7155 10354946 : replace_xmax = false,
7156 10354946 : freeze_xmax = false;
7157 : TransactionId xid;
7158 :
7159 10354946 : frz->xmax = HeapTupleHeaderGetRawXmax(tuple);
7160 10354946 : frz->t_infomask2 = tuple->t_infomask2;
7161 10354946 : frz->t_infomask = tuple->t_infomask;
7162 10354946 : frz->frzflags = 0;
7163 10354946 : frz->checkflags = 0;
7164 :
7165 : /*
7166 : * Process xmin, while keeping track of whether it's already frozen, or
7167 : * will become frozen iff our freeze plan is executed by caller (could be
7168 : * neither).
7169 : */
7170 10354946 : xid = HeapTupleHeaderGetXmin(tuple);
7171 10354946 : if (!TransactionIdIsNormal(xid))
7172 6411917 : xmin_already_frozen = true;
7173 : else
7174 : {
7175 3943029 : if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
7176 0 : ereport(ERROR,
7177 : (errcode(ERRCODE_DATA_CORRUPTED),
7178 : errmsg_internal("found xmin %u from before relfrozenxid %u",
7179 : xid, cutoffs->relfrozenxid)));
7180 :
7181 : /* Will set freeze_xmin flags in freeze plan below */
7182 3943029 : freeze_xmin = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
7183 :
7184 : /* Verify that xmin committed if and when freeze plan is executed */
7185 3943029 : if (freeze_xmin)
7186 : {
7187 3149178 : frz->checkflags |= HEAP_FREEZE_CHECK_XMIN_COMMITTED;
7188 3149178 : if (TransactionIdFollows(xid, pagefrz->FreezePageConflictXid))
7189 487530 : pagefrz->FreezePageConflictXid = xid;
7190 : }
7191 : }
7192 :
7193 : /*
7194 : * Old-style VACUUM FULL is gone, but we have to process xvac for as long
7195 : * as we support having MOVED_OFF/MOVED_IN tuples in the database
7196 : */
7197 10354946 : xid = HeapTupleHeaderGetXvac(tuple);
7198 10354946 : if (TransactionIdIsNormal(xid))
7199 : {
7200 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7201 : Assert(TransactionIdPrecedes(xid, cutoffs->OldestXmin));
7202 :
7203 : /*
7204 : * For Xvac, we always freeze proactively. This allows totally_frozen
7205 : * tracking to ignore xvac.
7206 : */
7207 0 : replace_xvac = pagefrz->freeze_required = true;
7208 :
7209 0 : if (TransactionIdFollows(xid, pagefrz->FreezePageConflictXid))
7210 0 : pagefrz->FreezePageConflictXid = xid;
7211 :
7212 : /* Will set replace_xvac flags in freeze plan below */
7213 : }
7214 :
7215 : /* Now process xmax */
7216 10354946 : xid = frz->xmax;
7217 10354946 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7218 : {
7219 : /* Raw xmax is a MultiXactId */
7220 : TransactionId newxmax;
7221 : uint16 flags;
7222 :
7223 : /*
7224 : * We will either remove xmax completely (in the "freeze_xmax" path),
7225 : * process xmax by replacing it (in the "replace_xmax" path), or
7226 : * perform no-op xmax processing. The only constraint is that the
7227 : * FreezeLimit/MultiXactCutoff postcondition must never be violated.
7228 : */
7229 6 : newxmax = FreezeMultiXactId(xid, tuple->t_infomask, cutoffs,
7230 : &flags, pagefrz);
7231 :
7232 6 : if (flags & FRM_NOOP)
7233 : {
7234 : /*
7235 : * xmax is a MultiXactId, and nothing about it changes for now.
7236 : * This is the only case where 'freeze_required' won't have been
7237 : * set for us by FreezeMultiXactId, as well as the only case where
7238 : * neither freeze_xmax nor replace_xmax are set (given a multi).
7239 : *
7240 : * This is a no-op, but the call to FreezeMultiXactId might have
7241 : * ratcheted back NewRelfrozenXid and/or NewRelminMxid trackers
7242 : * for us (the "freeze page" variants, specifically). That'll
7243 : * make it safe for our caller to freeze the page later on, while
7244 : * leaving this particular xmax undisturbed.
7245 : *
7246 : * FreezeMultiXactId is _not_ responsible for the "no freeze"
7247 : * NewRelfrozenXid/NewRelminMxid trackers, though -- that's our
7248 : * job. A call to heap_tuple_should_freeze for this same tuple
7249 : * will take place below if 'freeze_required' isn't set already.
7250 : * (This repeats work from FreezeMultiXactId, but allows "no
7251 : * freeze" tracker maintenance to happen in only one place.)
7252 : */
7253 : Assert(!MultiXactIdPrecedes(newxmax, cutoffs->MultiXactCutoff));
7254 : Assert(MultiXactIdIsValid(newxmax) && xid == newxmax);
7255 : }
7256 5 : else if (flags & FRM_RETURN_IS_XID)
7257 : {
7258 : /*
7259 : * xmax will become an updater Xid (original MultiXact's updater
7260 : * member Xid will be carried forward as a simple Xid in Xmax).
7261 : */
7262 : Assert(!TransactionIdPrecedes(newxmax, cutoffs->OldestXmin));
7263 :
7264 : /*
7265 : * NB -- some of these transformations are only valid because we
7266 : * know the return Xid is a tuple updater (i.e. not merely a
7267 : * locker.) Also note that the only reason we don't explicitly
7268 : * worry about HEAP_KEYS_UPDATED is because it lives in
7269 : * t_infomask2 rather than t_infomask.
7270 : */
7271 0 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7272 0 : frz->xmax = newxmax;
7273 0 : if (flags & FRM_MARK_COMMITTED)
7274 0 : frz->t_infomask |= HEAP_XMAX_COMMITTED;
7275 0 : replace_xmax = true;
7276 : }
7277 5 : else if (flags & FRM_RETURN_IS_MULTI)
7278 : {
7279 : uint16 newbits;
7280 : uint16 newbits2;
7281 :
7282 : /*
7283 : * xmax is an old MultiXactId that we have to replace with a new
7284 : * MultiXactId, to carry forward two or more original member XIDs.
7285 : */
7286 : Assert(!MultiXactIdPrecedes(newxmax, cutoffs->OldestMxact));
7287 :
7288 : /*
7289 : * We can't use GetMultiXactIdHintBits directly on the new multi
7290 : * here; that routine initializes the masks to all zeroes, which
7291 : * would lose other bits we need. Doing it this way ensures all
7292 : * unrelated bits remain untouched.
7293 : */
7294 1 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7295 1 : frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
7296 1 : GetMultiXactIdHintBits(newxmax, &newbits, &newbits2);
7297 1 : frz->t_infomask |= newbits;
7298 1 : frz->t_infomask2 |= newbits2;
7299 1 : frz->xmax = newxmax;
7300 1 : replace_xmax = true;
7301 : }
7302 : else
7303 : {
7304 : /*
7305 : * Freeze plan for tuple "freezes xmax" in the strictest sense:
7306 : * it'll leave nothing in xmax (neither an Xid nor a MultiXactId).
7307 : */
7308 : Assert(flags & FRM_INVALIDATE_XMAX);
7309 : Assert(!TransactionIdIsValid(newxmax));
7310 :
7311 : /* Will set freeze_xmax flags in freeze plan below */
7312 4 : freeze_xmax = true;
7313 : }
7314 :
7315 : /* MultiXactId processing forces freezing (barring FRM_NOOP case) */
7316 : Assert(pagefrz->freeze_required || (!freeze_xmax && !replace_xmax));
7317 : }
7318 10354940 : else if (TransactionIdIsNormal(xid))
7319 : {
7320 : /* Raw xmax is normal XID */
7321 4420222 : if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
7322 0 : ereport(ERROR,
7323 : (errcode(ERRCODE_DATA_CORRUPTED),
7324 : errmsg_internal("found xmax %u from before relfrozenxid %u",
7325 : xid, cutoffs->relfrozenxid)));
7326 :
7327 : /* Will set freeze_xmax flags in freeze plan below */
7328 4420222 : freeze_xmax = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
7329 :
7330 : /*
7331 : * Verify that xmax aborted if and when freeze plan is executed,
7332 : * provided it's from an update. (A lock-only xmax can be removed
7333 : * independent of this, since the lock is released at xact end.)
7334 : */
7335 4420222 : if (freeze_xmax && !HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask))
7336 1427 : frz->checkflags |= HEAP_FREEZE_CHECK_XMAX_ABORTED;
7337 : }
7338 5934718 : else if (!TransactionIdIsValid(xid))
7339 : {
7340 : /* Raw xmax is InvalidTransactionId XID */
7341 : Assert((tuple->t_infomask & HEAP_XMAX_IS_MULTI) == 0);
7342 5934718 : xmax_already_frozen = true;
7343 : }
7344 : else
7345 0 : ereport(ERROR,
7346 : (errcode(ERRCODE_DATA_CORRUPTED),
7347 : errmsg_internal("found raw xmax %u (infomask 0x%04x) not invalid and not multi",
7348 : xid, tuple->t_infomask)));
7349 :
7350 10354946 : if (freeze_xmin)
7351 : {
7352 : Assert(!xmin_already_frozen);
7353 :
7354 3149178 : frz->t_infomask |= HEAP_XMIN_FROZEN;
7355 : }
7356 10354946 : if (replace_xvac)
7357 : {
7358 : /*
7359 : * If a MOVED_OFF tuple is not dead, the xvac transaction must have
7360 : * failed; whereas a non-dead MOVED_IN tuple must mean the xvac
7361 : * transaction succeeded.
7362 : */
7363 : Assert(pagefrz->freeze_required);
7364 0 : if (tuple->t_infomask & HEAP_MOVED_OFF)
7365 0 : frz->frzflags |= XLH_INVALID_XVAC;
7366 : else
7367 0 : frz->frzflags |= XLH_FREEZE_XVAC;
7368 : }
7369 : if (replace_xmax)
7370 : {
7371 : Assert(!xmax_already_frozen && !freeze_xmax);
7372 : Assert(pagefrz->freeze_required);
7373 :
7374 : /* Already set replace_xmax flags in freeze plan earlier */
7375 : }
7376 10354946 : if (freeze_xmax)
7377 : {
7378 : Assert(!xmax_already_frozen && !replace_xmax);
7379 :
7380 2413 : frz->xmax = InvalidTransactionId;
7381 :
7382 : /*
7383 : * The tuple might be marked either XMAX_INVALID or XMAX_COMMITTED +
7384 : * LOCKED. Normalize to INVALID just to be sure no one gets confused.
7385 : * Also get rid of the HEAP_KEYS_UPDATED bit.
7386 : */
7387 2413 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7388 2413 : frz->t_infomask |= HEAP_XMAX_INVALID;
7389 2413 : frz->t_infomask2 &= ~HEAP_HOT_UPDATED;
7390 2413 : frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
7391 : }
7392 :
7393 : /*
7394 : * Determine if this tuple is already totally frozen, or will become
7395 : * totally frozen (provided caller executes freeze plans for the page)
7396 : */
7397 19913628 : *totally_frozen = ((freeze_xmin || xmin_already_frozen) &&
7398 9558682 : (freeze_xmax || xmax_already_frozen));
7399 :
7400 10354946 : if (!pagefrz->freeze_required && !(xmin_already_frozen &&
7401 : xmax_already_frozen))
7402 : {
7403 : /*
7404 : * So far no previous tuple from the page made freezing mandatory.
7405 : * Does this tuple force caller to freeze the entire page?
7406 : */
7407 6572494 : pagefrz->freeze_required =
7408 6572494 : heap_tuple_should_freeze(tuple, cutoffs,
7409 : &pagefrz->NoFreezePageRelfrozenXid,
7410 : &pagefrz->NoFreezePageRelminMxid);
7411 : }
7412 :
7413 : /* Tell caller if this tuple has a usable freeze plan set in *frz */
7414 10354946 : return freeze_xmin || replace_xvac || replace_xmax || freeze_xmax;
7415 : }
7416 :
7417 : /*
7418 : * Perform xmin/xmax XID status sanity checks before actually executing freeze
7419 : * plans.
7420 : *
7421 : * heap_prepare_freeze_tuple doesn't perform these checks directly because
7422 : * pg_xact lookups are relatively expensive. They shouldn't be repeated by
7423 : * successive VACUUMs that each decide against freezing the same page.
