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 10610555 : 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 10610555 : }
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 101738 : heap_scan_stream_read_next_parallel(ReadStream *stream,
253 : void *callback_private_data,
254 : void *per_buffer_data)
255 : {
256 101738 : HeapScanDesc scan = (HeapScanDesc) callback_private_data;
257 :
258 : Assert(ScanDirectionIsForward(scan->rs_dir));
259 : Assert(scan->rs_base.rs_parallel);
260 :
261 101738 : if (unlikely(!scan->rs_inited))
262 : {
263 : /* parallel scan */
264 1671 : table_block_parallelscan_startblock_init(scan->rs_base.rs_rd,
265 1671 : scan->rs_parallelworkerdata,
266 1671 : (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 3342 : scan->rs_prefetch_block = table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
272 1671 : scan->rs_parallelworkerdata,
273 1671 : (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel);
274 1671 : scan->rs_inited = true;
275 : }
276 : else
277 : {
278 100067 : scan->rs_prefetch_block = table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
279 100067 : scan->rs_parallelworkerdata, (ParallelBlockTableScanDesc)
280 100067 : scan->rs_base.rs_parallel);
281 : }
282 :
283 101738 : 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 4153764 : heap_scan_stream_read_next_serial(ReadStream *stream,
293 : void *callback_private_data,
294 : void *per_buffer_data)
295 : {
296 4153764 : HeapScanDesc scan = (HeapScanDesc) callback_private_data;
297 :
298 4153764 : if (unlikely(!scan->rs_inited))
299 : {
300 1207003 : scan->rs_prefetch_block = heapgettup_initial_block(scan, scan->rs_dir);
301 1207003 : scan->rs_inited = true;
302 : }
303 : else
304 2946761 : scan->rs_prefetch_block = heapgettup_advance_block(scan,
305 : scan->rs_prefetch_block,
306 : scan->rs_dir);
307 :
308 4153764 : 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 218331 : bitmapheap_stream_read_next(ReadStream *pgsr, void *private_data,
318 : void *per_buffer_data)
319 : {
320 218331 : TBMIterateResult *tbmres = per_buffer_data;
321 218331 : BitmapHeapScanDesc bscan = (BitmapHeapScanDesc) private_data;
322 218331 : HeapScanDesc hscan = (HeapScanDesc) bscan;
323 218331 : TableScanDesc sscan = &hscan->rs_base;
324 :
325 : for (;;)
326 : {
327 218331 : CHECK_FOR_INTERRUPTS();
328 :
329 : /* no more entries in the bitmap */
330 218331 : if (!tbm_iterate(&sscan->st.rs_tbmiterator, tbmres))
331 13395 : 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 204936 : if (!IsolationIsSerializable() &&
342 204827 : tbmres->blockno >= hscan->rs_nblocks)
343 0 : continue;
344 :
345 204936 : 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 1232139 : initscan(HeapScanDesc scan, ScanKey key, bool keep_startblock)
358 : {
359 1232139 : 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 1232139 : if (scan->rs_base.rs_parallel != NULL)
375 : {
376 2229 : bpscan = (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel;
377 2229 : scan->rs_nblocks = bpscan->phs_nblocks;
378 : }
379 : else
380 1229910 : 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 1232137 : if (!RelationUsesLocalBuffers(scan->rs_base.rs_rd) &&
395 1224802 : scan->rs_nblocks > NBuffers / 4)
396 : {
397 13819 : allow_strat = (scan->rs_base.rs_flags & SO_ALLOW_STRAT) != 0;
398 13819 : allow_sync = (scan->rs_base.rs_flags & SO_ALLOW_SYNC) != 0;
399 : }
400 : else
401 1218318 : allow_strat = allow_sync = false;
402 :
403 1232137 : if (allow_strat)
404 : {
405 : /* During a rescan, keep the previous strategy object. */
406 12495 : if (scan->rs_strategy == NULL)
407 12312 : scan->rs_strategy = GetAccessStrategy(BAS_BULKREAD);
408 : }
409 : else
410 : {
411 1219642 : if (scan->rs_strategy != NULL)
412 0 : FreeAccessStrategy(scan->rs_strategy);
413 1219642 : scan->rs_strategy = NULL;
414 : }
415 :
416 1232137 : if (scan->rs_base.rs_parallel != NULL)
417 : {
418 : /* For parallel scan, believe whatever ParallelTableScanDesc says. */
419 2229 : if (scan->rs_base.rs_parallel->phs_syncscan)
420 2 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
421 : else
422 2227 : 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 2229 : if (!keep_startblock)
432 2115 : scan->rs_startblock = InvalidBlockNumber;
433 : }
434 : else
435 : {
436 1229908 : 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 826556 : if (allow_sync && synchronize_seqscans)
444 50 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
445 : else
446 826506 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
447 : }
448 403352 : else if (allow_sync && synchronize_seqscans)
449 : {
450 72 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
451 72 : scan->rs_startblock = ss_get_location(scan->rs_base.rs_rd, scan->rs_nblocks);
452 : }
453 : else
454 : {
455 403280 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
456 403280 : scan->rs_startblock = 0;
457 : }
458 : }
459 :
460 1232137 : scan->rs_numblocks = InvalidBlockNumber;
461 1232137 : scan->rs_inited = false;
462 1232137 : scan->rs_ctup.t_data = NULL;
463 1232137 : ItemPointerSetInvalid(&scan->rs_ctup.t_self);
464 1232137 : scan->rs_cbuf = InvalidBuffer;
465 1232137 : scan->rs_cblock = InvalidBlockNumber;
466 1232137 : scan->rs_ntuples = 0;
467 1232137 : 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 1232137 : scan->rs_dir = ForwardScanDirection;
474 1232137 : 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 1232137 : if (key != NULL && scan->rs_base.rs_nkeys > 0)
482 228931 : 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 1232137 : if (scan->rs_base.rs_flags & SO_TYPE_SEQSCAN)
490 1208301 : pgstat_count_heap_scan(scan->rs_base.rs_rd);
491 1232137 : }
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 2865 : heap_setscanlimits(TableScanDesc sscan, BlockNumber startBlk, BlockNumber numBlks)
501 : {
502 2865 : 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 2865 : scan->rs_startblock = startBlk;
512 2865 : scan->rs_numblocks = numBlks;
513 2865 : }
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 2935434 : 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 2935434 : Oid relid = RelationGetRelid(scan->rs_base.rs_rd);
528 2935434 : int ntup = 0;
529 2935434 : 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 150537242 : for (OffsetNumber lineoff = FirstOffsetNumber; lineoff <= lines; lineoff++)
537 : {
538 147601808 : ItemId lpp = PageGetItemId(page, lineoff);
539 : HeapTuple tup;
540 :
541 147601808 : if (unlikely(!ItemIdIsNormal(lpp)))
542 34021105 : 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 113580703 : if (!all_visible || check_serializable)
551 : {
552 68031686 : tup = &batchmvcc.tuples[ntup];
553 :
554 68031686 : tup->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
555 68031686 : tup->t_len = ItemIdGetLength(lpp);
556 68031686 : tup->t_tableOid = relid;
557 68031686 : 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 113580703 : if (all_visible)
565 : {
566 45549017 : if (check_serializable)
567 : {
568 0 : batchmvcc.visible[ntup] = true;
569 : }
570 45549017 : scan->rs_vistuples[ntup] = lineoff;
571 : }
572 :
573 113580703 : 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 2935434 : if (all_visible)
584 1061908 : nvis = ntup;
585 : else
586 1873526 : nvis = HeapTupleSatisfiesMVCCBatch(snapshot, buffer,
587 : ntup,
588 : &batchmvcc,
589 1873526 : 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 2935434 : if (check_serializable)
596 : {
597 2045 : for (int i = 0; i < ntup; i++)
598 : {
599 1421 : HeapCheckForSerializableConflictOut(batchmvcc.visible[i],
600 : scan->rs_base.rs_rd,
601 : &batchmvcc.tuples[i],
602 : buffer, snapshot);
603 : }
604 : }
605 :
606 2935426 : 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 2935434 : heap_prepare_pagescan(TableScanDesc sscan)
617 : {
618 2935434 : HeapScanDesc scan = (HeapScanDesc) sscan;
619 2935434 : Buffer buffer = scan->rs_cbuf;
620 2935434 : 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 2935434 : snapshot = scan->rs_base.rs_snapshot;
632 :
633 : /*
634 : * Prune and repair fragmentation for the whole page, if possible.
635 : */
636 2935434 : heap_page_prune_opt(scan->rs_base.rs_rd, buffer);
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 2935434 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
644 :
645 2935434 : page = BufferGetPage(buffer);
646 2935434 : 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 2935434 : all_visible = PageIsAllVisible(page) && !snapshot->takenDuringRecovery;
669 : check_serializable =
670 2935434 : 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 2935434 : if (likely(all_visible))
679 : {
680 1061908 : if (likely(!check_serializable))
681 1061908 : 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 1873526 : if (likely(!check_serializable))
690 1872894 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
691 : block, lines, false, false);
692 : else
693 632 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
694 : block, lines, false, true);
695 : }
696 :
697 2935426 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
698 2935426 : }
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 4063082 : heap_fetch_next_buffer(HeapScanDesc scan, ScanDirection dir)
708 : {
709 : Assert(scan->rs_read_stream);
710 :
711 : /* release previous scan buffer, if any */
712 4063082 : if (BufferIsValid(scan->rs_cbuf))
713 : {
714 2854368 : ReleaseBuffer(scan->rs_cbuf);
715 2854368 : 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 4063082 : 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 4063081 : if (unlikely(scan->rs_dir != dir))
732 : {
733 77 : scan->rs_prefetch_block = scan->rs_cblock;
734 77 : read_stream_reset(scan->rs_read_stream);
735 : }
736 :
737 4063081 : scan->rs_dir = dir;
738 :
739 4063081 : scan->rs_cbuf = read_stream_next_buffer(scan->rs_read_stream, NULL);
740 4063056 : if (BufferIsValid(scan->rs_cbuf))
741 3023788 : scan->rs_cblock = BufferGetBlockNumber(scan->rs_cbuf);
742 4063056 : }
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 1207043 : 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 1207043 : if (scan->rs_nblocks == 0 || scan->rs_numblocks == 0)
759 710388 : return InvalidBlockNumber;
760 :
761 496655 : if (ScanDirectionIsForward(dir))
762 : {
763 496623 : 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 32 : 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 32 : if (scan->rs_numblocks != InvalidBlockNumber)
780 3 : return (scan->rs_startblock + scan->rs_numblocks - 1) % scan->rs_nblocks;
781 :
782 29 : if (scan->rs_startblock > 0)
783 0 : return scan->rs_startblock - 1;
784 :
785 29 : 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 92630 : 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 92630 : page = BufferGetPage(scan->rs_cbuf);
809 :
810 92630 : *linesleft = PageGetMaxOffsetNumber(page) - FirstOffsetNumber + 1;
811 :
812 92630 : if (ScanDirectionIsForward(dir))
813 92630 : *lineoff = FirstOffsetNumber;
814 : else
815 0 : *lineoff = (OffsetNumber) (*linesleft);
816 :
817 : /* lineoff now references the physically previous or next tid */
818 92630 : 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 7485319 : 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 7485319 : page = BufferGetPage(scan->rs_cbuf);
840 :
841 7485319 : if (ScanDirectionIsForward(dir))
842 : {
843 7485319 : *lineoff = OffsetNumberNext(scan->rs_coffset);
844 7485319 : *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 7485319 : 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 2947060 : heapgettup_advance_block(HeapScanDesc scan, BlockNumber block, ScanDirection dir)
877 : {
878 : Assert(scan->rs_base.rs_parallel == NULL);
879 :
880 2947060 : if (likely(ScanDirectionIsForward(dir)))
881 : {
882 2947001 : block++;
883 :
884 : /* wrap back to the start of the heap */
885 2947001 : if (block >= scan->rs_nblocks)
886 392692 : 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 2947001 : if (scan->rs_base.rs_flags & SO_ALLOW_SYNC)
901 11267 : ss_report_location(scan->rs_base.rs_rd, block);
902 :
903 : /* we're done if we're back at where we started */
904 2947001 : if (block == scan->rs_startblock)
905 392651 : return InvalidBlockNumber;
906 :
907 : /* check if the limit imposed by heap_setscanlimits() is met */
908 2554350 : if (scan->rs_numblocks != InvalidBlockNumber)
909 : {
910 2484 : if (--scan->rs_numblocks == 0)
911 1546 : return InvalidBlockNumber;
912 : }
913 :
914 2552804 : return block;
915 : }
916 : else
917 : {
918 : /* we're done if the last block is the start position */
919 59 : if (block == scan->rs_startblock)
920 59 : 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 7506428 : heapgettup(HeapScanDesc scan,
961 : ScanDirection dir,
962 : int nkeys,
963 : ScanKey key)
964 : {
965 7506428 : HeapTuple tuple = &(scan->rs_ctup);
966 : Page page;
967 : OffsetNumber lineoff;
968 : int linesleft;
969 :
970 7506428 : if (likely(scan->rs_inited))
971 : {
972 : /* continue from previously returned page/tuple */
973 7485319 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
974 7485319 : page = heapgettup_continue_page(scan, dir, &linesleft, &lineoff);
975 7485319 : 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 113589 : heap_fetch_next_buffer(scan, dir);
985 :
986 : /* did we run out of blocks to scan? */
987 113589 : if (!BufferIsValid(scan->rs_cbuf))
988 20959 : break;
989 :
990 : Assert(BufferGetBlockNumber(scan->rs_cbuf) == scan->rs_cblock);
991 :
992 92630 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
993 92630 : page = heapgettup_start_page(scan, dir, &linesleft, &lineoff);
994 7577949 : 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 7620366 : for (; linesleft > 0; linesleft--, lineoff += dir)
1004 : {
1005 : bool visible;
1006 7527886 : ItemId lpp = PageGetItemId(page, lineoff);
1007 :
1008 7527886 : if (!ItemIdIsNormal(lpp))
1009 37186 : continue;
1010 :
1011 7490700 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
1012 7490700 : tuple->t_len = ItemIdGetLength(lpp);
1013 7490700 : ItemPointerSet(&(tuple->t_self), scan->rs_cblock, lineoff);
1014 :
1015 7490700 : visible = HeapTupleSatisfiesVisibility(tuple,
1016 : scan->rs_base.rs_snapshot,
1017 : scan->rs_cbuf);
1018 :
1019 7490700 : 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 7490700 : if (!visible)
1025 5231 : continue;
1026 :
1027 : /* skip any tuples that don't match the scan key */
1028 7485469 : if (key != NULL &&
1029 0 : !HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
1030 : nkeys, key))
1031 0 : continue;
1032 :
1033 7485469 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1034 7485469 : scan->rs_coffset = lineoff;
1035 7485469 : 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 92480 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1043 : }
1044 :
1045 : /* end of scan */
1046 20959 : if (BufferIsValid(scan->rs_cbuf))
1047 0 : ReleaseBuffer(scan->rs_cbuf);
1048 :
1049 20959 : scan->rs_cbuf = InvalidBuffer;
1050 20959 : scan->rs_cblock = InvalidBlockNumber;
1051 20959 : scan->rs_prefetch_block = InvalidBlockNumber;
1052 20959 : tuple->t_data = NULL;
1053 20959 : 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 52339226 : heapgettup_pagemode(HeapScanDesc scan,
1071 : ScanDirection dir,
1072 : int nkeys,
1073 : ScanKey key)
1074 : {
1075 52339226 : HeapTuple tuple = &(scan->rs_ctup);
1076 : Page page;
1077 : uint32 lineindex;
1078 : uint32 linesleft;
1079 :
1080 52339226 : if (likely(scan->rs_inited))
1081 : {
1082 : /* continue from previously returned page/tuple */
1083 51151621 : page = BufferGetPage(scan->rs_cbuf);
1084 :
1085 51151621 : lineindex = scan->rs_cindex + dir;
1086 51151621 : if (ScanDirectionIsForward(dir))
1087 51151292 : linesleft = scan->rs_ntuples - lineindex;
1088 : else
1089 329 : linesleft = scan->rs_cindex;
1090 : /* lineindex now references the next or previous visible tid */
1091 :
1092 51151621 : 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 3949493 : heap_fetch_next_buffer(scan, dir);
1102 :
1103 : /* did we run out of blocks to scan? */
1104 3949467 : if (!BufferIsValid(scan->rs_cbuf))
1105 1018309 : break;
1106 :
1107 : Assert(BufferGetBlockNumber(scan->rs_cbuf) == scan->rs_cblock);
1108 :
1109 : /* prune the page and determine visible tuple offsets */
1110 2931158 : heap_prepare_pagescan((TableScanDesc) scan);
1111 2931150 : page = BufferGetPage(scan->rs_cbuf);
1112 2931150 : linesleft = scan->rs_ntuples;
1113 2931150 : lineindex = ScanDirectionIsForward(dir) ? 0 : linesleft - 1;
1114 :
1115 : /* block is the same for all tuples, set it once outside the loop */
1116 2931150 : ItemPointerSetBlockNumber(&tuple->t_self, scan->rs_cblock);
1117 :
1118 : /* lineindex now references the next or previous visible tid */
1119 54082771 : continue_page:
1120 :
1121 104154907 : for (; linesleft > 0; linesleft--, lineindex += dir)
1122 : {
1123 : ItemId lpp;
1124 : OffsetNumber lineoff;
1125 :
1126 : Assert(lineindex < scan->rs_ntuples);
1127 101393019 : lineoff = scan->rs_vistuples[lineindex];
1128 101393019 : lpp = PageGetItemId(page, lineoff);
1129 : Assert(ItemIdIsNormal(lpp));
1130 :
1131 101393019 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
1132 101393019 : tuple->t_len = ItemIdGetLength(lpp);
1133 101393019 : ItemPointerSetOffsetNumber(&tuple->t_self, lineoff);
1134 :
1135 : /* skip any tuples that don't match the scan key */
1136 101393019 : if (key != NULL &&
1137 50488232 : !HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
1138 : nkeys, key))
1139 50072136 : continue;
1140 :
1141 51320883 : scan->rs_cindex = lineindex;
1142 51320883 : return;
1143 : }
1144 : }
1145 :
1146 : /* end of scan */
1147 1018309 : if (BufferIsValid(scan->rs_cbuf))
1148 0 : ReleaseBuffer(scan->rs_cbuf);
1149 1018309 : scan->rs_cbuf = InvalidBuffer;
1150 1018309 : scan->rs_cblock = InvalidBlockNumber;
1151 1018309 : scan->rs_prefetch_block = InvalidBlockNumber;
1152 1018309 : tuple->t_data = NULL;
1153 1018309 : scan->rs_inited = false;
1154 : }
1155 :
1156 :
1157 : /* ----------------------------------------------------------------
1158 : * heap access method interface
1159 : * ----------------------------------------------------------------
1160 : */
1161 :
1162 :
1163 : TableScanDesc
1164 405469 : 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 405469 : RelationIncrementReferenceCount(relation);
1179 :
1180 : /*
1181 : * allocate and initialize scan descriptor
1182 : */
1183 405469 : if (flags & SO_TYPE_BITMAPSCAN)
1184 : {
1185 11288 : 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 11288 : scan = (HeapScanDesc) bscan;
1192 : }
1193 : else
1194 394181 : scan = (HeapScanDesc) palloc_object(HeapScanDescData);
1195 :
1196 405469 : scan->rs_base.rs_rd = relation;
1197 405469 : scan->rs_base.rs_snapshot = snapshot;
1198 405469 : scan->rs_base.rs_nkeys = nkeys;
1199 405469 : scan->rs_base.rs_flags = flags;
1200 405469 : scan->rs_base.rs_parallel = parallel_scan;
1201 405469 : scan->rs_strategy = NULL; /* set in initscan */
1202 405469 : scan->rs_cbuf = InvalidBuffer;
1203 :
1204 : /*
1205 : * Disable page-at-a-time mode if it's not a MVCC-safe snapshot.
1206 : */
1207 405469 : if (!(snapshot && IsMVCCSnapshot(snapshot)))
1208 29953 : scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
1209 :
1210 : /* Check that a historic snapshot is not used for non-catalog tables */
1211 405469 : if (snapshot &&
1212 396625 : IsHistoricMVCCSnapshot(snapshot) &&
1213 674 : !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 405469 : 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 383954 : PredicateLockRelation(relation, snapshot);
1243 : }
1244 :
1245 : /* we only need to set this up once */
1246 405469 : 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 405469 : if (parallel_scan != NULL)
1253 2115 : scan->rs_parallelworkerdata = palloc_object(ParallelBlockTableScanWorkerData);
1254 : else
1255 403354 : 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 405469 : if (nkeys > 0)
1262 228931 : scan->rs_base.rs_key = palloc_array(ScanKeyData, nkeys);
1263 : else
1264 176538 : scan->rs_base.rs_key = NULL;
1265 :
1266 405469 : initscan(scan, key, false);
1267 :
1268 405467 : 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 405467 : if (scan->rs_base.rs_flags & SO_TYPE_SEQSCAN ||
1276 21588 : scan->rs_base.rs_flags & SO_TYPE_TIDRANGESCAN)
1277 384869 : {
1278 : ReadStreamBlockNumberCB cb;
1279 :
1280 384869 : if (scan->rs_base.rs_parallel)
1281 2115 : cb = heap_scan_stream_read_next_parallel;
1282 : else
1283 382754 : 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 384869 : 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 20598 : else if (scan->rs_base.rs_flags & SO_TYPE_BITMAPSCAN)
1302 : {
1303 11288 : 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 :
1314 405467 : return (TableScanDesc) scan;
1315 : }
1316 :
1317 : void
1318 826670 : heap_rescan(TableScanDesc sscan, ScanKey key, bool set_params,
1319 : bool allow_strat, bool allow_sync, bool allow_pagemode)
1320 : {
1321 826670 : HeapScanDesc scan = (HeapScanDesc) sscan;
1322 :
1323 826670 : if (set_params)
1324 : {
1325 15 : if (allow_strat)
1326 15 : scan->rs_base.rs_flags |= SO_ALLOW_STRAT;
1327 : else
1328 0 : scan->rs_base.rs_flags &= ~SO_ALLOW_STRAT;
1329 :
1330 15 : if (allow_sync)
1331 6 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
1332 : else
1333 9 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
1334 :
1335 15 : if (allow_pagemode && scan->rs_base.rs_snapshot &&
1336 15 : IsMVCCSnapshot(scan->rs_base.rs_snapshot))
1337 15 : scan->rs_base.rs_flags |= SO_ALLOW_PAGEMODE;
1338 : else
1339 0 : scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
1340 : }
1341 :
1342 : /*
1343 : * unpin scan buffers
1344 : */
1345 826670 : if (BufferIsValid(scan->rs_cbuf))
1346 : {
1347 1682 : ReleaseBuffer(scan->rs_cbuf);
1348 1682 : scan->rs_cbuf = InvalidBuffer;
1349 : }
1350 :
1351 : /*
1352 : * SO_TYPE_BITMAPSCAN would be cleaned up here, but it does not hold any
1353 : * additional data vs a normal HeapScan
1354 : */
1355 :
1356 : /*
1357 : * The read stream is reset on rescan. This must be done before
1358 : * initscan(), as some state referred to by read_stream_reset() is reset
1359 : * in initscan().
1360 : */
1361 826670 : if (scan->rs_read_stream)
1362 826652 : read_stream_reset(scan->rs_read_stream);
1363 :
1364 : /*
1365 : * reinitialize scan descriptor
1366 : */
1367 826670 : initscan(scan, key, true);
1368 826670 : }
1369 :
1370 : void
1371 403049 : heap_endscan(TableScanDesc sscan)
1372 : {
1373 403049 : HeapScanDesc scan = (HeapScanDesc) sscan;
1374 :
1375 : /* Note: no locking manipulations needed */
1376 :
1377 : /*
1378 : * unpin scan buffers
1379 : */
1380 403049 : if (BufferIsValid(scan->rs_cbuf))
1381 166072 : ReleaseBuffer(scan->rs_cbuf);
1382 :
1383 : /*
1384 : * Must free the read stream before freeing the BufferAccessStrategy.
1385 : */
1386 403049 : if (scan->rs_read_stream)
1387 393792 : read_stream_end(scan->rs_read_stream);
1388 :
1389 : /*
1390 : * decrement relation reference count and free scan descriptor storage
1391 : */
1392 403049 : RelationDecrementReferenceCount(scan->rs_base.rs_rd);
1393 :
1394 403049 : if (scan->rs_base.rs_key)
1395 228903 : pfree(scan->rs_base.rs_key);
1396 :
1397 403049 : if (scan->rs_strategy != NULL)
1398 12302 : FreeAccessStrategy(scan->rs_strategy);
1399 :
1400 403049 : if (scan->rs_parallelworkerdata != NULL)
1401 2115 : pfree(scan->rs_parallelworkerdata);
1402 :
1403 403049 : if (scan->rs_base.rs_flags & SO_TEMP_SNAPSHOT)
1404 41606 : UnregisterSnapshot(scan->rs_base.rs_snapshot);
1405 :
1406 403049 : pfree(scan);
1407 403049 : }
1408 :
1409 : HeapTuple
1410 10027494 : heap_getnext(TableScanDesc sscan, ScanDirection direction)
1411 : {
1412 10027494 : HeapScanDesc scan = (HeapScanDesc) sscan;
1413 :
1414 : /*
1415 : * This is still widely used directly, without going through table AM, so
1416 : * add a safety check. It's possible we should, at a later point,
1417 : * downgrade this to an assert. The reason for checking the AM routine,
1418 : * rather than the AM oid, is that this allows to write regression tests
1419 : * that create another AM reusing the heap handler.
1420 : */
1421 10027494 : if (unlikely(sscan->rs_rd->rd_tableam != GetHeapamTableAmRoutine()))
1422 0 : ereport(ERROR,
1423 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1424 : errmsg_internal("only heap AM is supported")));
1425 :
1426 : /* Note: no locking manipulations needed */
1427 :
1428 10027494 : if (scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE)
1429 3001261 : heapgettup_pagemode(scan, direction,
1430 3001261 : scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1431 : else
1432 7026233 : heapgettup(scan, direction,
1433 7026233 : scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1434 :
1435 10027494 : if (scan->rs_ctup.t_data == NULL)
1436 68554 : return NULL;
1437 :
1438 : /*
1439 : * if we get here it means we have a new current scan tuple, so point to
1440 : * the proper return buffer and return the tuple.
1441 : */
1442 :
1443 9958940 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1444 :
1445 9958940 : return &scan->rs_ctup;
1446 : }
1447 :
1448 : bool
1449 49812431 : heap_getnextslot(TableScanDesc sscan, ScanDirection direction, TupleTableSlot *slot)
1450 : {
1451 49812431 : HeapScanDesc scan = (HeapScanDesc) sscan;
1452 :
1453 : /* Note: no locking manipulations needed */
1454 :
1455 49812431 : if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1456 49332236 : heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1457 : else
1458 480195 : heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1459 :
1460 49812405 : if (scan->rs_ctup.t_data == NULL)
1461 : {
1462 970610 : ExecClearTuple(slot);
1463 970610 : return false;
1464 : }
1465 :
1466 : /*
1467 : * if we get here it means we have a new current scan tuple, so point to
1468 : * the proper return buffer and return the tuple.
1469 : */
1470 :
1471 48841795 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1472 :
1473 48841795 : ExecStoreBufferHeapTuple(&scan->rs_ctup, slot,
1474 : scan->rs_cbuf);
1475 48841795 : return true;
1476 : }
1477 :
1478 : void
1479 1035 : heap_set_tidrange(TableScanDesc sscan, ItemPointer mintid,
1480 : ItemPointer maxtid)
1481 : {
1482 1035 : HeapScanDesc scan = (HeapScanDesc) sscan;
1483 : BlockNumber startBlk;
1484 : BlockNumber numBlks;
1485 : ItemPointerData highestItem;
1486 : ItemPointerData lowestItem;
1487 :
1488 : /*
1489 : * For relations without any pages, we can simply leave the TID range
1490 : * unset. There will be no tuples to scan, therefore no tuples outside
1491 : * the given TID range.
1492 : */
1493 1035 : if (scan->rs_nblocks == 0)
1494 24 : return;
1495 :
1496 : /*
1497 : * Set up some ItemPointers which point to the first and last possible
1498 : * tuples in the heap.
1499 : */
1500 1029 : ItemPointerSet(&highestItem, scan->rs_nblocks - 1, MaxOffsetNumber);
1501 1029 : ItemPointerSet(&lowestItem, 0, FirstOffsetNumber);
1502 :
1503 : /*
1504 : * If the given maximum TID is below the highest possible TID in the
1505 : * relation, then restrict the range to that, otherwise we scan to the end
1506 : * of the relation.
1507 : */
1508 1029 : if (ItemPointerCompare(maxtid, &highestItem) < 0)
1509 130 : ItemPointerCopy(maxtid, &highestItem);
1510 :
1511 : /*
1512 : * If the given minimum TID is above the lowest possible TID in the
1513 : * relation, then restrict the range to only scan for TIDs above that.
1514 : */
1515 1029 : if (ItemPointerCompare(mintid, &lowestItem) > 0)
1516 911 : ItemPointerCopy(mintid, &lowestItem);
1517 :
1518 : /*
1519 : * Check for an empty range and protect from would be negative results
1520 : * from the numBlks calculation below.
1521 : */
1522 1029 : if (ItemPointerCompare(&highestItem, &lowestItem) < 0)
1523 : {
1524 : /* Set an empty range of blocks to scan */
1525 18 : heap_setscanlimits(sscan, 0, 0);
1526 18 : return;
1527 : }
1528 :
1529 : /*
1530 : * Calculate the first block and the number of blocks we must scan. We
1531 : * could be more aggressive here and perform some more validation to try
1532 : * and further narrow the scope of blocks to scan by checking if the
1533 : * lowestItem has an offset above MaxOffsetNumber. In this case, we could
1534 : * advance startBlk by one. Likewise, if highestItem has an offset of 0
1535 : * we could scan one fewer blocks. However, such an optimization does not
1536 : * seem worth troubling over, currently.
1537 : */
1538 1011 : startBlk = ItemPointerGetBlockNumberNoCheck(&lowestItem);
1539 :
1540 1011 : numBlks = ItemPointerGetBlockNumberNoCheck(&highestItem) -
1541 1011 : ItemPointerGetBlockNumberNoCheck(&lowestItem) + 1;
1542 :
1543 : /* Set the start block and number of blocks to scan */
1544 1011 : heap_setscanlimits(sscan, startBlk, numBlks);
1545 :
1546 : /* Finally, set the TID range in sscan */
1547 1011 : ItemPointerCopy(&lowestItem, &sscan->st.tidrange.rs_mintid);
1548 1011 : ItemPointerCopy(&highestItem, &sscan->st.tidrange.rs_maxtid);
1549 : }
1550 :
1551 : bool
1552 5636 : heap_getnextslot_tidrange(TableScanDesc sscan, ScanDirection direction,
1553 : TupleTableSlot *slot)
1554 : {
1555 5636 : HeapScanDesc scan = (HeapScanDesc) sscan;
1556 5636 : ItemPointer mintid = &sscan->st.tidrange.rs_mintid;
1557 5636 : ItemPointer maxtid = &sscan->st.tidrange.rs_maxtid;
1558 :
1559 : /* Note: no locking manipulations needed */
1560 : for (;;)
1561 : {
1562 5729 : if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1563 5729 : heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1564 : else
1565 0 : heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1566 :
1567 5721 : if (scan->rs_ctup.t_data == NULL)
1568 : {
1569 104 : ExecClearTuple(slot);
1570 104 : return false;
1571 : }
1572 :
1573 : /*
1574 : * heap_set_tidrange will have used heap_setscanlimits to limit the
1575 : * range of pages we scan to only ones that can contain the TID range
1576 : * we're scanning for. Here we must filter out any tuples from these
1577 : * pages that are outside of that range.
1578 : */
1579 5617 : if (ItemPointerCompare(&scan->rs_ctup.t_self, mintid) < 0)
1580 : {
1581 93 : ExecClearTuple(slot);
1582 :
1583 : /*
1584 : * When scanning backwards, the TIDs will be in descending order.
1585 : * Future tuples in this direction will be lower still, so we can
1586 : * just return false to indicate there will be no more tuples.
1587 : */
1588 93 : if (ScanDirectionIsBackward(direction))
1589 0 : return false;
1590 :
1591 93 : continue;
1592 : }
1593 :
1594 : /*
1595 : * Likewise for the final page, we must filter out TIDs greater than
1596 : * maxtid.
1597 : */
1598 5524 : if (ItemPointerCompare(&scan->rs_ctup.t_self, maxtid) > 0)
1599 : {
1600 56 : ExecClearTuple(slot);
1601 :
1602 : /*
1603 : * When scanning forward, the TIDs will be in ascending order.
1604 : * Future tuples in this direction will be higher still, so we can
1605 : * just return false to indicate there will be no more tuples.
1606 : */
1607 56 : if (ScanDirectionIsForward(direction))
1608 56 : return false;
1609 0 : continue;
1610 : }
1611 :
1612 5468 : break;
1613 : }
1614 :
1615 : /*
1616 : * if we get here it means we have a new current scan tuple, so point to
1617 : * the proper return buffer and return the tuple.
1618 : */
1619 5468 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1620 :
1621 5468 : ExecStoreBufferHeapTuple(&scan->rs_ctup, slot, scan->rs_cbuf);
1622 5468 : return true;
1623 : }
1624 :
1625 : /*
1626 : * heap_fetch - retrieve tuple with given tid
1627 : *
1628 : * On entry, tuple->t_self is the TID to fetch. We pin the buffer holding
1629 : * the tuple, fill in the remaining fields of *tuple, and check the tuple
1630 : * against the specified snapshot.
1631 : *
1632 : * If successful (tuple found and passes snapshot time qual), then *userbuf
1633 : * is set to the buffer holding the tuple and true is returned. The caller
1634 : * must unpin the buffer when done with the tuple.
1635 : *
1636 : * If the tuple is not found (ie, item number references a deleted slot),
1637 : * then tuple->t_data is set to NULL, *userbuf is set to InvalidBuffer,
1638 : * and false is returned.
1639 : *
1640 : * If the tuple is found but fails the time qual check, then the behavior
1641 : * depends on the keep_buf parameter. If keep_buf is false, the results
1642 : * are the same as for the tuple-not-found case. If keep_buf is true,
1643 : * then tuple->t_data and *userbuf are returned as for the success case,
1644 : * and again the caller must unpin the buffer; but false is returned.
1645 : *
1646 : * heap_fetch does not follow HOT chains: only the exact TID requested will
1647 : * be fetched.
1648 : *
1649 : * It is somewhat inconsistent that we ereport() on invalid block number but
1650 : * return false on invalid item number. There are a couple of reasons though.
1651 : * One is that the caller can relatively easily check the block number for
1652 : * validity, but cannot check the item number without reading the page
1653 : * himself. Another is that when we are following a t_ctid link, we can be
1654 : * reasonably confident that the page number is valid (since VACUUM shouldn't
1655 : * truncate off the destination page without having killed the referencing
1656 : * tuple first), but the item number might well not be good.
1657 : */
1658 : bool
1659 182388 : heap_fetch(Relation relation,
1660 : Snapshot snapshot,
1661 : HeapTuple tuple,
1662 : Buffer *userbuf,
1663 : bool keep_buf)
1664 : {
1665 182388 : ItemPointer tid = &(tuple->t_self);
1666 : ItemId lp;
1667 : Buffer buffer;
1668 : Page page;
1669 : OffsetNumber offnum;
1670 : bool valid;
1671 :
1672 : /*
1673 : * Fetch and pin the appropriate page of the relation.
1674 : */
1675 182388 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
1676 :
1677 : /*
1678 : * Need share lock on buffer to examine tuple commit status.
1679 : */
1680 182380 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
1681 182380 : page = BufferGetPage(buffer);
1682 :
1683 : /*
1684 : * We'd better check for out-of-range offnum in case of VACUUM since the
1685 : * TID was obtained.
1686 : */
1687 182380 : offnum = ItemPointerGetOffsetNumber(tid);
1688 182380 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1689 : {
1690 3 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1691 3 : ReleaseBuffer(buffer);
1692 3 : *userbuf = InvalidBuffer;
1693 3 : tuple->t_data = NULL;
1694 3 : return false;
1695 : }
1696 :
1697 : /*
1698 : * get the item line pointer corresponding to the requested tid
1699 : */
1700 182377 : lp = PageGetItemId(page, offnum);
1701 :
1702 : /*
1703 : * Must check for deleted tuple.