7424 : */
7425 : void
7426 23996 : heap_pre_freeze_checks(Buffer buffer,
7427 : HeapTupleFreeze *tuples, int ntuples)
7428 : {
7429 23996 : Page page = BufferGetPage(buffer);
7430 :
7431 1092419 : for (int i = 0; i < ntuples; i++)
7432 : {
7433 1068423 : HeapTupleFreeze *frz = tuples + i;
7434 1068423 : ItemId itemid = PageGetItemId(page, frz->offset);
7435 : HeapTupleHeader htup;
7436 :
7437 1068423 : htup = (HeapTupleHeader) PageGetItem(page, itemid);
7438 :
7439 : /* Deliberately avoid relying on tuple hint bits here */
7440 1068423 : if (frz->checkflags & HEAP_FREEZE_CHECK_XMIN_COMMITTED)
7441 : {
7442 1068422 : TransactionId xmin = HeapTupleHeaderGetRawXmin(htup);
7443 :
7444 : Assert(!HeapTupleHeaderXminFrozen(htup));
7445 1068422 : if (unlikely(!TransactionIdDidCommit(xmin)))
7446 0 : ereport(ERROR,
7447 : (errcode(ERRCODE_DATA_CORRUPTED),
7448 : errmsg_internal("uncommitted xmin %u needs to be frozen",
7449 : xmin)));
7450 : }
7451 :
7452 : /*
7453 : * TransactionIdDidAbort won't work reliably in the presence of XIDs
7454 : * left behind by transactions that were in progress during a crash,
7455 : * so we can only check that xmax didn't commit
7456 : */
7457 1068423 : if (frz->checkflags & HEAP_FREEZE_CHECK_XMAX_ABORTED)
7458 : {
7459 414 : TransactionId xmax = HeapTupleHeaderGetRawXmax(htup);
7460 :
7461 : Assert(TransactionIdIsNormal(xmax));
7462 414 : if (unlikely(TransactionIdDidCommit(xmax)))
7463 0 : ereport(ERROR,
7464 : (errcode(ERRCODE_DATA_CORRUPTED),
7465 : errmsg_internal("cannot freeze committed xmax %u",
7466 : xmax)));
7467 : }
7468 : }
7469 23996 : }
7470 :
7471 : /*
7472 : * Helper which executes freezing of one or more heap tuples on a page on
7473 : * behalf of caller. Caller passes an array of tuple plans from
7474 : * heap_prepare_freeze_tuple. Caller must set 'offset' in each plan for us.
7475 : * Must be called in a critical section that also marks the buffer dirty and,
7476 : * if needed, emits WAL.
7477 : */
7478 : void
7479 23996 : heap_freeze_prepared_tuples(Buffer buffer, HeapTupleFreeze *tuples, int ntuples)
7480 : {
7481 23996 : Page page = BufferGetPage(buffer);
7482 :
7483 1092419 : for (int i = 0; i < ntuples; i++)
7484 : {
7485 1068423 : HeapTupleFreeze *frz = tuples + i;
7486 1068423 : ItemId itemid = PageGetItemId(page, frz->offset);
7487 : HeapTupleHeader htup;
7488 :
7489 1068423 : htup = (HeapTupleHeader) PageGetItem(page, itemid);
7490 1068423 : heap_execute_freeze_tuple(htup, frz);
7491 : }
7492 23996 : }
7493 :
7494 : /*
7495 : * heap_freeze_tuple
7496 : * Freeze tuple in place, without WAL logging.
7497 : *
7498 : * Useful for callers like CLUSTER that perform their own WAL logging.
7499 : */
7500 : bool
7501 466739 : heap_freeze_tuple(HeapTupleHeader tuple,
7502 : TransactionId relfrozenxid, TransactionId relminmxid,
7503 : TransactionId FreezeLimit, TransactionId MultiXactCutoff)
7504 : {
7505 : HeapTupleFreeze frz;
7506 : bool do_freeze;
7507 : bool totally_frozen;
7508 : struct VacuumCutoffs cutoffs;
7509 : HeapPageFreeze pagefrz;
7510 :
7511 466739 : cutoffs.relfrozenxid = relfrozenxid;
7512 466739 : cutoffs.relminmxid = relminmxid;
7513 466739 : cutoffs.OldestXmin = FreezeLimit;
7514 466739 : cutoffs.OldestMxact = MultiXactCutoff;
7515 466739 : cutoffs.FreezeLimit = FreezeLimit;
7516 466739 : cutoffs.MultiXactCutoff = MultiXactCutoff;
7517 :
7518 466739 : pagefrz.freeze_required = true;
7519 466739 : pagefrz.FreezePageRelfrozenXid = FreezeLimit;
7520 466739 : pagefrz.FreezePageRelminMxid = MultiXactCutoff;
7521 466739 : pagefrz.FreezePageConflictXid = InvalidTransactionId;
7522 466739 : pagefrz.NoFreezePageRelfrozenXid = FreezeLimit;
7523 466739 : pagefrz.NoFreezePageRelminMxid = MultiXactCutoff;
7524 :
7525 466739 : do_freeze = heap_prepare_freeze_tuple(tuple, &cutoffs,
7526 : &pagefrz, &frz, &totally_frozen);
7527 :
7528 : /*
7529 : * Note that because this is not a WAL-logged operation, we don't need to
7530 : * fill in the offset in the freeze record.
7531 : */
7532 :
7533 466739 : if (do_freeze)
7534 361791 : heap_execute_freeze_tuple(tuple, &frz);
7535 466739 : return do_freeze;
7536 : }
7537 :
7538 : /*
7539 : * For a given MultiXactId, return the hint bits that should be set in the
7540 : * tuple's infomask.
7541 : *
7542 : * Normally this should be called for a multixact that was just created, and
7543 : * so is on our local cache, so the GetMembers call is fast.
7544 : */
7545 : static void
7546 76862 : GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
7547 : uint16 *new_infomask2)
7548 : {
7549 : int nmembers;
7550 : MultiXactMember *members;
7551 : int i;
7552 76862 : uint16 bits = HEAP_XMAX_IS_MULTI;
7553 76862 : uint16 bits2 = 0;
7554 76862 : bool has_update = false;
7555 76862 : LockTupleMode strongest = LockTupleKeyShare;
7556 :
7557 : /*
7558 : * We only use this in multis we just created, so they cannot be values
7559 : * pre-pg_upgrade.
7560 : */
7561 76862 : nmembers = GetMultiXactIdMembers(multi, &members, false, false);
7562 :
7563 1472794 : for (i = 0; i < nmembers; i++)
7564 : {
7565 : LockTupleMode mode;
7566 :
7567 : /*
7568 : * Remember the strongest lock mode held by any member of the
7569 : * multixact.
7570 : */
7571 1395932 : mode = TUPLOCK_from_mxstatus(members[i].status);
7572 1395932 : if (mode > strongest)
7573 2924 : strongest = mode;
7574 :
7575 : /* See what other bits we need */
7576 1395932 : switch (members[i].status)
7577 : {
7578 1393512 : case MultiXactStatusForKeyShare:
7579 : case MultiXactStatusForShare:
7580 : case MultiXactStatusForNoKeyUpdate:
7581 1393512 : break;
7582 :
7583 53 : case MultiXactStatusForUpdate:
7584 53 : bits2 |= HEAP_KEYS_UPDATED;
7585 53 : break;
7586 :
7587 2357 : case MultiXactStatusNoKeyUpdate:
7588 2357 : has_update = true;
7589 2357 : break;
7590 :
7591 10 : case MultiXactStatusUpdate:
7592 10 : bits2 |= HEAP_KEYS_UPDATED;
7593 10 : has_update = true;
7594 10 : break;
7595 : }
7596 : }
7597 :
7598 76862 : if (strongest == LockTupleExclusive ||
7599 : strongest == LockTupleNoKeyExclusive)
7600 2448 : bits |= HEAP_XMAX_EXCL_LOCK;
7601 74414 : else if (strongest == LockTupleShare)
7602 473 : bits |= HEAP_XMAX_SHR_LOCK;
7603 73941 : else if (strongest == LockTupleKeyShare)
7604 73941 : bits |= HEAP_XMAX_KEYSHR_LOCK;
7605 :
7606 76862 : if (!has_update)
7607 74495 : bits |= HEAP_XMAX_LOCK_ONLY;
7608 :
7609 76862 : if (nmembers > 0)
7610 76862 : pfree(members);
7611 :
7612 76862 : *new_infomask = bits;
7613 76862 : *new_infomask2 = bits2;
7614 76862 : }
7615 :
7616 : /*
7617 : * MultiXactIdGetUpdateXid
7618 : *
7619 : * Given a multixact Xmax and corresponding infomask, which does not have the
7620 : * HEAP_XMAX_LOCK_ONLY bit set, obtain and return the Xid of the updating
7621 : * transaction.
7622 : *
7623 : * Caller is expected to check the status of the updating transaction, if
7624 : * necessary.
7625 : */
7626 : static TransactionId
7627 162089 : MultiXactIdGetUpdateXid(TransactionId xmax, uint16 t_infomask)
7628 : {
7629 162089 : TransactionId update_xact = InvalidTransactionId;
7630 : MultiXactMember *members;
7631 : int nmembers;
7632 :
7633 : Assert(!(t_infomask & HEAP_XMAX_LOCK_ONLY));
7634 : Assert(t_infomask & HEAP_XMAX_IS_MULTI);
7635 :
7636 : /*
7637 : * Since we know the LOCK_ONLY bit is not set, this cannot be a multi from
7638 : * pre-pg_upgrade.
7639 : */
7640 162089 : nmembers = GetMultiXactIdMembers(xmax, &members, false, false);
7641 :
7642 162089 : if (nmembers > 0)
7643 : {
7644 : int i;
7645 :
7646 245987 : for (i = 0; i < nmembers; i++)
7647 : {
7648 : /* Ignore lockers */
7649 245987 : if (!ISUPDATE_from_mxstatus(members[i].status))
7650 83898 : continue;
7651 :
7652 : /* there can be at most one updater */
7653 : Assert(update_xact == InvalidTransactionId);
7654 162089 : update_xact = members[i].xid;
7655 : #ifndef USE_ASSERT_CHECKING
7656 :
7657 : /*
7658 : * in an assert-enabled build, walk the whole array to ensure
7659 : * there's no other updater.
7660 : */
7661 162089 : break;
7662 : #endif
7663 : }
7664 :
7665 162089 : pfree(members);
7666 : }
7667 :
7668 162089 : return update_xact;
7669 : }
7670 :
7671 : /*
7672 : * HeapTupleGetUpdateXid
7673 : * As above, but use a HeapTupleHeader
7674 : *
7675 : * See also HeapTupleHeaderGetUpdateXid, which can be used without previously
7676 : * checking the hint bits.
7677 : */
7678 : TransactionId
7679 159950 : HeapTupleGetUpdateXid(const HeapTupleHeaderData *tup)
7680 : {
7681 159950 : return MultiXactIdGetUpdateXid(HeapTupleHeaderGetRawXmax(tup),
7682 159950 : tup->t_infomask);
7683 : }
7684 :
7685 : /*
7686 : * Does the given multixact conflict with the current transaction grabbing a
7687 : * tuple lock of the given strength?
7688 : *
7689 : * The passed infomask pairs up with the given multixact in the tuple header.
7690 : *
7691 : * If current_is_member is not NULL, it is set to 'true' if the current
7692 : * transaction is a member of the given multixact.
7693 : */
7694 : static bool
7695 218 : DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
7696 : LockTupleMode lockmode, bool *current_is_member)
7697 : {
7698 : int nmembers;
7699 : MultiXactMember *members;
7700 218 : bool result = false;
7701 218 : LOCKMODE wanted = tupleLockExtraInfo[lockmode].hwlock;
7702 :
7703 218 : if (HEAP_LOCKED_UPGRADED(infomask))
7704 0 : return false;
7705 :
7706 218 : nmembers = GetMultiXactIdMembers(multi, &members, false,
7707 218 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
7708 218 : if (nmembers >= 0)
7709 : {
7710 : int i;
7711 :
7712 2682 : for (i = 0; i < nmembers; i++)
7713 : {
7714 : TransactionId memxid;
7715 : LOCKMODE memlockmode;
7716 :
7717 2471 : if (result && (current_is_member == NULL || *current_is_member))
7718 : break;
7719 :
7720 2464 : memlockmode = LOCKMODE_from_mxstatus(members[i].status);
7721 :
7722 : /* ignore members from current xact (but track their presence) */
7723 2464 : memxid = members[i].xid;
7724 2464 : if (TransactionIdIsCurrentTransactionId(memxid))
7725 : {
7726 92 : if (current_is_member != NULL)
7727 78 : *current_is_member = true;
7728 92 : continue;
7729 : }
7730 2372 : else if (result)
7731 8 : continue;
7732 :
7733 : /* ignore members that don't conflict with the lock we want */
7734 2364 : if (!DoLockModesConflict(memlockmode, wanted))
7735 2325 : continue;
7736 :
7737 39 : if (ISUPDATE_from_mxstatus(members[i].status))
7738 : {
7739 : /* ignore aborted updaters */
7740 17 : if (TransactionIdDidAbort(memxid))
7741 1 : continue;
7742 : }
7743 : else
7744 : {
7745 : /* ignore lockers-only that are no longer in progress */
7746 22 : if (!TransactionIdIsInProgress(memxid))
7747 7 : continue;
7748 : }
7749 :
7750 : /*
7751 : * Whatever remains are either live lockers that conflict with our
7752 : * wanted lock, and updaters that are not aborted. Those conflict
7753 : * with what we want. Set up to return true, but keep going to
7754 : * look for the current transaction among the multixact members,
7755 : * if needed.