1704 : */
1705 182377 : if (!ItemIdIsNormal(lp))
1706 : {
1707 331 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1708 331 : ReleaseBuffer(buffer);
1709 331 : *userbuf = InvalidBuffer;
1710 331 : tuple->t_data = NULL;
1711 331 : return false;
1712 : }
1713 :
1714 : /*
1715 : * fill in *tuple fields
1716 : */
1717 182046 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
1718 182046 : tuple->t_len = ItemIdGetLength(lp);
1719 182046 : tuple->t_tableOid = RelationGetRelid(relation);
1720 :
1721 : /*
1722 : * check tuple visibility, then release lock
1723 : */
1724 182046 : valid = HeapTupleSatisfiesVisibility(tuple, snapshot, buffer);
1725 :
1726 182046 : if (valid)
1727 181992 : PredicateLockTID(relation, &(tuple->t_self), snapshot,
1728 181992 : HeapTupleHeaderGetXmin(tuple->t_data));
1729 :
1730 182046 : HeapCheckForSerializableConflictOut(valid, relation, tuple, buffer, snapshot);
1731 :
1732 182046 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1733 :
1734 182046 : if (valid)
1735 : {
1736 : /*
1737 : * All checks passed, so return the tuple as valid. Caller is now
1738 : * responsible for releasing the buffer.
1739 : */
1740 181992 : *userbuf = buffer;
1741 :
1742 181992 : return true;
1743 : }
1744 :
1745 : /* Tuple failed time qual, but maybe caller wants to see it anyway. */
1746 54 : if (keep_buf)
1747 35 : *userbuf = buffer;
1748 : else
1749 : {
1750 19 : ReleaseBuffer(buffer);
1751 19 : *userbuf = InvalidBuffer;
1752 19 : tuple->t_data = NULL;
1753 : }
1754 :
1755 54 : return false;
1756 : }
1757 :
1758 : /*
1759 : * heap_hot_search_buffer - search HOT chain for tuple satisfying snapshot
1760 : *
1761 : * On entry, *tid is the TID of a tuple (either a simple tuple, or the root
1762 : * of a HOT chain), and buffer is the buffer holding this tuple. We search
1763 : * for the first chain member satisfying the given snapshot. If one is
1764 : * found, we update *tid to reference that tuple's offset number, and
1765 : * return true. If no match, return false without modifying *tid.
1766 : *
1767 : * heapTuple is a caller-supplied buffer. When a match is found, we return
1768 : * the tuple here, in addition to updating *tid. If no match is found, the
1769 : * contents of this buffer on return are undefined.
1770 : *
1771 : * If all_dead is not NULL, we check non-visible tuples to see if they are
1772 : * globally dead; *all_dead is set true if all members of the HOT chain
1773 : * are vacuumable, false if not.
1774 : *
1775 : * Unlike heap_fetch, the caller must already have pin and (at least) share
1776 : * lock on the buffer; it is still pinned/locked at exit.
1777 : */
1778 : bool
1779 23496165 : heap_hot_search_buffer(ItemPointer tid, Relation relation, Buffer buffer,
1780 : Snapshot snapshot, HeapTuple heapTuple,
1781 : bool *all_dead, bool first_call)
1782 : {
1783 23496165 : Page page = BufferGetPage(buffer);
1784 23496165 : TransactionId prev_xmax = InvalidTransactionId;
1785 : BlockNumber blkno;
1786 : OffsetNumber offnum;
1787 : bool at_chain_start;
1788 : bool valid;
1789 : bool skip;
1790 23496165 : GlobalVisState *vistest = NULL;
1791 :
1792 : /* If this is not the first call, previous call returned a (live!) tuple */
1793 23496165 : if (all_dead)
1794 20090070 : *all_dead = first_call;
1795 :
1796 23496165 : blkno = ItemPointerGetBlockNumber(tid);
1797 23496165 : offnum = ItemPointerGetOffsetNumber(tid);
1798 23496165 : at_chain_start = first_call;
1799 23496165 : skip = !first_call;
1800 :
1801 : /* XXX: we should assert that a snapshot is pushed or registered */
1802 : Assert(TransactionIdIsValid(RecentXmin));
1803 : Assert(BufferGetBlockNumber(buffer) == blkno);
1804 :
1805 : /* Scan through possible multiple members of HOT-chain */
1806 : for (;;)
1807 1706908 : {
1808 : ItemId lp;
1809 :
1810 : /* check for bogus TID */
1811 25203073 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1812 : break;
1813 :
1814 25203073 : lp = PageGetItemId(page, offnum);
1815 :
1816 : /* check for unused, dead, or redirected items */
1817 25203073 : if (!ItemIdIsNormal(lp))
1818 : {
1819 : /* We should only see a redirect at start of chain */
1820 950987 : if (ItemIdIsRedirected(lp) && at_chain_start)
1821 : {
1822 : /* Follow the redirect */
1823 547924 : offnum = ItemIdGetRedirect(lp);
1824 547924 : at_chain_start = false;
1825 547924 : continue;
1826 : }
1827 : /* else must be end of chain */
1828 403063 : break;
1829 : }
1830 :
1831 : /*
1832 : * Update heapTuple to point to the element of the HOT chain we're
1833 : * currently investigating. Having t_self set correctly is important
1834 : * because the SSI checks and the *Satisfies routine for historical
1835 : * MVCC snapshots need the correct tid to decide about the visibility.
1836 : */
1837 24252086 : heapTuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
1838 24252086 : heapTuple->t_len = ItemIdGetLength(lp);
1839 24252086 : heapTuple->t_tableOid = RelationGetRelid(relation);
1840 24252086 : ItemPointerSet(&heapTuple->t_self, blkno, offnum);
1841 :
1842 : /*
1843 : * Shouldn't see a HEAP_ONLY tuple at chain start.
1844 : */
1845 24252086 : if (at_chain_start && HeapTupleIsHeapOnly(heapTuple))
1846 0 : break;
1847 :
1848 : /*
1849 : * The xmin should match the previous xmax value, else chain is
1850 : * broken.
1851 : */
1852 25411070 : if (TransactionIdIsValid(prev_xmax) &&
1853 1158984 : !TransactionIdEquals(prev_xmax,
1854 : HeapTupleHeaderGetXmin(heapTuple->t_data)))
1855 0 : break;
1856 :
1857 : /*
1858 : * When first_call is true (and thus, skip is initially false) we'll
1859 : * return the first tuple we find. But on later passes, heapTuple
1860 : * will initially be pointing to the tuple we returned last time.
1861 : * Returning it again would be incorrect (and would loop forever), so
1862 : * we skip it and return the next match we find.
1863 : */
1864 24252086 : if (!skip)
1865 : {
1866 : /* If it's visible per the snapshot, we must return it */
1867 24163188 : valid = HeapTupleSatisfiesVisibility(heapTuple, snapshot, buffer);
1868 24163188 : HeapCheckForSerializableConflictOut(valid, relation, heapTuple,
1869 : buffer, snapshot);
1870 :
1871 24163183 : if (valid)
1872 : {
1873 16866799 : ItemPointerSetOffsetNumber(tid, offnum);
1874 16866799 : PredicateLockTID(relation, &heapTuple->t_self, snapshot,
1875 16866799 : HeapTupleHeaderGetXmin(heapTuple->t_data));
1876 16866799 : if (all_dead)
1877 13773587 : *all_dead = false;
1878 16866799 : return true;
1879 : }
1880 : }
1881 7385282 : skip = false;
1882 :
1883 : /*
1884 : * If we can't see it, maybe no one else can either. At caller
1885 : * request, check whether all chain members are dead to all
1886 : * transactions.
1887 : *
1888 : * Note: if you change the criterion here for what is "dead", fix the
1889 : * planner's get_actual_variable_range() function to match.
1890 : */
1891 7385282 : if (all_dead && *all_dead)
1892 : {
1893 6462487 : if (!vistest)
1894 6338162 : vistest = GlobalVisTestFor(relation);
1895 :
1896 6462487 : if (!HeapTupleIsSurelyDead(heapTuple, vistest))
1897 6089800 : *all_dead = false;
1898 : }
1899 :
1900 : /*
1901 : * Check to see if HOT chain continues past this tuple; if so fetch
1902 : * the next offnum and loop around.
1903 : */
1904 7385282 : if (HeapTupleIsHotUpdated(heapTuple))
1905 : {
1906 : Assert(ItemPointerGetBlockNumber(&heapTuple->t_data->t_ctid) ==
1907 : blkno);
1908 1158984 : offnum = ItemPointerGetOffsetNumber(&heapTuple->t_data->t_ctid);
1909 1158984 : at_chain_start = false;
1910 1158984 : prev_xmax = HeapTupleHeaderGetUpdateXid(heapTuple->t_data);
1911 : }
1912 : else
1913 6226298 : break; /* end of chain */
1914 : }
1915 :
1916 6629361 : return false;
1917 : }
1918 :
1919 : /*
1920 : * heap_get_latest_tid - get the latest tid of a specified tuple
1921 : *
1922 : * Actually, this gets the latest version that is visible according to the
1923 : * scan's snapshot. Create a scan using SnapshotDirty to get the very latest,
1924 : * possibly uncommitted version.
1925 : *
1926 : * *tid is both an input and an output parameter: it is updated to
1927 : * show the latest version of the row. Note that it will not be changed
1928 : * if no version of the row passes the snapshot test.
1929 : */
1930 : void
1931 150 : heap_get_latest_tid(TableScanDesc sscan,
1932 : ItemPointer tid)
1933 : {
1934 150 : Relation relation = sscan->rs_rd;
1935 150 : Snapshot snapshot = sscan->rs_snapshot;
1936 : ItemPointerData ctid;
1937 : TransactionId priorXmax;
1938 :
1939 : /*
1940 : * table_tuple_get_latest_tid() verified that the passed in tid is valid.
1941 : * Assume that t_ctid links are valid however - there shouldn't be invalid
1942 : * ones in the table.
1943 : */
1944 : Assert(ItemPointerIsValid(tid));
1945 :
1946 : /*
1947 : * Loop to chase down t_ctid links. At top of loop, ctid is the tuple we
1948 : * need to examine, and *tid is the TID we will return if ctid turns out
1949 : * to be bogus.
1950 : *
1951 : * Note that we will loop until we reach the end of the t_ctid chain.
1952 : * Depending on the snapshot passed, there might be at most one visible
1953 : * version of the row, but we don't try to optimize for that.
1954 : */
1955 150 : ctid = *tid;
1956 150 : priorXmax = InvalidTransactionId; /* cannot check first XMIN */
1957 : for (;;)
1958 45 : {
1959 : Buffer buffer;
1960 : Page page;
1961 : OffsetNumber offnum;
1962 : ItemId lp;
1963 : HeapTupleData tp;
1964 : bool valid;
1965 :
1966 : /*
1967 : * Read, pin, and lock the page.
1968 : */
1969 195 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(&ctid));
1970 195 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
1971 195 : page = BufferGetPage(buffer);
1972 :
1973 : /*
1974 : * Check for bogus item number. This is not treated as an error
1975 : * condition because it can happen while following a t_ctid link. We
1976 : * just assume that the prior tid is OK and return it unchanged.
1977 : */
1978 195 : offnum = ItemPointerGetOffsetNumber(&ctid);
1979 195 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1980 : {
1981 0 : UnlockReleaseBuffer(buffer);
1982 0 : break;
1983 : }
1984 195 : lp = PageGetItemId(page, offnum);
1985 195 : if (!ItemIdIsNormal(lp))
1986 : {
1987 0 : UnlockReleaseBuffer(buffer);
1988 0 : break;
1989 : }
1990 :
1991 : /* OK to access the tuple */
1992 195 : tp.t_self = ctid;
1993 195 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
1994 195 : tp.t_len = ItemIdGetLength(lp);
1995 195 : tp.t_tableOid = RelationGetRelid(relation);
1996 :
1997 : /*
1998 : * After following a t_ctid link, we might arrive at an unrelated
1999 : * tuple. Check for XMIN match.
2000 : */
2001 240 : if (TransactionIdIsValid(priorXmax) &&
2002 45 : !TransactionIdEquals(priorXmax, HeapTupleHeaderGetXmin(tp.t_data)))
2003 : {
2004 0 : UnlockReleaseBuffer(buffer);
2005 0 : break;
2006 : }
2007 :
2008 : /*
2009 : * Check tuple visibility; if visible, set it as the new result
2010 : * candidate.
2011 : */
2012 195 : valid = HeapTupleSatisfiesVisibility(&tp, snapshot, buffer);
2013 195 : HeapCheckForSerializableConflictOut(valid, relation, &tp, buffer, snapshot);
2014 195 : if (valid)
2015 138 : *tid = ctid;
2016 :
2017 : /*
2018 : * If there's a valid t_ctid link, follow it, else we're done.
2019 : */
2020 276 : if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
2021 138 : HeapTupleHeaderIsOnlyLocked(tp.t_data) ||
2022 114 : HeapTupleHeaderIndicatesMovedPartitions(tp.t_data) ||
2023 57 : ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
2024 : {
2025 150 : UnlockReleaseBuffer(buffer);
2026 150 : break;
2027 : }
2028 :
2029 45 : ctid = tp.t_data->t_ctid;
2030 45 : priorXmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
2031 45 : UnlockReleaseBuffer(buffer);
2032 : } /* end of loop */
2033 150 : }
2034 :
2035 :
2036 : /*
2037 : * UpdateXmaxHintBits - update tuple hint bits after xmax transaction ends
2038 : *
2039 : * This is called after we have waited for the XMAX transaction to terminate.
2040 : * If the transaction aborted, we guarantee the XMAX_INVALID hint bit will
2041 : * be set on exit. If the transaction committed, we set the XMAX_COMMITTED
2042 : * hint bit if possible --- but beware that that may not yet be possible,
2043 : * if the transaction committed asynchronously.
2044 : *
2045 : * Note that if the transaction was a locker only, we set HEAP_XMAX_INVALID
2046 : * even if it commits.
2047 : *
2048 : * Hence callers should look only at XMAX_INVALID.
2049 : *
2050 : * Note this is not allowed for tuples whose xmax is a multixact.
2051 : */
2052 : static void
2053 226 : UpdateXmaxHintBits(HeapTupleHeader tuple, Buffer buffer, TransactionId xid)
2054 : {
2055 : Assert(TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple), xid));
2056 : Assert(!(tuple->t_infomask & HEAP_XMAX_IS_MULTI));
2057 :
2058 226 : if (!(tuple->t_infomask & (HEAP_XMAX_COMMITTED | HEAP_XMAX_INVALID)))
2059 : {
2060 400 : if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask) &&
2061 175 : TransactionIdDidCommit(xid))
2062 148 : HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_COMMITTED,
2063 : xid);
2064 : else
2065 77 : HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_INVALID,
2066 : InvalidTransactionId);
2067 : }
2068 226 : }
2069 :
2070 :
2071 : /*
2072 : * GetBulkInsertState - prepare status object for a bulk insert
2073 : */
2074 : BulkInsertState
2075 2732 : GetBulkInsertState(void)
2076 : {
2077 : BulkInsertState bistate;
2078 :
2079 2732 : bistate = (BulkInsertState) palloc_object(BulkInsertStateData);
2080 2732 : bistate->strategy = GetAccessStrategy(BAS_BULKWRITE);
2081 2732 : bistate->current_buf = InvalidBuffer;
2082 2732 : bistate->next_free = InvalidBlockNumber;
2083 2732 : bistate->last_free = InvalidBlockNumber;
2084 2732 : bistate->already_extended_by = 0;
2085 2732 : return bistate;
2086 : }
2087 :
2088 : /*
2089 : * FreeBulkInsertState - clean up after finishing a bulk insert
2090 : */
2091 : void
2092 2571 : FreeBulkInsertState(BulkInsertState bistate)
2093 : {
2094 2571 : if (bistate->current_buf != InvalidBuffer)
2095 2017 : ReleaseBuffer(bistate->current_buf);
2096 2571 : FreeAccessStrategy(bistate->strategy);
2097 2571 : pfree(bistate);
2098 2571 : }
2099 :
2100 : /*
2101 : * ReleaseBulkInsertStatePin - release a buffer currently held in bistate
2102 : */
2103 : void
2104 80758 : ReleaseBulkInsertStatePin(BulkInsertState bistate)
2105 : {
2106 80758 : if (bistate->current_buf != InvalidBuffer)
2107 30021 : ReleaseBuffer(bistate->current_buf);
2108 80758 : bistate->current_buf = InvalidBuffer;
2109 :
2110 : /*
2111 : * Despite the name, we also reset bulk relation extension state.
2112 : * Otherwise we can end up erroring out due to looking for free space in
2113 : * ->next_free of one partition, even though ->next_free was set when
2114 : * extending another partition. It could obviously also be bad for
2115 : * efficiency to look at existing blocks at offsets from another
2116 : * partition, even if we don't error out.
2117 : */
2118 80758 : bistate->next_free = InvalidBlockNumber;
2119 80758 : bistate->last_free = InvalidBlockNumber;
2120 80758 : }
2121 :
2122 :
2123 : /*
2124 : * heap_insert - insert tuple into a heap
2125 : *
2126 : * The new tuple is stamped with current transaction ID and the specified
2127 : * command ID.
2128 : *
2129 : * See table_tuple_insert for comments about most of the input flags, except
2130 : * that this routine directly takes a tuple rather than a slot.
2131 : *
2132 : * There's corresponding HEAP_INSERT_ options to all the TABLE_INSERT_
2133 : * options, and there additionally is HEAP_INSERT_SPECULATIVE which is used to
2134 : * implement table_tuple_insert_speculative().
2135 : *
2136 : * On return the header fields of *tup are updated to match the stored tuple;
2137 : * in particular tup->t_self receives the actual TID where the tuple was
2138 : * stored. But note that any toasting of fields within the tuple data is NOT
2139 : * reflected into *tup.
2140 : */
2141 : void
2142 8396028 : heap_insert(Relation relation, HeapTuple tup, CommandId cid,
2143 : int options, BulkInsertState bistate)
2144 : {
2145 8396028 : TransactionId xid = GetCurrentTransactionId();
2146 : HeapTuple heaptup;
2147 : Buffer buffer;
2148 8396022 : Buffer vmbuffer = InvalidBuffer;
2149 8396022 : bool all_visible_cleared = false;
2150 :
2151 : /* Cheap, simplistic check that the tuple matches the rel's rowtype. */
2152 : Assert(HeapTupleHeaderGetNatts(tup->t_data) <=
2153 : RelationGetNumberOfAttributes(relation));
2154 :
2155 8396022 : AssertHasSnapshotForToast(relation);
2156 :
2157 : /*
2158 : * Fill in tuple header fields and toast the tuple if necessary.
2159 : *
2160 : * Note: below this point, heaptup is the data we actually intend to store
2161 : * into the relation; tup is the caller's original untoasted data.
2162 : */
2163 8396022 : heaptup = heap_prepare_insert(relation, tup, xid, cid, options);
2164 :
2165 : /*
2166 : * Find buffer to insert this tuple into. If the page is all visible,
2167 : * this will also pin the requisite visibility map page.
2168 : */
2169 8396022 : buffer = RelationGetBufferForTuple(relation, heaptup->t_len,
2170 : InvalidBuffer, options, bistate,
2171 : &vmbuffer, NULL,
2172 : 0);
2173 :
2174 : /*
2175 : * We're about to do the actual insert -- but check for conflict first, to
2176 : * avoid possibly having to roll back work we've just done.
2177 : *
2178 : * This is safe without a recheck as long as there is no possibility of
2179 : * another process scanning the page between this check and the insert
2180 : * being visible to the scan (i.e., an exclusive buffer content lock is
2181 : * continuously held from this point until the tuple insert is visible).
2182 : *
2183 : * For a heap insert, we only need to check for table-level SSI locks. Our
2184 : * new tuple can't possibly conflict with existing tuple locks, and heap
2185 : * page locks are only consolidated versions of tuple locks; they do not
2186 : * lock "gaps" as index page locks do. So we don't need to specify a
2187 : * buffer when making the call, which makes for a faster check.
2188 : */
2189 8396022 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2190 :
2191 : /* NO EREPORT(ERROR) from here till changes are logged */
2192 8396010 : START_CRIT_SECTION();
2193 :
2194 8396010 : RelationPutHeapTuple(relation, buffer, heaptup,
2195 8396010 : (options & HEAP_INSERT_SPECULATIVE) != 0);
2196 :
2197 8396010 : if (PageIsAllVisible(BufferGetPage(buffer)))
2198 : {
2199 7228 : all_visible_cleared = true;
2200 7228 : PageClearAllVisible(BufferGetPage(buffer));
2201 7228 : visibilitymap_clear(relation,
2202 7228 : ItemPointerGetBlockNumber(&(heaptup->t_self)),
2203 : vmbuffer, VISIBILITYMAP_VALID_BITS);
2204 : }
2205 :
2206 : /*
2207 : * XXX Should we set PageSetPrunable on this page ?
2208 : *
2209 : * The inserting transaction may eventually abort thus making this tuple
2210 : * DEAD and hence available for pruning. Though we don't want to optimize
2211 : * for aborts, if no other tuple in this page is UPDATEd/DELETEd, the
2212 : * aborted tuple will never be pruned until next vacuum is triggered.
2213 : *
2214 : * If you do add PageSetPrunable here, add it in heap_xlog_insert too.
2215 : */
2216 :
2217 8396010 : MarkBufferDirty(buffer);
2218 :
2219 : /* XLOG stuff */
2220 8396010 : if (RelationNeedsWAL(relation))
2221 : {
2222 : xl_heap_insert xlrec;
2223 : xl_heap_header xlhdr;
2224 : XLogRecPtr recptr;
2225 7504302 : Page page = BufferGetPage(buffer);
2226 7504302 : uint8 info = XLOG_HEAP_INSERT;
2227 7504302 : int bufflags = 0;
2228 :
2229 : /*
2230 : * If this is a catalog, we need to transmit combo CIDs to properly
2231 : * decode, so log that as well.
2232 : */
2233 7504302 : if (RelationIsAccessibleInLogicalDecoding(relation))
2234 3434 : log_heap_new_cid(relation, heaptup);
2235 :
2236 : /*
2237 : * If this is the single and first tuple on page, we can reinit the
2238 : * page instead of restoring the whole thing. Set flag, and hide
2239 : * buffer references from XLogInsert.
2240 : */
2241 7601154 : if (ItemPointerGetOffsetNumber(&(heaptup->t_self)) == FirstOffsetNumber &&
2242 96852 : PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
2243 : {
2244 95963 : info |= XLOG_HEAP_INIT_PAGE;
2245 95963 : bufflags |= REGBUF_WILL_INIT;
2246 : }
2247 :
2248 7504302 : xlrec.offnum = ItemPointerGetOffsetNumber(&heaptup->t_self);
2249 7504302 : xlrec.flags = 0;
2250 7504302 : if (all_visible_cleared)
2251 7225 : xlrec.flags |= XLH_INSERT_ALL_VISIBLE_CLEARED;
2252 7504302 : if (options & HEAP_INSERT_SPECULATIVE)
2253 2112 : xlrec.flags |= XLH_INSERT_IS_SPECULATIVE;
2254 : Assert(ItemPointerGetBlockNumber(&heaptup->t_self) == BufferGetBlockNumber(buffer));
2255 :
2256 : /*
2257 : * For logical decoding, we need the tuple even if we're doing a full
2258 : * page write, so make sure it's included even if we take a full-page
2259 : * image. (XXX We could alternatively store a pointer into the FPW).
2260 : */
2261 7504302 : if (RelationIsLogicallyLogged(relation) &&
2262 250525 : !(options & HEAP_INSERT_NO_LOGICAL))
2263 : {
2264 250498 : xlrec.flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
2265 250498 : bufflags |= REGBUF_KEEP_DATA;
2266 :
2267 250498 : if (IsToastRelation(relation))
2268 1786 : xlrec.flags |= XLH_INSERT_ON_TOAST_RELATION;
2269 : }
2270 :
2271 7504302 : XLogBeginInsert();
2272 7504302 : XLogRegisterData(&xlrec, SizeOfHeapInsert);
2273 :
2274 7504302 : xlhdr.t_infomask2 = heaptup->t_data->t_infomask2;
2275 7504302 : xlhdr.t_infomask = heaptup->t_data->t_infomask;
2276 7504302 : xlhdr.t_hoff = heaptup->t_data->t_hoff;
2277 :
2278 : /*
2279 : * note we mark xlhdr as belonging to buffer; if XLogInsert decides to
2280 : * write the whole page to the xlog, we don't need to store
2281 : * xl_heap_header in the xlog.
2282 : */
2283 7504302 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD | bufflags);
2284 7504302 : XLogRegisterBufData(0, &xlhdr, SizeOfHeapHeader);
2285 : /* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
2286 7504302 : XLogRegisterBufData(0,
2287 7504302 : (char *) heaptup->t_data + SizeofHeapTupleHeader,
2288 7504302 : heaptup->t_len - SizeofHeapTupleHeader);
2289 :
2290 : /* filtering by origin on a row level is much more efficient */
2291 7504302 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
2292 :
2293 7504302 : recptr = XLogInsert(RM_HEAP_ID, info);
2294 :
2295 7504302 : PageSetLSN(page, recptr);
2296 : }
2297 :
2298 8396010 : END_CRIT_SECTION();
2299 :
2300 8396010 : UnlockReleaseBuffer(buffer);
2301 8396010 : if (vmbuffer != InvalidBuffer)
2302 7507 : ReleaseBuffer(vmbuffer);
2303 :
2304 : /*
2305 : * If tuple is cacheable, mark it for invalidation from the caches in case
2306 : * we abort. Note it is OK to do this after releasing the buffer, because
2307 : * the heaptup data structure is all in local memory, not in the shared
2308 : * buffer.
2309 : */
2310 8396010 : CacheInvalidateHeapTuple(relation, heaptup, NULL);
2311 :
2312 : /* Note: speculative insertions are counted too, even if aborted later */
2313 8396010 : pgstat_count_heap_insert(relation, 1);
2314 :
2315 : /*
2316 : * If heaptup is a private copy, release it. Don't forget to copy t_self
2317 : * back to the caller's image, too.
2318 : */
2319 8396010 : if (heaptup != tup)
2320 : {
2321 18463 : tup->t_self = heaptup->t_self;
2322 18463 : heap_freetuple(heaptup);
2323 : }
2324 8396010 : }
2325 :
2326 : /*
2327 : * Subroutine for heap_insert(). Prepares a tuple for insertion. This sets the
2328 : * tuple header fields and toasts the tuple if necessary. Returns a toasted
2329 : * version of the tuple if it was toasted, or the original tuple if not. Note
2330 : * that in any case, the header fields are also set in the original tuple.
2331 : */
2332 : static HeapTuple
2333 9904626 : heap_prepare_insert(Relation relation, HeapTuple tup, TransactionId xid,
2334 : CommandId cid, int options)
2335 : {
2336 : /*
2337 : * To allow parallel inserts, we need to ensure that they are safe to be
2338 : * performed in workers. We have the infrastructure to allow parallel
2339 : * inserts in general except for the cases where inserts generate a new
2340 : * CommandId (eg. inserts into a table having a foreign key column).
2341 : */
2342 9904626 : if (IsParallelWorker())
2343 0 : ereport(ERROR,
2344 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
2345 : errmsg("cannot insert tuples in a parallel worker")));
2346 :
2347 9904626 : tup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
2348 9904626 : tup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
2349 9904626 : tup->t_data->t_infomask |= HEAP_XMAX_INVALID;
2350 9904626 : HeapTupleHeaderSetXmin(tup->t_data, xid);
2351 9904626 : if (options & HEAP_INSERT_FROZEN)
2352 102088 : HeapTupleHeaderSetXminFrozen(tup->t_data);
2353 :
2354 9904626 : HeapTupleHeaderSetCmin(tup->t_data, cid);
2355 9904626 : HeapTupleHeaderSetXmax(tup->t_data, 0); /* for cleanliness */
2356 9904626 : tup->t_tableOid = RelationGetRelid(relation);
2357 :
2358 : /*
2359 : * If the new tuple is too big for storage or contains already toasted
2360 : * out-of-line attributes from some other relation, invoke the toaster.
2361 : */
2362 9904626 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
2363 31464 : relation->rd_rel->relkind != RELKIND_MATVIEW)
2364 : {
2365 : /* toast table entries should never be recursively toasted */
2366 : Assert(!HeapTupleHasExternal(tup));
2367 31416 : return tup;
2368 : }
2369 9873210 : else if (HeapTupleHasExternal(tup) || tup->t_len > TOAST_TUPLE_THRESHOLD)
2370 18515 : return heap_toast_insert_or_update(relation, tup, NULL, options);
2371 : else
2372 9854695 : return tup;
2373 : }
2374 :
2375 : /*
2376 : * Helper for heap_multi_insert() that computes the number of entire pages
2377 : * that inserting the remaining heaptuples requires. Used to determine how
2378 : * much the relation needs to be extended by.
2379 : */
2380 : static int
2381 381662 : heap_multi_insert_pages(HeapTuple *heaptuples, int done, int ntuples, Size saveFreeSpace)
2382 : {
2383 381662 : size_t page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
2384 381662 : int npages = 1;
2385 :
2386 2485967 : for (int i = done; i < ntuples; i++)
2387 : {
2388 2104305 : size_t tup_sz = sizeof(ItemIdData) + MAXALIGN(heaptuples[i]->t_len);
2389 :
2390 2104305 : if (page_avail < tup_sz)
2391 : {
2392 15654 : npages++;
2393 15654 : page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
2394 : }
2395 2104305 : page_avail -= tup_sz;
2396 : }
2397 :
2398 381662 : return npages;
2399 : }
2400 :
2401 : /*
2402 : * heap_multi_insert - insert multiple tuples into a heap
2403 : *
2404 : * This is like heap_insert(), but inserts multiple tuples in one operation.
2405 : * That's faster than calling heap_insert() in a loop, because when multiple
2406 : * tuples can be inserted on a single page, we can write just a single WAL
2407 : * record covering all of them, and only need to lock/unlock the page once.
2408 : *
2409 : * Note: this leaks memory into the current memory context. You can create a
2410 : * temporary context before calling this, if that's a problem.
2411 : */
2412 : void
2413 375163 : heap_multi_insert(Relation relation, TupleTableSlot **slots, int ntuples,
2414 : CommandId cid, int options, BulkInsertState bistate)
2415 : {
2416 375163 : TransactionId xid = GetCurrentTransactionId();
2417 : HeapTuple *heaptuples;
2418 : int i;
2419 : int ndone;
2420 : PGAlignedBlock scratch;
2421 : Page page;
2422 375163 : Buffer vmbuffer = InvalidBuffer;
2423 : bool needwal;
2424 : Size saveFreeSpace;
2425 375163 : bool need_tuple_data = RelationIsLogicallyLogged(relation);
2426 375163 : bool need_cids = RelationIsAccessibleInLogicalDecoding(relation);
2427 375163 : bool starting_with_empty_page = false;
2428 375163 : int npages = 0;
2429 375163 : int npages_used = 0;
2430 :
2431 : /* currently not needed (thus unsupported) for heap_multi_insert() */
2432 : Assert(!(options & HEAP_INSERT_NO_LOGICAL));
2433 :
2434 375163 : AssertHasSnapshotForToast(relation);
2435 :
2436 375163 : needwal = RelationNeedsWAL(relation);
2437 375163 : saveFreeSpace = RelationGetTargetPageFreeSpace(relation,
2438 : HEAP_DEFAULT_FILLFACTOR);
2439 :
2440 : /* Toast and set header data in all the slots */
2441 375163 : heaptuples = palloc(ntuples * sizeof(HeapTuple));
2442 1883767 : for (i = 0; i < ntuples; i++)
2443 : {
2444 : HeapTuple tuple;
2445 :
2446 1508604 : tuple = ExecFetchSlotHeapTuple(slots[i], true, NULL);
2447 1508604 : slots[i]->tts_tableOid = RelationGetRelid(relation);
2448 1508604 : tuple->t_tableOid = slots[i]->tts_tableOid;
2449 1508604 : heaptuples[i] = heap_prepare_insert(relation, tuple, xid, cid,
2450 : options);
2451 : }
2452 :
2453 : /*
2454 : * We're about to do the actual inserts -- but check for conflict first,
2455 : * to minimize the possibility of having to roll back work we've just
2456 : * done.
2457 : *
2458 : * A check here does not definitively prevent a serialization anomaly;
2459 : * that check MUST be done at least past the point of acquiring an
2460 : * exclusive buffer content lock on every buffer that will be affected,
2461 : * and MAY be done after all inserts are reflected in the buffers and
2462 : * those locks are released; otherwise there is a race condition. Since
2463 : * multiple buffers can be locked and unlocked in the loop below, and it
2464 : * would not be feasible to identify and lock all of those buffers before
2465 : * the loop, we must do a final check at the end.
2466 : *
2467 : * The check here could be omitted with no loss of correctness; it is
2468 : * present strictly as an optimization.
2469 : *
2470 : * For heap inserts, we only need to check for table-level SSI locks. Our
2471 : * new tuples can't possibly conflict with existing tuple locks, and heap
2472 : * page locks are only consolidated versions of tuple locks; they do not
2473 : * lock "gaps" as index page locks do. So we don't need to specify a
2474 : * buffer when making the call, which makes for a faster check.
2475 : */
2476 375163 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2477 :
2478 375163 : ndone = 0;
2479 765029 : while (ndone < ntuples)
2480 : {
2481 : Buffer buffer;
2482 389866 : bool all_visible_cleared = false;
2483 389866 : bool all_frozen_set = false;
2484 : int nthispage;
2485 :
2486 389866 : CHECK_FOR_INTERRUPTS();
2487 :
2488 : /*
2489 : * Compute number of pages needed to fit the to-be-inserted tuples in
2490 : * the worst case. This will be used to determine how much to extend
2491 : * the relation by in RelationGetBufferForTuple(), if needed. If we
2492 : * filled a prior page from scratch, we can just update our last
2493 : * computation, but if we started with a partially filled page,
2494 : * recompute from scratch, the number of potentially required pages
2495 : * can vary due to tuples needing to fit onto the page, page headers
2496 : * etc.
2497 : */
2498 389866 : if (ndone == 0 || !starting_with_empty_page)
2499 : {
2500 381662 : npages = heap_multi_insert_pages(heaptuples, ndone, ntuples,
2501 : saveFreeSpace);
2502 381662 : npages_used = 0;
2503 : }
2504 : else
2505 8204 : npages_used++;
2506 :
2507 : /*
2508 : * Find buffer where at least the next tuple will fit. If the page is
2509 : * all-visible, this will also pin the requisite visibility map page.
2510 : *
2511 : * Also pin visibility map page if COPY FREEZE inserts tuples into an
2512 : * empty page. See all_frozen_set below.
2513 : */
2514 389866 : buffer = RelationGetBufferForTuple(relation, heaptuples[ndone]->t_len,
2515 : InvalidBuffer, options, bistate,
2516 : &vmbuffer, NULL,
2517 : npages - npages_used);
2518 389866 : page = BufferGetPage(buffer);
2519 :
2520 389866 : starting_with_empty_page = PageGetMaxOffsetNumber(page) == 0;
2521 :
2522 389866 : if (starting_with_empty_page && (options & HEAP_INSERT_FROZEN))
2523 : {
2524 1661 : all_frozen_set = true;
2525 : /* Lock the vmbuffer before entering the critical section */
2526 1661 : LockBuffer(vmbuffer, BUFFER_LOCK_EXCLUSIVE);
2527 : }
2528 :
2529 : /* NO EREPORT(ERROR) from here till changes are logged */
2530 389866 : START_CRIT_SECTION();
2531 :
2532 : /*
2533 : * RelationGetBufferForTuple has ensured that the first tuple fits.
2534 : * Put that on the page, and then as many other tuples as fit.
2535 : */
2536 389866 : RelationPutHeapTuple(relation, buffer, heaptuples[ndone], false);
2537 :
2538 : /*
2539 : * For logical decoding we need combo CIDs to properly decode the
2540 : * catalog.
2541 : */
2542 389866 : if (needwal && need_cids)
2543 5063 : log_heap_new_cid(relation, heaptuples[ndone]);
2544 :
2545 1508604 : for (nthispage = 1; ndone + nthispage < ntuples; nthispage++)
2546 : {
2547 1133441 : HeapTuple heaptup = heaptuples[ndone + nthispage];
2548 :
2549 1133441 : if (PageGetHeapFreeSpace(page) < MAXALIGN(heaptup->t_len) + saveFreeSpace)
2550 14703 : break;
2551 :
2552 1118738 : RelationPutHeapTuple(relation, buffer, heaptup, false);
2553 :
2554 : /*
2555 : * For logical decoding we need combo CIDs to properly decode the
2556 : * catalog.
2557 : */
2558 1118738 : if (needwal && need_cids)
2559 4738 : log_heap_new_cid(relation, heaptup);
2560 : }
2561 :
2562 : /*
2563 : * If the page is all visible, need to clear that, unless we're only
2564 : * going to add further frozen rows to it.
2565 : *
2566 : * If we're only adding already frozen rows to a previously empty
2567 : * page, mark it as all-frozen and update the visibility map. We're
2568 : * already holding a pin on the vmbuffer.