7756 : */
7757 31 : result = true;
7758 : }
7759 218 : pfree(members);
7760 : }
7761 :
7762 218 : return result;
7763 : }
7764 :
7765 : /*
7766 : * Do_MultiXactIdWait
7767 : * Actual implementation for the two functions below.
7768 : *
7769 : * 'multi', 'status' and 'infomask' indicate what to sleep on (the status is
7770 : * needed to ensure we only sleep on conflicting members, and the infomask is
7771 : * used to optimize multixact access in case it's a lock-only multi); 'nowait'
7772 : * indicates whether to use conditional lock acquisition, to allow callers to
7773 : * fail if lock is unavailable. 'rel', 'ctid' and 'oper' are used to set up
7774 : * context information for error messages. 'remaining', if not NULL, receives
7775 : * the number of members that are still running, including any (non-aborted)
7776 : * subtransactions of our own transaction. 'logLockFailure' indicates whether
7777 : * to log details when a lock acquisition fails with 'nowait' enabled.
7778 : *
7779 : * We do this by sleeping on each member using XactLockTableWait. Any
7780 : * members that belong to the current backend are *not* waited for, however;
7781 : * this would not merely be useless but would lead to Assert failure inside
7782 : * XactLockTableWait. By the time this returns, it is certain that all
7783 : * transactions *of other backends* that were members of the MultiXactId
7784 : * that conflict with the requested status are dead (and no new ones can have
7785 : * been added, since it is not legal to add members to an existing
7786 : * MultiXactId).
7787 : *
7788 : * But by the time we finish sleeping, someone else may have changed the Xmax
7789 : * of the containing tuple, so the caller needs to iterate on us somehow.
7790 : *
7791 : * Note that in case we return false, the number of remaining members is
7792 : * not to be trusted.
7793 : */
7794 : static bool
7795 61 : Do_MultiXactIdWait(MultiXactId multi, MultiXactStatus status,
7796 : uint16 infomask, bool nowait,
7797 : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
7798 : int *remaining, bool logLockFailure)
7799 : {
7800 61 : bool result = true;
7801 : MultiXactMember *members;
7802 : int nmembers;
7803 61 : int remain = 0;
7804 :
7805 : /* for pre-pg_upgrade tuples, no need to sleep at all */
7806 61 : nmembers = HEAP_LOCKED_UPGRADED(infomask) ? -1 :
7807 61 : GetMultiXactIdMembers(multi, &members, false,
7808 61 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
7809 :
7810 61 : if (nmembers >= 0)
7811 : {
7812 : int i;
7813 :
7814 192 : for (i = 0; i < nmembers; i++)
7815 : {
7816 137 : TransactionId memxid = members[i].xid;
7817 137 : MultiXactStatus memstatus = members[i].status;
7818 :
7819 137 : if (TransactionIdIsCurrentTransactionId(memxid))
7820 : {
7821 25 : remain++;
7822 25 : continue;
7823 : }
7824 :
7825 112 : if (!DoLockModesConflict(LOCKMODE_from_mxstatus(memstatus),
7826 112 : LOCKMODE_from_mxstatus(status)))
7827 : {
7828 22 : if (remaining && TransactionIdIsInProgress(memxid))
7829 8 : remain++;
7830 22 : continue;
7831 : }
7832 :
7833 : /*
7834 : * This member conflicts with our multi, so we have to sleep (or
7835 : * return failure, if asked to avoid waiting.)
7836 : *
7837 : * Note that we don't set up an error context callback ourselves,
7838 : * but instead we pass the info down to XactLockTableWait. This
7839 : * might seem a bit wasteful because the context is set up and
7840 : * tore down for each member of the multixact, but in reality it
7841 : * should be barely noticeable, and it avoids duplicate code.
7842 : */
7843 90 : if (nowait)
7844 : {
7845 6 : result = ConditionalXactLockTableWait(memxid, logLockFailure);
7846 6 : if (!result)
7847 6 : break;
7848 : }
7849 : else
7850 84 : XactLockTableWait(memxid, rel, ctid, oper);
7851 : }
7852 :
7853 61 : pfree(members);
7854 : }
7855 :
7856 61 : if (remaining)
7857 10 : *remaining = remain;
7858 :
7859 61 : return result;
7860 : }
7861 :
7862 : /*
7863 : * MultiXactIdWait
7864 : * Sleep on a MultiXactId.
7865 : *
7866 : * By the time we finish sleeping, someone else may have changed the Xmax
7867 : * of the containing tuple, so the caller needs to iterate on us somehow.
7868 : *
7869 : * We return (in *remaining, if not NULL) the number of members that are still
7870 : * running, including any (non-aborted) subtransactions of our own transaction.
7871 : */
7872 : static void
7873 55 : MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
7874 : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
7875 : int *remaining)
7876 : {
7877 55 : (void) Do_MultiXactIdWait(multi, status, infomask, false,
7878 : rel, ctid, oper, remaining, false);
7879 55 : }
7880 :
7881 : /*
7882 : * ConditionalMultiXactIdWait
7883 : * As above, but only lock if we can get the lock without blocking.
7884 : *
7885 : * By the time we finish sleeping, someone else may have changed the Xmax
7886 : * of the containing tuple, so the caller needs to iterate on us somehow.
7887 : *
7888 : * If the multixact is now all gone, return true. Returns false if some
7889 : * transactions might still be running.
7890 : *
7891 : * We return (in *remaining, if not NULL) the number of members that are still
7892 : * running, including any (non-aborted) subtransactions of our own transaction.
7893 : */
7894 : static bool
7895 6 : ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status,
7896 : uint16 infomask, Relation rel, int *remaining,
7897 : bool logLockFailure)
7898 : {
7899 6 : return Do_MultiXactIdWait(multi, status, infomask, true,
7900 : rel, NULL, XLTW_None, remaining, logLockFailure);
7901 : }
7902 :
7903 : /*
7904 : * heap_tuple_needs_eventual_freeze
7905 : *
7906 : * Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
7907 : * will eventually require freezing (if tuple isn't removed by pruning first).
7908 : */
7909 : bool
7910 118489 : heap_tuple_needs_eventual_freeze(HeapTupleHeader tuple)
7911 : {
7912 : TransactionId xid;
7913 :
7914 : /*
7915 : * If xmin is a normal transaction ID, this tuple is definitely not
7916 : * frozen.
7917 : */
7918 118489 : xid = HeapTupleHeaderGetXmin(tuple);
7919 118489 : if (TransactionIdIsNormal(xid))
7920 3300 : return true;
7921 :
7922 : /*
7923 : * If xmax is a valid xact or multixact, this tuple is also not frozen.
7924 : */
7925 115189 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7926 : {
7927 : MultiXactId multi;
7928 :
7929 0 : multi = HeapTupleHeaderGetRawXmax(tuple);
7930 0 : if (MultiXactIdIsValid(multi))
7931 0 : return true;
7932 : }
7933 : else
7934 : {
7935 115189 : xid = HeapTupleHeaderGetRawXmax(tuple);
7936 115189 : if (TransactionIdIsNormal(xid))
7937 8 : return true;
7938 : }
7939 :
7940 115181 : if (tuple->t_infomask & HEAP_MOVED)
7941 : {
7942 0 : xid = HeapTupleHeaderGetXvac(tuple);
7943 0 : if (TransactionIdIsNormal(xid))
7944 0 : return true;
7945 : }
7946 :
7947 115181 : return false;
7948 : }
7949 :
7950 : /*
7951 : * heap_tuple_should_freeze
7952 : *
7953 : * Return value indicates if heap_prepare_freeze_tuple sibling function would
7954 : * (or should) force freezing of the heap page that contains caller's tuple.
7955 : * Tuple header XIDs/MXIDs < FreezeLimit/MultiXactCutoff trigger freezing.
7956 : * This includes (xmin, xmax, xvac) fields, as well as MultiXact member XIDs.
7957 : *
7958 : * The *NoFreezePageRelfrozenXid and *NoFreezePageRelminMxid input/output
7959 : * arguments help VACUUM track the oldest extant XID/MXID remaining in rel.
7960 : * Our working assumption is that caller won't decide to freeze this tuple.
7961 : * It's up to caller to only ratchet back its own top-level trackers after the
7962 : * point that it fully commits to not freezing the tuple/page in question.
7963 : */
7964 : bool
7965 6575538 : heap_tuple_should_freeze(HeapTupleHeader tuple,
7966 : const struct VacuumCutoffs *cutoffs,
7967 : TransactionId *NoFreezePageRelfrozenXid,
7968 : MultiXactId *NoFreezePageRelminMxid)
7969 : {
7970 : TransactionId xid;
7971 : MultiXactId multi;
7972 6575538 : bool freeze = false;
7973 :
7974 : /* First deal with xmin */
7975 6575538 : xid = HeapTupleHeaderGetXmin(tuple);
7976 6575538 : if (TransactionIdIsNormal(xid))
7977 : {
7978 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7979 2561288 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7980 24324 : *NoFreezePageRelfrozenXid = xid;
7981 2561288 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7982 22076 : freeze = true;
7983 : }
7984 :
7985 : /* Now deal with xmax */
7986 6575538 : xid = InvalidTransactionId;
7987 6575538 : multi = InvalidMultiXactId;
7988 6575538 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7989 2 : multi = HeapTupleHeaderGetRawXmax(tuple);
7990 : else
7991 6575536 : xid = HeapTupleHeaderGetRawXmax(tuple);
7992 :
7993 6575538 : if (TransactionIdIsNormal(xid))
7994 : {
7995 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7996 : /* xmax is a non-permanent XID */
7997 4343792 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7998 1 : *NoFreezePageRelfrozenXid = xid;
7999 4343792 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
8000 5 : freeze = true;
8001 : }
8002 2231746 : else if (!MultiXactIdIsValid(multi))
8003 : {
8004 : /* xmax is a permanent XID or invalid MultiXactId/XID */
8005 : }
8006 2 : else if (HEAP_LOCKED_UPGRADED(tuple->t_infomask))
8007 : {
8008 : /* xmax is a pg_upgrade'd MultiXact, which can't have updater XID */
8009 0 : if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
8010 0 : *NoFreezePageRelminMxid = multi;
8011 : /* heap_prepare_freeze_tuple always freezes pg_upgrade'd xmax */
8012 0 : freeze = true;
8013 : }
8014 : else
8015 : {
8016 : /* xmax is a MultiXactId that may have an updater XID */
8017 : MultiXactMember *members;
8018 : int nmembers;
8019 :
8020 : Assert(MultiXactIdPrecedesOrEquals(cutoffs->relminmxid, multi));
8021 2 : if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
8022 2 : *NoFreezePageRelminMxid = multi;
8023 2 : if (MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff))
8024 2 : freeze = true;
8025 :
8026 : /* need to check whether any member of the mxact is old */
8027 2 : nmembers = GetMultiXactIdMembers(multi, &members, false,
8028 2 : HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask));
8029 :
8030 5 : for (int i = 0; i < nmembers; i++)
8031 : {
8032 3 : xid = members[i].xid;
8033 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
8034 3 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
8035 0 : *NoFreezePageRelfrozenXid = xid;
8036 3 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
8037 0 : freeze = true;
8038 : }
8039 2 : if (nmembers > 0)
8040 1 : pfree(members);
8041 : }
8042 :
8043 6575538 : if (tuple->t_infomask & HEAP_MOVED)
8044 : {
8045 0 : xid = HeapTupleHeaderGetXvac(tuple);
8046 0 : if (TransactionIdIsNormal(xid))
8047 : {
8048 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
8049 0 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
8050 0 : *NoFreezePageRelfrozenXid = xid;
8051 : /* heap_prepare_freeze_tuple forces xvac freezing */
8052 0 : freeze = true;
8053 : }
8054 : }
8055 :
8056 6575538 : return freeze;
8057 : }
8058 :
8059 : /*
8060 : * Maintain snapshotConflictHorizon for caller by ratcheting forward its value
8061 : * using any committed XIDs contained in 'tuple', an obsolescent heap tuple
8062 : * that caller is in the process of physically removing, e.g. via HOT pruning
8063 : * or index deletion.