2569 : */
2570 389866 : if (PageIsAllVisible(page) && !(options & HEAP_INSERT_FROZEN))
2571 : {
2572 2902 : all_visible_cleared = true;
2573 2902 : PageClearAllVisible(page);
2574 2902 : visibilitymap_clear(relation,
2575 : BufferGetBlockNumber(buffer),
2576 : vmbuffer, VISIBILITYMAP_VALID_BITS);
2577 : }
2578 386964 : else if (all_frozen_set)
2579 : {
2580 1661 : PageSetAllVisible(page);
2581 1661 : PageClearPrunable(page);
2582 1661 : visibilitymap_set_vmbits(BufferGetBlockNumber(buffer),
2583 : vmbuffer,
2584 : VISIBILITYMAP_ALL_VISIBLE |
2585 : VISIBILITYMAP_ALL_FROZEN,
2586 : relation->rd_locator);
2587 : }
2588 :
2589 : /*
2590 : * XXX Should we set PageSetPrunable on this page ? See heap_insert()
2591 : */
2592 :
2593 389866 : MarkBufferDirty(buffer);
2594 :
2595 : /* XLOG stuff */
2596 389866 : if (needwal)
2597 : {
2598 : XLogRecPtr recptr;
2599 : xl_heap_multi_insert *xlrec;
2600 386043 : uint8 info = XLOG_HEAP2_MULTI_INSERT;
2601 : char *tupledata;
2602 : int totaldatalen;
2603 386043 : char *scratchptr = scratch.data;
2604 : bool init;
2605 386043 : int bufflags = 0;
2606 :
2607 : /*
2608 : * If the page was previously empty, we can reinit the page
2609 : * instead of restoring the whole thing.
2610 : */
2611 386043 : init = starting_with_empty_page;
2612 :
2613 : /* allocate xl_heap_multi_insert struct from the scratch area */
2614 386043 : xlrec = (xl_heap_multi_insert *) scratchptr;
2615 386043 : scratchptr += SizeOfHeapMultiInsert;
2616 :
2617 : /*
2618 : * Allocate offsets array. Unless we're reinitializing the page,
2619 : * in that case the tuples are stored in order starting at
2620 : * FirstOffsetNumber and we don't need to store the offsets
2621 : * explicitly.
2622 : */
2623 386043 : if (!init)
2624 372594 : scratchptr += nthispage * sizeof(OffsetNumber);
2625 :
2626 : /* the rest of the scratch space is used for tuple data */
2627 386043 : tupledata = scratchptr;
2628 :
2629 : /* check that the mutually exclusive flags are not both set */
2630 : Assert(!(all_visible_cleared && all_frozen_set));
2631 :
2632 386043 : xlrec->flags = 0;
2633 386043 : if (all_visible_cleared)
2634 2902 : xlrec->flags = XLH_INSERT_ALL_VISIBLE_CLEARED;
2635 :
2636 : /*
2637 : * We don't have to worry about including a conflict xid in the
2638 : * WAL record, as HEAP_INSERT_FROZEN intentionally violates
2639 : * visibility rules.
2640 : */
2641 386043 : if (all_frozen_set)
2642 17 : xlrec->flags = XLH_INSERT_ALL_FROZEN_SET;
2643 :
2644 386043 : xlrec->ntuples = nthispage;
2645 :
2646 : /*
2647 : * Write out an xl_multi_insert_tuple and the tuple data itself
2648 : * for each tuple.
2649 : */
2650 1689197 : for (i = 0; i < nthispage; i++)
2651 : {
2652 1303154 : HeapTuple heaptup = heaptuples[ndone + i];
2653 : xl_multi_insert_tuple *tuphdr;
2654 : int datalen;
2655 :
2656 1303154 : if (!init)
2657 777253 : xlrec->offsets[i] = ItemPointerGetOffsetNumber(&heaptup->t_self);
2658 : /* xl_multi_insert_tuple needs two-byte alignment. */
2659 1303154 : tuphdr = (xl_multi_insert_tuple *) SHORTALIGN(scratchptr);
2660 1303154 : scratchptr = ((char *) tuphdr) + SizeOfMultiInsertTuple;
2661 :
2662 1303154 : tuphdr->t_infomask2 = heaptup->t_data->t_infomask2;
2663 1303154 : tuphdr->t_infomask = heaptup->t_data->t_infomask;
2664 1303154 : tuphdr->t_hoff = heaptup->t_data->t_hoff;
2665 :
2666 : /* write bitmap [+ padding] [+ oid] + data */
2667 1303154 : datalen = heaptup->t_len - SizeofHeapTupleHeader;
2668 1303154 : memcpy(scratchptr,
2669 1303154 : (char *) heaptup->t_data + SizeofHeapTupleHeader,
2670 : datalen);
2671 1303154 : tuphdr->datalen = datalen;
2672 1303154 : scratchptr += datalen;
2673 : }
2674 386043 : totaldatalen = scratchptr - tupledata;
2675 : Assert((scratchptr - scratch.data) < BLCKSZ);
2676 :
2677 386043 : if (need_tuple_data)
2678 72 : xlrec->flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
2679 :
2680 : /*
2681 : * Signal that this is the last xl_heap_multi_insert record
2682 : * emitted by this call to heap_multi_insert(). Needed for logical
2683 : * decoding so it knows when to cleanup temporary data.
2684 : */
2685 386043 : if (ndone + nthispage == ntuples)
2686 374749 : xlrec->flags |= XLH_INSERT_LAST_IN_MULTI;
2687 :
2688 386043 : if (init)
2689 : {
2690 13449 : info |= XLOG_HEAP_INIT_PAGE;
2691 13449 : bufflags |= REGBUF_WILL_INIT;
2692 : }
2693 :
2694 : /*
2695 : * If we're doing logical decoding, include the new tuple data
2696 : * even if we take a full-page image of the page.
2697 : */
2698 386043 : if (need_tuple_data)
2699 72 : bufflags |= REGBUF_KEEP_DATA;
2700 :
2701 386043 : XLogBeginInsert();
2702 386043 : XLogRegisterData(xlrec, tupledata - scratch.data);
2703 386043 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD | bufflags);
2704 386043 : if (all_frozen_set)
2705 17 : XLogRegisterBuffer(1, vmbuffer, 0);
2706 :
2707 386043 : XLogRegisterBufData(0, tupledata, totaldatalen);
2708 :
2709 : /* filtering by origin on a row level is much more efficient */
2710 386043 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
2711 :
2712 386043 : recptr = XLogInsert(RM_HEAP2_ID, info);
2713 :
2714 386043 : PageSetLSN(page, recptr);
2715 386043 : if (all_frozen_set)
2716 : {
2717 : Assert(BufferIsDirty(vmbuffer));
2718 17 : PageSetLSN(BufferGetPage(vmbuffer), recptr);
2719 : }
2720 : }
2721 :
2722 389866 : END_CRIT_SECTION();
2723 :
2724 389866 : if (all_frozen_set)
2725 1661 : LockBuffer(vmbuffer, BUFFER_LOCK_UNLOCK);
2726 :
2727 389866 : UnlockReleaseBuffer(buffer);
2728 389866 : ndone += nthispage;
2729 :
2730 : /*
2731 : * NB: Only release vmbuffer after inserting all tuples - it's fairly
2732 : * likely that we'll insert into subsequent heap pages that are likely
2733 : * to use the same vm page.
2734 : */
2735 : }
2736 :
2737 : /* We're done with inserting all tuples, so release the last vmbuffer. */
2738 375163 : if (vmbuffer != InvalidBuffer)
2739 3003 : ReleaseBuffer(vmbuffer);
2740 :
2741 : /*
2742 : * We're done with the actual inserts. Check for conflicts again, to
2743 : * ensure that all rw-conflicts in to these inserts are detected. Without
2744 : * this final check, a sequential scan of the heap may have locked the
2745 : * table after the "before" check, missing one opportunity to detect the
2746 : * conflict, and then scanned the table before the new tuples were there,
2747 : * missing the other chance to detect the conflict.
2748 : *
2749 : * For heap inserts, we only need to check for table-level SSI locks. Our
2750 : * new tuples can't possibly conflict with existing tuple locks, and heap
2751 : * page locks are only consolidated versions of tuple locks; they do not
2752 : * lock "gaps" as index page locks do. So we don't need to specify a
2753 : * buffer when making the call.
2754 : */
2755 375163 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2756 :
2757 : /*
2758 : * If tuples are cacheable, mark them for invalidation from the caches in
2759 : * case we abort. Note it is OK to do this after releasing the buffer,
2760 : * because the heaptuples data structure is all in local memory, not in
2761 : * the shared buffer.
2762 : */
2763 375163 : if (IsCatalogRelation(relation))
2764 : {
2765 1280113 : for (i = 0; i < ntuples; i++)
2766 906184 : CacheInvalidateHeapTuple(relation, heaptuples[i], NULL);
2767 : }
2768 :
2769 : /* copy t_self fields back to the caller's slots */
2770 1883767 : for (i = 0; i < ntuples; i++)
2771 1508604 : slots[i]->tts_tid = heaptuples[i]->t_self;
2772 :
2773 375163 : pgstat_count_heap_insert(relation, ntuples);
2774 375163 : }
2775 :
2776 : /*
2777 : * simple_heap_insert - insert a tuple
2778 : *
2779 : * Currently, this routine differs from heap_insert only in supplying
2780 : * a default command ID and not allowing access to the speedup options.
2781 : *
2782 : * This should be used rather than using heap_insert directly in most places
2783 : * where we are modifying system catalogs.
2784 : */
2785 : void
2786 934872 : simple_heap_insert(Relation relation, HeapTuple tup)
2787 : {
2788 934872 : heap_insert(relation, tup, GetCurrentCommandId(true), 0, NULL);
2789 934872 : }
2790 :
2791 : /*
2792 : * Given infomask/infomask2, compute the bits that must be saved in the
2793 : * "infobits" field of xl_heap_delete, xl_heap_update, xl_heap_lock,
2794 : * xl_heap_lock_updated WAL records.
2795 : *
2796 : * See fix_infomask_from_infobits.
2797 : */
2798 : static uint8
2799 2063269 : compute_infobits(uint16 infomask, uint16 infomask2)
2800 : {
2801 : return
2802 2063269 : ((infomask & HEAP_XMAX_IS_MULTI) != 0 ? XLHL_XMAX_IS_MULTI : 0) |
2803 2063269 : ((infomask & HEAP_XMAX_LOCK_ONLY) != 0 ? XLHL_XMAX_LOCK_ONLY : 0) |
2804 2063269 : ((infomask & HEAP_XMAX_EXCL_LOCK) != 0 ? XLHL_XMAX_EXCL_LOCK : 0) |
2805 : /* note we ignore HEAP_XMAX_SHR_LOCK here */
2806 4126538 : ((infomask & HEAP_XMAX_KEYSHR_LOCK) != 0 ? XLHL_XMAX_KEYSHR_LOCK : 0) |
2807 : ((infomask2 & HEAP_KEYS_UPDATED) != 0 ?
2808 2063269 : XLHL_KEYS_UPDATED : 0);
2809 : }
2810 :
2811 : /*
2812 : * Given two versions of the same t_infomask for a tuple, compare them and
2813 : * return whether the relevant status for a tuple Xmax has changed. This is
2814 : * used after a buffer lock has been released and reacquired: we want to ensure
2815 : * that the tuple state continues to be the same it was when we previously
2816 : * examined it.
2817 : *
2818 : * Note the Xmax field itself must be compared separately.
2819 : */
2820 : static inline bool
2821 5385 : xmax_infomask_changed(uint16 new_infomask, uint16 old_infomask)
2822 : {
2823 5385 : const uint16 interesting =
2824 : HEAP_XMAX_IS_MULTI | HEAP_XMAX_LOCK_ONLY | HEAP_LOCK_MASK;
2825 :
2826 5385 : if ((new_infomask & interesting) != (old_infomask & interesting))
2827 18 : return true;
2828 :
2829 5367 : return false;
2830 : }
2831 :
2832 : /*
2833 : * heap_delete - delete a tuple
2834 : *
2835 : * See table_tuple_delete() for an explanation of the parameters, except that
2836 : * this routine directly takes a tuple rather than a slot.
2837 : *
2838 : * In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
2839 : * t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
2840 : * only for TM_SelfModified, since we cannot obtain cmax from a combo CID
2841 : * generated by another transaction).
2842 : */
2843 : TM_Result
2844 1527972 : heap_delete(Relation relation, const ItemPointerData *tid,
2845 : CommandId cid, Snapshot crosscheck, bool wait,
2846 : TM_FailureData *tmfd, bool changingPart)
2847 : {
2848 : TM_Result result;
2849 1527972 : TransactionId xid = GetCurrentTransactionId();
2850 : ItemId lp;
2851 : HeapTupleData tp;
2852 : Page page;
2853 : BlockNumber block;
2854 : Buffer buffer;
2855 1527972 : Buffer vmbuffer = InvalidBuffer;
2856 : TransactionId new_xmax;
2857 : uint16 new_infomask,
2858 : new_infomask2;
2859 1527972 : bool have_tuple_lock = false;
2860 : bool iscombo;
2861 1527972 : bool all_visible_cleared = false;
2862 1527972 : HeapTuple old_key_tuple = NULL; /* replica identity of the tuple */
2863 1527972 : bool old_key_copied = false;
2864 :
2865 : Assert(ItemPointerIsValid(tid));
2866 :
2867 1527972 : AssertHasSnapshotForToast(relation);
2868 :
2869 : /*
2870 : * Forbid this during a parallel operation, lest it allocate a combo CID.
2871 : * Other workers might need that combo CID for visibility checks, and we
2872 : * have no provision for broadcasting it to them.
2873 : */
2874 1527972 : if (IsInParallelMode())
2875 0 : ereport(ERROR,
2876 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
2877 : errmsg("cannot delete tuples during a parallel operation")));
2878 :
2879 1527972 : block = ItemPointerGetBlockNumber(tid);
2880 1527972 : buffer = ReadBuffer(relation, block);
2881 1527972 : page = BufferGetPage(buffer);
2882 :
2883 : /*
2884 : * Before locking the buffer, pin the visibility map page if it appears to
2885 : * be necessary. Since we haven't got the lock yet, someone else might be
2886 : * in the middle of changing this, so we'll need to recheck after we have
2887 : * the lock.
2888 : */
2889 1527972 : if (PageIsAllVisible(page))
2890 221 : visibilitymap_pin(relation, block, &vmbuffer);
2891 :
2892 1527972 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2893 :
2894 1527972 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
2895 : Assert(ItemIdIsNormal(lp));
2896 :
2897 1527972 : tp.t_tableOid = RelationGetRelid(relation);
2898 1527972 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
2899 1527972 : tp.t_len = ItemIdGetLength(lp);
2900 1527972 : tp.t_self = *tid;
2901 :
2902 1 : l1:
2903 :
2904 : /*
2905 : * If we didn't pin the visibility map page and the page has become all
2906 : * visible while we were busy locking the buffer, we'll have to unlock and
2907 : * re-lock, to avoid holding the buffer lock across an I/O. That's a bit
2908 : * unfortunate, but hopefully shouldn't happen often.
2909 : */
2910 1527973 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
2911 : {
2912 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2913 0 : visibilitymap_pin(relation, block, &vmbuffer);
2914 0 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2915 : }
2916 :
2917 1527973 : result = HeapTupleSatisfiesUpdate(&tp, cid, buffer);
2918 :
2919 1527973 : if (result == TM_Invisible)
2920 : {
2921 0 : UnlockReleaseBuffer(buffer);
2922 0 : ereport(ERROR,
2923 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2924 : errmsg("attempted to delete invisible tuple")));
2925 : }
2926 1527973 : else if (result == TM_BeingModified && wait)
2927 : {
2928 : TransactionId xwait;
2929 : uint16 infomask;
2930 :
2931 : /* must copy state data before unlocking buffer */
2932 40562 : xwait = HeapTupleHeaderGetRawXmax(tp.t_data);
2933 40562 : infomask = tp.t_data->t_infomask;
2934 :
2935 : /*
2936 : * Sleep until concurrent transaction ends -- except when there's a
2937 : * single locker and it's our own transaction. Note we don't care
2938 : * which lock mode the locker has, because we need the strongest one.
2939 : *
2940 : * Before sleeping, we need to acquire tuple lock to establish our
2941 : * priority for the tuple (see heap_lock_tuple). LockTuple will
2942 : * release us when we are next-in-line for the tuple.
2943 : *
2944 : * If we are forced to "start over" below, we keep the tuple lock;
2945 : * this arranges that we stay at the head of the line while rechecking
2946 : * tuple state.
2947 : */
2948 40562 : if (infomask & HEAP_XMAX_IS_MULTI)
2949 : {
2950 8 : bool current_is_member = false;
2951 :
2952 8 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
2953 : LockTupleExclusive, ¤t_is_member))
2954 : {
2955 8 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2956 :
2957 : /*
2958 : * Acquire the lock, if necessary (but skip it when we're
2959 : * requesting a lock and already have one; avoids deadlock).
2960 : */
2961 8 : if (!current_is_member)
2962 6 : heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
2963 : LockWaitBlock, &have_tuple_lock);
2964 :
2965 : /* wait for multixact */
2966 8 : MultiXactIdWait((MultiXactId) xwait, MultiXactStatusUpdate, infomask,
2967 : relation, &(tp.t_self), XLTW_Delete,
2968 : NULL);
2969 8 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2970 :
2971 : /*
2972 : * If xwait had just locked the tuple then some other xact
2973 : * could update this tuple before we get to this point. Check
2974 : * for xmax change, and start over if so.
2975 : *
2976 : * We also must start over if we didn't pin the VM page, and
2977 : * the page has become all visible.
2978 : */
2979 16 : if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
2980 16 : xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
2981 8 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
2982 : xwait))
2983 0 : goto l1;
2984 : }
2985 :
2986 : /*
2987 : * You might think the multixact is necessarily done here, but not
2988 : * so: it could have surviving members, namely our own xact or
2989 : * other subxacts of this backend. It is legal for us to delete
2990 : * the tuple in either case, however (the latter case is
2991 : * essentially a situation of upgrading our former shared lock to
2992 : * exclusive). We don't bother changing the on-disk hint bits
2993 : * since we are about to overwrite the xmax altogether.
2994 : */
2995 : }
2996 40554 : else if (!TransactionIdIsCurrentTransactionId(xwait))
2997 : {
2998 : /*
2999 : * Wait for regular transaction to end; but first, acquire tuple
3000 : * lock.
3001 : */
3002 52 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3003 52 : heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
3004 : LockWaitBlock, &have_tuple_lock);
3005 52 : XactLockTableWait(xwait, relation, &(tp.t_self), XLTW_Delete);
3006 48 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3007 :
3008 : /*
3009 : * xwait is done, but if xwait had just locked the tuple then some
3010 : * other xact could update this tuple before we get to this point.
3011 : * Check for xmax change, and start over if so.
3012 : *
3013 : * We also must start over if we didn't pin the VM page, and the
3014 : * page has become all visible.
3015 : */
3016 96 : if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
3017 95 : xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
3018 47 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
3019 : xwait))
3020 1 : goto l1;
3021 :
3022 : /* Otherwise check if it committed or aborted */
3023 47 : UpdateXmaxHintBits(tp.t_data, buffer, xwait);
3024 : }
3025 :
3026 : /*
3027 : * We may overwrite if previous xmax aborted, or if it committed but
3028 : * only locked the tuple without updating it.
3029 : */
3030 81094 : if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
3031 40568 : HEAP_XMAX_IS_LOCKED_ONLY(tp.t_data->t_infomask) ||
3032 31 : HeapTupleHeaderIsOnlyLocked(tp.t_data))
3033 40530 : result = TM_Ok;
3034 27 : else if (!ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
3035 23 : result = TM_Updated;
3036 : else
3037 4 : result = TM_Deleted;
3038 : }
3039 :
3040 : /* sanity check the result HeapTupleSatisfiesUpdate() and the logic above */
3041 : if (result != TM_Ok)
3042 : {
3043 : Assert(result == TM_SelfModified ||
3044 : result == TM_Updated ||
3045 : result == TM_Deleted ||
3046 : result == TM_BeingModified);
3047 : Assert(!(tp.t_data->t_infomask & HEAP_XMAX_INVALID));
3048 : Assert(result != TM_Updated ||
3049 : !ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid));
3050 : }
3051 :
3052 1527968 : if (crosscheck != InvalidSnapshot && result == TM_Ok)
3053 : {
3054 : /* Perform additional check for transaction-snapshot mode RI updates */
3055 1 : if (!HeapTupleSatisfiesVisibility(&tp, crosscheck, buffer))
3056 1 : result = TM_Updated;
3057 : }
3058 :
3059 1527968 : if (result != TM_Ok)
3060 : {
3061 62 : tmfd->ctid = tp.t_data->t_ctid;
3062 62 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
3063 62 : if (result == TM_SelfModified)
3064 21 : tmfd->cmax = HeapTupleHeaderGetCmax(tp.t_data);
3065 : else
3066 41 : tmfd->cmax = InvalidCommandId;
3067 62 : UnlockReleaseBuffer(buffer);
3068 62 : if (have_tuple_lock)
3069 27 : UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
3070 62 : if (vmbuffer != InvalidBuffer)
3071 0 : ReleaseBuffer(vmbuffer);
3072 62 : return result;
3073 : }
3074 :
3075 : /*
3076 : * We're about to do the actual delete -- check for conflict first, to
3077 : * avoid possibly having to roll back work we've just done.
3078 : *
3079 : * This is safe without a recheck as long as there is no possibility of
3080 : * another process scanning the page between this check and the delete
3081 : * being visible to the scan (i.e., an exclusive buffer content lock is
3082 : * continuously held from this point until the tuple delete is visible).
3083 : */
3084 1527906 : CheckForSerializableConflictIn(relation, tid, BufferGetBlockNumber(buffer));
3085 :
3086 : /* replace cid with a combo CID if necessary */
3087 1527892 : HeapTupleHeaderAdjustCmax(tp.t_data, &cid, &iscombo);
3088 :
3089 : /*
3090 : * Compute replica identity tuple before entering the critical section so
3091 : * we don't PANIC upon a memory allocation failure.
3092 : */
3093 1527892 : old_key_tuple = ExtractReplicaIdentity(relation, &tp, true, &old_key_copied);
3094 :
3095 : /*
3096 : * If this is the first possibly-multixact-able operation in the current
3097 : * transaction, set my per-backend OldestMemberMXactId setting. We can be
3098 : * certain that the transaction will never become a member of any older
3099 : * MultiXactIds than that. (We have to do this even if we end up just
3100 : * using our own TransactionId below, since some other backend could
3101 : * incorporate our XID into a MultiXact immediately afterwards.)
3102 : */
3103 1527892 : MultiXactIdSetOldestMember();
3104 :
3105 1527892 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(tp.t_data),
3106 1527892 : tp.t_data->t_infomask, tp.t_data->t_infomask2,
3107 : xid, LockTupleExclusive, true,
3108 : &new_xmax, &new_infomask, &new_infomask2);
3109 :
3110 1527892 : START_CRIT_SECTION();
3111 :
3112 : /*
3113 : * If this transaction commits, the tuple will become DEAD sooner or
3114 : * later. Set flag that this page is a candidate for pruning once our xid
3115 : * falls below the OldestXmin horizon. If the transaction finally aborts,
3116 : * the subsequent page pruning will be a no-op and the hint will be
3117 : * cleared.
3118 : */
3119 1527892 : PageSetPrunable(page, xid);
3120 :
3121 1527892 : if (PageIsAllVisible(page))
3122 : {
3123 221 : all_visible_cleared = true;
3124 221 : PageClearAllVisible(page);
3125 221 : visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
3126 : vmbuffer, VISIBILITYMAP_VALID_BITS);
3127 : }
3128 :
3129 : /* store transaction information of xact deleting the tuple */
3130 1527892 : tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
3131 1527892 : tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
3132 1527892 : tp.t_data->t_infomask |= new_infomask;
3133 1527892 : tp.t_data->t_infomask2 |= new_infomask2;
3134 1527892 : HeapTupleHeaderClearHotUpdated(tp.t_data);
3135 1527892 : HeapTupleHeaderSetXmax(tp.t_data, new_xmax);
3136 1527892 : HeapTupleHeaderSetCmax(tp.t_data, cid, iscombo);
3137 : /* Make sure there is no forward chain link in t_ctid */
3138 1527892 : tp.t_data->t_ctid = tp.t_self;
3139 :
3140 : /* Signal that this is actually a move into another partition */
3141 1527892 : if (changingPart)
3142 493 : HeapTupleHeaderSetMovedPartitions(tp.t_data);
3143 :
3144 1527892 : MarkBufferDirty(buffer);
3145 :
3146 : /*
3147 : * XLOG stuff
3148 : *
3149 : * NB: heap_abort_speculative() uses the same xlog record and replay
3150 : * routines.
3151 : */
3152 1527892 : if (RelationNeedsWAL(relation))
3153 : {
3154 : xl_heap_delete xlrec;
3155 : xl_heap_header xlhdr;
3156 : XLogRecPtr recptr;
3157 :
3158 : /*
3159 : * For logical decode we need combo CIDs to properly decode the
3160 : * catalog
3161 : */
3162 1465298 : if (RelationIsAccessibleInLogicalDecoding(relation))
3163 6294 : log_heap_new_cid(relation, &tp);
3164 :
3165 1465298 : xlrec.flags = 0;
3166 1465298 : if (all_visible_cleared)
3167 221 : xlrec.flags |= XLH_DELETE_ALL_VISIBLE_CLEARED;
3168 1465298 : if (changingPart)
3169 493 : xlrec.flags |= XLH_DELETE_IS_PARTITION_MOVE;
3170 2930596 : xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
3171 1465298 : tp.t_data->t_infomask2);
3172 1465298 : xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
3173 1465298 : xlrec.xmax = new_xmax;
3174 :
3175 1465298 : if (old_key_tuple != NULL)
3176 : {
3177 47020 : if (relation->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
3178 132 : xlrec.flags |= XLH_DELETE_CONTAINS_OLD_TUPLE;
3179 : else
3180 46888 : xlrec.flags |= XLH_DELETE_CONTAINS_OLD_KEY;
3181 : }
3182 :
3183 1465298 : XLogBeginInsert();
3184 1465298 : XLogRegisterData(&xlrec, SizeOfHeapDelete);
3185 :
3186 1465298 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
3187 :
3188 : /*
3189 : * Log replica identity of the deleted tuple if there is one
3190 : */
3191 1465298 : if (old_key_tuple != NULL)
3192 : {
3193 47020 : xlhdr.t_infomask2 = old_key_tuple->t_data->t_infomask2;
3194 47020 : xlhdr.t_infomask = old_key_tuple->t_data->t_infomask;
3195 47020 : xlhdr.t_hoff = old_key_tuple->t_data->t_hoff;
3196 :
3197 47020 : XLogRegisterData(&xlhdr, SizeOfHeapHeader);
3198 47020 : XLogRegisterData((char *) old_key_tuple->t_data
3199 : + SizeofHeapTupleHeader,
3200 47020 : old_key_tuple->t_len
3201 : - SizeofHeapTupleHeader);
3202 : }
3203 :
3204 : /* filtering by origin on a row level is much more efficient */
3205 1465298 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
3206 :
3207 1465298 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
3208 :
3209 1465298 : PageSetLSN(page, recptr);
3210 : }
3211 :
3212 1527892 : END_CRIT_SECTION();
3213 :
3214 1527892 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3215 :
3216 1527892 : if (vmbuffer != InvalidBuffer)
3217 221 : ReleaseBuffer(vmbuffer);
3218 :
3219 : /*
3220 : * If the tuple has toasted out-of-line attributes, we need to delete
3221 : * those items too. We have to do this before releasing the buffer
3222 : * because we need to look at the contents of the tuple, but it's OK to
3223 : * release the content lock on the buffer first.
3224 : */
3225 1530451 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
3226 2569 : relation->rd_rel->relkind != RELKIND_MATVIEW)
3227 : {
3228 : /* toast table entries should never be recursively toasted */
3229 : Assert(!HeapTupleHasExternal(&tp));
3230 : }
3231 1525333 : else if (HeapTupleHasExternal(&tp))
3232 294 : heap_toast_delete(relation, &tp, false);
3233 :
3234 : /*
3235 : * Mark tuple for invalidation from system caches at next command
3236 : * boundary. We have to do this before releasing the buffer because we
3237 : * need to look at the contents of the tuple.
3238 : */
3239 1527892 : CacheInvalidateHeapTuple(relation, &tp, NULL);
3240 :
3241 : /* Now we can release the buffer */
3242 1527892 : ReleaseBuffer(buffer);
3243 :
3244 : /*
3245 : * Release the lmgr tuple lock, if we had it.
3246 : */
3247 1527892 : if (have_tuple_lock)
3248 26 : UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
3249 :
3250 1527892 : pgstat_count_heap_delete(relation);
3251 :
3252 1527892 : if (old_key_tuple != NULL && old_key_copied)
3253 46889 : heap_freetuple(old_key_tuple);
3254 :
3255 1527892 : return TM_Ok;
3256 : }
3257 :
3258 : /*
3259 : * simple_heap_delete - delete a tuple
3260 : *
3261 : * This routine may be used to delete a tuple when concurrent updates of
3262 : * the target tuple are not expected (for example, because we have a lock
3263 : * on the relation associated with the tuple). Any failure is reported
3264 : * via ereport().
3265 : */
3266 : void
3267 661990 : simple_heap_delete(Relation relation, const ItemPointerData *tid)
3268 : {
3269 : TM_Result result;
3270 : TM_FailureData tmfd;
3271 :
3272 661990 : result = heap_delete(relation, tid,
3273 : GetCurrentCommandId(true), InvalidSnapshot,
3274 : true /* wait for commit */ ,
3275 : &tmfd, false /* changingPart */ );
3276 661990 : switch (result)
3277 : {
3278 0 : case TM_SelfModified:
3279 : /* Tuple was already updated in current command? */
3280 0 : elog(ERROR, "tuple already updated by self");
3281 : break;
3282 :
3283 661990 : case TM_Ok:
3284 : /* done successfully */
3285 661990 : break;
3286 :
3287 0 : case TM_Updated:
3288 0 : elog(ERROR, "tuple concurrently updated");
3289 : break;
3290 :
3291 0 : case TM_Deleted:
3292 0 : elog(ERROR, "tuple concurrently deleted");
3293 : break;
3294 :
3295 0 : default:
3296 0 : elog(ERROR, "unrecognized heap_delete status: %u", result);
3297 : break;
3298 : }
3299 661990 : }
3300 :
3301 : /*
3302 : * heap_update - replace a tuple
3303 : *
3304 : * See table_tuple_update() for an explanation of the parameters, except that
3305 : * this routine directly takes a tuple rather than a slot.
3306 : *
3307 : * In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
3308 : * t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
3309 : * only for TM_SelfModified, since we cannot obtain cmax from a combo CID
3310 : * generated by another transaction).
3311 : */
3312 : TM_Result
3313 311409 : heap_update(Relation relation, const ItemPointerData *otid, HeapTuple newtup,
3314 : CommandId cid, Snapshot crosscheck, bool wait,
3315 : TM_FailureData *tmfd, LockTupleMode *lockmode,
3316 : TU_UpdateIndexes *update_indexes)
3317 : {
3318 : TM_Result result;
3319 311409 : TransactionId xid = GetCurrentTransactionId();
3320 : Bitmapset *hot_attrs;
3321 : Bitmapset *sum_attrs;
3322 : Bitmapset *key_attrs;
3323 : Bitmapset *id_attrs;
3324 : Bitmapset *interesting_attrs;
3325 : Bitmapset *modified_attrs;
3326 : ItemId lp;
3327 : HeapTupleData oldtup;
3328 : HeapTuple heaptup;
3329 311409 : HeapTuple old_key_tuple = NULL;
3330 311409 : bool old_key_copied = false;
3331 : Page page;
3332 : BlockNumber block;
3333 : MultiXactStatus mxact_status;
3334 : Buffer buffer,
3335 : newbuf,
3336 311409 : vmbuffer = InvalidBuffer,
3337 311409 : vmbuffer_new = InvalidBuffer;
3338 : bool need_toast;
3339 : Size newtupsize,
3340 : pagefree;
3341 311409 : bool have_tuple_lock = false;
3342 : bool iscombo;
3343 311409 : bool use_hot_update = false;
3344 311409 : bool summarized_update = false;
3345 : bool key_intact;
3346 311409 : bool all_visible_cleared = false;
3347 311409 : bool all_visible_cleared_new = false;
3348 : bool checked_lockers;
3349 : bool locker_remains;
3350 311409 : bool id_has_external = false;
3351 : TransactionId xmax_new_tuple,
3352 : xmax_old_tuple;
3353 : uint16 infomask_old_tuple,
3354 : infomask2_old_tuple,
3355 : infomask_new_tuple,
3356 : infomask2_new_tuple;
3357 :
3358 : Assert(ItemPointerIsValid(otid));
3359 :
3360 : /* Cheap, simplistic check that the tuple matches the rel's rowtype. */
3361 : Assert(HeapTupleHeaderGetNatts(newtup->t_data) <=
3362 : RelationGetNumberOfAttributes(relation));
3363 :
3364 311409 : AssertHasSnapshotForToast(relation);
3365 :
3366 : /*
3367 : * Forbid this during a parallel operation, lest it allocate a combo CID.
3368 : * Other workers might need that combo CID for visibility checks, and we
3369 : * have no provision for broadcasting it to them.
3370 : */
3371 311409 : if (IsInParallelMode())
3372 0 : ereport(ERROR,
3373 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
3374 : errmsg("cannot update tuples during a parallel operation")));
3375 :
3376 : #ifdef USE_ASSERT_CHECKING
3377 : check_lock_if_inplace_updateable_rel(relation, otid, newtup);
3378 : #endif
3379 :
3380 : /*
3381 : * Fetch the list of attributes to be checked for various operations.
3382 : *
3383 : * For HOT considerations, this is wasted effort if we fail to update or
3384 : * have to put the new tuple on a different page. But we must compute the
3385 : * list before obtaining buffer lock --- in the worst case, if we are
3386 : * doing an update on one of the relevant system catalogs, we could
3387 : * deadlock if we try to fetch the list later. In any case, the relcache
3388 : * caches the data so this is usually pretty cheap.
3389 : *
3390 : * We also need columns used by the replica identity and columns that are
3391 : * considered the "key" of rows in the table.
3392 : *
3393 : * Note that we get copies of each bitmap, so we need not worry about
3394 : * relcache flush happening midway through.
3395 : */
3396 311409 : hot_attrs = RelationGetIndexAttrBitmap(relation,
3397 : INDEX_ATTR_BITMAP_HOT_BLOCKING);
3398 311409 : sum_attrs = RelationGetIndexAttrBitmap(relation,
3399 : INDEX_ATTR_BITMAP_SUMMARIZED);
3400 311409 : key_attrs = RelationGetIndexAttrBitmap(relation, INDEX_ATTR_BITMAP_KEY);
3401 311409 : id_attrs = RelationGetIndexAttrBitmap(relation,
3402 : INDEX_ATTR_BITMAP_IDENTITY_KEY);
3403 311409 : interesting_attrs = NULL;
3404 311409 : interesting_attrs = bms_add_members(interesting_attrs, hot_attrs);
3405 311409 : interesting_attrs = bms_add_members(interesting_attrs, sum_attrs);
3406 311409 : interesting_attrs = bms_add_members(interesting_attrs, key_attrs);
3407 311409 : interesting_attrs = bms_add_members(interesting_attrs, id_attrs);
3408 :
3409 311409 : block = ItemPointerGetBlockNumber(otid);
3410 311409 : INJECTION_POINT("heap_update-before-pin", NULL);
3411 311409 : buffer = ReadBuffer(relation, block);
3412 311409 : page = BufferGetPage(buffer);
3413 :
3414 : /*
3415 : * Before locking the buffer, pin the visibility map page if it appears to
3416 : * be necessary. Since we haven't got the lock yet, someone else might be
3417 : * in the middle of changing this, so we'll need to recheck after we have
3418 : * the lock.
3419 : */
3420 311409 : if (PageIsAllVisible(page))
3421 1747 : visibilitymap_pin(relation, block, &vmbuffer);
3422 :
3423 311409 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3424 :
3425 311409 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(otid));
3426 :
3427 : /*
3428 : * Usually, a buffer pin and/or snapshot blocks pruning of otid, ensuring
3429 : * we see LP_NORMAL here. When the otid origin is a syscache, we may have
3430 : * neither a pin nor a snapshot. Hence, we may see other LP_ states, each
3431 : * of which indicates concurrent pruning.
3432 : *
3433 : * Failing with TM_Updated would be most accurate. However, unlike other
3434 : * TM_Updated scenarios, we don't know the successor ctid in LP_UNUSED and
3435 : * LP_DEAD cases. While the distinction between TM_Updated and TM_Deleted
3436 : * does matter to SQL statements UPDATE and MERGE, those SQL statements
3437 : * hold a snapshot that ensures LP_NORMAL. Hence, the choice between
3438 : * TM_Updated and TM_Deleted affects only the wording of error messages.
3439 : * Settle on TM_Deleted, for two reasons. First, it avoids complicating
3440 : * the specification of when tmfd->ctid is valid. Second, it creates
3441 : * error log evidence that we took this branch.
3442 : *
3443 : * Since it's possible to see LP_UNUSED at otid, it's also possible to see
3444 : * LP_NORMAL for a tuple that replaced LP_UNUSED. If it's a tuple for an
3445 : * unrelated row, we'll fail with "duplicate key value violates unique".
3446 : * XXX if otid is the live, newer version of the newtup row, we'll discard
3447 : * changes originating in versions of this catalog row after the version
3448 : * the caller got from syscache. See syscache-update-pruned.spec.