8064 : *
8065 : * Caller must initialize its value to InvalidTransactionId, which is
8066 : * generally interpreted as "definitely no need for a recovery conflict".
8067 : * Final value must reflect all heap tuples that caller will physically remove
8068 : * (or remove TID references to) via its ongoing pruning/deletion operation.
8069 : * ResolveRecoveryConflictWithSnapshot() is passed the final value (taken from
8070 : * caller's WAL record) by REDO routine when it replays caller's operation.
8071 : */
8072 : void
8073 1895097 : HeapTupleHeaderAdvanceConflictHorizon(HeapTupleHeader tuple,
8074 : TransactionId *snapshotConflictHorizon)
8075 : {
8076 1895097 : TransactionId xmin = HeapTupleHeaderGetXmin(tuple);
8077 1895097 : TransactionId xmax = HeapTupleHeaderGetUpdateXid(tuple);
8078 1895097 : TransactionId xvac = HeapTupleHeaderGetXvac(tuple);
8079 :
8080 1895097 : if (tuple->t_infomask & HEAP_MOVED)
8081 : {
8082 0 : if (TransactionIdPrecedes(*snapshotConflictHorizon, xvac))
8083 0 : *snapshotConflictHorizon = xvac;
8084 : }
8085 :
8086 : /*
8087 : * Ignore tuples inserted by an aborted transaction or if the tuple was
8088 : * updated/deleted by the inserting transaction.
8089 : *
8090 : * Look for a committed hint bit, or if no xmin bit is set, check clog.
8091 : */
8092 1895097 : if (HeapTupleHeaderXminCommitted(tuple) ||
8093 122212 : (!HeapTupleHeaderXminInvalid(tuple) && TransactionIdDidCommit(xmin)))
8094 : {
8095 3378337 : if (xmax != xmin &&
8096 1591268 : TransactionIdFollows(xmax, *snapshotConflictHorizon))
8097 117433 : *snapshotConflictHorizon = xmax;
8098 : }
8099 1895097 : }
8100 :
8101 : #ifdef USE_PREFETCH
8102 : /*
8103 : * Helper function for heap_index_delete_tuples. Issues prefetch requests for
8104 : * prefetch_count buffers. The prefetch_state keeps track of all the buffers
8105 : * we can prefetch, and which have already been prefetched; each call to this
8106 : * function picks up where the previous call left off.
8107 : *
8108 : * Note: we expect the deltids array to be sorted in an order that groups TIDs
8109 : * by heap block, with all TIDs for each block appearing together in exactly
8110 : * one group.
8111 : */
8112 : static void
8113 25328 : index_delete_prefetch_buffer(Relation rel,
8114 : IndexDeletePrefetchState *prefetch_state,
8115 : int prefetch_count)
8116 : {
8117 25328 : BlockNumber cur_hblkno = prefetch_state->cur_hblkno;
8118 25328 : int count = 0;
8119 : int i;
8120 25328 : int ndeltids = prefetch_state->ndeltids;
8121 25328 : TM_IndexDelete *deltids = prefetch_state->deltids;
8122 :
8123 25328 : for (i = prefetch_state->next_item;
8124 822767 : i < ndeltids && count < prefetch_count;
8125 797439 : i++)
8126 : {
8127 797439 : ItemPointer htid = &deltids[i].tid;
8128 :
8129 1587319 : if (cur_hblkno == InvalidBlockNumber ||
8130 789880 : ItemPointerGetBlockNumber(htid) != cur_hblkno)
8131 : {
8132 22320 : cur_hblkno = ItemPointerGetBlockNumber(htid);
8133 22320 : PrefetchBuffer(rel, MAIN_FORKNUM, cur_hblkno);
8134 22320 : count++;
8135 : }
8136 : }
8137 :
8138 : /*
8139 : * Save the prefetch position so that next time we can continue from that
8140 : * position.
8141 : */
8142 25328 : prefetch_state->next_item = i;
8143 25328 : prefetch_state->cur_hblkno = cur_hblkno;
8144 25328 : }
8145 : #endif
8146 :
8147 : /*
8148 : * Helper function for heap_index_delete_tuples. Checks for index corruption
8149 : * involving an invalid TID in index AM caller's index page.
8150 : *
8151 : * This is an ideal place for these checks. The index AM must hold a buffer
8152 : * lock on the index page containing the TIDs we examine here, so we don't
8153 : * have to worry about concurrent VACUUMs at all. We can be sure that the
8154 : * index is corrupt when htid points directly to an LP_UNUSED item or
8155 : * heap-only tuple, which is not the case during standard index scans.
8156 : */
8157 : static inline void
8158 678057 : index_delete_check_htid(TM_IndexDeleteOp *delstate,
8159 : Page page, OffsetNumber maxoff,
8160 : const ItemPointerData *htid, TM_IndexStatus *istatus)
8161 : {
8162 678057 : OffsetNumber indexpagehoffnum = ItemPointerGetOffsetNumber(htid);
8163 : ItemId iid;
8164 :
8165 : Assert(OffsetNumberIsValid(istatus->idxoffnum));
8166 :
8167 678057 : if (unlikely(indexpagehoffnum > maxoff))
8168 0 : ereport(ERROR,
8169 : (errcode(ERRCODE_INDEX_CORRUPTED),
8170 : errmsg_internal("heap tid from index tuple (%u,%u) points past end of heap page line pointer array at offset %u of block %u in index \"%s\"",
8171 : ItemPointerGetBlockNumber(htid),
8172 : indexpagehoffnum,
8173 : istatus->idxoffnum, delstate->iblknum,
8174 : RelationGetRelationName(delstate->irel))));
8175 :
8176 678057 : iid = PageGetItemId(page, indexpagehoffnum);
8177 678057 : if (unlikely(!ItemIdIsUsed(iid)))
8178 0 : ereport(ERROR,
8179 : (errcode(ERRCODE_INDEX_CORRUPTED),
8180 : errmsg_internal("heap tid from index tuple (%u,%u) points to unused heap page item at offset %u of block %u in index \"%s\"",
8181 : ItemPointerGetBlockNumber(htid),
8182 : indexpagehoffnum,
8183 : istatus->idxoffnum, delstate->iblknum,
8184 : RelationGetRelationName(delstate->irel))));
8185 :
8186 678057 : if (ItemIdHasStorage(iid))
8187 : {
8188 : HeapTupleHeader htup;
8189 :
8190 : Assert(ItemIdIsNormal(iid));
8191 398205 : htup = (HeapTupleHeader) PageGetItem(page, iid);
8192 :
8193 398205 : if (unlikely(HeapTupleHeaderIsHeapOnly(htup)))
8194 0 : ereport(ERROR,
8195 : (errcode(ERRCODE_INDEX_CORRUPTED),
8196 : errmsg_internal("heap tid from index tuple (%u,%u) points to heap-only tuple at offset %u of block %u in index \"%s\"",
8197 : ItemPointerGetBlockNumber(htid),
8198 : indexpagehoffnum,
8199 : istatus->idxoffnum, delstate->iblknum,
8200 : RelationGetRelationName(delstate->irel))));
8201 : }
8202 678057 : }
8203 :
8204 : /*
8205 : * heapam implementation of tableam's index_delete_tuples interface.
8206 : *
8207 : * This helper function is called by index AMs during index tuple deletion.
8208 : * See tableam header comments for an explanation of the interface implemented
8209 : * here and a general theory of operation. Note that each call here is either
8210 : * a simple index deletion call, or a bottom-up index deletion call.
8211 : *
8212 : * It's possible for this to generate a fair amount of I/O, since we may be
8213 : * deleting hundreds of tuples from a single index block. To amortize that
8214 : * cost to some degree, this uses prefetching and combines repeat accesses to
8215 : * the same heap block.
8216 : */
8217 : TransactionId
8218 7559 : heap_index_delete_tuples(Relation rel, TM_IndexDeleteOp *delstate)
8219 : {
8220 : /* Initial assumption is that earlier pruning took care of conflict */
8221 7559 : TransactionId snapshotConflictHorizon = InvalidTransactionId;
8222 7559 : BlockNumber blkno = InvalidBlockNumber;
8223 7559 : Buffer buf = InvalidBuffer;
8224 7559 : Page page = NULL;
8225 7559 : OffsetNumber maxoff = InvalidOffsetNumber;
8226 : TransactionId priorXmax;
8227 : #ifdef USE_PREFETCH
8228 : IndexDeletePrefetchState prefetch_state;
8229 : int prefetch_distance;
8230 : #endif
8231 : SnapshotData SnapshotNonVacuumable;
8232 7559 : int finalndeltids = 0,
8233 7559 : nblocksaccessed = 0;
8234 :
8235 : /* State that's only used in bottom-up index deletion case */
8236 7559 : int nblocksfavorable = 0;
8237 7559 : int curtargetfreespace = delstate->bottomupfreespace,
8238 7559 : lastfreespace = 0,
8239 7559 : actualfreespace = 0;
8240 7559 : bool bottomup_final_block = false;
8241 :
8242 7559 : InitNonVacuumableSnapshot(SnapshotNonVacuumable, GlobalVisTestFor(rel));
8243 :
8244 : /* Sort caller's deltids array by TID for further processing */
8245 7559 : index_delete_sort(delstate);
8246 :
8247 : /*
8248 : * Bottom-up case: resort deltids array in an order attuned to where the
8249 : * greatest number of promising TIDs are to be found, and determine how
8250 : * many blocks from the start of sorted array should be considered
8251 : * favorable. This will also shrink the deltids array in order to
8252 : * eliminate completely unfavorable blocks up front.
8253 : */
8254 7559 : if (delstate->bottomup)
8255 2283 : nblocksfavorable = bottomup_sort_and_shrink(delstate);
8256 :
8257 : #ifdef USE_PREFETCH
8258 : /* Initialize prefetch state. */
8259 7559 : prefetch_state.cur_hblkno = InvalidBlockNumber;
8260 7559 : prefetch_state.next_item = 0;
8261 7559 : prefetch_state.ndeltids = delstate->ndeltids;
8262 7559 : prefetch_state.deltids = delstate->deltids;
8263 :
8264 : /*
8265 : * Determine the prefetch distance that we will attempt to maintain.
8266 : *
8267 : * Since the caller holds a buffer lock somewhere in rel, we'd better make
8268 : * sure that isn't a catalog relation before we call code that does
8269 : * syscache lookups, to avoid risk of deadlock.
8270 : */
8271 7559 : if (IsCatalogRelation(rel))
8272 5804 : prefetch_distance = maintenance_io_concurrency;
8273 : else
8274 : prefetch_distance =
8275 1755 : get_tablespace_maintenance_io_concurrency(rel->rd_rel->reltablespace);
8276 :
8277 : /* Cap initial prefetch distance for bottom-up deletion caller */
8278 7559 : if (delstate->bottomup)
8279 : {
8280 : Assert(nblocksfavorable >= 1);
8281 : Assert(nblocksfavorable <= BOTTOMUP_MAX_NBLOCKS);
8282 2283 : prefetch_distance = Min(prefetch_distance, nblocksfavorable);
8283 : }
8284 :
8285 : /* Start prefetching. */
8286 7559 : index_delete_prefetch_buffer(rel, &prefetch_state, prefetch_distance);
8287 : #endif
8288 :
8289 : /* Iterate over deltids, determine which to delete, check their horizon */
8290 : Assert(delstate->ndeltids > 0);
8291 685616 : for (int i = 0; i < delstate->ndeltids; i++)
8292 : {
8293 680340 : TM_IndexDelete *ideltid = &delstate->deltids[i];
8294 680340 : TM_IndexStatus *istatus = delstate->status + ideltid->id;
8295 680340 : ItemPointer htid = &ideltid->tid;
8296 : OffsetNumber offnum;
8297 :
8298 : /*
8299 : * Read buffer, and perform required extra steps each time a new block
8300 : * is encountered. Avoid refetching if it's the same block as the one
8301 : * from the last htid.
8302 : */
8303 1353121 : if (blkno == InvalidBlockNumber ||
8304 672781 : ItemPointerGetBlockNumber(htid) != blkno)
8305 : {
8306 : /*
8307 : * Consider giving up early for bottom-up index deletion caller
8308 : * first. (Only prefetch next-next block afterwards, when it
8309 : * becomes clear that we're at least going to access the next
8310 : * block in line.)
8311 : *
8312 : * Sometimes the first block frees so much space for bottom-up
8313 : * caller that the deletion process can end without accessing any
8314 : * more blocks. It is usually necessary to access 2 or 3 blocks
8315 : * per bottom-up deletion operation, though.
8316 : */
8317 20052 : if (delstate->bottomup)
8318 : {
8319 : /*
8320 : * We often allow caller to delete a few additional items
8321 : * whose entries we reached after the point that space target
8322 : * from caller was satisfied. The cost of accessing the page
8323 : * was already paid at that point, so it made sense to finish
8324 : * it off. When that happened, we finalize everything here
8325 : * (by finishing off the whole bottom-up deletion operation
8326 : * without needlessly paying the cost of accessing any more
8327 : * blocks).
8328 : */
8329 4943 : if (bottomup_final_block)
8330 187 : break;
8331 :
8332 : /*
8333 : * Give up when we didn't enable our caller to free any
8334 : * additional space as a result of processing the page that we
8335 : * just finished up with. This rule is the main way in which
8336 : * we keep the cost of bottom-up deletion under control.
8337 : */
8338 4756 : if (nblocksaccessed >= 1 && actualfreespace == lastfreespace)
8339 2096 : break;
8340 2660 : lastfreespace = actualfreespace; /* for next time */
8341 :
8342 : /*
8343 : * Deletion operation (which is bottom-up) will definitely
8344 : * access the next block in line. Prepare for that now.
8345 : *
8346 : * Decay target free space so that we don't hang on for too
8347 : * long with a marginal case. (Space target is only truly
8348 : * helpful when it allows us to recognize that we don't need
8349 : * to access more than 1 or 2 blocks to satisfy caller due to
8350 : * agreeable workload characteristics.)
8351 : *
8352 : * We are a bit more patient when we encounter contiguous
8353 : * blocks, though: these are treated as favorable blocks. The
8354 : * decay process is only applied when the next block in line
8355 : * is not a favorable/contiguous block. This is not an
8356 : * exception to the general rule; we still insist on finding
8357 : * at least one deletable item per block accessed. See
8358 : * bottomup_nblocksfavorable() for full details of the theory
8359 : * behind favorable blocks and heap block locality in general.
8360 : *
8361 : * Note: The first block in line is always treated as a
8362 : * favorable block, so the earliest possible point that the
8363 : * decay can be applied is just before we access the second
8364 : * block in line. The Assert() verifies this for us.
8365 : */
8366 : Assert(nblocksaccessed > 0 || nblocksfavorable > 0);
8367 2660 : if (nblocksfavorable > 0)
8368 2473 : nblocksfavorable--;
8369 : else
8370 187 : curtargetfreespace /= 2;
8371 : }
8372 :
8373 : /* release old buffer */
8374 17769 : if (BufferIsValid(buf))
8375 10210 : UnlockReleaseBuffer(buf);
8376 :
8377 17769 : blkno = ItemPointerGetBlockNumber(htid);
8378 17769 : buf = ReadBuffer(rel, blkno);
8379 17769 : nblocksaccessed++;
8380 : Assert(!delstate->bottomup ||
8381 : nblocksaccessed <= BOTTOMUP_MAX_NBLOCKS);
8382 :
8383 : #ifdef USE_PREFETCH
8384 :
8385 : /*
8386 : * To maintain the prefetch distance, prefetch one more page for
8387 : * each page we read.
8388 : */
8389 17769 : index_delete_prefetch_buffer(rel, &prefetch_state, 1);
8390 : #endif
8391 :
8392 17769 : LockBuffer(buf, BUFFER_LOCK_SHARE);
8393 :
8394 17769 : page = BufferGetPage(buf);
8395 17769 : maxoff = PageGetMaxOffsetNumber(page);
8396 : }
8397 :
8398 : /*
8399 : * In passing, detect index corruption involving an index page with a
8400 : * TID that points to a location in the heap that couldn't possibly be
8401 : * correct. We only do this with actual TIDs from caller's index page
8402 : * (not items reached by traversing through a HOT chain).
8403 : */
8404 678057 : index_delete_check_htid(delstate, page, maxoff, htid, istatus);
8405 :
8406 678057 : if (istatus->knowndeletable)
8407 : Assert(!delstate->bottomup && !istatus->promising);
8408 : else
8409 : {
8410 515553 : ItemPointerData tmp = *htid;
8411 : HeapTupleData heapTuple;
8412 :
8413 : /* Are any tuples from this HOT chain non-vacuumable? */
8414 515553 : if (heap_hot_search_buffer(&tmp, rel, buf, &SnapshotNonVacuumable,
8415 : &heapTuple, NULL, true))
8416 287306 : continue; /* can't delete entry */
8417 :
8418 : /* Caller will delete, since whole HOT chain is vacuumable */
8419 228247 : istatus->knowndeletable = true;
8420 :
8421 : /* Maintain index free space info for bottom-up deletion case */
8422 228247 : if (delstate->bottomup)
8423 : {
8424 : Assert(istatus->freespace > 0);
8425 10231 : actualfreespace += istatus->freespace;
8426 10231 : if (actualfreespace >= curtargetfreespace)
8427 2951 : bottomup_final_block = true;
8428 : }
8429 : }
8430 :
8431 : /*
8432 : * Maintain snapshotConflictHorizon value for deletion operation as a
8433 : * whole by advancing current value using heap tuple headers. This is
8434 : * loosely based on the logic for pruning a HOT chain.
8435 : */
8436 390751 : offnum = ItemPointerGetOffsetNumber(htid);
8437 390751 : priorXmax = InvalidTransactionId; /* cannot check first XMIN */
8438 : for (;;)
8439 22270 : {
8440 : ItemId lp;
8441 : HeapTupleHeader htup;
8442 :
8443 : /* Sanity check (pure paranoia) */
8444 413021 : if (offnum < FirstOffsetNumber)
8445 0 : break;
8446 :
8447 : /*
8448 : * An offset past the end of page's line pointer array is possible
8449 : * when the array was truncated
8450 : */
8451 413021 : if (offnum > maxoff)
8452 0 : break;
8453 :
8454 413021 : lp = PageGetItemId(page, offnum);
8455 413021 : if (ItemIdIsRedirected(lp))
8456 : {
8457 10003 : offnum = ItemIdGetRedirect(lp);
8458 10003 : continue;
8459 : }
8460 :
8461 : /*
8462 : * We'll often encounter LP_DEAD line pointers (especially with an
8463 : * entry marked knowndeletable by our caller up front). No heap
8464 : * tuple headers get examined for an htid that leads us to an
8465 : * LP_DEAD item. This is okay because the earlier pruning
8466 : * operation that made the line pointer LP_DEAD in the first place
8467 : * must have considered the original tuple header as part of
8468 : * generating its own snapshotConflictHorizon value.
8469 : *
8470 : * Relying on XLOG_HEAP2_PRUNE_VACUUM_SCAN records like this is
8471 : * the same strategy that index vacuuming uses in all cases. Index
8472 : * VACUUM WAL records don't even have a snapshotConflictHorizon
8473 : * field of their own for this reason.
8474 : */
8475 403018 : if (!ItemIdIsNormal(lp))
8476 252649 : break;
8477 :
8478 150369 : htup = (HeapTupleHeader) PageGetItem(page, lp);
8479 :
8480 : /*
8481 : * Check the tuple XMIN against prior XMAX, if any
8482 : */
8483 162636 : if (TransactionIdIsValid(priorXmax) &&
8484 12267 : !TransactionIdEquals(HeapTupleHeaderGetXmin(htup), priorXmax))
8485 0 : break;
8486 :
8487 150369 : HeapTupleHeaderAdvanceConflictHorizon(htup,
8488 : &snapshotConflictHorizon);
8489 :
8490 : /*
8491 : * If the tuple is not HOT-updated, then we are at the end of this
8492 : * HOT-chain. No need to visit later tuples from the same update
8493 : * chain (they get their own index entries) -- just move on to
8494 : * next htid from index AM caller.
8495 : */
8496 150369 : if (!HeapTupleHeaderIsHotUpdated(htup))
8497 138102 : break;
8498 :
8499 : /* Advance to next HOT chain member */
8500 : Assert(ItemPointerGetBlockNumber(&htup->t_ctid) == blkno);
8501 12267 : offnum = ItemPointerGetOffsetNumber(&htup->t_ctid);
8502 12267 : priorXmax = HeapTupleHeaderGetUpdateXid(htup);
8503 : }
8504 :
8505 : /* Enable further/final shrinking of deltids for caller */
8506 390751 : finalndeltids = i + 1;
8507 : }
8508 :
8509 7559 : UnlockReleaseBuffer(buf);
8510 :
8511 : /*
8512 : * Shrink deltids array to exclude non-deletable entries at the end. This
8513 : * is not just a minor optimization. Final deltids array size might be
8514 : * zero for a bottom-up caller. Index AM is explicitly allowed to rely on
8515 : * ndeltids being zero in all cases with zero total deletable entries.
8516 : */
8517 : Assert(finalndeltids > 0 || delstate->bottomup);
8518 7559 : delstate->ndeltids = finalndeltids;
8519 :
8520 7559 : return snapshotConflictHorizon;
8521 : }
8522 :
8523 : /*
8524 : * Specialized inlineable comparison function for index_delete_sort()
8525 : */
8526 : static inline int
8527 15554235 : index_delete_sort_cmp(TM_IndexDelete *deltid1, TM_IndexDelete *deltid2)
8528 : {
8529 15554235 : ItemPointer tid1 = &deltid1->tid;
8530 15554235 : ItemPointer tid2 = &deltid2->tid;
8531 :
8532 : {
8533 15554235 : BlockNumber blk1 = ItemPointerGetBlockNumber(tid1);
8534 15554235 : BlockNumber blk2 = ItemPointerGetBlockNumber(tid2);
8535 :
8536 15554235 : if (blk1 != blk2)
8537 6489243 : return (blk1 < blk2) ? -1 : 1;
8538 : }
8539 : {
8540 9064992 : OffsetNumber pos1 = ItemPointerGetOffsetNumber(tid1);
8541 9064992 : OffsetNumber pos2 = ItemPointerGetOffsetNumber(tid2);
8542 :
8543 9064992 : if (pos1 != pos2)
8544 9064992 : return (pos1 < pos2) ? -1 : 1;
8545 : }
8546 :
8547 : Assert(false);
8548 :
8549 0 : return 0;
8550 : }
8551 :
8552 : /*
8553 : * Sort deltids array from delstate by TID. This prepares it for further
8554 : * processing by heap_index_delete_tuples().
8555 : *
8556 : * This operation becomes a noticeable consumer of CPU cycles with some
8557 : * workloads, so we go to the trouble of specialization/micro optimization.
8558 : * We use shellsort for this because it's easy to specialize, compiles to
8559 : * relatively few instructions, and is adaptive to presorted inputs/subsets
8560 : * (which are typical here).
8561 : */
8562 : static void
8563 7559 : index_delete_sort(TM_IndexDeleteOp *delstate)
8564 : {
8565 7559 : TM_IndexDelete *deltids = delstate->deltids;
8566 7559 : int ndeltids = delstate->ndeltids;
8567 :
8568 : /*
8569 : * Shellsort gap sequence (taken from Sedgewick-Incerpi paper).
8570 : *
8571 : * This implementation is fast with array sizes up to ~4500. This covers
8572 : * all supported BLCKSZ values.
8573 : */
8574 7559 : const int gaps[9] = {1968, 861, 336, 112, 48, 21, 7, 3, 1};
8575 :
8576 : /* Think carefully before changing anything here -- keep swaps cheap */
8577 : StaticAssertDecl(sizeof(TM_IndexDelete) <= 8,
8578 : "element size exceeds 8 bytes");
8579 :
8580 75590 : for (int g = 0; g < lengthof(gaps); g++)
8581 : {
8582 9370333 : for (int hi = gaps[g], i = hi; i < ndeltids; i++)
8583 : {
8584 9302302 : TM_IndexDelete d = deltids[i];
8585 9302302 : int j = i;
8586 :
8587 16021147 : while (j >= hi && index_delete_sort_cmp(&deltids[j - hi], &d) >= 0)
8588 : {
8589 6718845 : deltids[j] = deltids[j - hi];
8590 6718845 : j -= hi;
8591 : }
8592 9302302 : deltids[j] = d;
8593 : }
8594 : }
8595 7559 : }
8596 :
8597 : /*
8598 : * Returns how many blocks should be considered favorable/contiguous for a
8599 : * bottom-up index deletion pass. This is a number of heap blocks that starts
8600 : * from and includes the first block in line.
8601 : *
8602 : * There is always at least one favorable block during bottom-up index
8603 : * deletion. In the worst case (i.e. with totally random heap blocks) the
8604 : * first block in line (the only favorable block) can be thought of as a
8605 : * degenerate array of contiguous blocks that consists of a single block.
8606 : * heap_index_delete_tuples() will expect this.
8607 : *
8608 : * Caller passes blockgroups, a description of the final order that deltids
8609 : * will be sorted in for heap_index_delete_tuples() bottom-up index deletion
8610 : * processing. Note that deltids need not actually be sorted just yet (caller
8611 : * only passes deltids to us so that we can interpret blockgroups).