3449 : */
3450 311409 : if (!ItemIdIsNormal(lp))
3451 : {
3452 : Assert(RelationSupportsSysCache(RelationGetRelid(relation)));
3453 :
3454 1 : UnlockReleaseBuffer(buffer);
3455 : Assert(!have_tuple_lock);
3456 1 : if (vmbuffer != InvalidBuffer)
3457 1 : ReleaseBuffer(vmbuffer);
3458 1 : tmfd->ctid = *otid;
3459 1 : tmfd->xmax = InvalidTransactionId;
3460 1 : tmfd->cmax = InvalidCommandId;
3461 1 : *update_indexes = TU_None;
3462 :
3463 1 : bms_free(hot_attrs);
3464 1 : bms_free(sum_attrs);
3465 1 : bms_free(key_attrs);
3466 1 : bms_free(id_attrs);
3467 : /* modified_attrs not yet initialized */
3468 1 : bms_free(interesting_attrs);
3469 1 : return TM_Deleted;
3470 : }
3471 :
3472 : /*
3473 : * Fill in enough data in oldtup for HeapDetermineColumnsInfo to work
3474 : * properly.
3475 : */
3476 311408 : oldtup.t_tableOid = RelationGetRelid(relation);
3477 311408 : oldtup.t_data = (HeapTupleHeader) PageGetItem(page, lp);
3478 311408 : oldtup.t_len = ItemIdGetLength(lp);
3479 311408 : oldtup.t_self = *otid;
3480 :
3481 : /* the new tuple is ready, except for this: */
3482 311408 : newtup->t_tableOid = RelationGetRelid(relation);
3483 :
3484 : /*
3485 : * Determine columns modified by the update. Additionally, identify
3486 : * whether any of the unmodified replica identity key attributes in the
3487 : * old tuple is externally stored or not. This is required because for
3488 : * such attributes the flattened value won't be WAL logged as part of the
3489 : * new tuple so we must include it as part of the old_key_tuple. See
3490 : * ExtractReplicaIdentity.
3491 : */
3492 311408 : modified_attrs = HeapDetermineColumnsInfo(relation, interesting_attrs,
3493 : id_attrs, &oldtup,
3494 : newtup, &id_has_external);
3495 :
3496 : /*
3497 : * If we're not updating any "key" column, we can grab a weaker lock type.
3498 : * This allows for more concurrency when we are running simultaneously
3499 : * with foreign key checks.
3500 : *
3501 : * Note that if a column gets detoasted while executing the update, but
3502 : * the value ends up being the same, this test will fail and we will use
3503 : * the stronger lock. This is acceptable; the important case to optimize
3504 : * is updates that don't manipulate key columns, not those that
3505 : * serendipitously arrive at the same key values.
3506 : */
3507 311408 : if (!bms_overlap(modified_attrs, key_attrs))
3508 : {
3509 307075 : *lockmode = LockTupleNoKeyExclusive;
3510 307075 : mxact_status = MultiXactStatusNoKeyUpdate;
3511 307075 : key_intact = true;
3512 :
3513 : /*
3514 : * If this is the first possibly-multixact-able operation in the
3515 : * current transaction, set my per-backend OldestMemberMXactId
3516 : * setting. We can be certain that the transaction will never become a
3517 : * member of any older MultiXactIds than that. (We have to do this
3518 : * even if we end up just using our own TransactionId below, since
3519 : * some other backend could incorporate our XID into a MultiXact
3520 : * immediately afterwards.)
3521 : */
3522 307075 : MultiXactIdSetOldestMember();
3523 : }
3524 : else
3525 : {
3526 4333 : *lockmode = LockTupleExclusive;
3527 4333 : mxact_status = MultiXactStatusUpdate;
3528 4333 : key_intact = false;
3529 : }
3530 :
3531 : /*
3532 : * Note: beyond this point, use oldtup not otid to refer to old tuple.
3533 : * otid may very well point at newtup->t_self, which we will overwrite
3534 : * with the new tuple's location, so there's great risk of confusion if we
3535 : * use otid anymore.
3536 : */
3537 :
3538 1 : l2:
3539 311409 : checked_lockers = false;
3540 311409 : locker_remains = false;
3541 311409 : result = HeapTupleSatisfiesUpdate(&oldtup, cid, buffer);
3542 :
3543 : /* see below about the "no wait" case */
3544 : Assert(result != TM_BeingModified || wait);
3545 :
3546 311409 : if (result == TM_Invisible)
3547 : {
3548 0 : UnlockReleaseBuffer(buffer);
3549 0 : ereport(ERROR,
3550 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3551 : errmsg("attempted to update invisible tuple")));
3552 : }
3553 311409 : else if (result == TM_BeingModified && wait)
3554 : {
3555 : TransactionId xwait;
3556 : uint16 infomask;
3557 36114 : bool can_continue = false;
3558 :
3559 : /*
3560 : * XXX note that we don't consider the "no wait" case here. This
3561 : * isn't a problem currently because no caller uses that case, but it
3562 : * should be fixed if such a caller is introduced. It wasn't a
3563 : * problem previously because this code would always wait, but now
3564 : * that some tuple locks do not conflict with one of the lock modes we
3565 : * use, it is possible that this case is interesting to handle
3566 : * specially.
3567 : *
3568 : * This may cause failures with third-party code that calls
3569 : * heap_update directly.
3570 : */
3571 :
3572 : /* must copy state data before unlocking buffer */
3573 36114 : xwait = HeapTupleHeaderGetRawXmax(oldtup.t_data);
3574 36114 : infomask = oldtup.t_data->t_infomask;
3575 :
3576 : /*
3577 : * Now we have to do something about the existing locker. If it's a
3578 : * multi, sleep on it; we might be awakened before it is completely
3579 : * gone (or even not sleep at all in some cases); we need to preserve
3580 : * it as locker, unless it is gone completely.
3581 : *
3582 : * If it's not a multi, we need to check for sleeping conditions
3583 : * before actually going to sleep. If the update doesn't conflict
3584 : * with the locks, we just continue without sleeping (but making sure
3585 : * it is preserved).
3586 : *
3587 : * Before sleeping, we need to acquire tuple lock to establish our
3588 : * priority for the tuple (see heap_lock_tuple). LockTuple will
3589 : * release us when we are next-in-line for the tuple. Note we must
3590 : * not acquire the tuple lock until we're sure we're going to sleep;
3591 : * otherwise we're open for race conditions with other transactions
3592 : * holding the tuple lock which sleep on us.
3593 : *
3594 : * If we are forced to "start over" below, we keep the tuple lock;
3595 : * this arranges that we stay at the head of the line while rechecking
3596 : * tuple state.
3597 : */
3598 36114 : if (infomask & HEAP_XMAX_IS_MULTI)
3599 : {
3600 : TransactionId update_xact;
3601 : int remain;
3602 179 : bool current_is_member = false;
3603 :
3604 179 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
3605 : *lockmode, ¤t_is_member))
3606 : {
3607 8 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3608 :
3609 : /*
3610 : * Acquire the lock, if necessary (but skip it when we're
3611 : * requesting a lock and already have one; avoids deadlock).
3612 : */
3613 8 : if (!current_is_member)
3614 0 : heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3615 : LockWaitBlock, &have_tuple_lock);
3616 :
3617 : /* wait for multixact */
3618 8 : MultiXactIdWait((MultiXactId) xwait, mxact_status, infomask,
3619 : relation, &oldtup.t_self, XLTW_Update,
3620 : &remain);
3621 8 : checked_lockers = true;
3622 8 : locker_remains = remain != 0;
3623 8 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3624 :
3625 : /*
3626 : * If xwait had just locked the tuple then some other xact
3627 : * could update this tuple before we get to this point. Check
3628 : * for xmax change, and start over if so.
3629 : */
3630 8 : if (xmax_infomask_changed(oldtup.t_data->t_infomask,
3631 8 : infomask) ||
3632 8 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3633 : xwait))
3634 0 : goto l2;
3635 : }
3636 :
3637 : /*
3638 : * Note that the multixact may not be done by now. It could have
3639 : * surviving members; our own xact or other subxacts of this
3640 : * backend, and also any other concurrent transaction that locked
3641 : * the tuple with LockTupleKeyShare if we only got
3642 : * LockTupleNoKeyExclusive. If this is the case, we have to be
3643 : * careful to mark the updated tuple with the surviving members in
3644 : * Xmax.
3645 : *
3646 : * Note that there could have been another update in the
3647 : * MultiXact. In that case, we need to check whether it committed
3648 : * or aborted. If it aborted we are safe to update it again;
3649 : * otherwise there is an update conflict, and we have to return
3650 : * TableTuple{Deleted, Updated} below.
3651 : *
3652 : * In the LockTupleExclusive case, we still need to preserve the
3653 : * surviving members: those would include the tuple locks we had
3654 : * before this one, which are important to keep in case this
3655 : * subxact aborts.
3656 : */
3657 179 : if (!HEAP_XMAX_IS_LOCKED_ONLY(oldtup.t_data->t_infomask))
3658 8 : update_xact = HeapTupleGetUpdateXid(oldtup.t_data);
3659 : else
3660 171 : update_xact = InvalidTransactionId;
3661 :
3662 : /*
3663 : * There was no UPDATE in the MultiXact; or it aborted. No
3664 : * TransactionIdIsInProgress() call needed here, since we called
3665 : * MultiXactIdWait() above.
3666 : */
3667 187 : if (!TransactionIdIsValid(update_xact) ||
3668 8 : TransactionIdDidAbort(update_xact))
3669 172 : can_continue = true;
3670 : }
3671 35935 : else if (TransactionIdIsCurrentTransactionId(xwait))
3672 : {
3673 : /*
3674 : * The only locker is ourselves; we can avoid grabbing the tuple
3675 : * lock here, but must preserve our locking information.
3676 : */
3677 35827 : checked_lockers = true;
3678 35827 : locker_remains = true;
3679 35827 : can_continue = true;
3680 : }
3681 108 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) && key_intact)
3682 : {
3683 : /*
3684 : * If it's just a key-share locker, and we're not changing the key
3685 : * columns, we don't need to wait for it to end; but we need to
3686 : * preserve it as locker.
3687 : */
3688 29 : checked_lockers = true;
3689 29 : locker_remains = true;
3690 29 : can_continue = true;
3691 : }
3692 : else
3693 : {
3694 : /*
3695 : * Wait for regular transaction to end; but first, acquire tuple
3696 : * lock.
3697 : */
3698 79 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3699 79 : heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3700 : LockWaitBlock, &have_tuple_lock);
3701 79 : XactLockTableWait(xwait, relation, &oldtup.t_self,
3702 : XLTW_Update);
3703 79 : checked_lockers = true;
3704 79 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3705 :
3706 : /*
3707 : * xwait is done, but if xwait had just locked the tuple then some
3708 : * other xact could update this tuple before we get to this point.
3709 : * Check for xmax change, and start over if so.
3710 : */
3711 157 : if (xmax_infomask_changed(oldtup.t_data->t_infomask, infomask) ||
3712 78 : !TransactionIdEquals(xwait,
3713 : HeapTupleHeaderGetRawXmax(oldtup.t_data)))
3714 1 : goto l2;
3715 :
3716 : /* Otherwise check if it committed or aborted */
3717 78 : UpdateXmaxHintBits(oldtup.t_data, buffer, xwait);
3718 78 : if (oldtup.t_data->t_infomask & HEAP_XMAX_INVALID)
3719 22 : can_continue = true;
3720 : }
3721 :
3722 36113 : if (can_continue)
3723 36050 : result = TM_Ok;
3724 63 : else if (!ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid))
3725 58 : result = TM_Updated;
3726 : else
3727 5 : result = TM_Deleted;
3728 : }
3729 :
3730 : /* Sanity check the result HeapTupleSatisfiesUpdate() and the logic above */
3731 : if (result != TM_Ok)
3732 : {
3733 : Assert(result == TM_SelfModified ||
3734 : result == TM_Updated ||
3735 : result == TM_Deleted ||
3736 : result == TM_BeingModified);
3737 : Assert(!(oldtup.t_data->t_infomask & HEAP_XMAX_INVALID));
3738 : Assert(result != TM_Updated ||
3739 : !ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid));
3740 : }
3741 :
3742 311408 : if (crosscheck != InvalidSnapshot && result == TM_Ok)
3743 : {
3744 : /* Perform additional check for transaction-snapshot mode RI updates */
3745 1 : if (!HeapTupleSatisfiesVisibility(&oldtup, crosscheck, buffer))
3746 1 : result = TM_Updated;
3747 : }
3748 :
3749 311408 : if (result != TM_Ok)
3750 : {
3751 162 : tmfd->ctid = oldtup.t_data->t_ctid;
3752 162 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(oldtup.t_data);
3753 162 : if (result == TM_SelfModified)
3754 52 : tmfd->cmax = HeapTupleHeaderGetCmax(oldtup.t_data);
3755 : else
3756 110 : tmfd->cmax = InvalidCommandId;
3757 162 : UnlockReleaseBuffer(buffer);
3758 162 : if (have_tuple_lock)
3759 56 : UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
3760 162 : if (vmbuffer != InvalidBuffer)
3761 0 : ReleaseBuffer(vmbuffer);
3762 162 : *update_indexes = TU_None;
3763 :
3764 162 : bms_free(hot_attrs);
3765 162 : bms_free(sum_attrs);
3766 162 : bms_free(key_attrs);
3767 162 : bms_free(id_attrs);
3768 162 : bms_free(modified_attrs);
3769 162 : bms_free(interesting_attrs);
3770 162 : return result;
3771 : }
3772 :
3773 : /*
3774 : * If we didn't pin the visibility map page and the page has become all
3775 : * visible while we were busy locking the buffer, or during some
3776 : * subsequent window during which we had it unlocked, we'll have to unlock
3777 : * and re-lock, to avoid holding the buffer lock across an I/O. That's a
3778 : * bit unfortunate, especially since we'll now have to recheck whether the
3779 : * tuple has been locked or updated under us, but hopefully it won't
3780 : * happen very often.
3781 : */
3782 311246 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
3783 : {
3784 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3785 0 : visibilitymap_pin(relation, block, &vmbuffer);
3786 0 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3787 0 : goto l2;
3788 : }
3789 :
3790 : /* Fill in transaction status data */
3791 :
3792 : /*
3793 : * If the tuple we're updating is locked, we need to preserve the locking
3794 : * info in the old tuple's Xmax. Prepare a new Xmax value for this.
3795 : */
3796 311246 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3797 311246 : oldtup.t_data->t_infomask,
3798 311246 : oldtup.t_data->t_infomask2,
3799 : xid, *lockmode, true,
3800 : &xmax_old_tuple, &infomask_old_tuple,
3801 : &infomask2_old_tuple);
3802 :
3803 : /*
3804 : * And also prepare an Xmax value for the new copy of the tuple. If there
3805 : * was no xmax previously, or there was one but all lockers are now gone,
3806 : * then use InvalidTransactionId; otherwise, get the xmax from the old
3807 : * tuple. (In rare cases that might also be InvalidTransactionId and yet
3808 : * not have the HEAP_XMAX_INVALID bit set; that's fine.)
3809 : */
3810 347274 : if ((oldtup.t_data->t_infomask & HEAP_XMAX_INVALID) ||
3811 72056 : HEAP_LOCKED_UPGRADED(oldtup.t_data->t_infomask) ||
3812 35857 : (checked_lockers && !locker_remains))
3813 275218 : xmax_new_tuple = InvalidTransactionId;
3814 : else
3815 36028 : xmax_new_tuple = HeapTupleHeaderGetRawXmax(oldtup.t_data);
3816 :
3817 311246 : if (!TransactionIdIsValid(xmax_new_tuple))
3818 : {
3819 275218 : infomask_new_tuple = HEAP_XMAX_INVALID;
3820 275218 : infomask2_new_tuple = 0;
3821 : }
3822 : else
3823 : {
3824 : /*
3825 : * If we found a valid Xmax for the new tuple, then the infomask bits
3826 : * to use on the new tuple depend on what was there on the old one.
3827 : * Note that since we're doing an update, the only possibility is that
3828 : * the lockers had FOR KEY SHARE lock.
3829 : */
3830 36028 : if (oldtup.t_data->t_infomask & HEAP_XMAX_IS_MULTI)
3831 : {
3832 172 : GetMultiXactIdHintBits(xmax_new_tuple, &infomask_new_tuple,
3833 : &infomask2_new_tuple);
3834 : }
3835 : else
3836 : {
3837 35856 : infomask_new_tuple = HEAP_XMAX_KEYSHR_LOCK | HEAP_XMAX_LOCK_ONLY;
3838 35856 : infomask2_new_tuple = 0;
3839 : }
3840 : }
3841 :
3842 : /*
3843 : * Prepare the new tuple with the appropriate initial values of Xmin and
3844 : * Xmax, as well as initial infomask bits as computed above.
3845 : */
3846 311246 : newtup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
3847 311246 : newtup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
3848 311246 : HeapTupleHeaderSetXmin(newtup->t_data, xid);
3849 311246 : HeapTupleHeaderSetCmin(newtup->t_data, cid);
3850 311246 : newtup->t_data->t_infomask |= HEAP_UPDATED | infomask_new_tuple;
3851 311246 : newtup->t_data->t_infomask2 |= infomask2_new_tuple;
3852 311246 : HeapTupleHeaderSetXmax(newtup->t_data, xmax_new_tuple);
3853 :
3854 : /*
3855 : * Replace cid with a combo CID if necessary. Note that we already put
3856 : * the plain cid into the new tuple.
3857 : */
3858 311246 : HeapTupleHeaderAdjustCmax(oldtup.t_data, &cid, &iscombo);
3859 :
3860 : /*
3861 : * If the toaster needs to be activated, OR if the new tuple will not fit
3862 : * on the same page as the old, then we need to release the content lock
3863 : * (but not the pin!) on the old tuple's buffer while we are off doing
3864 : * TOAST and/or table-file-extension work. We must mark the old tuple to
3865 : * show that it's locked, else other processes may try to update it
3866 : * themselves.
3867 : *
3868 : * We need to invoke the toaster if there are already any out-of-line
3869 : * toasted values present, or if the new tuple is over-threshold.
3870 : */
3871 311246 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
3872 0 : relation->rd_rel->relkind != RELKIND_MATVIEW)
3873 : {
3874 : /* toast table entries should never be recursively toasted */
3875 : Assert(!HeapTupleHasExternal(&oldtup));
3876 : Assert(!HeapTupleHasExternal(newtup));
3877 0 : need_toast = false;
3878 : }
3879 : else
3880 933364 : need_toast = (HeapTupleHasExternal(&oldtup) ||
3881 622118 : HeapTupleHasExternal(newtup) ||
3882 310848 : newtup->t_len > TOAST_TUPLE_THRESHOLD);
3883 :
3884 311246 : pagefree = PageGetHeapFreeSpace(page);
3885 :
3886 311246 : newtupsize = MAXALIGN(newtup->t_len);
3887 :
3888 311246 : if (need_toast || newtupsize > pagefree)
3889 149728 : {
3890 : TransactionId xmax_lock_old_tuple;
3891 : uint16 infomask_lock_old_tuple,
3892 : infomask2_lock_old_tuple;
3893 149728 : bool cleared_all_frozen = false;
3894 :
3895 : /*
3896 : * To prevent concurrent sessions from updating the tuple, we have to
3897 : * temporarily mark it locked, while we release the page-level lock.
3898 : *
3899 : * To satisfy the rule that any xid potentially appearing in a buffer
3900 : * written out to disk, we unfortunately have to WAL log this
3901 : * temporary modification. We can reuse xl_heap_lock for this
3902 : * purpose. If we crash/error before following through with the
3903 : * actual update, xmax will be of an aborted transaction, allowing
3904 : * other sessions to proceed.
3905 : */
3906 :
3907 : /*
3908 : * Compute xmax / infomask appropriate for locking the tuple. This has
3909 : * to be done separately from the combo that's going to be used for
3910 : * updating, because the potentially created multixact would otherwise
3911 : * be wrong.
3912 : */
3913 149728 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3914 149728 : oldtup.t_data->t_infomask,
3915 149728 : oldtup.t_data->t_infomask2,
3916 : xid, *lockmode, false,
3917 : &xmax_lock_old_tuple, &infomask_lock_old_tuple,
3918 : &infomask2_lock_old_tuple);
3919 :
3920 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(infomask_lock_old_tuple));
3921 :
3922 149728 : START_CRIT_SECTION();
3923 :
3924 : /* Clear obsolete visibility flags ... */
3925 149728 : oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
3926 149728 : oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
3927 149728 : HeapTupleClearHotUpdated(&oldtup);
3928 : /* ... and store info about transaction updating this tuple */
3929 : Assert(TransactionIdIsValid(xmax_lock_old_tuple));
3930 149728 : HeapTupleHeaderSetXmax(oldtup.t_data, xmax_lock_old_tuple);
3931 149728 : oldtup.t_data->t_infomask |= infomask_lock_old_tuple;
3932 149728 : oldtup.t_data->t_infomask2 |= infomask2_lock_old_tuple;
3933 149728 : HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
3934 :
3935 : /* temporarily make it look not-updated, but locked */
3936 149728 : oldtup.t_data->t_ctid = oldtup.t_self;
3937 :
3938 : /*
3939 : * Clear all-frozen bit on visibility map if needed. We could
3940 : * immediately reset ALL_VISIBLE, but given that the WAL logging
3941 : * overhead would be unchanged, that doesn't seem necessarily
3942 : * worthwhile.
3943 : */
3944 150657 : if (PageIsAllVisible(page) &&
3945 929 : visibilitymap_clear(relation, block, vmbuffer,
3946 : VISIBILITYMAP_ALL_FROZEN))
3947 776 : cleared_all_frozen = true;
3948 :
3949 149728 : MarkBufferDirty(buffer);
3950 :
3951 149728 : if (RelationNeedsWAL(relation))
3952 : {
3953 : xl_heap_lock xlrec;
3954 : XLogRecPtr recptr;
3955 :
3956 139599 : XLogBeginInsert();
3957 139599 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
3958 :
3959 139599 : xlrec.offnum = ItemPointerGetOffsetNumber(&oldtup.t_self);
3960 139599 : xlrec.xmax = xmax_lock_old_tuple;
3961 279198 : xlrec.infobits_set = compute_infobits(oldtup.t_data->t_infomask,
3962 139599 : oldtup.t_data->t_infomask2);
3963 139599 : xlrec.flags =
3964 139599 : cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
3965 139599 : XLogRegisterData(&xlrec, SizeOfHeapLock);
3966 139599 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
3967 139599 : PageSetLSN(page, recptr);
3968 : }
3969 :
3970 149728 : END_CRIT_SECTION();
3971 :
3972 149728 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3973 :
3974 : /*
3975 : * Let the toaster do its thing, if needed.
3976 : *
3977 : * Note: below this point, heaptup is the data we actually intend to
3978 : * store into the relation; newtup is the caller's original untoasted
3979 : * data.
3980 : */
3981 149728 : if (need_toast)
3982 : {
3983 : /* Note we always use WAL and FSM during updates */
3984 1651 : heaptup = heap_toast_insert_or_update(relation, newtup, &oldtup, 0);
3985 1651 : newtupsize = MAXALIGN(heaptup->t_len);
3986 : }
3987 : else
3988 148077 : heaptup = newtup;
3989 :
3990 : /*
3991 : * Now, do we need a new page for the tuple, or not? This is a bit
3992 : * tricky since someone else could have added tuples to the page while
3993 : * we weren't looking. We have to recheck the available space after
3994 : * reacquiring the buffer lock. But don't bother to do that if the
3995 : * former amount of free space is still not enough; it's unlikely
3996 : * there's more free now than before.
3997 : *
3998 : * What's more, if we need to get a new page, we will need to acquire
3999 : * buffer locks on both old and new pages. To avoid deadlock against
4000 : * some other backend trying to get the same two locks in the other
4001 : * order, we must be consistent about the order we get the locks in.
4002 : * We use the rule "lock the lower-numbered page of the relation
4003 : * first". To implement this, we must do RelationGetBufferForTuple
4004 : * while not holding the lock on the old page, and we must rely on it
4005 : * to get the locks on both pages in the correct order.
4006 : *
4007 : * Another consideration is that we need visibility map page pin(s) if
4008 : * we will have to clear the all-visible flag on either page. If we
4009 : * call RelationGetBufferForTuple, we rely on it to acquire any such
4010 : * pins; but if we don't, we have to handle that here. Hence we need
4011 : * a loop.
4012 : */
4013 : for (;;)
4014 : {
4015 149728 : if (newtupsize > pagefree)
4016 : {
4017 : /* It doesn't fit, must use RelationGetBufferForTuple. */
4018 149159 : newbuf = RelationGetBufferForTuple(relation, heaptup->t_len,
4019 : buffer, 0, NULL,
4020 : &vmbuffer_new, &vmbuffer,
4021 : 0);
4022 : /* We're all done. */
4023 149159 : break;
4024 : }
4025 : /* Acquire VM page pin if needed and we don't have it. */
4026 569 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
4027 0 : visibilitymap_pin(relation, block, &vmbuffer);
4028 : /* Re-acquire the lock on the old tuple's page. */
4029 569 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
4030 : /* Re-check using the up-to-date free space */
4031 569 : pagefree = PageGetHeapFreeSpace(page);
4032 569 : if (newtupsize > pagefree ||
4033 569 : (vmbuffer == InvalidBuffer && PageIsAllVisible(page)))
4034 : {
4035 : /*
4036 : * Rats, it doesn't fit anymore, or somebody just now set the
4037 : * all-visible flag. We must now unlock and loop to avoid
4038 : * deadlock. Fortunately, this path should seldom be taken.
4039 : */
4040 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
4041 : }
4042 : else
4043 : {
4044 : /* We're all done. */
4045 569 : newbuf = buffer;
4046 569 : break;
4047 : }
4048 : }
4049 : }
4050 : else
4051 : {
4052 : /* No TOAST work needed, and it'll fit on same page */
4053 161518 : newbuf = buffer;
4054 161518 : heaptup = newtup;
4055 : }
4056 :
4057 : /*
4058 : * We're about to do the actual update -- check for conflict first, to
4059 : * avoid possibly having to roll back work we've just done.
4060 : *
4061 : * This is safe without a recheck as long as there is no possibility of
4062 : * another process scanning the pages between this check and the update
4063 : * being visible to the scan (i.e., exclusive buffer content lock(s) are
4064 : * continuously held from this point until the tuple update is visible).
4065 : *
4066 : * For the new tuple the only check needed is at the relation level, but
4067 : * since both tuples are in the same relation and the check for oldtup
4068 : * will include checking the relation level, there is no benefit to a
4069 : * separate check for the new tuple.
4070 : */
4071 311246 : CheckForSerializableConflictIn(relation, &oldtup.t_self,
4072 : BufferGetBlockNumber(buffer));
4073 :
4074 : /*
4075 : * At this point newbuf and buffer are both pinned and locked, and newbuf
4076 : * has enough space for the new tuple. If they are the same buffer, only
4077 : * one pin is held.
4078 : */
4079 :
4080 311234 : if (newbuf == buffer)
4081 : {
4082 : /*
4083 : * Since the new tuple is going into the same page, we might be able
4084 : * to do a HOT update. Check if any of the index columns have been
4085 : * changed.
4086 : */
4087 162075 : if (!bms_overlap(modified_attrs, hot_attrs))
4088 : {
4089 149515 : use_hot_update = true;
4090 :
4091 : /*
4092 : * If none of the columns that are used in hot-blocking indexes
4093 : * were updated, we can apply HOT, but we do still need to check
4094 : * if we need to update the summarizing indexes, and update those
4095 : * indexes if the columns were updated, or we may fail to detect
4096 : * e.g. value bound changes in BRIN minmax indexes.
4097 : */
4098 149515 : if (bms_overlap(modified_attrs, sum_attrs))
4099 1641 : summarized_update = true;
4100 : }
4101 : }
4102 : else
4103 : {
4104 : /* Set a hint that the old page could use prune/defrag */
4105 149159 : PageSetFull(page);
4106 : }
4107 :
4108 : /*
4109 : * Compute replica identity tuple before entering the critical section so
4110 : * we don't PANIC upon a memory allocation failure.
4111 : * ExtractReplicaIdentity() will return NULL if nothing needs to be
4112 : * logged. Pass old key required as true only if the replica identity key
4113 : * columns are modified or it has external data.
4114 : */
4115 311234 : old_key_tuple = ExtractReplicaIdentity(relation, &oldtup,
4116 311234 : bms_overlap(modified_attrs, id_attrs) ||
4117 : id_has_external,
4118 311234 : &old_key_copied);
4119 :
4120 : /* NO EREPORT(ERROR) from here till changes are logged */
4121 311234 : START_CRIT_SECTION();
4122 :
4123 : /*
4124 : * If this transaction commits, the old tuple will become DEAD sooner or
4125 : * later. Set flag that this page is a candidate for pruning once our xid
4126 : * falls below the OldestXmin horizon. If the transaction finally aborts,
4127 : * the subsequent page pruning will be a no-op and the hint will be
4128 : * cleared.
4129 : *
4130 : * XXX Should we set hint on newbuf as well? If the transaction aborts,
4131 : * there would be a prunable tuple in the newbuf; but for now we choose
4132 : * not to optimize for aborts. Note that heap_xlog_update must be kept in
4133 : * sync if this decision changes.
4134 : */
4135 311234 : PageSetPrunable(page, xid);
4136 :
4137 311234 : if (use_hot_update)
4138 : {
4139 : /* Mark the old tuple as HOT-updated */
4140 149515 : HeapTupleSetHotUpdated(&oldtup);
4141 : /* And mark the new tuple as heap-only */
4142 149515 : HeapTupleSetHeapOnly(heaptup);
4143 : /* Mark the caller's copy too, in case different from heaptup */
4144 149515 : HeapTupleSetHeapOnly(newtup);
4145 : }
4146 : else
4147 : {
4148 : /* Make sure tuples are correctly marked as not-HOT */
4149 161719 : HeapTupleClearHotUpdated(&oldtup);
4150 161719 : HeapTupleClearHeapOnly(heaptup);
4151 161719 : HeapTupleClearHeapOnly(newtup);
4152 : }
4153 :
4154 311234 : RelationPutHeapTuple(relation, newbuf, heaptup, false); /* insert new tuple */
4155 :
4156 :
4157 : /* Clear obsolete visibility flags, possibly set by ourselves above... */
4158 311234 : oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
4159 311234 : oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
4160 : /* ... and store info about transaction updating this tuple */
4161 : Assert(TransactionIdIsValid(xmax_old_tuple));
4162 311234 : HeapTupleHeaderSetXmax(oldtup.t_data, xmax_old_tuple);
4163 311234 : oldtup.t_data->t_infomask |= infomask_old_tuple;
4164 311234 : oldtup.t_data->t_infomask2 |= infomask2_old_tuple;
4165 311234 : HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
4166 :
4167 : /* record address of new tuple in t_ctid of old one */
4168 311234 : oldtup.t_data->t_ctid = heaptup->t_self;
4169 :
4170 : /* clear PD_ALL_VISIBLE flags, reset all visibilitymap bits */
4171 311234 : if (PageIsAllVisible(BufferGetPage(buffer)))
4172 : {
4173 1746 : all_visible_cleared = true;
4174 1746 : PageClearAllVisible(BufferGetPage(buffer));
4175 1746 : visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
4176 : vmbuffer, VISIBILITYMAP_VALID_BITS);
4177 : }
4178 311234 : if (newbuf != buffer && PageIsAllVisible(BufferGetPage(newbuf)))
4179 : {
4180 712 : all_visible_cleared_new = true;
4181 712 : PageClearAllVisible(BufferGetPage(newbuf));
4182 712 : visibilitymap_clear(relation, BufferGetBlockNumber(newbuf),
4183 : vmbuffer_new, VISIBILITYMAP_VALID_BITS);
4184 : }
4185 :
4186 311234 : if (newbuf != buffer)
4187 149159 : MarkBufferDirty(newbuf);
4188 311234 : MarkBufferDirty(buffer);
4189 :
4190 : /* XLOG stuff */
4191 311234 : if (RelationNeedsWAL(relation))
4192 : {
4193 : XLogRecPtr recptr;
4194 :
4195 : /*
4196 : * For logical decoding we need combo CIDs to properly decode the
4197 : * catalog.
4198 : */
4199 299869 : if (RelationIsAccessibleInLogicalDecoding(relation))
4200 : {
4201 2569 : log_heap_new_cid(relation, &oldtup);
4202 2569 : log_heap_new_cid(relation, heaptup);
4203 : }
4204 :
4205 299869 : recptr = log_heap_update(relation, buffer,
4206 : newbuf, &oldtup, heaptup,
4207 : old_key_tuple,
4208 : all_visible_cleared,
4209 : all_visible_cleared_new);
4210 299869 : if (newbuf != buffer)
4211 : {
4212 139036 : PageSetLSN(BufferGetPage(newbuf), recptr);
4213 : }
4214 299869 : PageSetLSN(BufferGetPage(buffer), recptr);
4215 : }
4216 :
4217 311234 : END_CRIT_SECTION();
4218 :
4219 311234 : if (newbuf != buffer)
4220 149159 : LockBuffer(newbuf, BUFFER_LOCK_UNLOCK);
4221 311234 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
4222 :
4223 : /*
4224 : * Mark old tuple for invalidation from system caches at next command
4225 : * boundary, and mark the new tuple for invalidation in case we abort. We
4226 : * have to do this before releasing the buffer because oldtup is in the
4227 : * buffer. (heaptup is all in local memory, but it's necessary to process
4228 : * both tuple versions in one call to inval.c so we can avoid redundant
4229 : * sinval messages.)
4230 : */
4231 311234 : CacheInvalidateHeapTuple(relation, &oldtup, heaptup);
4232 :
4233 : /* Now we can release the buffer(s) */
4234 311234 : if (newbuf != buffer)
4235 149159 : ReleaseBuffer(newbuf);
4236 311234 : ReleaseBuffer(buffer);
4237 311234 : if (BufferIsValid(vmbuffer_new))
4238 712 : ReleaseBuffer(vmbuffer_new);
4239 311234 : if (BufferIsValid(vmbuffer))
4240 1746 : ReleaseBuffer(vmbuffer);
4241 :
4242 : /*
4243 : * Release the lmgr tuple lock, if we had it.
4244 : */
4245 311234 : if (have_tuple_lock)
4246 22 : UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
4247 :
4248 311234 : pgstat_count_heap_update(relation, use_hot_update, newbuf != buffer);
4249 :
4250 : /*
4251 : * If heaptup is a private copy, release it. Don't forget to copy t_self
4252 : * back to the caller's image, too.
4253 : */
4254 311234 : if (heaptup != newtup)
4255 : {
4256 1603 : newtup->t_self = heaptup->t_self;
4257 1603 : heap_freetuple(heaptup);
4258 : }
4259 :
4260 : /*
4261 : * If it is a HOT update, the update may still need to update summarized
4262 : * indexes, lest we fail to update those summaries and get incorrect
4263 : * results (for example, minmax bounds of the block may change with this
4264 : * update).
4265 : */
4266 311234 : if (use_hot_update)
4267 : {
4268 149515 : if (summarized_update)
4269 1641 : *update_indexes = TU_Summarizing;
4270 : else
4271 147874 : *update_indexes = TU_None;
4272 : }
4273 : else
4274 161719 : *update_indexes = TU_All;
4275 :
4276 311234 : if (old_key_tuple != NULL && old_key_copied)
4277 84 : heap_freetuple(old_key_tuple);
4278 :
4279 311234 : bms_free(hot_attrs);
4280 311234 : bms_free(sum_attrs);
4281 311234 : bms_free(key_attrs);
4282 311234 : bms_free(id_attrs);
4283 311234 : bms_free(modified_attrs);
4284 311234 : bms_free(interesting_attrs);
4285 :
4286 311234 : return TM_Ok;
4287 : }
4288 :
4289 : #ifdef USE_ASSERT_CHECKING
4290 : /*
4291 : * Confirm adequate lock held during heap_update(), per rules from
4292 : * README.tuplock section "Locking to write inplace-updated tables".
4293 : */
4294 : static void
4295 : check_lock_if_inplace_updateable_rel(Relation relation,
4296 : const ItemPointerData *otid,
4297 : HeapTuple newtup)
4298 : {
4299 : /* LOCKTAG_TUPLE acceptable for any catalog */
4300 : switch (RelationGetRelid(relation))
4301 : {
4302 : case RelationRelationId:
4303 : case DatabaseRelationId:
4304 : {
4305 : LOCKTAG tuptag;
4306 :
4307 : SET_LOCKTAG_TUPLE(tuptag,
4308 : relation->rd_lockInfo.lockRelId.dbId,
4309 : relation->rd_lockInfo.lockRelId.relId,
4310 : ItemPointerGetBlockNumber(otid),
4311 : ItemPointerGetOffsetNumber(otid));
4312 : if (LockHeldByMe(&tuptag, InplaceUpdateTupleLock, false))
4313 : return;
4314 : }
4315 : break;
4316 : default:
4317 : Assert(!IsInplaceUpdateRelation(relation));
4318 : return;
4319 : }
4320 :
4321 : switch (RelationGetRelid(relation))
4322 : {
4323 : case RelationRelationId:
4324 : {
4325 : /* LOCKTAG_TUPLE or LOCKTAG_RELATION ok */
4326 : Form_pg_class classForm = (Form_pg_class) GETSTRUCT(newtup);
4327 : Oid relid = classForm->oid;
4328 : Oid dbid;
4329 : LOCKTAG tag;
4330 :
4331 : if (IsSharedRelation(relid))
4332 : dbid = InvalidOid;
4333 : else
4334 : dbid = MyDatabaseId;
4335 :
4336 : if (classForm->relkind == RELKIND_INDEX)
4337 : {
4338 : Relation irel = index_open(relid, AccessShareLock);
4339 :
4340 : SET_LOCKTAG_RELATION(tag, dbid, irel->rd_index->indrelid);
4341 : index_close(irel, AccessShareLock);
4342 : }
4343 : else
4344 : SET_LOCKTAG_RELATION(tag, dbid, relid);
4345 :
4346 : if (!LockHeldByMe(&tag, ShareUpdateExclusiveLock, false) &&
4347 : !LockHeldByMe(&tag, ShareRowExclusiveLock, true))
4348 : elog(WARNING,
4349 : "missing lock for relation \"%s\" (OID %u, relkind %c) @ TID (%u,%u)",
4350 : NameStr(classForm->relname),
4351 : relid,
4352 : classForm->relkind,
4353 : ItemPointerGetBlockNumber(otid),
4354 : ItemPointerGetOffsetNumber(otid));
4355 : }
4356 : break;
4357 : case DatabaseRelationId:
4358 : {
4359 : /* LOCKTAG_TUPLE required */
4360 : Form_pg_database dbForm = (Form_pg_database) GETSTRUCT(newtup);
4361 :
4362 : elog(WARNING,
4363 : "missing lock on database \"%s\" (OID %u) @ TID (%u,%u)",
4364 : NameStr(dbForm->datname),
4365 : dbForm->oid,
4366 : ItemPointerGetBlockNumber(otid),
4367 : ItemPointerGetOffsetNumber(otid));
4368 : }
4369 : break;
4370 : }
4371 : }
4372 :
4373 : /*
4374 : * Confirm adequate relation lock held, per rules from README.tuplock section
4375 : * "Locking to write inplace-updated tables".
4376 : */
4377 : static void
4378 : check_inplace_rel_lock(HeapTuple oldtup)
4379 : {
4380 : Form_pg_class classForm = (Form_pg_class) GETSTRUCT(oldtup);
4381 : Oid relid = classForm->oid;
4382 : Oid dbid;
4383 : LOCKTAG tag;
4384 :
4385 : if (IsSharedRelation(relid))
4386 : dbid = InvalidOid;
4387 : else
4388 : dbid = MyDatabaseId;
4389 :
4390 : if (classForm->relkind == RELKIND_INDEX)
4391 : {
4392 : Relation irel = index_open(relid, AccessShareLock);
4393 :
4394 : SET_LOCKTAG_RELATION(tag, dbid, irel->rd_index->indrelid);
4395 : index_close(irel, AccessShareLock);
4396 : }
4397 : else
4398 : SET_LOCKTAG_RELATION(tag, dbid, relid);
4399 :
4400 : if (!LockHeldByMe(&tag, ShareUpdateExclusiveLock, true))
4401 : elog(WARNING,
4402 : "missing lock for relation \"%s\" (OID %u, relkind %c) @ TID (%u,%u)",
4403 : NameStr(classForm->relname),
4404 : relid,
4405 : classForm->relkind,
4406 : ItemPointerGetBlockNumber(&oldtup->t_self),
4407 : ItemPointerGetOffsetNumber(&oldtup->t_self));
4408 : }
4409 : #endif
4410 :
4411 : /*
4412 : * Check if the specified attribute's values are the same. Subroutine for
4413 : * HeapDetermineColumnsInfo.
4414 : */
4415 : static bool
4416 759123 : heap_attr_equals(TupleDesc tupdesc, int attrnum, Datum value1, Datum value2,
4417 : bool isnull1, bool isnull2)
4418 : {
4419 : /*
4420 : * If one value is NULL and other is not, then they are certainly not
4421 : * equal
4422 : */
4423 759123 : if (isnull1 != isnull2)
4424 45 : return false;
4425 :
4426 : /*
4427 : * If both are NULL, they can be considered equal.
4428 : */
4429 759078 : if (isnull1)
4430 4991 : return true;
4431 :
4432 : /*
4433 : * We do simple binary comparison of the two datums. This may be overly
4434 : * strict because there can be multiple binary representations for the
4435 : * same logical value. But we should be OK as long as there are no false
4436 : * positives. Using a type-specific equality operator is messy because
4437 : * there could be multiple notions of equality in different operator
4438 : * classes; furthermore, we cannot safely invoke user-defined functions
4439 : * while holding exclusive buffer lock.
4440 : */
4441 754087 : if (attrnum <= 0)
4442 : {
4443 : /* The only allowed system columns are OIDs, so do this */
4444 0 : return (DatumGetObjectId(value1) == DatumGetObjectId(value2));
4445 : }
4446 : else
4447 : {
4448 : CompactAttribute *att;
4449 :
4450 : Assert(attrnum <= tupdesc->natts);
4451 754087 : att = TupleDescCompactAttr(tupdesc, attrnum - 1);
4452 754087 : return datumIsEqual(value1, value2, att->attbyval, att->attlen);
4453 : }
4454 : }
4455 :
4456 : /*
4457 : * Check which columns are being updated.
4458 : *
4459 : * Given an updated tuple, determine (and return into the output bitmapset),
4460 : * from those listed as interesting, the set of columns that changed.
4461 : *
4462 : * has_external indicates if any of the unmodified attributes (from those
4463 : * listed as interesting) of the old tuple is a member of external_cols and is
4464 : * stored externally.
4465 : */
4466 : static Bitmapset *
4467 311408 : HeapDetermineColumnsInfo(Relation relation,
4468 : Bitmapset *interesting_cols,
4469 : Bitmapset *external_cols,
4470 : HeapTuple oldtup, HeapTuple newtup,
4471 : bool *has_external)
4472 : {
4473 : int attidx;
4474 311408 : Bitmapset *modified = NULL;
4475 311408 : TupleDesc tupdesc = RelationGetDescr(relation);
4476 :
4477 311408 : attidx = -1;
4478 1070531 : while ((attidx = bms_next_member(interesting_cols, attidx)) >= 0)
4479 : {
4480 : /* attidx is zero-based, attrnum is the normal attribute number */
4481 759123 : AttrNumber attrnum = attidx + FirstLowInvalidHeapAttributeNumber;
4482 : Datum value1,
4483 : value2;
4484 : bool isnull1,
4485 : isnull2;
4486 :
4487 : /*
4488 : * If it's a whole-tuple reference, say "not equal". It's not really
4489 : * worth supporting this case, since it could only succeed after a
4490 : * no-op update, which is hardly a case worth optimizing for.
4491 : */
4492 759123 : if (attrnum == 0)
4493 : {
4494 0 : modified = bms_add_member(modified, attidx);
4495 729113 : continue;
4496 : }
4497 :
4498 : /*
4499 : * Likewise, automatically say "not equal" for any system attribute
4500 : * other than tableOID; we cannot expect these to be consistent in a
4501 : * HOT chain, or even to be set correctly yet in the new tuple.
4502 : */
4503 759123 : if (attrnum < 0)
4504 : {
4505 0 : if (attrnum != TableOidAttributeNumber)
4506 : {
4507 0 : modified = bms_add_member(modified, attidx);
4508 0 : continue;
4509 : }
4510 : }
4511 :
4512 : /*
4513 : * Extract the corresponding values. XXX this is pretty inefficient
4514 : * if there are many indexed columns. Should we do a single
4515 : * heap_deform_tuple call on each tuple, instead? But that doesn't
4516 : * work for system columns ...
4517 : */
4518 759123 : value1 = heap_getattr(oldtup, attrnum, tupdesc, &isnull1);
4519 759123 : value2 = heap_getattr(newtup, attrnum, tupdesc, &isnull2);
4520 :
4521 759123 : if (!heap_attr_equals(tupdesc, attrnum, value1,
4522 : value2, isnull1, isnull2))
4523 : {
4524 27331 : modified = bms_add_member(modified, attidx);
4525 27331 : continue;
4526 : }
4527 :
4528 : /*
4529 : * No need to check attributes that can't be stored externally. Note
4530 : * that system attributes can't be stored externally.
4531 : */
4532 731792 : if (attrnum < 0 || isnull1 ||
4533 726801 : TupleDescCompactAttr(tupdesc, attrnum - 1)->attlen != -1)
4534 701782 : continue;
4535 :
4536 : /*
4537 : * Check if the old tuple's attribute is stored externally and is a
4538 : * member of external_cols.
4539 : */
4540 30015 : if (VARATT_IS_EXTERNAL((varlena *) DatumGetPointer(value1)) &&
4541 5 : bms_is_member(attidx, external_cols))
4542 2 : *has_external = true;
4543 : }
4544 :
4545 311408 : return modified;
4546 : }
4547 :
4548 : /*
4549 : * simple_heap_update - replace a tuple
4550 : *
4551 : * This routine may be used to update a tuple when concurrent updates of
4552 : * the target tuple are not expected (for example, because we have a lock
4553 : * on the relation associated with the tuple). Any failure is reported
4554 : * via ereport().
4555 : */
4556 : void
4557 115649 : simple_heap_update(Relation relation, const ItemPointerData *otid, HeapTuple tup,
4558 : TU_UpdateIndexes *update_indexes)
4559 : {
4560 : TM_Result result;
4561 : TM_FailureData tmfd;
4562 : LockTupleMode lockmode;
4563 :
4564 115649 : result = heap_update(relation, otid, tup,
4565 : GetCurrentCommandId(true), InvalidSnapshot,
4566 : true /* wait for commit */ ,
4567 : &tmfd, &lockmode, update_indexes);
4568 115649 : switch (result)
4569 : {
4570 0 : case TM_SelfModified:
4571 : /* Tuple was already updated in current command? */
4572 0 : elog(ERROR, "tuple already updated by self");
4573 : break;
4574 :
4575 115648 : case TM_Ok:
4576 : /* done successfully */
4577 115648 : break;
4578 :
4579 0 : case TM_Updated:
4580 0 : elog(ERROR, "tuple concurrently updated");
4581 : break;
4582 :
4583 1 : case TM_Deleted:
4584 1 : elog(ERROR, "tuple concurrently deleted");
4585 : break;
4586 :
4587 0 : default:
4588 0 : elog(ERROR, "unrecognized heap_update status: %u", result);
4589 : break;
4590 : }
4591 115648 : }
4592 :
4593 :
4594 : /*
4595 : * Return the MultiXactStatus corresponding to the given tuple lock mode.
4596 : */
4597 : static MultiXactStatus
4598 115444 : get_mxact_status_for_lock(LockTupleMode mode, bool is_update)
4599 : {
4600 : int retval;
4601 :
4602 115444 : if (is_update)
4603 215 : retval = tupleLockExtraInfo[mode].updstatus;
4604 : else
4605 115229 : retval = tupleLockExtraInfo[mode].lockstatus;
4606 :
4607 115444 : if (retval == -1)
4608 0 : elog(ERROR, "invalid lock tuple mode %d/%s", mode,
4609 : is_update ? "true" : "false");
4610 :
4611 115444 : return (MultiXactStatus) retval;
4612 : }
4613 :
4614 : /*
4615 : * heap_lock_tuple - lock a tuple in shared or exclusive mode
4616 : *
4617 : * Note that this acquires a buffer pin, which the caller must release.
4618 : *
4619 : * Input parameters:
4620 : * relation: relation containing tuple (caller must hold suitable lock)
4621 : * cid: current command ID (used for visibility test, and stored into
4622 : * tuple's cmax if lock is successful)
4623 : * mode: indicates if shared or exclusive tuple lock is desired
4624 : * wait_policy: what to do if tuple lock is not available
4625 : * follow_updates: if true, follow the update chain to also lock descendant
4626 : * tuples.
4627 : *
4628 : * Output parameters:
4629 : * *tuple: all fields filled in
4630 : * *buffer: set to buffer holding tuple (pinned but not locked at exit)
4631 : * *tmfd: filled in failure cases (see below)
4632 : *
4633 : * Function results are the same as the ones for table_tuple_lock().
4634 : *
4635 : * In the failure cases other than TM_Invisible, the routine fills
4636 : * *tmfd with the tuple's t_ctid, t_xmax (resolving a possible MultiXact,
4637 : * if necessary), and t_cmax (the last only for TM_SelfModified,
4638 : * since we cannot obtain cmax from a combo CID generated by another
4639 : * transaction).
4640 : * See comments for struct TM_FailureData for additional info.
4641 : *
4642 : * See README.tuplock for a thorough explanation of this mechanism.
4643 : */
4644 : TM_Result
4645 158484 : heap_lock_tuple(Relation relation, HeapTuple tuple,
4646 : CommandId cid, LockTupleMode mode, LockWaitPolicy wait_policy,
4647 : bool follow_updates,
4648 : Buffer *buffer, TM_FailureData *tmfd)
4649 : {
4650 : TM_Result result;
4651 158484 : ItemPointer tid = &(tuple->t_self);
4652 : ItemId lp;
4653 : Page page;
4654 158484 : Buffer vmbuffer = InvalidBuffer;
4655 : BlockNumber block;
4656 : TransactionId xid,
4657 : xmax;
4658 : uint16 old_infomask,
4659 : new_infomask,
4660 : new_infomask2;
4661 158484 : bool first_time = true;
4662 158484 : bool skip_tuple_lock = false;
4663 158484 : bool have_tuple_lock = false;
4664 158484 : bool cleared_all_frozen = false;
4665 :
4666 158484 : *buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
4667 158484 : block = ItemPointerGetBlockNumber(tid);
4668 :
4669 : /*
4670 : * Before locking the buffer, pin the visibility map page if it appears to
4671 : * be necessary. Since we haven't got the lock yet, someone else might be
4672 : * in the middle of changing this, so we'll need to recheck after we have
4673 : * the lock.
4674 : */
4675 158484 : if (PageIsAllVisible(BufferGetPage(*buffer)))
4676 1666 : visibilitymap_pin(relation, block, &vmbuffer);
4677 :
4678 158484 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4679 :
4680 158484 : page = BufferGetPage(*buffer);
4681 158484 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
4682 : Assert(ItemIdIsNormal(lp));
4683 :
4684 158484 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
4685 158484 : tuple->t_len = ItemIdGetLength(lp);
4686 158484 : tuple->t_tableOid = RelationGetRelid(relation);
4687 :
4688 18 : l3:
4689 158502 : result = HeapTupleSatisfiesUpdate(tuple, cid, *buffer);
4690 :
4691 158502 : if (result == TM_Invisible)
4692 : {
4693 : /*
4694 : * This is possible, but only when locking a tuple for ON CONFLICT DO
4695 : * SELECT/UPDATE. We return this value here rather than throwing an
4696 : * error in order to give that case the opportunity to throw a more
4697 : * specific error.
4698 : */
4699 21 : result = TM_Invisible;
4700 21 : goto out_locked;
4701 : }
4702 158481 : else if (result == TM_BeingModified ||
4703 77268 : result == TM_Updated ||
4704 : result == TM_Deleted)
4705 : {
4706 : TransactionId xwait;
4707 : uint16 infomask;
4708 : uint16 infomask2;
4709 : bool require_sleep;
4710 : ItemPointerData t_ctid;
4711 :
4712 : /* must copy state data before unlocking buffer */
4713 81215 : xwait = HeapTupleHeaderGetRawXmax(tuple->t_data);
4714 81215 : infomask = tuple->t_data->t_infomask;
4715 81215 : infomask2 = tuple->t_data->t_infomask2;
4716 81215 : ItemPointerCopy(&tuple->t_data->t_ctid, &t_ctid);
4717 :
4718 81215 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
4719 :
4720 : /*
4721 : * If any subtransaction of the current top transaction already holds
4722 : * a lock as strong as or stronger than what we're requesting, we
4723 : * effectively hold the desired lock already. We *must* succeed
4724 : * without trying to take the tuple lock, else we will deadlock
4725 : * against anyone wanting to acquire a stronger lock.
4726 : *
4727 : * Note we only do this the first time we loop on the HTSU result;
4728 : * there is no point in testing in subsequent passes, because
4729 : * evidently our own transaction cannot have acquired a new lock after
4730 : * the first time we checked.
4731 : */
4732 81215 : if (first_time)
4733 : {
4734 81203 : first_time = false;
4735 :
4736 81203 : if (infomask & HEAP_XMAX_IS_MULTI)
4737 : {
4738 : int i;
4739 : int nmembers;
4740 : MultiXactMember *members;
4741 :
4742 : /*
4743 : * We don't need to allow old multixacts here; if that had
4744 : * been the case, HeapTupleSatisfiesUpdate would have returned
4745 : * MayBeUpdated and we wouldn't be here.
4746 : */
4747 : nmembers =
4748 73307 : GetMultiXactIdMembers(xwait, &members, false,
4749 73307 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
4750 :
4751 1422696 : for (i = 0; i < nmembers; i++)
4752 : {
4753 : /* only consider members of our own transaction */
4754 1349403 : if (!TransactionIdIsCurrentTransactionId(members[i].xid))
4755 1349354 : continue;
4756 :
4757 49 : if (TUPLOCK_from_mxstatus(members[i].status) >= mode)
4758 : {
4759 14 : pfree(members);
4760 14 : result = TM_Ok;
4761 14 : goto out_unlocked;
4762 : }
4763 : else
4764 : {
4765 : /*
4766 : * Disable acquisition of the heavyweight tuple lock.
4767 : * Otherwise, when promoting a weaker lock, we might
4768 : * deadlock with another locker that has acquired the
4769 : * heavyweight tuple lock and is waiting for our
4770 : * transaction to finish.
4771 : *
4772 : * Note that in this case we still need to wait for
4773 : * the multixact if required, to avoid acquiring
4774 : * conflicting locks.
4775 : */
4776 35 : skip_tuple_lock = true;
4777 : }
4778 : }
4779 :
4780 73293 : if (members)
4781 73293 : pfree(members);
4782 : }
4783 7896 : else if (TransactionIdIsCurrentTransactionId(xwait))
4784 : {
4785 6577 : switch (mode)
4786 : {
4787 175 : case LockTupleKeyShare:
4788 : Assert(HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) ||
4789 : HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4790 : HEAP_XMAX_IS_EXCL_LOCKED(infomask));
4791 175 : result = TM_Ok;
4792 175 : goto out_unlocked;
4793 25 : case LockTupleShare:
4794 31 : if (HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4795 6 : HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4796 : {
4797 19 : result = TM_Ok;
4798 19 : goto out_unlocked;
4799 : }
4800 6 : break;
4801 72 : case LockTupleNoKeyExclusive:
4802 72 : if (HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4803 : {
4804 60 : result = TM_Ok;
4805 60 : goto out_unlocked;
4806 : }
4807 12 : break;
4808 6305 : case LockTupleExclusive:
4809 6305 : if (HEAP_XMAX_IS_EXCL_LOCKED(infomask) &&
4810 1265 : infomask2 & HEAP_KEYS_UPDATED)
4811 : {
4812 1244 : result = TM_Ok;
4813 1244 : goto out_unlocked;
4814 : }
4815 5061 : break;
4816 : }
4817 : }
4818 : }
4819 :
4820 : /*
4821 : * Initially assume that we will have to wait for the locking
4822 : * transaction(s) to finish. We check various cases below in which
4823 : * this can be turned off.
4824 : */
4825 79703 : require_sleep = true;
4826 79703 : if (mode == LockTupleKeyShare)
4827 : {
4828 : /*
4829 : * If we're requesting KeyShare, and there's no update present, we
4830 : * don't need to wait. Even if there is an update, we can still
4831 : * continue if the key hasn't been modified.
4832 : *
4833 : * However, if there are updates, we need to walk the update chain
4834 : * to mark future versions of the row as locked, too. That way,
4835 : * if somebody deletes that future version, we're protected
4836 : * against the key going away. This locking of future versions
4837 : * could block momentarily, if a concurrent transaction is
4838 : * deleting a key; or it could return a value to the effect that
4839 : * the transaction deleting the key has already committed. So we
4840 : * do this before re-locking the buffer; otherwise this would be
4841 : * prone to deadlocks.
4842 : *
4843 : * Note that the TID we're locking was grabbed before we unlocked
4844 : * the buffer. For it to change while we're not looking, the
4845 : * other properties we're testing for below after re-locking the
4846 : * buffer would also change, in which case we would restart this
4847 : * loop above.
4848 : */
4849 73865 : if (!(infomask2 & HEAP_KEYS_UPDATED))
4850 : {
4851 : bool updated;
4852 :
4853 73822 : updated = !HEAP_XMAX_IS_LOCKED_ONLY(infomask);
4854 :
4855 : /*
4856 : * If there are updates, follow the update chain; bail out if
4857 : * that cannot be done.
4858 : */
4859 73822 : if (follow_updates && updated &&
4860 2169 : !ItemPointerEquals(&tuple->t_self, &t_ctid))
4861 : {
4862 : TM_Result res;
4863 :
4864 2169 : res = heap_lock_updated_tuple(relation,
4865 : infomask, xwait, &t_ctid,
4866 : GetCurrentTransactionId(),
4867 : mode);
4868 2169 : if (res != TM_Ok)
4869 : {
4870 6 : result = res;
4871 : /* recovery code expects to have buffer lock held */
4872 6 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4873 199 : goto failed;
4874 : }
4875 : }
4876 :
4877 73816 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4878 :
4879 : /*
4880 : * Make sure it's still an appropriate lock, else start over.
4881 : * Also, if it wasn't updated before we released the lock, but
4882 : * is updated now, we start over too; the reason is that we
4883 : * now need to follow the update chain to lock the new
4884 : * versions.
4885 : */
4886 73816 : if (!HeapTupleHeaderIsOnlyLocked(tuple->t_data) &&
4887 2151 : ((tuple->t_data->t_infomask2 & HEAP_KEYS_UPDATED) ||
4888 2151 : !updated))
4889 18 : goto l3;
4890 :
4891 : /* Things look okay, so we can skip sleeping */
4892 73816 : require_sleep = false;
4893 :
4894 : /*
4895 : * Note we allow Xmax to change here; other updaters/lockers
4896 : * could have modified it before we grabbed the buffer lock.
4897 : * However, this is not a problem, because with the recheck we
4898 : * just did we ensure that they still don't conflict with the
4899 : * lock we want.
4900 : */
4901 : }
4902 : }
4903 5838 : else if (mode == LockTupleShare)
4904 : {
4905 : /*
4906 : * If we're requesting Share, we can similarly avoid sleeping if
4907 : * there's no update and no exclusive lock present.
4908 : */
4909 443 : if (HEAP_XMAX_IS_LOCKED_ONLY(infomask) &&
4910 443 : !HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4911 : {
4912 437 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4913 :
4914 : /*
4915 : * Make sure it's still an appropriate lock, else start over.
4916 : * See above about allowing xmax to change.
4917 : */
4918 874 : if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
4919 437 : HEAP_XMAX_IS_EXCL_LOCKED(tuple->t_data->t_infomask))
4920 0 : goto l3;
4921 437 : require_sleep = false;
4922 : }
4923 : }
4924 5395 : else if (mode == LockTupleNoKeyExclusive)
4925 : {
4926 : /*
4927 : * If we're requesting NoKeyExclusive, we might also be able to
4928 : * avoid sleeping; just ensure that there no conflicting lock
4929 : * already acquired.
4930 : */
4931 172 : if (infomask & HEAP_XMAX_IS_MULTI)
4932 : {
4933 26 : if (!DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
4934 : mode, NULL))
4935 : {
4936 : /*
4937 : * No conflict, but if the xmax changed under us in the
4938 : * meantime, start over.
4939 : */
4940 13 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4941 26 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4942 13 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4943 : xwait))
4944 0 : goto l3;
4945 :
4946 : /* otherwise, we're good */
4947 13 : require_sleep = false;
4948 : }
4949 : }
4950 146 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask))
4951 : {
4952 18 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4953 :
4954 : /* if the xmax changed in the meantime, start over */
4955 36 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4956 18 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4957 : xwait))
4958 0 : goto l3;
4959 : /* otherwise, we're good */
4960 18 : require_sleep = false;
4961 : }
4962 : }
4963 :
4964 : /*
4965 : * As a check independent from those above, we can also avoid sleeping
4966 : * if the current transaction is the sole locker of the tuple. Note
4967 : * that the strength of the lock already held is irrelevant; this is
4968 : * not about recording the lock in Xmax (which will be done regardless
4969 : * of this optimization, below). Also, note that the cases where we
4970 : * hold a lock stronger than we are requesting are already handled
4971 : * above by not doing anything.
4972 : *
4973 : * Note we only deal with the non-multixact case here; MultiXactIdWait
4974 : * is well equipped to deal with this situation on its own.
4975 : */
4976 85069 : if (require_sleep && !(infomask & HEAP_XMAX_IS_MULTI) &&
4977 5372 : TransactionIdIsCurrentTransactionId(xwait))
4978 : {
4979 : /* ... but if the xmax changed in the meantime, start over */
4980 5061 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4981 10122 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4982 5061 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4983 : xwait))
4984 0 : goto l3;
4985 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask));
4986 5061 : require_sleep = false;
4987 : }
4988 :
4989 : /*
4990 : * Time to sleep on the other transaction/multixact, if necessary.
4991 : *
4992 : * If the other transaction is an update/delete that's already
4993 : * committed, then sleeping cannot possibly do any good: if we're
4994 : * required to sleep, get out to raise an error instead.
4995 : *
4996 : * By here, we either have already acquired the buffer exclusive lock,
4997 : * or we must wait for the locking transaction or multixact; so below
4998 : * we ensure that we grab buffer lock after the sleep.
4999 : */
5000 79697 : if (require_sleep && (result == TM_Updated || result == TM_Deleted))
5001 : {
5002 155 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5003 155 : goto failed;
5004 : }
5005 79542 : else if (require_sleep)
5006 : {
5007 : /*
5008 : * Acquire tuple lock to establish our priority for the tuple, or
5009 : * die trying. LockTuple will release us when we are next-in-line
5010 : * for the tuple. We must do this even if we are share-locking,
5011 : * but not if we already have a weaker lock on the tuple.
5012 : *
5013 : * If we are forced to "start over" below, we keep the tuple lock;
5014 : * this arranges that we stay at the head of the line while
5015 : * rechecking tuple state.
5016 : */
5017 197 : if (!skip_tuple_lock &&
5018 181 : !heap_acquire_tuplock(relation, tid, mode, wait_policy,
5019 : &have_tuple_lock))
5020 : {
5021 : /*
5022 : * This can only happen if wait_policy is Skip and the lock
5023 : * couldn't be obtained.
5024 : */
5025 1 : result = TM_WouldBlock;
5026 : /* recovery code expects to have buffer lock held */
5027 1 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5028 1 : goto failed;
5029 : }
5030 :
5031 195 : if (infomask & HEAP_XMAX_IS_MULTI)
5032 : {
5033 40 : MultiXactStatus status = get_mxact_status_for_lock(mode, false);
5034 :
5035 : /* We only ever lock tuples, never update them */
5036 40 : if (status >= MultiXactStatusNoKeyUpdate)
5037 0 : elog(ERROR, "invalid lock mode in heap_lock_tuple");
5038 :
5039 : /* wait for multixact to end, or die trying */
5040 40 : switch (wait_policy)
5041 : {
5042 36 : case LockWaitBlock:
5043 36 : MultiXactIdWait((MultiXactId) xwait, status, infomask,
5044 36 : relation, &tuple->t_self, XLTW_Lock, NULL);
5045 36 : break;
5046 2 : case LockWaitSkip:
5047 2 : if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
5048 : status, infomask, relation,
5049 : NULL, false))
5050 : {
5051 2 : result = TM_WouldBlock;
5052 : /* recovery code expects to have buffer lock held */
5053 2 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5054 2 : goto failed;
5055 : }
5056 0 : break;
5057 2 : case LockWaitError:
5058 2 : if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
5059 : status, infomask, relation,
5060 : NULL, log_lock_failures))
5061 2 : ereport(ERROR,
5062 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5063 : errmsg("could not obtain lock on row in relation \"%s\"",
5064 : RelationGetRelationName(relation))));
5065 :
5066 0 : break;
5067 : }
5068 :
5069 : /*
5070 : * Of course, the multixact might not be done here: if we're
5071 : * requesting a light lock mode, other transactions with light
5072 : * locks could still be alive, as well as locks owned by our
5073 : * own xact or other subxacts of this backend. We need to
5074 : * preserve the surviving MultiXact members. Note that it
5075 : * isn't absolutely necessary in the latter case, but doing so
5076 : * is simpler.
5077 : */
5078 : }
5079 : else
5080 : {
5081 : /* wait for regular transaction to end, or die trying */
5082 155 : switch (wait_policy)
5083 : {
5084 116 : case LockWaitBlock:
5085 116 : XactLockTableWait(xwait, relation, &tuple->t_self,
5086 : XLTW_Lock);
5087 116 : break;
5088 33 : case LockWaitSkip:
5089 33 : if (!ConditionalXactLockTableWait(xwait, false))
5090 : {
5091 33 : result = TM_WouldBlock;
5092 : /* recovery code expects to have buffer lock held */
5093 33 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5094 33 : goto failed;
5095 : }
5096 0 : break;
5097 6 : case LockWaitError:
5098 6 : if (!ConditionalXactLockTableWait(xwait, log_lock_failures))
5099 6 : ereport(ERROR,
5100 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5101 : errmsg("could not obtain lock on row in relation \"%s\"",
5102 : RelationGetRelationName(relation))));
5103 0 : break;
5104 : }
5105 : }
5106 :
5107 : /* if there are updates, follow the update chain */
5108 152 : if (follow_updates && !HEAP_XMAX_IS_LOCKED_ONLY(infomask) &&
5109 58 : !ItemPointerEquals(&tuple->t_self, &t_ctid))
5110 : {
5111 : TM_Result res;
5112 :
5113 45 : res = heap_lock_updated_tuple(relation,
5114 : infomask, xwait, &t_ctid,
5115 : GetCurrentTransactionId(),
5116 : mode);
5117 45 : if (res != TM_Ok)
5118 : {
5119 2 : result = res;
5120 : /* recovery code expects to have buffer lock held */
5121 2 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5122 2 : goto failed;
5123 : }
5124 : }
5125 :
5126 150 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5127 :
5128 : /*
5129 : * xwait is done, but if xwait had just locked the tuple then some
5130 : * other xact could update this tuple before we get to this point.
5131 : * Check for xmax change, and start over if so.
5132 : */
5133 284 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
5134 134 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
5135 : xwait))
5136 18 : goto l3;
5137 :
5138 132 : if (!(infomask & HEAP_XMAX_IS_MULTI))
5139 : {
5140 : /*
5141 : * Otherwise check if it committed or aborted. Note we cannot
5142 : * be here if the tuple was only locked by somebody who didn't
5143 : * conflict with us; that would have been handled above. So
5144 : * that transaction must necessarily be gone by now. But
5145 : * don't check for this in the multixact case, because some
5146 : * locker transactions might still be running.
5147 : */
5148 101 : UpdateXmaxHintBits(tuple->t_data, *buffer, xwait);
5149 : }
5150 : }
5151 :
5152 : /* By here, we're certain that we hold buffer exclusive lock again */
5153 :
5154 : /*
5155 : * We may lock if previous xmax aborted, or if it committed but only
5156 : * locked the tuple without updating it; or if we didn't have to wait
5157 : * at all for whatever reason.
5158 : */
5159 79477 : if (!require_sleep ||
5160 228 : (tuple->t_data->t_infomask & HEAP_XMAX_INVALID) ||
5161 177 : HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
5162 81 : HeapTupleHeaderIsOnlyLocked(tuple->t_data))
5163 79402 : result = TM_Ok;
5164 75 : else if (!ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid))
5165 57 : result = TM_Updated;
5166 : else
5167 18 : result = TM_Deleted;
5168 : }
5169 :
5170 77266 : failed:
5171 156942 : if (result != TM_Ok)
5172 : {
5173 : Assert(result == TM_SelfModified || result == TM_Updated ||
5174 : result == TM_Deleted || result == TM_WouldBlock);
5175 :
5176 : /*
5177 : * When locking a tuple under LockWaitSkip semantics and we fail with
5178 : * TM_WouldBlock above, it's possible for concurrent transactions to
5179 : * release the lock and set HEAP_XMAX_INVALID in the meantime. So
5180 : * this assert is slightly different from the equivalent one in
5181 : * heap_delete and heap_update.
5182 : */
5183 : Assert((result == TM_WouldBlock) ||
5184 : !(tuple->t_data->t_infomask & HEAP_XMAX_INVALID));
5185 : Assert(result != TM_Updated ||
5186 : !ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid));
5187 280 : tmfd->ctid = tuple->t_data->t_ctid;
5188 280 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(tuple->t_data);
5189 280 : if (result == TM_SelfModified)
5190 6 : tmfd->cmax = HeapTupleHeaderGetCmax(tuple->t_data);
5191 : else
5192 274 : tmfd->cmax = InvalidCommandId;
5193 280 : goto out_locked;
5194 : }
5195 :
5196 : /*
5197 : * If we didn't pin the visibility map page and the page has become all
5198 : * visible while we were busy locking the buffer, or during some
5199 : * subsequent window during which we had it unlocked, we'll have to unlock
5200 : * and re-lock, to avoid holding the buffer lock across I/O. That's a bit
5201 : * unfortunate, especially since we'll now have to recheck whether the
5202 : * tuple has been locked or updated under us, but hopefully it won't
5203 : * happen very often.
5204 : */
5205 156662 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
5206 : {
5207 0 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
5208 0 : visibilitymap_pin(relation, block, &vmbuffer);
5209 0 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5210 0 : goto l3;
5211 : }
5212 :
5213 156662 : xmax = HeapTupleHeaderGetRawXmax(tuple->t_data);
5214 156662 : old_infomask = tuple->t_data->t_infomask;
5215 :
5216 : /*
5217 : * If this is the first possibly-multixact-able operation in the current
5218 : * transaction, set my per-backend OldestMemberMXactId setting. We can be
5219 : * certain that the transaction will never become a member of any older
5220 : * MultiXactIds than that. (We have to do this even if we end up just
5221 : * using our own TransactionId below, since some other backend could
5222 : * incorporate our XID into a MultiXact immediately afterwards.)
5223 : */
5224 156662 : MultiXactIdSetOldestMember();
5225 :
5226 : /*
5227 : * Compute the new xmax and infomask to store into the tuple. Note we do
5228 : * not modify the tuple just yet, because that would leave it in the wrong
5229 : * state if multixact.c elogs.
5230 : */
5231 156662 : compute_new_xmax_infomask(xmax, old_infomask, tuple->t_data->t_infomask2,
5232 : GetCurrentTransactionId(), mode, false,
5233 : &xid, &new_infomask, &new_infomask2);
5234 :
5235 156662 : START_CRIT_SECTION();
5236 :
5237 : /*
5238 : * Store transaction information of xact locking the tuple.
5239 : *
5240 : * Note: Cmax is meaningless in this context, so don't set it; this avoids
5241 : * possibly generating a useless combo CID. Moreover, if we're locking a
5242 : * previously updated tuple, it's important to preserve the Cmax.
5243 : *
5244 : * Also reset the HOT UPDATE bit, but only if there's no update; otherwise
5245 : * we would break the HOT chain.
5246 : */
5247 156662 : tuple->t_data->t_infomask &= ~HEAP_XMAX_BITS;
5248 156662 : tuple->t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
5249 156662 : tuple->t_data->t_infomask |= new_infomask;
5250 156662 : tuple->t_data->t_infomask2 |= new_infomask2;
5251 156662 : if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
5252 154515 : HeapTupleHeaderClearHotUpdated(tuple->t_data);
5253 156662 : HeapTupleHeaderSetXmax(tuple->t_data, xid);
5254 :
5255 : /*
5256 : * Make sure there is no forward chain link in t_ctid. Note that in the
5257 : * cases where the tuple has been updated, we must not overwrite t_ctid,
5258 : * because it was set by the updater. Moreover, if the tuple has been
5259 : * updated, we need to follow the update chain to lock the new versions of
5260 : * the tuple as well.
5261 : */
5262 156662 : if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
5263 154515 : tuple->t_data->t_ctid = *tid;
5264 :
5265 : /* Clear only the all-frozen bit on visibility map if needed */
5266 158328 : if (PageIsAllVisible(page) &&
5267 1666 : visibilitymap_clear(relation, block, vmbuffer,
5268 : VISIBILITYMAP_ALL_FROZEN))
5269 14 : cleared_all_frozen = true;
5270 :
5271 :
5272 156662 : MarkBufferDirty(*buffer);
5273 :
5274 : /*
5275 : * XLOG stuff. You might think that we don't need an XLOG record because
5276 : * there is no state change worth restoring after a crash. You would be
5277 : * wrong however: we have just written either a TransactionId or a
5278 : * MultiXactId that may never have been seen on disk before, and we need
5279 : * to make sure that there are XLOG entries covering those ID numbers.