8612 : *
8613 : * You might guess that the existence of contiguous blocks cannot matter much,
8614 : * since in general the main factor that determines which blocks we visit is
8615 : * the number of promising TIDs, which is a fixed hint from the index AM.
8616 : * We're not really targeting the general case, though -- the actual goal is
8617 : * to adapt our behavior to a wide variety of naturally occurring conditions.
8618 : * The effects of most of the heuristics we apply are only noticeable in the
8619 : * aggregate, over time and across many _related_ bottom-up index deletion
8620 : * passes.
8621 : *
8622 : * Deeming certain blocks favorable allows heapam to recognize and adapt to
8623 : * workloads where heap blocks visited during bottom-up index deletion can be
8624 : * accessed contiguously, in the sense that each newly visited block is the
8625 : * neighbor of the block that bottom-up deletion just finished processing (or
8626 : * close enough to it). It will likely be cheaper to access more favorable
8627 : * blocks sooner rather than later (e.g. in this pass, not across a series of
8628 : * related bottom-up passes). Either way it is probably only a matter of time
8629 : * (or a matter of further correlated version churn) before all blocks that
8630 : * appear together as a single large batch of favorable blocks get accessed by
8631 : * _some_ bottom-up pass. Large batches of favorable blocks tend to either
8632 : * appear almost constantly or not even once (it all depends on per-index
8633 : * workload characteristics).
8634 : *
8635 : * Note that the blockgroups sort order applies a power-of-two bucketing
8636 : * scheme that creates opportunities for contiguous groups of blocks to get
8637 : * batched together, at least with workloads that are naturally amenable to
8638 : * being driven by heap block locality. This doesn't just enhance the spatial
8639 : * locality of bottom-up heap block processing in the obvious way. It also
8640 : * enables temporal locality of access, since sorting by heap block number
8641 : * naturally tends to make the bottom-up processing order deterministic.
8642 : *
8643 : * Consider the following example to get a sense of how temporal locality
8644 : * might matter: There is a heap relation with several indexes, each of which
8645 : * is low to medium cardinality. It is subject to constant non-HOT updates.
8646 : * The updates are skewed (in one part of the primary key, perhaps). None of
8647 : * the indexes are logically modified by the UPDATE statements (if they were
8648 : * then bottom-up index deletion would not be triggered in the first place).
8649 : * Naturally, each new round of index tuples (for each heap tuple that gets a
8650 : * heap_update() call) will have the same heap TID in each and every index.
8651 : * Since these indexes are low cardinality and never get logically modified,
8652 : * heapam processing during bottom-up deletion passes will access heap blocks
8653 : * in approximately sequential order. Temporal locality of access occurs due
8654 : * to bottom-up deletion passes behaving very similarly across each of the
8655 : * indexes at any given moment. This keeps the number of buffer misses needed
8656 : * to visit heap blocks to a minimum.
8657 : */
8658 : static int
8659 2283 : bottomup_nblocksfavorable(IndexDeleteCounts *blockgroups, int nblockgroups,
8660 : TM_IndexDelete *deltids)
8661 : {
8662 2283 : int64 lastblock = -1;
8663 2283 : int nblocksfavorable = 0;
8664 :
8665 : Assert(nblockgroups >= 1);
8666 : Assert(nblockgroups <= BOTTOMUP_MAX_NBLOCKS);
8667 :
8668 : /*
8669 : * We tolerate heap blocks that will be accessed only slightly out of
8670 : * physical order. Small blips occur when a pair of almost-contiguous
8671 : * blocks happen to fall into different buckets (perhaps due only to a
8672 : * small difference in npromisingtids that the bucketing scheme didn't
8673 : * quite manage to ignore). We effectively ignore these blips by applying
8674 : * a small tolerance. The precise tolerance we use is a little arbitrary,
8675 : * but it works well enough in practice.
8676 : */
8677 7206 : for (int b = 0; b < nblockgroups; b++)
8678 : {
8679 6895 : IndexDeleteCounts *group = blockgroups + b;
8680 6895 : TM_IndexDelete *firstdtid = deltids + group->ifirsttid;
8681 6895 : BlockNumber block = ItemPointerGetBlockNumber(&firstdtid->tid);
8682 :
8683 6895 : if (lastblock != -1 &&
8684 4612 : ((int64) block < lastblock - BOTTOMUP_TOLERANCE_NBLOCKS ||
8685 3990 : (int64) block > lastblock + BOTTOMUP_TOLERANCE_NBLOCKS))
8686 : break;
8687 :
8688 4923 : nblocksfavorable++;
8689 4923 : lastblock = block;
8690 : }
8691 :
8692 : /* Always indicate that there is at least 1 favorable block */
8693 : Assert(nblocksfavorable >= 1);
8694 :
8695 2283 : return nblocksfavorable;
8696 : }
8697 :
8698 : /*
8699 : * qsort comparison function for bottomup_sort_and_shrink()
8700 : */
8701 : static int
8702 258955 : bottomup_sort_and_shrink_cmp(const void *arg1, const void *arg2)
8703 : {
8704 258955 : const IndexDeleteCounts *group1 = (const IndexDeleteCounts *) arg1;
8705 258955 : const IndexDeleteCounts *group2 = (const IndexDeleteCounts *) arg2;
8706 :
8707 : /*
8708 : * Most significant field is npromisingtids (which we invert the order of
8709 : * so as to sort in desc order).
8710 : *
8711 : * Caller should have already normalized npromisingtids fields into
8712 : * power-of-two values (buckets).
8713 : */
8714 258955 : if (group1->npromisingtids > group2->npromisingtids)
8715 11657 : return -1;
8716 247298 : if (group1->npromisingtids < group2->npromisingtids)
8717 14725 : return 1;
8718 :
8719 : /*
8720 : * Tiebreak: desc ntids sort order.
8721 : *
8722 : * We cannot expect power-of-two values for ntids fields. We should
8723 : * behave as if they were already rounded up for us instead.
8724 : */
8725 232573 : if (group1->ntids != group2->ntids)
8726 : {
8727 166068 : uint32 ntids1 = pg_nextpower2_32((uint32) group1->ntids);
8728 166068 : uint32 ntids2 = pg_nextpower2_32((uint32) group2->ntids);
8729 :
8730 166068 : if (ntids1 > ntids2)
8731 23804 : return -1;
8732 142264 : if (ntids1 < ntids2)
8733 31246 : return 1;
8734 : }
8735 :
8736 : /*
8737 : * Tiebreak: asc offset-into-deltids-for-block (offset to first TID for
8738 : * block in deltids array) order.
8739 : *
8740 : * This is equivalent to sorting in ascending heap block number order
8741 : * (among otherwise equal subsets of the array). This approach allows us
8742 : * to avoid accessing the out-of-line TID. (We rely on the assumption
8743 : * that the deltids array was sorted in ascending heap TID order when
8744 : * these offsets to the first TID from each heap block group were formed.)
8745 : */
8746 177523 : if (group1->ifirsttid > group2->ifirsttid)
8747 88208 : return 1;
8748 89315 : if (group1->ifirsttid < group2->ifirsttid)
8749 89315 : return -1;
8750 :
8751 0 : pg_unreachable();
8752 :
8753 : return 0;
8754 : }
8755 :
8756 : /*
8757 : * heap_index_delete_tuples() helper function for bottom-up deletion callers.
8758 : *
8759 : * Sorts deltids array in the order needed for useful processing by bottom-up
8760 : * deletion. The array should already be sorted in TID order when we're
8761 : * called. The sort process groups heap TIDs from deltids into heap block
8762 : * groupings. Earlier/more-promising groups/blocks are usually those that are
8763 : * known to have the most "promising" TIDs.
8764 : *
8765 : * Sets new size of deltids array (ndeltids) in state. deltids will only have
8766 : * TIDs from the BOTTOMUP_MAX_NBLOCKS most promising heap blocks when we
8767 : * return. This often means that deltids will be shrunk to a small fraction
8768 : * of its original size (we eliminate many heap blocks from consideration for
8769 : * caller up front).
8770 : *
8771 : * Returns the number of "favorable" blocks. See bottomup_nblocksfavorable()
8772 : * for a definition and full details.
8773 : */
8774 : static int
8775 2283 : bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate)
8776 : {
8777 : IndexDeleteCounts *blockgroups;
8778 : TM_IndexDelete *reordereddeltids;
8779 2283 : BlockNumber curblock = InvalidBlockNumber;
8780 2283 : int nblockgroups = 0;
8781 2283 : int ncopied = 0;
8782 2283 : int nblocksfavorable = 0;
8783 :
8784 : Assert(delstate->bottomup);
8785 : Assert(delstate->ndeltids > 0);
8786 :
8787 : /* Calculate per-heap-block count of TIDs */
8788 2283 : blockgroups = palloc_array(IndexDeleteCounts, delstate->ndeltids);
8789 1133689 : for (int i = 0; i < delstate->ndeltids; i++)
8790 : {
8791 1131406 : TM_IndexDelete *ideltid = &delstate->deltids[i];
8792 1131406 : TM_IndexStatus *istatus = delstate->status + ideltid->id;
8793 1131406 : ItemPointer htid = &ideltid->tid;
8794 1131406 : bool promising = istatus->promising;
8795 :
8796 1131406 : if (curblock != ItemPointerGetBlockNumber(htid))
8797 : {
8798 : /* New block group */
8799 48740 : nblockgroups++;
8800 :
8801 : Assert(curblock < ItemPointerGetBlockNumber(htid) ||
8802 : !BlockNumberIsValid(curblock));
8803 :
8804 48740 : curblock = ItemPointerGetBlockNumber(htid);
8805 48740 : blockgroups[nblockgroups - 1].ifirsttid = i;
8806 48740 : blockgroups[nblockgroups - 1].ntids = 1;
8807 48740 : blockgroups[nblockgroups - 1].npromisingtids = 0;
8808 : }
8809 : else
8810 : {
8811 1082666 : blockgroups[nblockgroups - 1].ntids++;
8812 : }
8813 :
8814 1131406 : if (promising)
8815 140871 : blockgroups[nblockgroups - 1].npromisingtids++;
8816 : }
8817 :
8818 : /*
8819 : * We're about ready to sort block groups to determine the optimal order
8820 : * for visiting heap blocks. But before we do, round the number of
8821 : * promising tuples for each block group up to the next power-of-two,
8822 : * unless it is very low (less than 4), in which case we round up to 4.
8823 : * npromisingtids is far too noisy to trust when choosing between a pair
8824 : * of block groups that both have very low values.
8825 : *
8826 : * This scheme divides heap blocks/block groups into buckets. Each bucket
8827 : * contains blocks that have _approximately_ the same number of promising
8828 : * TIDs as each other. The goal is to ignore relatively small differences
8829 : * in the total number of promising entries, so that the whole process can
8830 : * give a little weight to heapam factors (like heap block locality)
8831 : * instead. This isn't a trade-off, really -- we have nothing to lose. It
8832 : * would be foolish to interpret small differences in npromisingtids
8833 : * values as anything more than noise.
8834 : *
8835 : * We tiebreak on nhtids when sorting block group subsets that have the
8836 : * same npromisingtids, but this has the same issues as npromisingtids,
8837 : * and so nhtids is subject to the same power-of-two bucketing scheme. The
8838 : * only reason that we don't fix nhtids in the same way here too is that
8839 : * we'll need accurate nhtids values after the sort. We handle nhtids
8840 : * bucketization dynamically instead (in the sort comparator).
8841 : *
8842 : * See bottomup_nblocksfavorable() for a full explanation of when and how
8843 : * heap locality/favorable blocks can significantly influence when and how
8844 : * heap blocks are accessed.