5280 : * Else the same IDs might be re-used after a crash, which would be
5281 : * disastrous if this page made it to disk before the crash. Essentially
5282 : * we have to enforce the WAL log-before-data rule even in this case.
5283 : * (Also, in a PITR log-shipping or 2PC environment, we have to have XLOG
5284 : * entries for everything anyway.)
5285 : */
5286 156662 : if (RelationNeedsWAL(relation))
5287 : {
5288 : xl_heap_lock xlrec;
5289 : XLogRecPtr recptr;
5290 :
5291 156310 : XLogBeginInsert();
5292 156310 : XLogRegisterBuffer(0, *buffer, REGBUF_STANDARD);
5293 :
5294 156310 : xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
5295 156310 : xlrec.xmax = xid;
5296 312620 : xlrec.infobits_set = compute_infobits(new_infomask,
5297 156310 : tuple->t_data->t_infomask2);
5298 156310 : xlrec.flags = cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
5299 156310 : XLogRegisterData(&xlrec, SizeOfHeapLock);
5300 :
5301 : /* we don't decode row locks atm, so no need to log the origin */
5302 :
5303 156310 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
5304 :
5305 156310 : PageSetLSN(page, recptr);
5306 : }
5307 :
5308 156662 : END_CRIT_SECTION();
5309 :
5310 156662 : result = TM_Ok;
5311 :
5312 156963 : out_locked:
5313 156963 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
5314 :
5315 158475 : out_unlocked:
5316 158475 : if (BufferIsValid(vmbuffer))
5317 1666 : ReleaseBuffer(vmbuffer);
5318 :
5319 : /*
5320 : * Don't update the visibility map here. Locking a tuple doesn't change
5321 : * visibility info.
5322 : */
5323 :
5324 : /*
5325 : * Now that we have successfully marked the tuple as locked, we can
5326 : * release the lmgr tuple lock, if we had it.
5327 : */
5328 158475 : if (have_tuple_lock)
5329 166 : UnlockTupleTuplock(relation, tid, mode);
5330 :
5331 158475 : return result;
5332 : }
5333 :
5334 : /*
5335 : * Acquire heavyweight lock on the given tuple, in preparation for acquiring
5336 : * its normal, Xmax-based tuple lock.
5337 : *
5338 : * have_tuple_lock is an input and output parameter: on input, it indicates
5339 : * whether the lock has previously been acquired (and this function does
5340 : * nothing in that case). If this function returns success, have_tuple_lock
5341 : * has been flipped to true.
5342 : *
5343 : * Returns false if it was unable to obtain the lock; this can only happen if
5344 : * wait_policy is Skip.
5345 : */
5346 : static bool
5347 318 : heap_acquire_tuplock(Relation relation, const ItemPointerData *tid, LockTupleMode mode,
5348 : LockWaitPolicy wait_policy, bool *have_tuple_lock)
5349 : {
5350 318 : if (*have_tuple_lock)
5351 9 : return true;
5352 :
5353 309 : switch (wait_policy)
5354 : {
5355 268 : case LockWaitBlock:
5356 268 : LockTupleTuplock(relation, tid, mode);
5357 268 : break;
5358 :
5359 34 : case LockWaitSkip:
5360 34 : if (!ConditionalLockTupleTuplock(relation, tid, mode, false))
5361 1 : return false;
5362 33 : break;
5363 :
5364 7 : case LockWaitError:
5365 7 : if (!ConditionalLockTupleTuplock(relation, tid, mode, log_lock_failures))
5366 1 : ereport(ERROR,
5367 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5368 : errmsg("could not obtain lock on row in relation \"%s\"",
5369 : RelationGetRelationName(relation))));
5370 6 : break;
5371 : }
5372 307 : *have_tuple_lock = true;
5373 :
5374 307 : return true;
5375 : }
5376 :
5377 : /*
5378 : * Given an original set of Xmax and infomask, and a transaction (identified by
5379 : * add_to_xmax) acquiring a new lock of some mode, compute the new Xmax and
5380 : * corresponding infomasks to use on the tuple.
5381 : *
5382 : * Note that this might have side effects such as creating a new MultiXactId.
5383 : *
5384 : * Most callers will have called HeapTupleSatisfiesUpdate before this function;
5385 : * that will have set the HEAP_XMAX_INVALID bit if the xmax was a MultiXactId
5386 : * but it was not running anymore. There is a race condition, which is that the
5387 : * MultiXactId may have finished since then, but that uncommon case is handled
5388 : * either here, or within MultiXactIdExpand.
5389 : *
5390 : * There is a similar race condition possible when the old xmax was a regular
5391 : * TransactionId. We test TransactionIdIsInProgress again just to narrow the
5392 : * window, but it's still possible to end up creating an unnecessary
5393 : * MultiXactId. Fortunately this is harmless.
5394 : */
5395 : static void
5396 2147709 : compute_new_xmax_infomask(TransactionId xmax, uint16 old_infomask,
5397 : uint16 old_infomask2, TransactionId add_to_xmax,
5398 : LockTupleMode mode, bool is_update,
5399 : TransactionId *result_xmax, uint16 *result_infomask,
5400 : uint16 *result_infomask2)
5401 : {
5402 : TransactionId new_xmax;
5403 : uint16 new_infomask,
5404 : new_infomask2;
5405 :
5406 : Assert(TransactionIdIsCurrentTransactionId(add_to_xmax));
5407 :
5408 104034 : l5:
5409 2251743 : new_infomask = 0;
5410 2251743 : new_infomask2 = 0;
5411 2251743 : if (old_infomask & HEAP_XMAX_INVALID)
5412 : {
5413 : /*
5414 : * No previous locker; we just insert our own TransactionId.
5415 : *
5416 : * Note that it's critical that this case be the first one checked,
5417 : * because there are several blocks below that come back to this one
5418 : * to implement certain optimizations; old_infomask might contain
5419 : * other dirty bits in those cases, but we don't really care.
5420 : */
5421 2071079 : if (is_update)
5422 : {
5423 1838923 : new_xmax = add_to_xmax;
5424 1838923 : if (mode == LockTupleExclusive)
5425 1564265 : new_infomask2 |= HEAP_KEYS_UPDATED;
5426 : }
5427 : else
5428 : {
5429 232156 : new_infomask |= HEAP_XMAX_LOCK_ONLY;
5430 232156 : switch (mode)
5431 : {
5432 2638 : case LockTupleKeyShare:
5433 2638 : new_xmax = add_to_xmax;
5434 2638 : new_infomask |= HEAP_XMAX_KEYSHR_LOCK;
5435 2638 : break;
5436 749 : case LockTupleShare:
5437 749 : new_xmax = add_to_xmax;
5438 749 : new_infomask |= HEAP_XMAX_SHR_LOCK;
5439 749 : break;
5440 133017 : case LockTupleNoKeyExclusive:
5441 133017 : new_xmax = add_to_xmax;
5442 133017 : new_infomask |= HEAP_XMAX_EXCL_LOCK;
5443 133017 : break;
5444 95752 : case LockTupleExclusive:
5445 95752 : new_xmax = add_to_xmax;
5446 95752 : new_infomask |= HEAP_XMAX_EXCL_LOCK;
5447 95752 : new_infomask2 |= HEAP_KEYS_UPDATED;
5448 95752 : break;
5449 0 : default:
5450 0 : new_xmax = InvalidTransactionId; /* silence compiler */
5451 0 : elog(ERROR, "invalid lock mode");
5452 : }
5453 : }
5454 : }
5455 180664 : else if (old_infomask & HEAP_XMAX_IS_MULTI)
5456 : {
5457 : MultiXactStatus new_status;
5458 :
5459 : /*
5460 : * Currently we don't allow XMAX_COMMITTED to be set for multis, so
5461 : * cross-check.
5462 : */
5463 : Assert(!(old_infomask & HEAP_XMAX_COMMITTED));
5464 :
5465 : /*
5466 : * A multixact together with LOCK_ONLY set but neither lock bit set
5467 : * (i.e. a pg_upgraded share locked tuple) cannot possibly be running
5468 : * anymore. This check is critical for databases upgraded by
5469 : * pg_upgrade; both MultiXactIdIsRunning and MultiXactIdExpand assume
5470 : * that such multis are never passed.
5471 : */
5472 75570 : if (HEAP_LOCKED_UPGRADED(old_infomask))
5473 : {
5474 0 : old_infomask &= ~HEAP_XMAX_IS_MULTI;
5475 0 : old_infomask |= HEAP_XMAX_INVALID;
5476 0 : goto l5;
5477 : }
5478 :
5479 : /*
5480 : * If the XMAX is already a MultiXactId, then we need to expand it to
5481 : * include add_to_xmax; but if all the members were lockers and are
5482 : * all gone, we can do away with the IS_MULTI bit and just set
5483 : * add_to_xmax as the only locker/updater. If all lockers are gone
5484 : * and we have an updater that aborted, we can also do without a
5485 : * multi.
5486 : *
5487 : * The cost of doing GetMultiXactIdMembers would be paid by
5488 : * MultiXactIdExpand if we weren't to do this, so this check is not
5489 : * incurring extra work anyhow.
5490 : */
5491 75570 : if (!MultiXactIdIsRunning(xmax, HEAP_XMAX_IS_LOCKED_ONLY(old_infomask)))
5492 : {
5493 23 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) ||
5494 8 : !TransactionIdDidCommit(MultiXactIdGetUpdateXid(xmax,
5495 : old_infomask)))
5496 : {
5497 : /*
5498 : * Reset these bits and restart; otherwise fall through to
5499 : * create a new multi below.
5500 : */
5501 23 : old_infomask &= ~HEAP_XMAX_IS_MULTI;
5502 23 : old_infomask |= HEAP_XMAX_INVALID;
5503 23 : goto l5;
5504 : }
5505 : }
5506 :
5507 75547 : new_status = get_mxact_status_for_lock(mode, is_update);
5508 :
5509 75547 : new_xmax = MultiXactIdExpand((MultiXactId) xmax, add_to_xmax,
5510 : new_status);
5511 75547 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5512 : }
5513 105094 : else if (old_infomask & HEAP_XMAX_COMMITTED)
5514 : {
5515 : /*
5516 : * It's a committed update, so we need to preserve him as updater of
5517 : * the tuple.
5518 : */
5519 : MultiXactStatus status;
5520 : MultiXactStatus new_status;
5521 :
5522 13 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5523 0 : status = MultiXactStatusUpdate;
5524 : else
5525 13 : status = MultiXactStatusNoKeyUpdate;
5526 :
5527 13 : new_status = get_mxact_status_for_lock(mode, is_update);
5528 :
5529 : /*
5530 : * since it's not running, it's obviously impossible for the old
5531 : * updater to be identical to the current one, so we need not check
5532 : * for that case as we do in the block above.
5533 : */
5534 13 : new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5535 13 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5536 : }
5537 105081 : else if (TransactionIdIsInProgress(xmax))
5538 : {
5539 : /*
5540 : * If the XMAX is a valid, in-progress TransactionId, then we need to
5541 : * create a new MultiXactId that includes both the old locker or
5542 : * updater and our own TransactionId.
5543 : */
5544 : MultiXactStatus new_status;
5545 : MultiXactStatus old_status;
5546 : LockTupleMode old_mode;
5547 :
5548 105072 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
5549 : {
5550 105046 : if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
5551 5670 : old_status = MultiXactStatusForKeyShare;
5552 99376 : else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
5553 433 : old_status = MultiXactStatusForShare;
5554 98943 : else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
5555 : {
5556 98943 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5557 92795 : old_status = MultiXactStatusForUpdate;
5558 : else
5559 6148 : old_status = MultiXactStatusForNoKeyUpdate;
5560 : }
5561 : else
5562 : {
5563 : /*
5564 : * LOCK_ONLY can be present alone only when a page has been
5565 : * upgraded by pg_upgrade. But in that case,
5566 : * TransactionIdIsInProgress() should have returned false. We
5567 : * assume it's no longer locked in this case.
5568 : */
5569 0 : elog(WARNING, "LOCK_ONLY found for Xid in progress %u", xmax);
5570 0 : old_infomask |= HEAP_XMAX_INVALID;
5571 0 : old_infomask &= ~HEAP_XMAX_LOCK_ONLY;
5572 0 : goto l5;
5573 : }
5574 : }
5575 : else
5576 : {
5577 : /* it's an update, but which kind? */
5578 26 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5579 0 : old_status = MultiXactStatusUpdate;
5580 : else
5581 26 : old_status = MultiXactStatusNoKeyUpdate;
5582 : }
5583 :
5584 105072 : old_mode = TUPLOCK_from_mxstatus(old_status);
5585 :
5586 : /*
5587 : * If the lock to be acquired is for the same TransactionId as the
5588 : * existing lock, there's an optimization possible: consider only the
5589 : * strongest of both locks as the only one present, and restart.
5590 : */
5591 105072 : if (xmax == add_to_xmax)
5592 : {
5593 : /*
5594 : * Note that it's not possible for the original tuple to be
5595 : * updated: we wouldn't be here because the tuple would have been
5596 : * invisible and we wouldn't try to update it. As a subtlety,
5597 : * this code can also run when traversing an update chain to lock
5598 : * future versions of a tuple. But we wouldn't be here either,
5599 : * because the add_to_xmax would be different from the original
5600 : * updater.
5601 : */
5602 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
5603 :
5604 : /* acquire the strongest of both */
5605 104003 : if (mode < old_mode)
5606 52226 : mode = old_mode;
5607 : /* mustn't touch is_update */
5608 :
5609 104003 : old_infomask |= HEAP_XMAX_INVALID;
5610 104003 : goto l5;
5611 : }
5612 :
5613 : /* otherwise, just fall back to creating a new multixact */
5614 1069 : new_status = get_mxact_status_for_lock(mode, is_update);
5615 1069 : new_xmax = MultiXactIdCreate(xmax, old_status,
5616 : add_to_xmax, new_status);
5617 1069 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5618 : }
5619 14 : else if (!HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) &&
5620 5 : TransactionIdDidCommit(xmax))
5621 1 : {
5622 : /*
5623 : * It's a committed update, so we gotta preserve him as updater of the
5624 : * tuple.
5625 : */
5626 : MultiXactStatus status;
5627 : MultiXactStatus new_status;
5628 :
5629 1 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5630 0 : status = MultiXactStatusUpdate;
5631 : else
5632 1 : status = MultiXactStatusNoKeyUpdate;
5633 :
5634 1 : new_status = get_mxact_status_for_lock(mode, is_update);
5635 :
5636 : /*
5637 : * since it's not running, it's obviously impossible for the old
5638 : * updater to be identical to the current one, so we need not check
5639 : * for that case as we do in the block above.
5640 : */
5641 1 : new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5642 1 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5643 : }
5644 : else
5645 : {
5646 : /*
5647 : * Can get here iff the locking/updating transaction was running when
5648 : * the infomask was extracted from the tuple, but finished before
5649 : * TransactionIdIsInProgress got to run. Deal with it as if there was
5650 : * no locker at all in the first place.
5651 : */
5652 8 : old_infomask |= HEAP_XMAX_INVALID;
5653 8 : goto l5;
5654 : }
5655 :
5656 2147709 : *result_infomask = new_infomask;
5657 2147709 : *result_infomask2 = new_infomask2;
5658 2147709 : *result_xmax = new_xmax;
5659 2147709 : }
5660 :
5661 : /*
5662 : * Subroutine for heap_lock_updated_tuple_rec.
5663 : *
5664 : * Given a hypothetical multixact status held by the transaction identified
5665 : * with the given xid, does the current transaction need to wait, fail, or can
5666 : * it continue if it wanted to acquire a lock of the given mode? "needwait"
5667 : * is set to true if waiting is necessary; if it can continue, then TM_Ok is
5668 : * returned. If the lock is already held by the current transaction, return
5669 : * TM_SelfModified. In case of a conflict with another transaction, a
5670 : * different HeapTupleSatisfiesUpdate return code is returned.
5671 : *
5672 : * The held status is said to be hypothetical because it might correspond to a
5673 : * lock held by a single Xid, i.e. not a real MultiXactId; we express it this
5674 : * way for simplicity of API.
5675 : */
5676 : static TM_Result
5677 38774 : test_lockmode_for_conflict(MultiXactStatus status, TransactionId xid,
5678 : LockTupleMode mode, HeapTuple tup,
5679 : bool *needwait)
5680 : {
5681 : MultiXactStatus wantedstatus;
5682 :
5683 38774 : *needwait = false;
5684 38774 : wantedstatus = get_mxact_status_for_lock(mode, false);
5685 :
5686 : /*
5687 : * Note: we *must* check TransactionIdIsInProgress before
5688 : * TransactionIdDidAbort/Commit; see comment at top of heapam_visibility.c
5689 : * for an explanation.
5690 : */
5691 38774 : if (TransactionIdIsCurrentTransactionId(xid))
5692 : {
5693 : /*
5694 : * The tuple has already been locked by our own transaction. This is
5695 : * very rare but can happen if multiple transactions are trying to
5696 : * lock an ancient version of the same tuple.
5697 : */
5698 0 : return TM_SelfModified;
5699 : }
5700 38774 : else if (TransactionIdIsInProgress(xid))
5701 : {
5702 : /*
5703 : * If the locking transaction is running, what we do depends on
5704 : * whether the lock modes conflict: if they do, then we must wait for
5705 : * it to finish; otherwise we can fall through to lock this tuple
5706 : * version without waiting.
5707 : */
5708 36539 : if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
5709 36539 : LOCKMODE_from_mxstatus(wantedstatus)))
5710 : {
5711 8 : *needwait = true;
5712 : }
5713 :
5714 : /*
5715 : * If we set needwait above, then this value doesn't matter;
5716 : * otherwise, this value signals to caller that it's okay to proceed.
5717 : */
5718 36539 : return TM_Ok;
5719 : }
5720 2235 : else if (TransactionIdDidAbort(xid))
5721 206 : return TM_Ok;
5722 2029 : else if (TransactionIdDidCommit(xid))
5723 : {
5724 : /*
5725 : * The other transaction committed. If it was only a locker, then the
5726 : * lock is completely gone now and we can return success; but if it
5727 : * was an update, then what we do depends on whether the two lock
5728 : * modes conflict. If they conflict, then we must report error to
5729 : * caller. But if they don't, we can fall through to allow the current
5730 : * transaction to lock the tuple.
5731 : *
5732 : * Note: the reason we worry about ISUPDATE here is because as soon as
5733 : * a transaction ends, all its locks are gone and meaningless, and
5734 : * thus we can ignore them; whereas its updates persist. In the
5735 : * TransactionIdIsInProgress case, above, we don't need to check
5736 : * because we know the lock is still "alive" and thus a conflict needs
5737 : * always be checked.
5738 : */
5739 2029 : if (!ISUPDATE_from_mxstatus(status))
5740 2020 : return TM_Ok;
5741 :
5742 9 : if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
5743 9 : LOCKMODE_from_mxstatus(wantedstatus)))
5744 : {
5745 : /* bummer */
5746 8 : if (!ItemPointerEquals(&tup->t_self, &tup->t_data->t_ctid))
5747 6 : return TM_Updated;
5748 : else
5749 2 : return TM_Deleted;
5750 : }
5751 :
5752 1 : return TM_Ok;
5753 : }
5754 :
5755 : /* Not in progress, not aborted, not committed -- must have crashed */
5756 0 : return TM_Ok;
5757 : }
5758 :
5759 :
5760 : /*
5761 : * Recursive part of heap_lock_updated_tuple
5762 : *
5763 : * Fetch the tuple pointed to by tid in rel, and mark it as locked by the given
5764 : * xid with the given mode; if this tuple is updated, recurse to lock the new
5765 : * version as well.
5766 : */
5767 : static TM_Result
5768 2212 : heap_lock_updated_tuple_rec(Relation rel, TransactionId priorXmax,
5769 : const ItemPointerData *tid, TransactionId xid,
5770 : LockTupleMode mode)
5771 : {
5772 : TM_Result result;
5773 : ItemPointerData tupid;
5774 : HeapTupleData mytup;
5775 : Buffer buf;
5776 : uint16 new_infomask,
5777 : new_infomask2,
5778 : old_infomask,
5779 : old_infomask2;
5780 : TransactionId xmax,
5781 : new_xmax;
5782 2212 : bool cleared_all_frozen = false;
5783 : bool pinned_desired_page;
5784 2212 : Buffer vmbuffer = InvalidBuffer;
5785 : BlockNumber block;
5786 :
5787 2212 : ItemPointerCopy(tid, &tupid);
5788 :
5789 : for (;;)
5790 : {
5791 2215 : new_infomask = 0;
5792 2215 : new_xmax = InvalidTransactionId;
5793 2215 : block = ItemPointerGetBlockNumber(&tupid);
5794 2215 : ItemPointerCopy(&tupid, &(mytup.t_self));
5795 :
5796 2215 : if (!heap_fetch(rel, SnapshotAny, &mytup, &buf, false))
5797 : {
5798 : /*
5799 : * if we fail to find the updated version of the tuple, it's
5800 : * because it was vacuumed/pruned away after its creator
5801 : * transaction aborted. So behave as if we got to the end of the
5802 : * chain, and there's no further tuple to lock: return success to
5803 : * caller.
5804 : */
5805 0 : result = TM_Ok;
5806 0 : goto out_unlocked;
5807 : }
5808 :
5809 2215 : l4:
5810 2223 : CHECK_FOR_INTERRUPTS();
5811 :
5812 : /*
5813 : * Before locking the buffer, pin the visibility map page if it
5814 : * appears to be necessary. Since we haven't got the lock yet,
5815 : * someone else might be in the middle of changing this, so we'll need
5816 : * to recheck after we have the lock.
5817 : */
5818 2223 : if (PageIsAllVisible(BufferGetPage(buf)))
5819 : {
5820 0 : visibilitymap_pin(rel, block, &vmbuffer);
5821 0 : pinned_desired_page = true;
5822 : }
5823 : else
5824 2223 : pinned_desired_page = false;
5825 :
5826 2223 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
5827 :
5828 : /*
5829 : * If we didn't pin the visibility map page and the page has become
5830 : * all visible while we were busy locking the buffer, we'll have to
5831 : * unlock and re-lock, to avoid holding the buffer lock across I/O.
5832 : * That's a bit unfortunate, but hopefully shouldn't happen often.
5833 : *
5834 : * Note: in some paths through this function, we will reach here
5835 : * holding a pin on a vm page that may or may not be the one matching
5836 : * this page. If this page isn't all-visible, we won't use the vm
5837 : * page, but we hold onto such a pin till the end of the function.
5838 : */
5839 2223 : if (!pinned_desired_page && PageIsAllVisible(BufferGetPage(buf)))
5840 : {
5841 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5842 0 : visibilitymap_pin(rel, block, &vmbuffer);
5843 0 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
5844 : }
5845 :
5846 : /*
5847 : * Check the tuple XMIN against prior XMAX, if any. If we reached the
5848 : * end of the chain, we're done, so return success.
5849 : */
5850 4446 : if (TransactionIdIsValid(priorXmax) &&
5851 2223 : !TransactionIdEquals(HeapTupleHeaderGetXmin(mytup.t_data),
5852 : priorXmax))
5853 : {
5854 2 : result = TM_Ok;
5855 2 : goto out_locked;
5856 : }
5857 :
5858 : /*
5859 : * Also check Xmin: if this tuple was created by an aborted
5860 : * (sub)transaction, then we already locked the last live one in the
5861 : * chain, thus we're done, so return success.
5862 : */
5863 2221 : if (TransactionIdDidAbort(HeapTupleHeaderGetXmin(mytup.t_data)))
5864 : {
5865 24 : result = TM_Ok;
5866 24 : goto out_locked;
5867 : }
5868 :
5869 2197 : old_infomask = mytup.t_data->t_infomask;
5870 2197 : old_infomask2 = mytup.t_data->t_infomask2;
5871 2197 : xmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
5872 :
5873 : /*
5874 : * If this tuple version has been updated or locked by some concurrent
5875 : * transaction(s), what we do depends on whether our lock mode
5876 : * conflicts with what those other transactions hold, and also on the
5877 : * status of them.
5878 : */
5879 2197 : if (!(old_infomask & HEAP_XMAX_INVALID))
5880 : {
5881 : TransactionId rawxmax;
5882 : bool needwait;
5883 :
5884 2138 : rawxmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
5885 2138 : if (old_infomask & HEAP_XMAX_IS_MULTI)
5886 : {
5887 : int nmembers;
5888 : int i;
5889 : MultiXactMember *members;
5890 :
5891 : /*
5892 : * We don't need a test for pg_upgrade'd tuples: this is only
5893 : * applied to tuples after the first in an update chain. Said
5894 : * first tuple in the chain may well be locked-in-9.2-and-
5895 : * pg_upgraded, but that one was already locked by our caller,
5896 : * not us; and any subsequent ones cannot be because our
5897 : * caller must necessarily have obtained a snapshot later than
5898 : * the pg_upgrade itself.
5899 : */
5900 : Assert(!HEAP_LOCKED_UPGRADED(mytup.t_data->t_infomask));
5901 :
5902 2109 : nmembers = GetMultiXactIdMembers(rawxmax, &members, false,
5903 2109 : HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
5904 40854 : for (i = 0; i < nmembers; i++)
5905 : {
5906 38745 : result = test_lockmode_for_conflict(members[i].status,
5907 38745 : members[i].xid,
5908 : mode,
5909 : &mytup,
5910 : &needwait);
5911 :
5912 : /*
5913 : * If the tuple was already locked by ourselves in a
5914 : * previous iteration of this (say heap_lock_tuple was
5915 : * forced to restart the locking loop because of a change
5916 : * in xmax), then we hold the lock already on this tuple
5917 : * version and we don't need to do anything; and this is
5918 : * not an error condition either. We just need to skip
5919 : * this tuple and continue locking the next version in the
5920 : * update chain.
5921 : */
5922 38745 : if (result == TM_SelfModified)
5923 : {
5924 0 : pfree(members);
5925 0 : goto next;
5926 : }
5927 :
5928 38745 : if (needwait)
5929 : {
5930 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5931 0 : XactLockTableWait(members[i].xid, rel,
5932 : &mytup.t_self,
5933 : XLTW_LockUpdated);
5934 0 : pfree(members);
5935 0 : goto l4;
5936 : }
5937 38745 : if (result != TM_Ok)
5938 : {
5939 0 : pfree(members);
5940 0 : goto out_locked;
5941 : }
5942 : }
5943 2109 : if (members)
5944 2109 : pfree(members);
5945 : }
5946 : else
5947 : {
5948 : MultiXactStatus status;
5949 :
5950 : /*
5951 : * For a non-multi Xmax, we first need to compute the
5952 : * corresponding MultiXactStatus by using the infomask bits.
5953 : */
5954 29 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
5955 : {
5956 10 : if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
5957 10 : status = MultiXactStatusForKeyShare;
5958 0 : else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
5959 0 : status = MultiXactStatusForShare;
5960 0 : else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
5961 : {
5962 0 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5963 0 : status = MultiXactStatusForUpdate;
5964 : else
5965 0 : status = MultiXactStatusForNoKeyUpdate;
5966 : }
5967 : else
5968 : {
5969 : /*
5970 : * LOCK_ONLY present alone (a pg_upgraded tuple marked
5971 : * as share-locked in the old cluster) shouldn't be
5972 : * seen in the middle of an update chain.
5973 : */
5974 0 : elog(ERROR, "invalid lock status in tuple");
5975 : }
5976 : }
5977 : else
5978 : {
5979 : /* it's an update, but which kind? */
5980 19 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5981 14 : status = MultiXactStatusUpdate;
5982 : else
5983 5 : status = MultiXactStatusNoKeyUpdate;
5984 : }
5985 :
5986 29 : result = test_lockmode_for_conflict(status, rawxmax, mode,
5987 : &mytup, &needwait);
5988 :
5989 : /*
5990 : * If the tuple was already locked by ourselves in a previous
5991 : * iteration of this (say heap_lock_tuple was forced to
5992 : * restart the locking loop because of a change in xmax), then
5993 : * we hold the lock already on this tuple version and we don't
5994 : * need to do anything; and this is not an error condition
5995 : * either. We just need to skip this tuple and continue
5996 : * locking the next version in the update chain.
5997 : */
5998 29 : if (result == TM_SelfModified)
5999 0 : goto next;
6000 :
6001 29 : if (needwait)
6002 : {
6003 8 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
6004 8 : XactLockTableWait(rawxmax, rel, &mytup.t_self,
6005 : XLTW_LockUpdated);
6006 8 : goto l4;
6007 : }
6008 21 : if (result != TM_Ok)
6009 : {
6010 8 : goto out_locked;
6011 : }
6012 : }
6013 : }
6014 :
6015 : /* compute the new Xmax and infomask values for the tuple ... */
6016 2181 : compute_new_xmax_infomask(xmax, old_infomask, mytup.t_data->t_infomask2,
6017 : xid, mode, false,
6018 : &new_xmax, &new_infomask, &new_infomask2);
6019 :
6020 2181 : if (PageIsAllVisible(BufferGetPage(buf)) &&
6021 0 : visibilitymap_clear(rel, block, vmbuffer,
6022 : VISIBILITYMAP_ALL_FROZEN))
6023 0 : cleared_all_frozen = true;
6024 :
6025 2181 : START_CRIT_SECTION();
6026 :
6027 : /* ... and set them */
6028 2181 : HeapTupleHeaderSetXmax(mytup.t_data, new_xmax);
6029 2181 : mytup.t_data->t_infomask &= ~HEAP_XMAX_BITS;
6030 2181 : mytup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
6031 2181 : mytup.t_data->t_infomask |= new_infomask;
6032 2181 : mytup.t_data->t_infomask2 |= new_infomask2;
6033 :
6034 2181 : MarkBufferDirty(buf);
6035 :
6036 : /* XLOG stuff */
6037 2181 : if (RelationNeedsWAL(rel))
6038 : {
6039 : xl_heap_lock_updated xlrec;
6040 : XLogRecPtr recptr;
6041 2181 : Page page = BufferGetPage(buf);
6042 :
6043 2181 : XLogBeginInsert();
6044 2181 : XLogRegisterBuffer(0, buf, REGBUF_STANDARD);
6045 :
6046 2181 : xlrec.offnum = ItemPointerGetOffsetNumber(&mytup.t_self);
6047 2181 : xlrec.xmax = new_xmax;
6048 2181 : xlrec.infobits_set = compute_infobits(new_infomask, new_infomask2);
6049 2181 : xlrec.flags =
6050 2181 : cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
6051 :
6052 2181 : XLogRegisterData(&xlrec, SizeOfHeapLockUpdated);
6053 :
6054 2181 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_LOCK_UPDATED);
6055 :
6056 2181 : PageSetLSN(page, recptr);
6057 : }
6058 :
6059 2181 : END_CRIT_SECTION();
6060 :
6061 2181 : next:
6062 : /* if we find the end of update chain, we're done. */
6063 4362 : if (mytup.t_data->t_infomask & HEAP_XMAX_INVALID ||
6064 4362 : HeapTupleHeaderIndicatesMovedPartitions(mytup.t_data) ||
6065 2185 : ItemPointerEquals(&mytup.t_self, &mytup.t_data->t_ctid) ||
6066 4 : HeapTupleHeaderIsOnlyLocked(mytup.t_data))
6067 : {
6068 2178 : result = TM_Ok;
6069 2178 : goto out_locked;
6070 : }
6071 :
6072 : /* tail recursion */
6073 3 : priorXmax = HeapTupleHeaderGetUpdateXid(mytup.t_data);
6074 3 : ItemPointerCopy(&(mytup.t_data->t_ctid), &tupid);
6075 3 : UnlockReleaseBuffer(buf);
6076 : }
6077 :
6078 : result = TM_Ok;
6079 :
6080 2212 : out_locked:
6081 2212 : UnlockReleaseBuffer(buf);
6082 :
6083 2212 : out_unlocked:
6084 2212 : if (vmbuffer != InvalidBuffer)
6085 0 : ReleaseBuffer(vmbuffer);
6086 :
6087 2212 : return result;
6088 : }
6089 :
6090 : /*
6091 : * heap_lock_updated_tuple
6092 : * Follow update chain when locking an updated tuple, acquiring locks (row
6093 : * marks) on the updated versions.
6094 : *
6095 : * 'prior_infomask', 'prior_raw_xmax' and 'prior_ctid' are the corresponding
6096 : * fields from the initial tuple. We will lock the tuples starting from the
6097 : * one that 'prior_ctid' points to. Note: This function does not lock the
6098 : * initial tuple itself.
6099 : *
6100 : * This function doesn't check visibility, it just unconditionally marks the
6101 : * tuple(s) as locked. If any tuple in the updated chain is being deleted
6102 : * concurrently (or updated with the key being modified), sleep until the
6103 : * transaction doing it is finished.
6104 : *
6105 : * Note that we don't acquire heavyweight tuple locks on the tuples we walk
6106 : * when we have to wait for other transactions to release them, as opposed to
6107 : * what heap_lock_tuple does. The reason is that having more than one
6108 : * transaction walking the chain is probably uncommon enough that risk of
6109 : * starvation is not likely: one of the preconditions for being here is that
6110 : * the snapshot in use predates the update that created this tuple (because we
6111 : * started at an earlier version of the tuple), but at the same time such a
6112 : * transaction cannot be using repeatable read or serializable isolation
6113 : * levels, because that would lead to a serializability failure.
6114 : */
6115 : static TM_Result
6116 2214 : heap_lock_updated_tuple(Relation rel,
6117 : uint16 prior_infomask,
6118 : TransactionId prior_raw_xmax,
6119 : const ItemPointerData *prior_ctid,
6120 : TransactionId xid, LockTupleMode mode)
6121 : {
6122 2214 : INJECTION_POINT("heap_lock_updated_tuple", NULL);
6123 :
6124 : /*
6125 : * If the tuple has moved into another partition (effectively a delete)
6126 : * stop here.
6127 : */
6128 2214 : if (!ItemPointerIndicatesMovedPartitions(prior_ctid))
6129 : {
6130 : TransactionId prior_xmax;
6131 :
6132 : /*
6133 : * If this is the first possibly-multixact-able operation in the
6134 : * current transaction, set my per-backend OldestMemberMXactId
6135 : * setting. We can be certain that the transaction will never become a
6136 : * member of any older MultiXactIds than that. (We have to do this
6137 : * even if we end up just using our own TransactionId below, since
6138 : * some other backend could incorporate our XID into a MultiXact
6139 : * immediately afterwards.)
6140 : */
6141 2212 : MultiXactIdSetOldestMember();
6142 :
6143 4424 : prior_xmax = (prior_infomask & HEAP_XMAX_IS_MULTI) ?
6144 2212 : MultiXactIdGetUpdateXid(prior_raw_xmax, prior_infomask) : prior_raw_xmax;
6145 2212 : return heap_lock_updated_tuple_rec(rel, prior_xmax, prior_ctid, xid, mode);
6146 : }
6147 :
6148 : /* nothing to lock */
6149 2 : return TM_Ok;
6150 : }
6151 :
6152 : /*
6153 : * heap_finish_speculative - mark speculative insertion as successful
6154 : *
6155 : * To successfully finish a speculative insertion we have to clear speculative
6156 : * token from tuple. To do so the t_ctid field, which will contain a
6157 : * speculative token value, is modified in place to point to the tuple itself,
6158 : * which is characteristic of a newly inserted ordinary tuple.
6159 : *
6160 : * NB: It is not ok to commit without either finishing or aborting a
6161 : * speculative insertion. We could treat speculative tuples of committed
6162 : * transactions implicitly as completed, but then we would have to be prepared
6163 : * to deal with speculative tokens on committed tuples. That wouldn't be
6164 : * difficult - no-one looks at the ctid field of a tuple with invalid xmax -
6165 : * but clearing the token at completion isn't very expensive either.
6166 : * An explicit confirmation WAL record also makes logical decoding simpler.
6167 : */
6168 : void
6169 2119 : heap_finish_speculative(Relation relation, const ItemPointerData *tid)
6170 : {
6171 : Buffer buffer;
6172 : Page page;
6173 : OffsetNumber offnum;
6174 : ItemId lp;
6175 : HeapTupleHeader htup;
6176 :
6177 2119 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
6178 2119 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6179 2119 : page = BufferGetPage(buffer);
6180 :
6181 2119 : offnum = ItemPointerGetOffsetNumber(tid);
6182 2119 : if (offnum < 1 || offnum > PageGetMaxOffsetNumber(page))
6183 0 : elog(ERROR, "offnum out of range");
6184 2119 : lp = PageGetItemId(page, offnum);
6185 2119 : if (!ItemIdIsNormal(lp))
6186 0 : elog(ERROR, "invalid lp");
6187 :
6188 2119 : htup = (HeapTupleHeader) PageGetItem(page, lp);
6189 :
6190 : /* NO EREPORT(ERROR) from here till changes are logged */
6191 2119 : START_CRIT_SECTION();
6192 :
6193 : Assert(HeapTupleHeaderIsSpeculative(htup));
6194 :
6195 2119 : MarkBufferDirty(buffer);
6196 :
6197 : /*
6198 : * Replace the speculative insertion token with a real t_ctid, pointing to
6199 : * itself like it does on regular tuples.
6200 : */
6201 2119 : htup->t_ctid = *tid;
6202 :
6203 : /* XLOG stuff */
6204 2119 : if (RelationNeedsWAL(relation))
6205 : {
6206 : xl_heap_confirm xlrec;
6207 : XLogRecPtr recptr;
6208 :
6209 2102 : xlrec.offnum = ItemPointerGetOffsetNumber(tid);
6210 :
6211 2102 : XLogBeginInsert();
6212 :
6213 : /* We want the same filtering on this as on a plain insert */
6214 2102 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
6215 :
6216 2102 : XLogRegisterData(&xlrec, SizeOfHeapConfirm);
6217 2102 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
6218 :
6219 2102 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_CONFIRM);
6220 :
6221 2102 : PageSetLSN(page, recptr);
6222 : }
6223 :
6224 2119 : END_CRIT_SECTION();
6225 :
6226 2119 : UnlockReleaseBuffer(buffer);
6227 2119 : }
6228 :
6229 : /*
6230 : * heap_abort_speculative - kill a speculatively inserted tuple
6231 : *
6232 : * Marks a tuple that was speculatively inserted in the same command as dead,
6233 : * by setting its xmin as invalid. That makes it immediately appear as dead
6234 : * to all transactions, including our own. In particular, it makes
6235 : * HeapTupleSatisfiesDirty() regard the tuple as dead, so that another backend
6236 : * inserting a duplicate key value won't unnecessarily wait for our whole
6237 : * transaction to finish (it'll just wait for our speculative insertion to
6238 : * finish).
6239 : *
6240 : * Killing the tuple prevents "unprincipled deadlocks", which are deadlocks
6241 : * that arise due to a mutual dependency that is not user visible. By
6242 : * definition, unprincipled deadlocks cannot be prevented by the user
6243 : * reordering lock acquisition in client code, because the implementation level
6244 : * lock acquisitions are not under the user's direct control. If speculative
6245 : * inserters did not take this precaution, then under high concurrency they
6246 : * could deadlock with each other, which would not be acceptable.
6247 : *
6248 : * This is somewhat redundant with heap_delete, but we prefer to have a
6249 : * dedicated routine with stripped down requirements. Note that this is also
6250 : * used to delete the TOAST tuples created during speculative insertion.
6251 : *
6252 : * This routine does not affect logical decoding as it only looks at
6253 : * confirmation records.
6254 : */
6255 : void
6256 16 : heap_abort_speculative(Relation relation, const ItemPointerData *tid)
6257 : {
6258 16 : TransactionId xid = GetCurrentTransactionId();
6259 : ItemId lp;
6260 : HeapTupleData tp;
6261 : Page page;
6262 : BlockNumber block;
6263 : Buffer buffer;
6264 :
6265 : Assert(ItemPointerIsValid(tid));
6266 :
6267 16 : block = ItemPointerGetBlockNumber(tid);
6268 16 : buffer = ReadBuffer(relation, block);
6269 16 : page = BufferGetPage(buffer);
6270 :
6271 16 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6272 :
6273 : /*
6274 : * Page can't be all visible, we just inserted into it, and are still
6275 : * running.
6276 : */
6277 : Assert(!PageIsAllVisible(page));
6278 :
6279 16 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
6280 : Assert(ItemIdIsNormal(lp));
6281 :
6282 16 : tp.t_tableOid = RelationGetRelid(relation);
6283 16 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
6284 16 : tp.t_len = ItemIdGetLength(lp);
6285 16 : tp.t_self = *tid;
6286 :
6287 : /*
6288 : * Sanity check that the tuple really is a speculatively inserted tuple,
6289 : * inserted by us.
6290 : */
6291 16 : if (tp.t_data->t_choice.t_heap.t_xmin != xid)
6292 0 : elog(ERROR, "attempted to kill a tuple inserted by another transaction");
6293 16 : if (!(IsToastRelation(relation) || HeapTupleHeaderIsSpeculative(tp.t_data)))
6294 0 : elog(ERROR, "attempted to kill a non-speculative tuple");
6295 : Assert(!HeapTupleHeaderIsHeapOnly(tp.t_data));
6296 :
6297 : /*
6298 : * No need to check for serializable conflicts here. There is never a
6299 : * need for a combo CID, either. No need to extract replica identity, or
6300 : * do anything special with infomask bits.
6301 : */
6302 :
6303 16 : START_CRIT_SECTION();
6304 :
6305 : /*
6306 : * The tuple will become DEAD immediately. Flag that this page is a
6307 : * candidate for pruning by setting xmin to TransactionXmin. While not
6308 : * immediately prunable, it is the oldest xid we can cheaply determine
6309 : * that's safe against wraparound / being older than the table's
6310 : * relfrozenxid. To defend against the unlikely case of a new relation
6311 : * having a newer relfrozenxid than our TransactionXmin, use relfrozenxid
6312 : * if so (vacuum can't subsequently move relfrozenxid to beyond
6313 : * TransactionXmin, so there's no race here).
6314 : */
6315 : Assert(TransactionIdIsValid(TransactionXmin));
6316 : {
6317 16 : TransactionId relfrozenxid = relation->rd_rel->relfrozenxid;
6318 : TransactionId prune_xid;
6319 :
6320 16 : if (TransactionIdPrecedes(TransactionXmin, relfrozenxid))
6321 0 : prune_xid = relfrozenxid;
6322 : else
6323 16 : prune_xid = TransactionXmin;
6324 16 : PageSetPrunable(page, prune_xid);
6325 : }
6326 :
6327 : /* store transaction information of xact deleting the tuple */
6328 16 : tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
6329 16 : tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
6330 :
6331 : /*
6332 : * Set the tuple header xmin to InvalidTransactionId. This makes the
6333 : * tuple immediately invisible everyone. (In particular, to any
6334 : * transactions waiting on the speculative token, woken up later.)
6335 : */
6336 16 : HeapTupleHeaderSetXmin(tp.t_data, InvalidTransactionId);
6337 :
6338 : /* Clear the speculative insertion token too */
6339 16 : tp.t_data->t_ctid = tp.t_self;
6340 :
6341 16 : MarkBufferDirty(buffer);
6342 :
6343 : /*
6344 : * XLOG stuff
6345 : *
6346 : * The WAL records generated here match heap_delete(). The same recovery
6347 : * routines are used.
6348 : */
6349 16 : if (RelationNeedsWAL(relation))
6350 : {
6351 : xl_heap_delete xlrec;
6352 : XLogRecPtr recptr;
6353 :
6354 12 : xlrec.flags = XLH_DELETE_IS_SUPER;
6355 24 : xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
6356 12 : tp.t_data->t_infomask2);
6357 12 : xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
6358 12 : xlrec.xmax = xid;
6359 :
6360 12 : XLogBeginInsert();
6361 12 : XLogRegisterData(&xlrec, SizeOfHeapDelete);
6362 12 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
6363 :
6364 : /* No replica identity & replication origin logged */
6365 :
6366 12 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
6367 :
6368 12 : PageSetLSN(page, recptr);
6369 : }
6370 :
6371 16 : END_CRIT_SECTION();
6372 :
6373 16 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6374 :
6375 16 : if (HeapTupleHasExternal(&tp))
6376 : {
6377 : Assert(!IsToastRelation(relation));
6378 1 : heap_toast_delete(relation, &tp, true);
6379 : }
6380 :
6381 : /*
6382 : * Never need to mark tuple for invalidation, since catalogs don't support
6383 : * speculative insertion
6384 : */
6385 :
6386 : /* Now we can release the buffer */
6387 16 : ReleaseBuffer(buffer);
6388 :
6389 : /* count deletion, as we counted the insertion too */
6390 16 : pgstat_count_heap_delete(relation);
6391 16 : }
6392 :
6393 : /*
6394 : * heap_inplace_lock - protect inplace update from concurrent heap_update()
6395 : *
6396 : * Evaluate whether the tuple's state is compatible with a no-key update.
6397 : * Current transaction rowmarks are fine, as is KEY SHARE from any
6398 : * transaction. If compatible, return true with the buffer exclusive-locked,
6399 : * and the caller must release that by calling
6400 : * heap_inplace_update_and_unlock(), calling heap_inplace_unlock(), or raising
6401 : * an error. Otherwise, call release_callback(arg), wait for blocking
6402 : * transactions to end, and return false.
6403 : *
6404 : * Since this is intended for system catalogs and SERIALIZABLE doesn't cover
6405 : * DDL, this doesn't guarantee any particular predicate locking.
6406 : *
6407 : * heap_delete() is a rarer source of blocking transactions (xwait). We'll
6408 : * wait for such a transaction just like for the normal heap_update() case.
6409 : * Normal concurrent DROP commands won't cause that, because all inplace
6410 : * updaters take some lock that conflicts with DROP. An explicit SQL "DELETE
6411 : * FROM pg_class" can cause it. By waiting, if the concurrent transaction
6412 : * executed both "DELETE FROM pg_class" and "INSERT INTO pg_class", our caller
6413 : * can find the successor tuple.
6414 : *
6415 : * Readers of inplace-updated fields expect changes to those fields are
6416 : * durable. For example, vac_truncate_clog() reads datfrozenxid from
6417 : * pg_database tuples via catalog snapshots. A future snapshot must not
6418 : * return a lower datfrozenxid for the same database OID (lower in the
6419 : * FullTransactionIdPrecedes() sense). We achieve that since no update of a
6420 : * tuple can start while we hold a lock on its buffer. In cases like
6421 : * BEGIN;GRANT;CREATE INDEX;COMMIT we're inplace-updating a tuple visible only
6422 : * to this transaction. ROLLBACK then is one case where it's okay to lose
6423 : * inplace updates. (Restoring relhasindex=false on ROLLBACK is fine, since
6424 : * any concurrent CREATE INDEX would have blocked, then inplace-updated the
6425 : * committed tuple.)
6426 : *
6427 : * In principle, we could avoid waiting by overwriting every tuple in the
6428 : * updated tuple chain. Reader expectations permit updating a tuple only if
6429 : * it's aborted, is the tail of the chain, or we already updated the tuple
6430 : * referenced in its t_ctid. Hence, we would need to overwrite the tuples in
6431 : * order from tail to head. That would imply either (a) mutating all tuples
6432 : * in one critical section or (b) accepting a chance of partial completion.
6433 : * Partial completion of a relfrozenxid update would have the weird
6434 : * consequence that the table's next VACUUM could see the table's relfrozenxid
6435 : * move forward between vacuum_get_cutoffs() and finishing.
6436 : */
6437 : bool
6438 210302 : heap_inplace_lock(Relation relation,
6439 : HeapTuple oldtup_ptr, Buffer buffer,
6440 : void (*release_callback) (void *), void *arg)
6441 : {
6442 210302 : HeapTupleData oldtup = *oldtup_ptr; /* minimize diff vs. heap_update() */
6443 : TM_Result result;
6444 : bool ret;
6445 :
6446 : #ifdef USE_ASSERT_CHECKING
6447 : if (RelationGetRelid(relation) == RelationRelationId)
6448 : check_inplace_rel_lock(oldtup_ptr);
6449 : #endif
6450 :
6451 : Assert(BufferIsValid(buffer));
6452 :
6453 : /*
6454 : * Register shared cache invals if necessary. Other sessions may finish
6455 : * inplace updates of this tuple between this step and LockTuple(). Since
6456 : * inplace updates don't change cache keys, that's harmless.
6457 : *
6458 : * While it's tempting to register invals only after confirming we can
6459 : * return true, the following obstacle precludes reordering steps that
6460 : * way. Registering invals might reach a CatalogCacheInitializeCache()
6461 : * that locks "buffer". That would hang indefinitely if running after our
6462 : * own LockBuffer(). Hence, we must register invals before LockBuffer().
6463 : */
6464 210302 : CacheInvalidateHeapTupleInplace(relation, oldtup_ptr);
6465 :
6466 210302 : LockTuple(relation, &oldtup.t_self, InplaceUpdateTupleLock);
6467 210302 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6468 :
6469 : /*----------
6470 : * Interpret HeapTupleSatisfiesUpdate() like heap_update() does, except:
6471 : *
6472 : * - wait unconditionally
6473 : * - already locked tuple above, since inplace needs that unconditionally
6474 : * - don't recheck header after wait: simpler to defer to next iteration
6475 : * - don't try to continue even if the updater aborts: likewise
6476 : * - no crosscheck
6477 : */
6478 210302 : result = HeapTupleSatisfiesUpdate(&oldtup, GetCurrentCommandId(false),
6479 : buffer);
6480 :
6481 210302 : if (result == TM_Invisible)
6482 : {
6483 : /* no known way this can happen */
6484 0 : ereport(ERROR,
6485 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6486 : errmsg_internal("attempted to overwrite invisible tuple")));
6487 : }
6488 210302 : else if (result == TM_SelfModified)
6489 : {
6490 : /*
6491 : * CREATE INDEX might reach this if an expression is silly enough to
6492 : * call e.g. SELECT ... FROM pg_class FOR SHARE. C code of other SQL
6493 : * statements might get here after a heap_update() of the same row, in
6494 : * the absence of an intervening CommandCounterIncrement().
6495 : */
6496 0 : ereport(ERROR,
6497 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6498 : errmsg("tuple to be updated was already modified by an operation triggered by the current command")));
6499 : }
6500 210302 : else if (result == TM_BeingModified)
6501 : {
6502 : TransactionId xwait;
6503 : uint16 infomask;
6504 :
6505 67 : xwait = HeapTupleHeaderGetRawXmax(oldtup.t_data);
6506 67 : infomask = oldtup.t_data->t_infomask;
6507 :
6508 67 : if (infomask & HEAP_XMAX_IS_MULTI)
6509 : {
6510 5 : LockTupleMode lockmode = LockTupleNoKeyExclusive;
6511 5 : MultiXactStatus mxact_status = MultiXactStatusNoKeyUpdate;
6512 : int remain;
6513 :
6514 5 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
6515 : lockmode, NULL))
6516 : {
6517 2 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6518 2 : release_callback(arg);
6519 2 : ret = false;
6520 2 : MultiXactIdWait((MultiXactId) xwait, mxact_status, infomask,
6521 : relation, &oldtup.t_self, XLTW_Update,
6522 : &remain);
6523 : }
6524 : else
6525 3 : ret = true;
6526 : }
6527 62 : else if (TransactionIdIsCurrentTransactionId(xwait))
6528 1 : ret = true;
6529 61 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask))
6530 1 : ret = true;
6531 : else
6532 : {
6533 60 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6534 60 : release_callback(arg);
6535 60 : ret = false;
6536 60 : XactLockTableWait(xwait, relation, &oldtup.t_self,
6537 : XLTW_Update);
6538 : }
6539 : }
6540 : else
6541 : {
6542 210235 : ret = (result == TM_Ok);
6543 210235 : if (!ret)
6544 : {
6545 1 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6546 1 : release_callback(arg);
6547 : }
6548 : }
6549 :
6550 : /*
6551 : * GetCatalogSnapshot() relies on invalidation messages to know when to
6552 : * take a new snapshot. COMMIT of xwait is responsible for sending the
6553 : * invalidation. We're not acquiring heavyweight locks sufficient to
6554 : * block if not yet sent, so we must take a new snapshot to ensure a later
6555 : * attempt has a fair chance. While we don't need this if xwait aborted,
6556 : * don't bother optimizing that.
6557 : */
6558 210302 : if (!ret)
6559 : {
6560 63 : UnlockTuple(relation, &oldtup.t_self, InplaceUpdateTupleLock);
6561 63 : ForgetInplace_Inval();
6562 63 : InvalidateCatalogSnapshot();
6563 : }
6564 210302 : return ret;
6565 : }
6566 :
6567 : /*
6568 : * heap_inplace_update_and_unlock - core of systable_inplace_update_finish
6569 : *
6570 : * The tuple cannot change size, and therefore its header fields and null
6571 : * bitmap (if any) don't change either.
6572 : *
6573 : * Since we hold LOCKTAG_TUPLE, no updater has a local copy of this tuple.
6574 : */
6575 : void
6576 83742 : heap_inplace_update_and_unlock(Relation relation,
6577 : HeapTuple oldtup, HeapTuple tuple,
6578 : Buffer buffer)
6579 : {
6580 83742 : HeapTupleHeader htup = oldtup->t_data;
6581 : uint32 oldlen;
6582 : uint32 newlen;
6583 : char *dst;
6584 : char *src;
6585 83742 : int nmsgs = 0;
6586 83742 : SharedInvalidationMessage *invalMessages = NULL;
6587 83742 : bool RelcacheInitFileInval = false;
6588 :
6589 : Assert(ItemPointerEquals(&oldtup->t_self, &tuple->t_self));
6590 83742 : oldlen = oldtup->t_len - htup->t_hoff;
6591 83742 : newlen = tuple->t_len - tuple->t_data->t_hoff;
6592 83742 : if (oldlen != newlen || htup->t_hoff != tuple->t_data->t_hoff)
6593 0 : elog(ERROR, "wrong tuple length");
6594 :
6595 83742 : dst = (char *) htup + htup->t_hoff;
6596 83742 : src = (char *) tuple->t_data + tuple->t_data->t_hoff;
6597 :
6598 : /* Like RecordTransactionCommit(), log only if needed */
6599 83742 : if (XLogStandbyInfoActive())
6600 58439 : nmsgs = inplaceGetInvalidationMessages(&invalMessages,
6601 : &RelcacheInitFileInval);
6602 :
6603 : /*
6604 : * Unlink relcache init files as needed. If unlinking, acquire
6605 : * RelCacheInitLock until after associated invalidations. By doing this
6606 : * in advance, if we checkpoint and then crash between inplace
6607 : * XLogInsert() and inval, we don't rely on StartupXLOG() ->
6608 : * RelationCacheInitFileRemove(). That uses elevel==LOG, so replay would
6609 : * neglect to PANIC on EIO.
6610 : */
6611 83742 : PreInplace_Inval();
6612 :
6613 : /*----------
6614 : * NO EREPORT(ERROR) from here till changes are complete
6615 : *
6616 : * Our buffer lock won't stop a reader having already pinned and checked
6617 : * visibility for this tuple. Hence, we write WAL first, then mutate the
6618 : * buffer. Like in MarkBufferDirtyHint() or RecordTransactionCommit(),
6619 : * checkpoint delay makes that acceptable. With the usual order of
6620 : * changes, a crash after memcpy() and before XLogInsert() could allow
6621 : * datfrozenxid to overtake relfrozenxid:
6622 : *
6623 : * ["D" is a VACUUM (ONLY_DATABASE_STATS)]
6624 : * ["R" is a VACUUM tbl]
6625 : * D: vac_update_datfrozenxid() -> systable_beginscan(pg_class)
6626 : * D: systable_getnext() returns pg_class tuple of tbl
6627 : * R: memcpy() into pg_class tuple of tbl
6628 : * D: raise pg_database.datfrozenxid, XLogInsert(), finish
6629 : * [crash]
6630 : * [recovery restores datfrozenxid w/o relfrozenxid]
6631 : *
6632 : * Mimic MarkBufferDirtyHint() subroutine XLogSaveBufferForHint().
6633 : * Specifically, use DELAY_CHKPT_START, and copy the buffer to the stack.
6634 : * The stack copy facilitates a FPI of the post-mutation block before we
6635 : * accept other sessions seeing it. DELAY_CHKPT_START allows us to
6636 : * XLogInsert() before MarkBufferDirty(). Since XLogSaveBufferForHint()
6637 : * can operate under BUFFER_LOCK_SHARED, it can't avoid DELAY_CHKPT_START.
6638 : * This function, however, likely could avoid it with the following order
6639 : * of operations: MarkBufferDirty(), XLogInsert(), memcpy(). Opt to use
6640 : * DELAY_CHKPT_START here, too, as a way to have fewer distinct code
6641 : * patterns to analyze. Inplace update isn't so frequent that it should
6642 : * pursue the small optimization of skipping DELAY_CHKPT_START.
6643 : */
6644 : Assert((MyProc->delayChkptFlags & DELAY_CHKPT_START) == 0);
6645 83742 : START_CRIT_SECTION();
6646 83742 : MyProc->delayChkptFlags |= DELAY_CHKPT_START;
6647 :
6648 : /* XLOG stuff */
6649 83742 : if (RelationNeedsWAL(relation))
6650 : {
6651 : xl_heap_inplace xlrec;
6652 : PGAlignedBlock copied_buffer;
6653 83738 : char *origdata = (char *) BufferGetBlock(buffer);
6654 83738 : Page page = BufferGetPage(buffer);
6655 83738 : uint16 lower = ((PageHeader) page)->pd_lower;
6656 83738 : uint16 upper = ((PageHeader) page)->pd_upper;
6657 : uintptr_t dst_offset_in_block;
6658 : RelFileLocator rlocator;
6659 : ForkNumber forkno;
6660 : BlockNumber blkno;
6661 : XLogRecPtr recptr;
6662 :
6663 83738 : xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
6664 83738 : xlrec.dbId = MyDatabaseId;
6665 83738 : xlrec.tsId = MyDatabaseTableSpace;
6666 83738 : xlrec.relcacheInitFileInval = RelcacheInitFileInval;
6667 83738 : xlrec.nmsgs = nmsgs;
6668 :
6669 83738 : XLogBeginInsert();
6670 83738 : XLogRegisterData(&xlrec, MinSizeOfHeapInplace);
6671 83738 : if (nmsgs != 0)
6672 43496 : XLogRegisterData(invalMessages,
6673 : nmsgs * sizeof(SharedInvalidationMessage));
6674 :
6675 : /* register block matching what buffer will look like after changes */
6676 83738 : memcpy(copied_buffer.data, origdata, lower);
6677 83738 : memcpy(copied_buffer.data + upper, origdata + upper, BLCKSZ - upper);
6678 83738 : dst_offset_in_block = dst - origdata;
6679 83738 : memcpy(copied_buffer.data + dst_offset_in_block, src, newlen);
6680 83738 : BufferGetTag(buffer, &rlocator, &forkno, &blkno);
6681 : Assert(forkno == MAIN_FORKNUM);
6682 83738 : XLogRegisterBlock(0, &rlocator, forkno, blkno, copied_buffer.data,
6683 : REGBUF_STANDARD);
6684 83738 : XLogRegisterBufData(0, src, newlen);
6685 :
6686 : /* inplace updates aren't decoded atm, don't log the origin */
6687 :
6688 83738 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_INPLACE);
6689 :
6690 83738 : PageSetLSN(page, recptr);
6691 : }
6692 :
6693 83742 : memcpy(dst, src, newlen);
6694 :
6695 83742 : MarkBufferDirty(buffer);
6696 :
6697 83742 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6698 :
6699 : /*
6700 : * Send invalidations to shared queue. SearchSysCacheLocked1() assumes we
6701 : * do this before UnlockTuple().
6702 : */
6703 83742 : AtInplace_Inval();
6704 :
6705 83742 : MyProc->delayChkptFlags &= ~DELAY_CHKPT_START;
6706 83742 : END_CRIT_SECTION();
6707 83742 : UnlockTuple(relation, &tuple->t_self, InplaceUpdateTupleLock);
6708 :
6709 83742 : AcceptInvalidationMessages(); /* local processing of just-sent inval */
6710 :
6711 : /*
6712 : * Queue a transactional inval, for logical decoding and for third-party
6713 : * code that might have been relying on it since long before inplace
6714 : * update adopted immediate invalidation. See README.tuplock section
6715 : * "Reading inplace-updated columns" for logical decoding details.
6716 : */
6717 83742 : if (!IsBootstrapProcessingMode())
6718 68799 : CacheInvalidateHeapTuple(relation, tuple, NULL);
6719 83742 : }
6720 :
6721 : /*
6722 : * heap_inplace_unlock - reverse of heap_inplace_lock
6723 : */
6724 : void
6725 126497 : heap_inplace_unlock(Relation relation,
6726 : HeapTuple oldtup, Buffer buffer)
6727 : {
6728 126497 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6729 126497 : UnlockTuple(relation, &oldtup->t_self, InplaceUpdateTupleLock);
6730 126497 : ForgetInplace_Inval();
6731 126497 : }
6732 :
6733 : #define FRM_NOOP 0x0001
6734 : #define FRM_INVALIDATE_XMAX 0x0002
6735 : #define FRM_RETURN_IS_XID 0x0004
6736 : #define FRM_RETURN_IS_MULTI 0x0008
6737 : #define FRM_MARK_COMMITTED 0x0010
6738 :
6739 : /*
6740 : * FreezeMultiXactId
6741 : * Determine what to do during freezing when a tuple is marked by a
6742 : * MultiXactId.
6743 : *
6744 : * "flags" is an output value; it's used to tell caller what to do on return.
6745 : * "pagefrz" is an input/output value, used to manage page level freezing.
6746 : *
6747 : * Possible values that we can set in "flags":
6748 : * FRM_NOOP
6749 : * don't do anything -- keep existing Xmax
6750 : * FRM_INVALIDATE_XMAX
6751 : * mark Xmax as InvalidTransactionId and set XMAX_INVALID flag.
6752 : * FRM_RETURN_IS_XID
6753 : * The Xid return value is a single update Xid to set as xmax.
6754 : * FRM_MARK_COMMITTED
6755 : * Xmax can be marked as HEAP_XMAX_COMMITTED
6756 : * FRM_RETURN_IS_MULTI
6757 : * The return value is a new MultiXactId to set as new Xmax.
6758 : * (caller must obtain proper infomask bits using GetMultiXactIdHintBits)
6759 : *
6760 : * Caller delegates control of page freezing to us. In practice we always
6761 : * force freezing of caller's page unless FRM_NOOP processing is indicated.
6762 : * We help caller ensure that XIDs < FreezeLimit and MXIDs < MultiXactCutoff
6763 : * can never be left behind. We freely choose when and how to process each
6764 : * Multi, without ever violating the cutoff postconditions for freezing.
6765 : *
6766 : * It's useful to remove Multis on a proactive timeline (relative to freezing
6767 : * XIDs) to keep MultiXact member SLRU buffer misses to a minimum. It can also
6768 : * be cheaper in the short run, for us, since we too can avoid SLRU buffer
6769 : * misses through eager processing.
6770 : *
6771 : * NB: Creates a _new_ MultiXactId when FRM_RETURN_IS_MULTI is set, though only
6772 : * when FreezeLimit and/or MultiXactCutoff cutoffs leave us with no choice.
6773 : * This can usually be put off, which is usually enough to avoid it altogether.
6774 : * Allocating new multis during VACUUM should be avoided on general principle;
6775 : * only VACUUM can advance relminmxid, so allocating new Multis here comes with
6776 : * its own special risks.
6777 : *
6778 : * NB: Caller must maintain "no freeze" NewRelfrozenXid/NewRelminMxid trackers
6779 : * using heap_tuple_should_freeze when we haven't forced page-level freezing.
6780 : *
6781 : * NB: Caller should avoid needlessly calling heap_tuple_should_freeze when we
6782 : * have already forced page-level freezing, since that might incur the same
6783 : * SLRU buffer misses that we specifically intended to avoid by freezing.
6784 : */
6785 : static TransactionId
6786 7 : FreezeMultiXactId(MultiXactId multi, uint16 t_infomask,
6787 : const struct VacuumCutoffs *cutoffs, uint16 *flags,
6788 : HeapPageFreeze *pagefrz)
6789 : {
6790 : TransactionId newxmax;
6791 : MultiXactMember *members;
6792 : int nmembers;
6793 : bool need_replace;
6794 : int nnewmembers;
6795 : MultiXactMember *newmembers;
6796 : bool has_lockers;
6797 : TransactionId update_xid;
6798 : bool update_committed;
6799 : TransactionId FreezePageRelfrozenXid;
6800 :
6801 7 : *flags = 0;
6802 :
6803 : /* We should only be called in Multis */
6804 : Assert(t_infomask & HEAP_XMAX_IS_MULTI);
6805 :
6806 14 : if (!MultiXactIdIsValid(multi) ||
6807 7 : HEAP_LOCKED_UPGRADED(t_infomask))
6808 : {
6809 0 : *flags |= FRM_INVALIDATE_XMAX;
6810 0 : pagefrz->freeze_required = true;
6811 0 : return InvalidTransactionId;
6812 : }
6813 7 : else if (MultiXactIdPrecedes(multi, cutoffs->relminmxid))
6814 0 : ereport(ERROR,
6815 : (errcode(ERRCODE_DATA_CORRUPTED),
6816 : errmsg_internal("found multixact %u from before relminmxid %u",
6817 : multi, cutoffs->relminmxid)));
6818 7 : else if (MultiXactIdPrecedes(multi, cutoffs->OldestMxact))
6819 : {
6820 : TransactionId update_xact;
6821 :
6822 : /*
6823 : * This old multi cannot possibly have members still running, but
6824 : * verify just in case. If it was a locker only, it can be removed
6825 : * without any further consideration; but if it contained an update,
6826 : * we might need to preserve it.
6827 : */
6828 5 : if (MultiXactIdIsRunning(multi,
6829 5 : HEAP_XMAX_IS_LOCKED_ONLY(t_infomask)))
6830 0 : ereport(ERROR,
6831 : (errcode(ERRCODE_DATA_CORRUPTED),
6832 : errmsg_internal("multixact %u from before multi freeze cutoff %u found to be still running",
6833 : multi, cutoffs->OldestMxact)));
6834 :
6835 5 : if (HEAP_XMAX_IS_LOCKED_ONLY(t_infomask))
6836 : {
6837 5 : *flags |= FRM_INVALIDATE_XMAX;
6838 5 : pagefrz->freeze_required = true;
6839 5 : return InvalidTransactionId;
6840 : }
6841 :
6842 : /* replace multi with single XID for its updater? */
6843 0 : update_xact = MultiXactIdGetUpdateXid(multi, t_infomask);
6844 0 : if (TransactionIdPrecedes(update_xact, cutoffs->relfrozenxid))
6845 0 : ereport(ERROR,
6846 : (errcode(ERRCODE_DATA_CORRUPTED),
6847 : errmsg_internal("multixact %u contains update XID %u from before relfrozenxid %u",
6848 : multi, update_xact,
6849 : cutoffs->relfrozenxid)));
6850 0 : else if (TransactionIdPrecedes(update_xact, cutoffs->OldestXmin))
6851 : {
6852 : /*
6853 : * Updater XID has to have aborted (otherwise the tuple would have
6854 : * been pruned away instead, since updater XID is < OldestXmin).
6855 : * Just remove xmax.
6856 : */
6857 0 : if (TransactionIdDidCommit(update_xact))
6858 0 : ereport(ERROR,
6859 : (errcode(ERRCODE_DATA_CORRUPTED),
6860 : errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
6861 : multi, update_xact,
6862 : cutoffs->OldestXmin)));
6863 0 : *flags |= FRM_INVALIDATE_XMAX;
6864 0 : pagefrz->freeze_required = true;
6865 0 : return InvalidTransactionId;
6866 : }
6867 :
6868 : /* Have to keep updater XID as new xmax */
6869 0 : *flags |= FRM_RETURN_IS_XID;
6870 0 : pagefrz->freeze_required = true;
6871 0 : return update_xact;
6872 : }
6873 :
6874 : /*
6875 : * Some member(s) of this Multi may be below FreezeLimit xid cutoff, so we
6876 : * need to walk the whole members array to figure out what to do, if
6877 : * anything.
6878 : */
6879 : nmembers =
6880 2 : GetMultiXactIdMembers(multi, &members, false,
6881 2 : HEAP_XMAX_IS_LOCKED_ONLY(t_infomask));
6882 2 : if (nmembers <= 0)
6883 : {
6884 : /* Nothing worth keeping */
6885 0 : *flags |= FRM_INVALIDATE_XMAX;
6886 0 : pagefrz->freeze_required = true;
6887 0 : return InvalidTransactionId;
6888 : }
6889 :
6890 : /*
6891 : * The FRM_NOOP case is the only case where we might need to ratchet back
6892 : * FreezePageRelfrozenXid or FreezePageRelminMxid. It is also the only
6893 : * case where our caller might ratchet back its NoFreezePageRelfrozenXid
6894 : * or NoFreezePageRelminMxid "no freeze" trackers to deal with a multi.
6895 : * FRM_NOOP handling should result in the NewRelfrozenXid/NewRelminMxid
6896 : * trackers managed by VACUUM being ratcheting back by xmax to the degree
6897 : * required to make it safe to leave xmax undisturbed, independent of
6898 : * whether or not page freezing is triggered somewhere else.
6899 : *
6900 : * Our policy is to force freezing in every case other than FRM_NOOP,
6901 : * which obviates the need to maintain either set of trackers, anywhere.
6902 : * Every other case will reliably execute a freeze plan for xmax that
6903 : * either replaces xmax with an XID/MXID >= OldestXmin/OldestMxact, or
6904 : * sets xmax to an InvalidTransactionId XID, rendering xmax fully frozen.
6905 : * (VACUUM's NewRelfrozenXid/NewRelminMxid trackers are initialized with
6906 : * OldestXmin/OldestMxact, so later values never need to be tracked here.)
6907 : */
6908 2 : need_replace = false;
6909 2 : FreezePageRelfrozenXid = pagefrz->FreezePageRelfrozenXid;
6910 4 : for (int i = 0; i < nmembers; i++)
6911 : {
6912 3 : TransactionId xid = members[i].xid;
6913 :
6914 : Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
6915 :
6916 3 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
6917 : {
6918 : /* Can't violate the FreezeLimit postcondition */
6919 1 : need_replace = true;
6920 1 : break;
6921 : }
6922 2 : if (TransactionIdPrecedes(xid, FreezePageRelfrozenXid))
6923 0 : FreezePageRelfrozenXid = xid;
6924 : }
6925 :
6926 : /* Can't violate the MultiXactCutoff postcondition, either */
6927 2 : if (!need_replace)
6928 1 : need_replace = MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff);
6929 :
6930 2 : if (!need_replace)
6931 : {
6932 : /*
6933 : * vacuumlazy.c might ratchet back NewRelminMxid, NewRelfrozenXid, or
6934 : * both together to make it safe to retain this particular multi after
6935 : * freezing its page
6936 : */
6937 1 : *flags |= FRM_NOOP;
6938 1 : pagefrz->FreezePageRelfrozenXid = FreezePageRelfrozenXid;
6939 1 : if (MultiXactIdPrecedes(multi, pagefrz->FreezePageRelminMxid))
6940 0 : pagefrz->FreezePageRelminMxid = multi;
6941 1 : pfree(members);
6942 1 : return multi;
6943 : }
6944 :
6945 : /*
6946 : * Do a more thorough second pass over the multi to figure out which
6947 : * member XIDs actually need to be kept. Checking the precise status of
6948 : * individual members might even show that we don't need to keep anything.
6949 : * That is quite possible even though the Multi must be >= OldestMxact,
6950 : * since our second pass only keeps member XIDs when it's truly necessary;
6951 : * even member XIDs >= OldestXmin often won't be kept by second pass.
6952 : */
6953 1 : nnewmembers = 0;
6954 1 : newmembers = palloc_array(MultiXactMember, nmembers);
6955 1 : has_lockers = false;
6956 1 : update_xid = InvalidTransactionId;
6957 1 : update_committed = false;
6958 :
6959 : /*
6960 : * Determine whether to keep each member xid, or to ignore it instead
6961 : */
6962 3 : for (int i = 0; i < nmembers; i++)
6963 : {
6964 2 : TransactionId xid = members[i].xid;
6965 2 : MultiXactStatus mstatus = members[i].status;
6966 :
6967 : Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
6968 :
6969 2 : if (!ISUPDATE_from_mxstatus(mstatus))
6970 : {
6971 : /*
6972 : * Locker XID (not updater XID). We only keep lockers that are
6973 : * still running.
6974 : */
6975 4 : if (TransactionIdIsCurrentTransactionId(xid) ||
6976 2 : TransactionIdIsInProgress(xid))
6977 : {
6978 1 : if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
6979 0 : ereport(ERROR,
6980 : (errcode(ERRCODE_DATA_CORRUPTED),
6981 : errmsg_internal("multixact %u contains running locker XID %u from before removable cutoff %u",
6982 : multi, xid,
6983 : cutoffs->OldestXmin)));
6984 1 : newmembers[nnewmembers++] = members[i];
6985 1 : has_lockers = true;
6986 : }
6987 :
6988 2 : continue;
6989 : }
6990 :
6991 : /*
6992 : * Updater XID (not locker XID). Should we keep it?
6993 : *
6994 : * Since the tuple wasn't totally removed when vacuum pruned, the
6995 : * update Xid cannot possibly be older than OldestXmin cutoff unless
6996 : * the updater XID aborted. If the updater transaction is known
6997 : * aborted or crashed then it's okay to ignore it, otherwise not.
6998 : *
6999 : * In any case the Multi should never contain two updaters, whatever
7000 : * their individual commit status. Check for that first, in passing.
7001 : */
7002 0 : if (TransactionIdIsValid(update_xid))
7003 0 : ereport(ERROR,
7004 : (errcode(ERRCODE_DATA_CORRUPTED),
7005 : errmsg_internal("multixact %u has two or more updating members",
7006 : multi),
7007 : errdetail_internal("First updater XID=%u second updater XID=%u.",
7008 : update_xid, xid)));
7009 :
7010 : /*
7011 : * As with all tuple visibility routines, it's critical to test
7012 : * TransactionIdIsInProgress before TransactionIdDidCommit, because of
7013 : * race conditions explained in detail in heapam_visibility.c.