8845 : */
8846 51023 : for (int b = 0; b < nblockgroups; b++)
8847 : {
8848 48740 : IndexDeleteCounts *group = blockgroups + b;
8849 :
8850 : /* Better off falling back on nhtids with low npromisingtids */
8851 48740 : if (group->npromisingtids <= 4)
8852 41761 : group->npromisingtids = 4;
8853 : else
8854 6979 : group->npromisingtids =
8855 6979 : pg_nextpower2_32((uint32) group->npromisingtids);
8856 : }
8857 :
8858 : /* Sort groups and rearrange caller's deltids array */
8859 2283 : qsort(blockgroups, nblockgroups, sizeof(IndexDeleteCounts),
8860 : bottomup_sort_and_shrink_cmp);
8861 2283 : reordereddeltids = palloc(delstate->ndeltids * sizeof(TM_IndexDelete));
8862 :
8863 2283 : nblockgroups = Min(BOTTOMUP_MAX_NBLOCKS, nblockgroups);
8864 : /* Determine number of favorable blocks at the start of final deltids */
8865 2283 : nblocksfavorable = bottomup_nblocksfavorable(blockgroups, nblockgroups,
8866 : delstate->deltids);
8867 :
8868 15351 : for (int b = 0; b < nblockgroups; b++)
8869 : {
8870 13068 : IndexDeleteCounts *group = blockgroups + b;
8871 13068 : TM_IndexDelete *firstdtid = delstate->deltids + group->ifirsttid;
8872 :
8873 13068 : memcpy(reordereddeltids + ncopied, firstdtid,
8874 13068 : sizeof(TM_IndexDelete) * group->ntids);
8875 13068 : ncopied += group->ntids;
8876 : }
8877 :
8878 : /* Copy final grouped and sorted TIDs back into start of caller's array */
8879 2283 : memcpy(delstate->deltids, reordereddeltids,
8880 : sizeof(TM_IndexDelete) * ncopied);
8881 2283 : delstate->ndeltids = ncopied;
8882 :
8883 2283 : pfree(reordereddeltids);
8884 2283 : pfree(blockgroups);
8885 :
8886 2283 : return nblocksfavorable;
8887 : }
8888 :
8889 : /*
8890 : * Perform XLogInsert for a heap-visible operation. 'block' is the block
8891 : * being marked all-visible, and vm_buffer is the buffer containing the
8892 : * corresponding visibility map block. Both should have already been modified
8893 : * and dirtied.
8894 : *
8895 : * snapshotConflictHorizon comes from the largest xmin on the page being
8896 : * marked all-visible. REDO routine uses it to generate recovery conflicts.
8897 : *
8898 : * If checksums or wal_log_hints are enabled, we may also generate a full-page
8899 : * image of heap_buffer. Otherwise, we optimize away the FPI (by specifying
8900 : * REGBUF_NO_IMAGE for the heap buffer), in which case the caller should *not*
8901 : * update the heap page's LSN.
8902 : */
8903 : XLogRecPtr
8904 41721 : log_heap_visible(Relation rel, Buffer heap_buffer, Buffer vm_buffer,
8905 : TransactionId snapshotConflictHorizon, uint8 vmflags)
8906 : {
8907 : xl_heap_visible xlrec;
8908 : XLogRecPtr recptr;
8909 : uint8 flags;
8910 :
8911 : Assert(BufferIsValid(heap_buffer));
8912 : Assert(BufferIsValid(vm_buffer));
8913 :
8914 41721 : xlrec.snapshotConflictHorizon = snapshotConflictHorizon;
8915 41721 : xlrec.flags = vmflags;
8916 41721 : if (RelationIsAccessibleInLogicalDecoding(rel))
8917 54 : xlrec.flags |= VISIBILITYMAP_XLOG_CATALOG_REL;
8918 41721 : XLogBeginInsert();
8919 41721 : XLogRegisterData(&xlrec, SizeOfHeapVisible);
8920 :
8921 41721 : XLogRegisterBuffer(0, vm_buffer, 0);
8922 :
8923 41721 : flags = REGBUF_STANDARD;
8924 41721 : if (!XLogHintBitIsNeeded())
8925 3246 : flags |= REGBUF_NO_IMAGE;
8926 41721 : XLogRegisterBuffer(1, heap_buffer, flags);
8927 :
8928 41721 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_VISIBLE);
8929 :
8930 41721 : return recptr;
8931 : }
8932 :
8933 : /*
8934 : * Perform XLogInsert for a heap-update operation. Caller must already
8935 : * have modified the buffer(s) and marked them dirty.
8936 : */
8937 : static XLogRecPtr
8938 324625 : log_heap_update(Relation reln, Buffer oldbuf,
8939 : Buffer newbuf, HeapTuple oldtup, HeapTuple newtup,
8940 : HeapTuple old_key_tuple,
8941 : bool all_visible_cleared, bool new_all_visible_cleared)
8942 : {
8943 : xl_heap_update xlrec;
8944 : xl_heap_header xlhdr;
8945 : xl_heap_header xlhdr_idx;
8946 : uint8 info;
8947 : uint16 prefix_suffix[2];
8948 324625 : uint16 prefixlen = 0,
8949 324625 : suffixlen = 0;
8950 : XLogRecPtr recptr;
8951 324625 : Page page = BufferGetPage(newbuf);
8952 324625 : bool need_tuple_data = RelationIsLogicallyLogged(reln);
8953 : bool init;
8954 : int bufflags;
8955 :
8956 : /* Caller should not call me on a non-WAL-logged relation */
8957 : Assert(RelationNeedsWAL(reln));
8958 :
8959 324625 : XLogBeginInsert();
8960 :
8961 324625 : if (HeapTupleIsHeapOnly(newtup))
8962 162275 : info = XLOG_HEAP_HOT_UPDATE;
8963 : else
8964 162350 : info = XLOG_HEAP_UPDATE;
8965 :
8966 : /*
8967 : * If the old and new tuple are on the same page, we only need to log the
8968 : * parts of the new tuple that were changed. That saves on the amount of
8969 : * WAL we need to write. Currently, we just count any unchanged bytes in
8970 : * the beginning and end of the tuple. That's quick to check, and
8971 : * perfectly covers the common case that only one field is updated.
8972 : *
8973 : * We could do this even if the old and new tuple are on different pages,
8974 : * but only if we don't make a full-page image of the old page, which is
8975 : * difficult to know in advance. Also, if the old tuple is corrupt for
8976 : * some reason, it would allow the corruption to propagate the new page,
8977 : * so it seems best to avoid. Under the general assumption that most
8978 : * updates tend to create the new tuple version on the same page, there
8979 : * isn't much to be gained by doing this across pages anyway.
8980 : *
8981 : * Skip this if we're taking a full-page image of the new page, as we
8982 : * don't include the new tuple in the WAL record in that case. Also
8983 : * disable if effective_wal_level='logical', as logical decoding needs to
8984 : * be able to read the new tuple in whole from the WAL record alone.
8985 : */
8986 324625 : if (oldbuf == newbuf && !need_tuple_data &&
8987 166168 : !XLogCheckBufferNeedsBackup(newbuf))
8988 : {
8989 165616 : char *oldp = (char *) oldtup->t_data + oldtup->t_data->t_hoff;
8990 165616 : char *newp = (char *) newtup->t_data + newtup->t_data->t_hoff;
8991 165616 : int oldlen = oldtup->t_len - oldtup->t_data->t_hoff;
8992 165616 : int newlen = newtup->t_len - newtup->t_data->t_hoff;
8993 :
8994 : /* Check for common prefix between old and new tuple */
8995 13888096 : for (prefixlen = 0; prefixlen < Min(oldlen, newlen); prefixlen++)
8996 : {
8997 13861563 : if (newp[prefixlen] != oldp[prefixlen])
8998 139083 : break;
8999 : }
9000 :
9001 : /*
9002 : * Storing the length of the prefix takes 2 bytes, so we need to save
9003 : * at least 3 bytes or there's no point.
9004 : */
9005 165616 : if (prefixlen < 3)
9006 22615 : prefixlen = 0;
9007 :
9008 : /* Same for suffix */
9009 6100592 : for (suffixlen = 0; suffixlen < Min(oldlen, newlen) - prefixlen; suffixlen++)
9010 : {
9011 6073741 : if (newp[newlen - suffixlen - 1] != oldp[oldlen - suffixlen - 1])
9012 138765 : break;
9013 : }
9014 165616 : if (suffixlen < 3)
9015 39569 : suffixlen = 0;
9016 : }
9017 :
9018 : /* Prepare main WAL data chain */
9019 324625 : xlrec.flags = 0;
9020 324625 : if (all_visible_cleared)
9021 2114 : xlrec.flags |= XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED;
9022 324625 : if (new_all_visible_cleared)
9023 787 : xlrec.flags |= XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED;
9024 324625 : if (prefixlen > 0)
9025 143001 : xlrec.flags |= XLH_UPDATE_PREFIX_FROM_OLD;
9026 324625 : if (suffixlen > 0)
9027 126047 : xlrec.flags |= XLH_UPDATE_SUFFIX_FROM_OLD;
9028 324625 : if (need_tuple_data)
9029 : {
9030 47025 : xlrec.flags |= XLH_UPDATE_CONTAINS_NEW_TUPLE;
9031 47025 : if (old_key_tuple)
9032 : {
9033 147 : if (reln->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
9034 65 : xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_TUPLE;
9035 : else
9036 82 : xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_KEY;
9037 : }
9038 : }
9039 :
9040 : /* If new tuple is the single and first tuple on page... */
9041 328554 : if (ItemPointerGetOffsetNumber(&(newtup->t_self)) == FirstOffsetNumber &&
9042 3929 : PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
9043 : {
9044 3724 : info |= XLOG_HEAP_INIT_PAGE;
9045 3724 : init = true;
9046 : }
9047 : else
9048 320901 : init = false;
9049 :
9050 : /* Prepare WAL data for the old page */
9051 324625 : xlrec.old_offnum = ItemPointerGetOffsetNumber(&oldtup->t_self);
9052 324625 : xlrec.old_xmax = HeapTupleHeaderGetRawXmax(oldtup->t_data);
9053 649250 : xlrec.old_infobits_set = compute_infobits(oldtup->t_data->t_infomask,
9054 324625 : oldtup->t_data->t_infomask2);
9055 :
9056 : /* Prepare WAL data for the new page */
9057 324625 : xlrec.new_offnum = ItemPointerGetOffsetNumber(&newtup->t_self);
9058 324625 : xlrec.new_xmax = HeapTupleHeaderGetRawXmax(newtup->t_data);
9059 :
9060 324625 : bufflags = REGBUF_STANDARD;
9061 324625 : if (init)
9062 3724 : bufflags |= REGBUF_WILL_INIT;
9063 324625 : if (need_tuple_data)
9064 47025 : bufflags |= REGBUF_KEEP_DATA;
9065 :
9066 324625 : XLogRegisterBuffer(0, newbuf, bufflags);
9067 324625 : if (oldbuf != newbuf)
9068 146512 : XLogRegisterBuffer(1, oldbuf, REGBUF_STANDARD);
9069 :
9070 324625 : XLogRegisterData(&xlrec, SizeOfHeapUpdate);
9071 :
9072 : /*
9073 : * Prepare WAL data for the new tuple.
9074 : */
9075 324625 : if (prefixlen > 0 || suffixlen > 0)
9076 : {
9077 165050 : if (prefixlen > 0 && suffixlen > 0)
9078 : {
9079 103998 : prefix_suffix[0] = prefixlen;
9080 103998 : prefix_suffix[1] = suffixlen;
9081 103998 : XLogRegisterBufData(0, &prefix_suffix, sizeof(uint16) * 2);
9082 : }
9083 61052 : else if (prefixlen > 0)
9084 : {
9085 39003 : XLogRegisterBufData(0, &prefixlen, sizeof(uint16));
9086 : }
9087 : else
9088 : {
9089 22049 : XLogRegisterBufData(0, &suffixlen, sizeof(uint16));
9090 : }
9091 : }
9092 :
9093 324625 : xlhdr.t_infomask2 = newtup->t_data->t_infomask2;
9094 324625 : xlhdr.t_infomask = newtup->t_data->t_infomask;
9095 324625 : xlhdr.t_hoff = newtup->t_data->t_hoff;
9096 : Assert(SizeofHeapTupleHeader + prefixlen + suffixlen <= newtup->t_len);
9097 :
9098 : /*
9099 : * PG73FORMAT: write bitmap [+ padding] [+ oid] + data
9100 : *
9101 : * The 'data' doesn't include the common prefix or suffix.
9102 : */
9103 324625 : XLogRegisterBufData(0, &xlhdr, SizeOfHeapHeader);
9104 324625 : if (prefixlen == 0)
9105 : {
9106 181624 : XLogRegisterBufData(0,
9107 181624 : (char *) newtup->t_data + SizeofHeapTupleHeader,
9108 181624 : newtup->t_len - SizeofHeapTupleHeader - suffixlen);
9109 : }
9110 : else
9111 : {
9112 : /*
9113 : * Have to write the null bitmap and data after the common prefix as
9114 : * two separate rdata entries.
9115 : */
9116 : /* bitmap [+ padding] [+ oid] */
9117 143001 : if (newtup->t_data->t_hoff - SizeofHeapTupleHeader > 0)
9118 : {
9119 143001 : XLogRegisterBufData(0,
9120 143001 : (char *) newtup->t_data + SizeofHeapTupleHeader,
9121 143001 : newtup->t_data->t_hoff - SizeofHeapTupleHeader);
9122 : }
9123 :
9124 : /* data after common prefix */
9125 143001 : XLogRegisterBufData(0,
9126 143001 : (char *) newtup->t_data + newtup->t_data->t_hoff + prefixlen,
9127 143001 : newtup->t_len - newtup->t_data->t_hoff - prefixlen - suffixlen);
9128 : }
9129 :
9130 : /* We need to log a tuple identity */
9131 324625 : if (need_tuple_data && old_key_tuple)
9132 : {
9133 : /* don't really need this, but its more comfy to decode */
9134 147 : xlhdr_idx.t_infomask2 = old_key_tuple->t_data->t_infomask2;
9135 147 : xlhdr_idx.t_infomask = old_key_tuple->t_data->t_infomask;
9136 147 : xlhdr_idx.t_hoff = old_key_tuple->t_data->t_hoff;
9137 :
9138 147 : XLogRegisterData(&xlhdr_idx, SizeOfHeapHeader);
9139 :
9140 : /* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
9141 147 : XLogRegisterData((char *) old_key_tuple->t_data + SizeofHeapTupleHeader,
9142 147 : old_key_tuple->t_len - SizeofHeapTupleHeader);
9143 : }
9144 :
9145 : /* filtering by origin on a row level is much more efficient */
9146 324625 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
9147 :
9148 324625 : recptr = XLogInsert(RM_HEAP_ID, info);
9149 :
9150 324625 : return recptr;
9151 : }
9152 :
9153 : /*
9154 : * Perform XLogInsert of an XLOG_HEAP2_NEW_CID record
9155 : *
9156 : * This is only used when effective_wal_level is logical, and only for
9157 : * catalog tuples.
9158 : */
9159 : static XLogRecPtr
9160 25088 : log_heap_new_cid(Relation relation, HeapTuple tup)
9161 : {
9162 : xl_heap_new_cid xlrec;
9163 :
9164 : XLogRecPtr recptr;
9165 25088 : HeapTupleHeader hdr = tup->t_data;
9166 :
9167 : Assert(ItemPointerIsValid(&tup->t_self));
9168 : Assert(tup->t_tableOid != InvalidOid);
9169 :
9170 25088 : xlrec.top_xid = GetTopTransactionId();
9171 25088 : xlrec.target_locator = relation->rd_locator;
9172 25088 : xlrec.target_tid = tup->t_self;
9173 :
9174 : /*
9175 : * If the tuple got inserted & deleted in the same TX we definitely have a
9176 : * combo CID, set cmin and cmax.
9177 : */
9178 25088 : if (hdr->t_infomask & HEAP_COMBOCID)
9179 : {
9180 : Assert(!(hdr->t_infomask & HEAP_XMAX_INVALID));
9181 : Assert(!HeapTupleHeaderXminInvalid(hdr));
9182 2027 : xlrec.cmin = HeapTupleHeaderGetCmin(hdr);
9183 2027 : xlrec.cmax = HeapTupleHeaderGetCmax(hdr);
9184 2027 : xlrec.combocid = HeapTupleHeaderGetRawCommandId(hdr);
9185 : }
9186 : /* No combo CID, so only cmin or cmax can be set by this TX */
9187 : else
9188 : {
9189 : /*
9190 : * Tuple inserted.
9191 : *
9192 : * We need to check for LOCK ONLY because multixacts might be
9193 : * transferred to the new tuple in case of FOR KEY SHARE updates in
9194 : * which case there will be an xmax, although the tuple just got
9195 : * inserted.
9196 : */
9197 30061 : if (hdr->t_infomask & HEAP_XMAX_INVALID ||
9198 7000 : HEAP_XMAX_IS_LOCKED_ONLY(hdr->t_infomask))
9199 : {
9200 16062 : xlrec.cmin = HeapTupleHeaderGetRawCommandId(hdr);
9201 16062 : xlrec.cmax = InvalidCommandId;
9202 : }
9203 : /* Tuple from a different tx updated or deleted. */
9204 : else
9205 : {
9206 6999 : xlrec.cmin = InvalidCommandId;
9207 6999 : xlrec.cmax = HeapTupleHeaderGetRawCommandId(hdr);
9208 : }
9209 23061 : xlrec.combocid = InvalidCommandId;
9210 : }
9211 :
9212 : /*
9213 : * Note that we don't need to register the buffer here, because this
9214 : * operation does not modify the page. The insert/update/delete that
9215 : * called us certainly did, but that's WAL-logged separately.
9216 : */
9217 25088 : XLogBeginInsert();
9218 25088 : XLogRegisterData(&xlrec, SizeOfHeapNewCid);
9219 :
9220 : /* will be looked at irrespective of origin */
9221 :
9222 25088 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_NEW_CID);
9223 :
9224 25088 : return recptr;
9225 : }
9226 :
9227 : /*
9228 : * Build a heap tuple representing the configured REPLICA IDENTITY to represent
9229 : * the old tuple in an UPDATE or DELETE.
9230 : *
9231 : * Returns NULL if there's no need to log an identity or if there's no suitable
9232 : * key defined.
9233 : *
9234 : * Pass key_required true if any replica identity columns changed value, or if
9235 : * any of them have any external data. Delete must always pass true.
9236 : *
9237 : * *copy is set to true if the returned tuple is a modified copy rather than
9238 : * the same tuple that was passed in.
9239 : */
9240 : static HeapTuple
9241 2181001 : ExtractReplicaIdentity(Relation relation, HeapTuple tp, bool key_required,
9242 : bool *copy)
9243 : {
9244 2181001 : TupleDesc desc = RelationGetDescr(relation);
9245 2181001 : char replident = relation->rd_rel->relreplident;
9246 : Bitmapset *idattrs;
9247 : HeapTuple key_tuple;
9248 : bool nulls[MaxHeapAttributeNumber];
9249 : Datum values[MaxHeapAttributeNumber];
9250 :
9251 2181001 : *copy = false;
9252 :
9253 2181001 : if (!RelationIsLogicallyLogged(relation))
9254 2080708 : return NULL;
9255 :
9256 100293 : if (replident == REPLICA_IDENTITY_NOTHING)
9257 228 : return NULL;
9258 :
9259 100065 : if (replident == REPLICA_IDENTITY_FULL)
9260 : {
9261 : /*
9262 : * When logging the entire old tuple, it very well could contain
9263 : * toasted columns. If so, force them to be inlined.
9264 : */
9265 197 : if (HeapTupleHasExternal(tp))
9266 : {
9267 4 : *copy = true;
9268 4 : tp = toast_flatten_tuple(tp, desc);
9269 : }
9270 197 : return tp;
9271 : }
9272 :
9273 : /* if the key isn't required and we're only logging the key, we're done */
9274 99868 : if (!key_required)
9275 46878 : return NULL;
9276 :
9277 : /* find out the replica identity columns */
9278 52990 : idattrs = RelationGetIndexAttrBitmap(relation,
9279 : INDEX_ATTR_BITMAP_IDENTITY_KEY);
9280 :
9281 : /*
9282 : * If there's no defined replica identity columns, treat as !key_required.
9283 : * (This case should not be reachable from heap_update, since that should
9284 : * calculate key_required accurately. But heap_delete just passes
9285 : * constant true for key_required, so we can hit this case in deletes.)
9286 : */
9287 52990 : if (bms_is_empty(idattrs))
9288 6021 : return NULL;
9289 :
9290 : /*
9291 : * Construct a new tuple containing only the replica identity columns,
9292 : * with nulls elsewhere. While we're at it, assert that the replica
9293 : * identity columns aren't null.
9294 : */
9295 46969 : heap_deform_tuple(tp, desc, values, nulls);
9296 :
9297 150902 : for (int i = 0; i < desc->natts; i++)
9298 : {
9299 103933 : if (bms_is_member(i + 1 - FirstLowInvalidHeapAttributeNumber,
9300 : idattrs))
9301 : Assert(!nulls[i]);
9302 : else
9303 56952 : nulls[i] = true;
9304 : }
9305 :
9306 46969 : key_tuple = heap_form_tuple(desc, values, nulls);
9307 46969 : *copy = true;
9308 :
9309 46969 : bms_free(idattrs);
9310 :
9311 : /*
9312 : * If the tuple, which by here only contains indexed columns, still has
9313 : * toasted columns, force them to be inlined. This is somewhat unlikely
9314 : * since there's limits on the size of indexed columns, so we don't
9315 : * duplicate toast_flatten_tuple()s functionality in the above loop over
9316 : * the indexed columns, even if it would be more efficient.
9317 : */
9318 46969 : if (HeapTupleHasExternal(key_tuple))
9319 : {
9320 4 : HeapTuple oldtup = key_tuple;
9321 :
9322 4 : key_tuple = toast_flatten_tuple(oldtup, desc);
9323 4 : heap_freetuple(oldtup);
9324 : }
9325 :
9326 46969 : return key_tuple;
9327 : }
9328 :
9329 : /*
9330 : * HeapCheckForSerializableConflictOut
9331 : * We are reading a tuple. If it's not visible, there may be a
9332 : * rw-conflict out with the inserter. Otherwise, if it is visible to us
9333 : * but has been deleted, there may be a rw-conflict out with the deleter.
9334 : *
9335 : * We will determine the top level xid of the writing transaction with which
9336 : * we may be in conflict, and ask CheckForSerializableConflictOut() to check
9337 : * for overlap with our own transaction.
9338 : *
9339 : * This function should be called just about anywhere in heapam.c where a
9340 : * tuple has been read. The caller must hold at least a shared lock on the
9341 : * buffer, because this function might set hint bits on the tuple. There is
9342 : * currently no known reason to call this function from an index AM.
9343 : */
9344 : void
9345 39809497 : HeapCheckForSerializableConflictOut(bool visible, Relation relation,
9346 : HeapTuple tuple, Buffer buffer,
9347 : Snapshot snapshot)
9348 : {
9349 : TransactionId xid;
9350 : HTSV_Result htsvResult;
9351 :
9352 39809497 : if (!CheckForSerializableConflictOutNeeded(relation, snapshot))
9353 39784089 : return;
9354 :
9355 : /*
9356 : * Check to see whether the tuple has been written to by a concurrent
9357 : * transaction, either to create it not visible to us, or to delete it
9358 : * while it is visible to us. The "visible" bool indicates whether the
9359 : * tuple is visible to us, while HeapTupleSatisfiesVacuum checks what else
9360 : * is going on with it.
9361 : *
9362 : * In the event of a concurrently inserted tuple that also happens to have
9363 : * been concurrently updated (by a separate transaction), the xmin of the
9364 : * tuple will be used -- not the updater's xid.
9365 : */
9366 25408 : htsvResult = HeapTupleSatisfiesVacuum(tuple, TransactionXmin, buffer);
9367 25408 : switch (htsvResult)
9368 : {
9369 24610 : case HEAPTUPLE_LIVE:
9370 24610 : if (visible)
9371 24597 : return;
9372 13 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9373 13 : break;
9374 361 : case HEAPTUPLE_RECENTLY_DEAD:
9375 : case HEAPTUPLE_DELETE_IN_PROGRESS:
9376 361 : if (visible)
9377 286 : xid = HeapTupleHeaderGetUpdateXid(tuple->t_data);
9378 : else
9379 75 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9380 :
9381 361 : if (TransactionIdPrecedes(xid, TransactionXmin))
9382 : {
9383 : /* This is like the HEAPTUPLE_DEAD case */
9384 : Assert(!visible);
9385 70 : return;
9386 : }
9387 291 : break;
9388 327 : case HEAPTUPLE_INSERT_IN_PROGRESS:
9389 327 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9390 327 : break;
9391 110 : case HEAPTUPLE_DEAD:
9392 : Assert(!visible);
9393 110 : return;
9394 0 : default:
9395 :
9396 : /*
9397 : * The only way to get to this default clause is if a new value is
9398 : * added to the enum type without adding it to this switch
9399 : * statement. That's a bug, so elog.
9400 : */
9401 0 : elog(ERROR, "unrecognized return value from HeapTupleSatisfiesVacuum: %u", htsvResult);
9402 :
9403 : /*
9404 : * In spite of having all enum values covered and calling elog on
9405 : * this default, some compilers think this is a code path which
9406 : * allows xid to be used below without initialization. Silence
9407 : * that warning.
9408 : */
9409 : xid = InvalidTransactionId;
9410 : }
9411 :
9412 : Assert(TransactionIdIsValid(xid));
9413 : Assert(TransactionIdFollowsOrEquals(xid, TransactionXmin));
9414 :
9415 : /*
9416 : * Find top level xid. Bail out if xid is too early to be a conflict, or
9417 : * if it's our own xid.
9418 : */
9419 631 : if (TransactionIdEquals(xid, GetTopTransactionIdIfAny()))
9420 64 : return;
9421 567 : xid = SubTransGetTopmostTransaction(xid);
9422 567 : if (TransactionIdPrecedes(xid, TransactionXmin))
9423 0 : return;
9424 :
9425 567 : CheckForSerializableConflictOut(relation, xid, snapshot);
9426 : }
|