7014 : */
7015 0 : if (TransactionIdIsCurrentTransactionId(xid) ||
7016 0 : TransactionIdIsInProgress(xid))
7017 0 : update_xid = xid;
7018 0 : else if (TransactionIdDidCommit(xid))
7019 : {
7020 : /*
7021 : * The transaction committed, so we can tell caller to set
7022 : * HEAP_XMAX_COMMITTED. (We can only do this because we know the
7023 : * transaction is not running.)
7024 : */
7025 0 : update_committed = true;
7026 0 : update_xid = xid;
7027 : }
7028 : else
7029 : {
7030 : /*
7031 : * Not in progress, not committed -- must be aborted or crashed;
7032 : * we can ignore it.
7033 : */
7034 0 : continue;
7035 : }
7036 :
7037 : /*
7038 : * We determined that updater must be kept -- add it to pending new
7039 : * members list
7040 : */
7041 0 : if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
7042 0 : ereport(ERROR,
7043 : (errcode(ERRCODE_DATA_CORRUPTED),
7044 : errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
7045 : multi, xid, cutoffs->OldestXmin)));
7046 0 : newmembers[nnewmembers++] = members[i];
7047 : }
7048 :
7049 1 : pfree(members);
7050 :
7051 : /*
7052 : * Determine what to do with caller's multi based on information gathered
7053 : * during our second pass
7054 : */
7055 1 : if (nnewmembers == 0)
7056 : {
7057 : /* Nothing worth keeping */
7058 0 : *flags |= FRM_INVALIDATE_XMAX;
7059 0 : newxmax = InvalidTransactionId;
7060 : }
7061 1 : else if (TransactionIdIsValid(update_xid) && !has_lockers)
7062 : {
7063 : /*
7064 : * If there's a single member and it's an update, pass it back alone
7065 : * without creating a new Multi. (XXX we could do this when there's a
7066 : * single remaining locker, too, but that would complicate the API too
7067 : * much; moreover, the case with the single updater is more
7068 : * interesting, because those are longer-lived.)
7069 : */
7070 : Assert(nnewmembers == 1);
7071 0 : *flags |= FRM_RETURN_IS_XID;
7072 0 : if (update_committed)
7073 0 : *flags |= FRM_MARK_COMMITTED;
7074 0 : newxmax = update_xid;
7075 : }
7076 : else
7077 : {
7078 : /*
7079 : * Create a new multixact with the surviving members of the previous
7080 : * one, to set as new Xmax in the tuple
7081 : */
7082 1 : newxmax = MultiXactIdCreateFromMembers(nnewmembers, newmembers);
7083 1 : *flags |= FRM_RETURN_IS_MULTI;
7084 : }
7085 :
7086 1 : pfree(newmembers);
7087 :
7088 1 : pagefrz->freeze_required = true;
7089 1 : return newxmax;
7090 : }
7091 :
7092 : /*
7093 : * heap_prepare_freeze_tuple
7094 : *
7095 : * Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
7096 : * are older than the OldestXmin and/or OldestMxact freeze cutoffs. If so,
7097 : * setup enough state (in the *frz output argument) to enable caller to
7098 : * process this tuple as part of freezing its page, and return true. Return
7099 : * false if nothing can be changed about the tuple right now.
7100 : *
7101 : * Also sets *totally_frozen to true if the tuple will be totally frozen once
7102 : * caller executes returned freeze plan (or if the tuple was already totally
7103 : * frozen by an earlier VACUUM). This indicates that there are no remaining
7104 : * XIDs or MultiXactIds that will need to be processed by a future VACUUM.
7105 : *
7106 : * VACUUM caller must assemble HeapTupleFreeze freeze plan entries for every
7107 : * tuple that we returned true for, and then execute freezing. Caller must
7108 : * initialize pagefrz fields for page as a whole before first call here for
7109 : * each heap page.
7110 : *
7111 : * VACUUM caller decides on whether or not to freeze the page as a whole.
7112 : * We'll often prepare freeze plans for a page that caller just discards.
7113 : * However, VACUUM doesn't always get to make a choice; it must freeze when
7114 : * pagefrz.freeze_required is set, to ensure that any XIDs < FreezeLimit (and
7115 : * MXIDs < MultiXactCutoff) can never be left behind. We help to make sure
7116 : * that VACUUM always follows that rule.
7117 : *
7118 : * We sometimes force freezing of xmax MultiXactId values long before it is
7119 : * strictly necessary to do so just to ensure the FreezeLimit postcondition.
7120 : * It's worth processing MultiXactIds proactively when it is cheap to do so,
7121 : * and it's convenient to make that happen by piggy-backing it on the "force
7122 : * freezing" mechanism. Conversely, we sometimes delay freezing MultiXactIds
7123 : * because it is expensive right now (though only when it's still possible to
7124 : * do so without violating the FreezeLimit/MultiXactCutoff postcondition).
7125 : *
7126 : * It is assumed that the caller has checked the tuple with
7127 : * HeapTupleSatisfiesVacuum() and determined that it is not HEAPTUPLE_DEAD
7128 : * (else we should be removing the tuple, not freezing it).
7129 : *
7130 : * NB: This function has side effects: it might allocate a new MultiXactId.
7131 : * It will be set as tuple's new xmax when our *frz output is processed within
7132 : * heap_execute_freeze_tuple later on. If the tuple is in a shared buffer
7133 : * then caller had better have an exclusive lock on it already.
7134 : */
7135 : bool
7136 25379299 : heap_prepare_freeze_tuple(HeapTupleHeader tuple,
7137 : const struct VacuumCutoffs *cutoffs,
7138 : HeapPageFreeze *pagefrz,
7139 : HeapTupleFreeze *frz, bool *totally_frozen)
7140 : {
7141 25379299 : bool xmin_already_frozen = false,
7142 25379299 : xmax_already_frozen = false;
7143 25379299 : bool freeze_xmin = false,
7144 25379299 : replace_xvac = false,
7145 25379299 : replace_xmax = false,
7146 25379299 : freeze_xmax = false;
7147 : TransactionId xid;
7148 :
7149 25379299 : frz->xmax = HeapTupleHeaderGetRawXmax(tuple);
7150 25379299 : frz->t_infomask2 = tuple->t_infomask2;
7151 25379299 : frz->t_infomask = tuple->t_infomask;
7152 25379299 : frz->frzflags = 0;
7153 25379299 : frz->checkflags = 0;
7154 :
7155 : /*
7156 : * Process xmin, while keeping track of whether it's already frozen, or
7157 : * will become frozen iff our freeze plan is executed by caller (could be
7158 : * neither).
7159 : */
7160 25379299 : xid = HeapTupleHeaderGetXmin(tuple);
7161 25379299 : if (!TransactionIdIsNormal(xid))
7162 22491437 : xmin_already_frozen = true;
7163 : else
7164 : {
7165 2887862 : if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
7166 0 : ereport(ERROR,
7167 : (errcode(ERRCODE_DATA_CORRUPTED),
7168 : errmsg_internal("found xmin %u from before relfrozenxid %u",
7169 : xid, cutoffs->relfrozenxid)));
7170 :
7171 : /* Will set freeze_xmin flags in freeze plan below */
7172 2887862 : freeze_xmin = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
7173 :
7174 : /* Verify that xmin committed if and when freeze plan is executed */
7175 2887862 : if (freeze_xmin)
7176 2277673 : frz->checkflags |= HEAP_FREEZE_CHECK_XMIN_COMMITTED;
7177 : }
7178 :
7179 : /*
7180 : * Old-style VACUUM FULL is gone, but we have to process xvac for as long
7181 : * as we support having MOVED_OFF/MOVED_IN tuples in the database
7182 : */
7183 25379299 : xid = HeapTupleHeaderGetXvac(tuple);
7184 25379299 : if (TransactionIdIsNormal(xid))
7185 : {
7186 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7187 : Assert(TransactionIdPrecedes(xid, cutoffs->OldestXmin));
7188 :
7189 : /*
7190 : * For Xvac, we always freeze proactively. This allows totally_frozen
7191 : * tracking to ignore xvac.
7192 : */
7193 0 : replace_xvac = pagefrz->freeze_required = true;
7194 :
7195 : /* Will set replace_xvac flags in freeze plan below */
7196 : }
7197 :
7198 : /* Now process xmax */
7199 25379299 : xid = frz->xmax;
7200 25379299 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7201 : {
7202 : /* Raw xmax is a MultiXactId */
7203 : TransactionId newxmax;
7204 : uint16 flags;
7205 :
7206 : /*
7207 : * We will either remove xmax completely (in the "freeze_xmax" path),
7208 : * process xmax by replacing it (in the "replace_xmax" path), or
7209 : * perform no-op xmax processing. The only constraint is that the
7210 : * FreezeLimit/MultiXactCutoff postcondition must never be violated.
7211 : */
7212 7 : newxmax = FreezeMultiXactId(xid, tuple->t_infomask, cutoffs,
7213 : &flags, pagefrz);
7214 :
7215 7 : if (flags & FRM_NOOP)
7216 : {
7217 : /*
7218 : * xmax is a MultiXactId, and nothing about it changes for now.
7219 : * This is the only case where 'freeze_required' won't have been
7220 : * set for us by FreezeMultiXactId, as well as the only case where
7221 : * neither freeze_xmax nor replace_xmax are set (given a multi).
7222 : *
7223 : * This is a no-op, but the call to FreezeMultiXactId might have
7224 : * ratcheted back NewRelfrozenXid and/or NewRelminMxid trackers
7225 : * for us (the "freeze page" variants, specifically). That'll
7226 : * make it safe for our caller to freeze the page later on, while
7227 : * leaving this particular xmax undisturbed.
7228 : *
7229 : * FreezeMultiXactId is _not_ responsible for the "no freeze"
7230 : * NewRelfrozenXid/NewRelminMxid trackers, though -- that's our
7231 : * job. A call to heap_tuple_should_freeze for this same tuple
7232 : * will take place below if 'freeze_required' isn't set already.
7233 : * (This repeats work from FreezeMultiXactId, but allows "no
7234 : * freeze" tracker maintenance to happen in only one place.)
7235 : */
7236 : Assert(!MultiXactIdPrecedes(newxmax, cutoffs->MultiXactCutoff));
7237 : Assert(MultiXactIdIsValid(newxmax) && xid == newxmax);
7238 : }
7239 6 : else if (flags & FRM_RETURN_IS_XID)
7240 : {
7241 : /*
7242 : * xmax will become an updater Xid (original MultiXact's updater
7243 : * member Xid will be carried forward as a simple Xid in Xmax).
7244 : */
7245 : Assert(!TransactionIdPrecedes(newxmax, cutoffs->OldestXmin));
7246 :
7247 : /*
7248 : * NB -- some of these transformations are only valid because we
7249 : * know the return Xid is a tuple updater (i.e. not merely a
7250 : * locker.) Also note that the only reason we don't explicitly
7251 : * worry about HEAP_KEYS_UPDATED is because it lives in
7252 : * t_infomask2 rather than t_infomask.
7253 : */
7254 0 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7255 0 : frz->xmax = newxmax;
7256 0 : if (flags & FRM_MARK_COMMITTED)
7257 0 : frz->t_infomask |= HEAP_XMAX_COMMITTED;
7258 0 : replace_xmax = true;
7259 : }
7260 6 : else if (flags & FRM_RETURN_IS_MULTI)
7261 : {
7262 : uint16 newbits;
7263 : uint16 newbits2;
7264 :
7265 : /*
7266 : * xmax is an old MultiXactId that we have to replace with a new
7267 : * MultiXactId, to carry forward two or more original member XIDs.
7268 : */
7269 : Assert(!MultiXactIdPrecedes(newxmax, cutoffs->OldestMxact));
7270 :
7271 : /*
7272 : * We can't use GetMultiXactIdHintBits directly on the new multi
7273 : * here; that routine initializes the masks to all zeroes, which
7274 : * would lose other bits we need. Doing it this way ensures all
7275 : * unrelated bits remain untouched.
7276 : */
7277 1 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7278 1 : frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
7279 1 : GetMultiXactIdHintBits(newxmax, &newbits, &newbits2);
7280 1 : frz->t_infomask |= newbits;
7281 1 : frz->t_infomask2 |= newbits2;
7282 1 : frz->xmax = newxmax;
7283 1 : replace_xmax = true;
7284 : }
7285 : else
7286 : {
7287 : /*
7288 : * Freeze plan for tuple "freezes xmax" in the strictest sense:
7289 : * it'll leave nothing in xmax (neither an Xid nor a MultiXactId).
7290 : */
7291 : Assert(flags & FRM_INVALIDATE_XMAX);
7292 : Assert(!TransactionIdIsValid(newxmax));
7293 :
7294 : /* Will set freeze_xmax flags in freeze plan below */
7295 5 : freeze_xmax = true;
7296 : }
7297 :
7298 : /* MultiXactId processing forces freezing (barring FRM_NOOP case) */
7299 : Assert(pagefrz->freeze_required || (!freeze_xmax && !replace_xmax));
7300 : }
7301 25379292 : else if (TransactionIdIsNormal(xid))
7302 : {
7303 : /* Raw xmax is normal XID */
7304 8001579 : if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
7305 0 : ereport(ERROR,
7306 : (errcode(ERRCODE_DATA_CORRUPTED),
7307 : errmsg_internal("found xmax %u from before relfrozenxid %u",
7308 : xid, cutoffs->relfrozenxid)));
7309 :
7310 : /* Will set freeze_xmax flags in freeze plan below */
7311 8001579 : freeze_xmax = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
7312 :
7313 : /*
7314 : * Verify that xmax aborted if and when freeze plan is executed,
7315 : * provided it's from an update. (A lock-only xmax can be removed
7316 : * independent of this, since the lock is released at xact end.)
7317 : */
7318 8001579 : if (freeze_xmax && !HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask))
7319 599 : frz->checkflags |= HEAP_FREEZE_CHECK_XMAX_ABORTED;
7320 : }
7321 17377713 : else if (!TransactionIdIsValid(xid))
7322 : {
7323 : /* Raw xmax is InvalidTransactionId XID */
7324 : Assert((tuple->t_infomask & HEAP_XMAX_IS_MULTI) == 0);
7325 17377713 : xmax_already_frozen = true;
7326 : }
7327 : else
7328 0 : ereport(ERROR,
7329 : (errcode(ERRCODE_DATA_CORRUPTED),
7330 : errmsg_internal("found raw xmax %u (infomask 0x%04x) not invalid and not multi",
7331 : xid, tuple->t_infomask)));
7332 :
7333 25379299 : if (freeze_xmin)
7334 : {
7335 : Assert(!xmin_already_frozen);
7336 :
7337 2277673 : frz->t_infomask |= HEAP_XMIN_FROZEN;
7338 : }
7339 25379299 : if (replace_xvac)
7340 : {
7341 : /*
7342 : * If a MOVED_OFF tuple is not dead, the xvac transaction must have
7343 : * failed; whereas a non-dead MOVED_IN tuple must mean the xvac
7344 : * transaction succeeded.
7345 : */
7346 : Assert(pagefrz->freeze_required);
7347 0 : if (tuple->t_infomask & HEAP_MOVED_OFF)
7348 0 : frz->frzflags |= XLH_INVALID_XVAC;
7349 : else
7350 0 : frz->frzflags |= XLH_FREEZE_XVAC;
7351 : }
7352 : if (replace_xmax)
7353 : {
7354 : Assert(!xmax_already_frozen && !freeze_xmax);
7355 : Assert(pagefrz->freeze_required);
7356 :
7357 : /* Already set replace_xmax flags in freeze plan earlier */
7358 : }
7359 25379299 : if (freeze_xmax)
7360 : {
7361 : Assert(!xmax_already_frozen && !replace_xmax);
7362 :
7363 1479 : frz->xmax = InvalidTransactionId;
7364 :
7365 : /*
7366 : * The tuple might be marked either XMAX_INVALID or XMAX_COMMITTED +
7367 : * LOCKED. Normalize to INVALID just to be sure no one gets confused.
7368 : * Also get rid of the HEAP_KEYS_UPDATED bit.
7369 : */
7370 1479 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7371 1479 : frz->t_infomask |= HEAP_XMAX_INVALID;
7372 1479 : frz->t_infomask2 &= ~HEAP_HOT_UPDATED;
7373 1479 : frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
7374 : }
7375 :
7376 : /*
7377 : * Determine if this tuple is already totally frozen, or will become
7378 : * totally frozen (provided caller executes freeze plans for the page)
7379 : */
7380 50146930 : *totally_frozen = ((freeze_xmin || xmin_already_frozen) &&
7381 24767631 : (freeze_xmax || xmax_already_frozen));
7382 :
7383 25379299 : if (!pagefrz->freeze_required && !(xmin_already_frozen &&
7384 : xmax_already_frozen))
7385 : {
7386 : /*
7387 : * So far no previous tuple from the page made freezing mandatory.
7388 : * Does this tuple force caller to freeze the entire page?
7389 : */
7390 9380121 : pagefrz->freeze_required =
7391 9380121 : heap_tuple_should_freeze(tuple, cutoffs,
7392 : &pagefrz->NoFreezePageRelfrozenXid,
7393 : &pagefrz->NoFreezePageRelminMxid);
7394 : }
7395 :
7396 : /* Tell caller if this tuple has a usable freeze plan set in *frz */
7397 25379299 : return freeze_xmin || replace_xvac || replace_xmax || freeze_xmax;
7398 : }
7399 :
7400 : /*
7401 : * Perform xmin/xmax XID status sanity checks before actually executing freeze
7402 : * plans.
7403 : *
7404 : * heap_prepare_freeze_tuple doesn't perform these checks directly because
7405 : * pg_xact lookups are relatively expensive. They shouldn't be repeated by
7406 : * successive VACUUMs that each decide against freezing the same page.
7407 : */
7408 : void
7409 22255 : heap_pre_freeze_checks(Buffer buffer,
7410 : HeapTupleFreeze *tuples, int ntuples)
7411 : {
7412 22255 : Page page = BufferGetPage(buffer);
7413 :
7414 977018 : for (int i = 0; i < ntuples; i++)
7415 : {
7416 954763 : HeapTupleFreeze *frz = tuples + i;
7417 954763 : ItemId itemid = PageGetItemId(page, frz->offset);
7418 : HeapTupleHeader htup;
7419 :
7420 954763 : htup = (HeapTupleHeader) PageGetItem(page, itemid);
7421 :
7422 : /* Deliberately avoid relying on tuple hint bits here */
7423 954763 : if (frz->checkflags & HEAP_FREEZE_CHECK_XMIN_COMMITTED)
7424 : {
7425 954762 : TransactionId xmin = HeapTupleHeaderGetRawXmin(htup);
7426 :
7427 : Assert(!HeapTupleHeaderXminFrozen(htup));
7428 954762 : if (unlikely(!TransactionIdDidCommit(xmin)))
7429 0 : ereport(ERROR,
7430 : (errcode(ERRCODE_DATA_CORRUPTED),
7431 : errmsg_internal("uncommitted xmin %u needs to be frozen",
7432 : xmin)));
7433 : }
7434 :
7435 : /*
7436 : * TransactionIdDidAbort won't work reliably in the presence of XIDs
7437 : * left behind by transactions that were in progress during a crash,
7438 : * so we can only check that xmax didn't commit
7439 : */
7440 954763 : if (frz->checkflags & HEAP_FREEZE_CHECK_XMAX_ABORTED)
7441 : {
7442 122 : TransactionId xmax = HeapTupleHeaderGetRawXmax(htup);
7443 :
7444 : Assert(TransactionIdIsNormal(xmax));
7445 122 : if (unlikely(TransactionIdDidCommit(xmax)))
7446 0 : ereport(ERROR,
7447 : (errcode(ERRCODE_DATA_CORRUPTED),
7448 : errmsg_internal("cannot freeze committed xmax %u",
7449 : xmax)));
7450 : }
7451 : }
7452 22255 : }
7453 :
7454 : /*
7455 : * Helper which executes freezing of one or more heap tuples on a page on
7456 : * behalf of caller. Caller passes an array of tuple plans from
7457 : * heap_prepare_freeze_tuple. Caller must set 'offset' in each plan for us.
7458 : * Must be called in a critical section that also marks the buffer dirty and,
7459 : * if needed, emits WAL.
7460 : */
7461 : void
7462 22255 : heap_freeze_prepared_tuples(Buffer buffer, HeapTupleFreeze *tuples, int ntuples)
7463 : {
7464 22255 : Page page = BufferGetPage(buffer);
7465 :
7466 977018 : for (int i = 0; i < ntuples; i++)
7467 : {
7468 954763 : HeapTupleFreeze *frz = tuples + i;
7469 954763 : ItemId itemid = PageGetItemId(page, frz->offset);
7470 : HeapTupleHeader htup;
7471 :
7472 954763 : htup = (HeapTupleHeader) PageGetItem(page, itemid);
7473 954763 : heap_execute_freeze_tuple(htup, frz);
7474 : }
7475 22255 : }
7476 :
7477 : /*
7478 : * heap_freeze_tuple
7479 : * Freeze tuple in place, without WAL logging.
7480 : *
7481 : * Useful for callers like CLUSTER that perform their own WAL logging.
7482 : */
7483 : bool
7484 360402 : heap_freeze_tuple(HeapTupleHeader tuple,
7485 : TransactionId relfrozenxid, TransactionId relminmxid,
7486 : TransactionId FreezeLimit, TransactionId MultiXactCutoff)
7487 : {
7488 : HeapTupleFreeze frz;
7489 : bool do_freeze;
7490 : bool totally_frozen;
7491 : struct VacuumCutoffs cutoffs;
7492 : HeapPageFreeze pagefrz;
7493 :
7494 360402 : cutoffs.relfrozenxid = relfrozenxid;
7495 360402 : cutoffs.relminmxid = relminmxid;
7496 360402 : cutoffs.OldestXmin = FreezeLimit;
7497 360402 : cutoffs.OldestMxact = MultiXactCutoff;
7498 360402 : cutoffs.FreezeLimit = FreezeLimit;
7499 360402 : cutoffs.MultiXactCutoff = MultiXactCutoff;
7500 :
7501 360402 : pagefrz.freeze_required = true;
7502 360402 : pagefrz.FreezePageRelfrozenXid = FreezeLimit;
7503 360402 : pagefrz.FreezePageRelminMxid = MultiXactCutoff;
7504 360402 : pagefrz.NoFreezePageRelfrozenXid = FreezeLimit;
7505 360402 : pagefrz.NoFreezePageRelminMxid = MultiXactCutoff;
7506 :
7507 360402 : do_freeze = heap_prepare_freeze_tuple(tuple, &cutoffs,
7508 : &pagefrz, &frz, &totally_frozen);
7509 :
7510 : /*
7511 : * Note that because this is not a WAL-logged operation, we don't need to
7512 : * fill in the offset in the freeze record.
7513 : */
7514 :
7515 360402 : if (do_freeze)
7516 267402 : heap_execute_freeze_tuple(tuple, &frz);
7517 360402 : return do_freeze;
7518 : }
7519 :
7520 : /*
7521 : * For a given MultiXactId, return the hint bits that should be set in the
7522 : * tuple's infomask.
7523 : *
7524 : * Normally this should be called for a multixact that was just created, and
7525 : * so is on our local cache, so the GetMembers call is fast.
7526 : */
7527 : static void
7528 76803 : GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
7529 : uint16 *new_infomask2)
7530 : {
7531 : int nmembers;
7532 : MultiXactMember *members;
7533 : int i;
7534 76803 : uint16 bits = HEAP_XMAX_IS_MULTI;
7535 76803 : uint16 bits2 = 0;
7536 76803 : bool has_update = false;
7537 76803 : LockTupleMode strongest = LockTupleKeyShare;
7538 :
7539 : /*
7540 : * We only use this in multis we just created, so they cannot be values
7541 : * pre-pg_upgrade.
7542 : */
7543 76803 : nmembers = GetMultiXactIdMembers(multi, &members, false, false);
7544 :
7545 1472619 : for (i = 0; i < nmembers; i++)
7546 : {
7547 : LockTupleMode mode;
7548 :
7549 : /*
7550 : * Remember the strongest lock mode held by any member of the
7551 : * multixact.
7552 : */
7553 1395816 : mode = TUPLOCK_from_mxstatus(members[i].status);
7554 1395816 : if (mode > strongest)
7555 2888 : strongest = mode;
7556 :
7557 : /* See what other bits we need */
7558 1395816 : switch (members[i].status)
7559 : {
7560 1393398 : case MultiXactStatusForKeyShare:
7561 : case MultiXactStatusForShare:
7562 : case MultiXactStatusForNoKeyUpdate:
7563 1393398 : break;
7564 :
7565 52 : case MultiXactStatusForUpdate:
7566 52 : bits2 |= HEAP_KEYS_UPDATED;
7567 52 : break;
7568 :
7569 2356 : case MultiXactStatusNoKeyUpdate:
7570 2356 : has_update = true;
7571 2356 : break;
7572 :
7573 10 : case MultiXactStatusUpdate:
7574 10 : bits2 |= HEAP_KEYS_UPDATED;
7575 10 : has_update = true;
7576 10 : break;
7577 : }
7578 : }
7579 :
7580 76803 : if (strongest == LockTupleExclusive ||
7581 : strongest == LockTupleNoKeyExclusive)
7582 2446 : bits |= HEAP_XMAX_EXCL_LOCK;
7583 74357 : else if (strongest == LockTupleShare)
7584 439 : bits |= HEAP_XMAX_SHR_LOCK;
7585 73918 : else if (strongest == LockTupleKeyShare)
7586 73918 : bits |= HEAP_XMAX_KEYSHR_LOCK;
7587 :
7588 76803 : if (!has_update)
7589 74437 : bits |= HEAP_XMAX_LOCK_ONLY;
7590 :
7591 76803 : if (nmembers > 0)
7592 76803 : pfree(members);
7593 :
7594 76803 : *new_infomask = bits;
7595 76803 : *new_infomask2 = bits2;
7596 76803 : }
7597 :
7598 : /*
7599 : * MultiXactIdGetUpdateXid
7600 : *
7601 : * Given a multixact Xmax and corresponding infomask, which does not have the
7602 : * HEAP_XMAX_LOCK_ONLY bit set, obtain and return the Xid of the updating
7603 : * transaction.
7604 : *
7605 : * Caller is expected to check the status of the updating transaction, if
7606 : * necessary.
7607 : */
7608 : static TransactionId
7609 162018 : MultiXactIdGetUpdateXid(TransactionId xmax, uint16 t_infomask)
7610 : {
7611 162018 : TransactionId update_xact = InvalidTransactionId;
7612 : MultiXactMember *members;
7613 : int nmembers;
7614 :
7615 : Assert(!(t_infomask & HEAP_XMAX_LOCK_ONLY));
7616 : Assert(t_infomask & HEAP_XMAX_IS_MULTI);
7617 :
7618 : /*
7619 : * Since we know the LOCK_ONLY bit is not set, this cannot be a multi from
7620 : * pre-pg_upgrade.
7621 : */
7622 162018 : nmembers = GetMultiXactIdMembers(xmax, &members, false, false);
7623 :
7624 162018 : if (nmembers > 0)
7625 : {
7626 : int i;
7627 :
7628 245778 : for (i = 0; i < nmembers; i++)
7629 : {
7630 : /* Ignore lockers */
7631 245778 : if (!ISUPDATE_from_mxstatus(members[i].status))
7632 83760 : continue;
7633 :
7634 : /* there can be at most one updater */
7635 : Assert(update_xact == InvalidTransactionId);
7636 162018 : update_xact = members[i].xid;
7637 : #ifndef USE_ASSERT_CHECKING
7638 :
7639 : /*
7640 : * in an assert-enabled build, walk the whole array to ensure
7641 : * there's no other updater.
7642 : */
7643 162018 : break;
7644 : #endif
7645 : }
7646 :
7647 162018 : pfree(members);
7648 : }
7649 :
7650 162018 : return update_xact;
7651 : }
7652 :
7653 : /*
7654 : * HeapTupleGetUpdateXid
7655 : * As above, but use a HeapTupleHeader
7656 : *
7657 : * See also HeapTupleHeaderGetUpdateXid, which can be used without previously
7658 : * checking the hint bits.
7659 : */
7660 : TransactionId
7661 159880 : HeapTupleGetUpdateXid(const HeapTupleHeaderData *tup)
7662 : {
7663 159880 : return MultiXactIdGetUpdateXid(HeapTupleHeaderGetRawXmax(tup),
7664 159880 : tup->t_infomask);
7665 : }
7666 :
7667 : /*
7668 : * Does the given multixact conflict with the current transaction grabbing a
7669 : * tuple lock of the given strength?
7670 : *
7671 : * The passed infomask pairs up with the given multixact in the tuple header.
7672 : *
7673 : * If current_is_member is not NULL, it is set to 'true' if the current
7674 : * transaction is a member of the given multixact.
7675 : */
7676 : static bool
7677 218 : DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
7678 : LockTupleMode lockmode, bool *current_is_member)
7679 : {
7680 : int nmembers;
7681 : MultiXactMember *members;
7682 218 : bool result = false;
7683 218 : LOCKMODE wanted = tupleLockExtraInfo[lockmode].hwlock;
7684 :
7685 218 : if (HEAP_LOCKED_UPGRADED(infomask))
7686 0 : return false;
7687 :
7688 218 : nmembers = GetMultiXactIdMembers(multi, &members, false,
7689 218 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
7690 218 : if (nmembers >= 0)
7691 : {
7692 : int i;
7693 :
7694 2682 : for (i = 0; i < nmembers; i++)
7695 : {
7696 : TransactionId memxid;
7697 : LOCKMODE memlockmode;
7698 :
7699 2471 : if (result && (current_is_member == NULL || *current_is_member))
7700 : break;
7701 :
7702 2464 : memlockmode = LOCKMODE_from_mxstatus(members[i].status);
7703 :
7704 : /* ignore members from current xact (but track their presence) */
7705 2464 : memxid = members[i].xid;
7706 2464 : if (TransactionIdIsCurrentTransactionId(memxid))
7707 : {
7708 92 : if (current_is_member != NULL)
7709 78 : *current_is_member = true;
7710 92 : continue;
7711 : }
7712 2372 : else if (result)
7713 8 : continue;
7714 :
7715 : /* ignore members that don't conflict with the lock we want */
7716 2364 : if (!DoLockModesConflict(memlockmode, wanted))
7717 2325 : continue;
7718 :
7719 39 : if (ISUPDATE_from_mxstatus(members[i].status))
7720 : {
7721 : /* ignore aborted updaters */
7722 17 : if (TransactionIdDidAbort(memxid))
7723 1 : continue;
7724 : }
7725 : else
7726 : {
7727 : /* ignore lockers-only that are no longer in progress */
7728 22 : if (!TransactionIdIsInProgress(memxid))
7729 7 : continue;
7730 : }
7731 :
7732 : /*
7733 : * Whatever remains are either live lockers that conflict with our
7734 : * wanted lock, and updaters that are not aborted. Those conflict
7735 : * with what we want. Set up to return true, but keep going to
7736 : * look for the current transaction among the multixact members,
7737 : * if needed.
7738 : */
7739 31 : result = true;
7740 : }
7741 218 : pfree(members);
7742 : }
7743 :
7744 218 : return result;
7745 : }
7746 :
7747 : /*
7748 : * Do_MultiXactIdWait
7749 : * Actual implementation for the two functions below.
7750 : *
7751 : * 'multi', 'status' and 'infomask' indicate what to sleep on (the status is
7752 : * needed to ensure we only sleep on conflicting members, and the infomask is
7753 : * used to optimize multixact access in case it's a lock-only multi); 'nowait'
7754 : * indicates whether to use conditional lock acquisition, to allow callers to
7755 : * fail if lock is unavailable. 'rel', 'ctid' and 'oper' are used to set up
7756 : * context information for error messages. 'remaining', if not NULL, receives
7757 : * the number of members that are still running, including any (non-aborted)
7758 : * subtransactions of our own transaction. 'logLockFailure' indicates whether
7759 : * to log details when a lock acquisition fails with 'nowait' enabled.
7760 : *
7761 : * We do this by sleeping on each member using XactLockTableWait. Any
7762 : * members that belong to the current backend are *not* waited for, however;
7763 : * this would not merely be useless but would lead to Assert failure inside
7764 : * XactLockTableWait. By the time this returns, it is certain that all
7765 : * transactions *of other backends* that were members of the MultiXactId
7766 : * that conflict with the requested status are dead (and no new ones can have
7767 : * been added, since it is not legal to add members to an existing
7768 : * MultiXactId).
7769 : *
7770 : * But by the time we finish sleeping, someone else may have changed the Xmax
7771 : * of the containing tuple, so the caller needs to iterate on us somehow.
7772 : *
7773 : * Note that in case we return false, the number of remaining members is
7774 : * not to be trusted.
7775 : */
7776 : static bool
7777 58 : Do_MultiXactIdWait(MultiXactId multi, MultiXactStatus status,
7778 : uint16 infomask, bool nowait,
7779 : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
7780 : int *remaining, bool logLockFailure)
7781 : {
7782 58 : bool result = true;
7783 : MultiXactMember *members;
7784 : int nmembers;
7785 58 : int remain = 0;
7786 :
7787 : /* for pre-pg_upgrade tuples, no need to sleep at all */
7788 58 : nmembers = HEAP_LOCKED_UPGRADED(infomask) ? -1 :
7789 58 : GetMultiXactIdMembers(multi, &members, false,
7790 58 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
7791 :
7792 58 : if (nmembers >= 0)
7793 : {
7794 : int i;
7795 :
7796 187 : for (i = 0; i < nmembers; i++)
7797 : {
7798 133 : TransactionId memxid = members[i].xid;
7799 133 : MultiXactStatus memstatus = members[i].status;
7800 :
7801 133 : if (TransactionIdIsCurrentTransactionId(memxid))
7802 : {
7803 24 : remain++;
7804 24 : continue;
7805 : }
7806 :
7807 109 : if (!DoLockModesConflict(LOCKMODE_from_mxstatus(memstatus),
7808 109 : LOCKMODE_from_mxstatus(status)))
7809 : {
7810 22 : if (remaining && TransactionIdIsInProgress(memxid))
7811 8 : remain++;
7812 22 : continue;
7813 : }
7814 :
7815 : /*
7816 : * This member conflicts with our multi, so we have to sleep (or
7817 : * return failure, if asked to avoid waiting.)
7818 : *
7819 : * Note that we don't set up an error context callback ourselves,
7820 : * but instead we pass the info down to XactLockTableWait. This
7821 : * might seem a bit wasteful because the context is set up and
7822 : * tore down for each member of the multixact, but in reality it
7823 : * should be barely noticeable, and it avoids duplicate code.
7824 : */
7825 87 : if (nowait)
7826 : {
7827 4 : result = ConditionalXactLockTableWait(memxid, logLockFailure);
7828 4 : if (!result)
7829 4 : break;
7830 : }
7831 : else
7832 83 : XactLockTableWait(memxid, rel, ctid, oper);
7833 : }
7834 :
7835 58 : pfree(members);
7836 : }
7837 :
7838 58 : if (remaining)
7839 10 : *remaining = remain;
7840 :
7841 58 : return result;
7842 : }
7843 :
7844 : /*
7845 : * MultiXactIdWait
7846 : * Sleep on a MultiXactId.
7847 : *
7848 : * By the time we finish sleeping, someone else may have changed the Xmax
7849 : * of the containing tuple, so the caller needs to iterate on us somehow.
7850 : *
7851 : * We return (in *remaining, if not NULL) the number of members that are still
7852 : * running, including any (non-aborted) subtransactions of our own transaction.
7853 : */
7854 : static void
7855 54 : MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
7856 : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
7857 : int *remaining)
7858 : {
7859 54 : (void) Do_MultiXactIdWait(multi, status, infomask, false,
7860 : rel, ctid, oper, remaining, false);
7861 54 : }
7862 :
7863 : /*
7864 : * ConditionalMultiXactIdWait
7865 : * As above, but only lock if we can get the lock without blocking.
7866 : *
7867 : * By the time we finish sleeping, someone else may have changed the Xmax
7868 : * of the containing tuple, so the caller needs to iterate on us somehow.
7869 : *
7870 : * If the multixact is now all gone, return true. Returns false if some
7871 : * transactions might still be running.
7872 : *
7873 : * We return (in *remaining, if not NULL) the number of members that are still
7874 : * running, including any (non-aborted) subtransactions of our own transaction.
7875 : */
7876 : static bool
7877 4 : ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status,
7878 : uint16 infomask, Relation rel, int *remaining,
7879 : bool logLockFailure)
7880 : {
7881 4 : return Do_MultiXactIdWait(multi, status, infomask, true,
7882 : rel, NULL, XLTW_None, remaining, logLockFailure);
7883 : }
7884 :
7885 : /*
7886 : * heap_tuple_needs_eventual_freeze
7887 : *
7888 : * Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
7889 : * will eventually require freezing (if tuple isn't removed by pruning first).
7890 : */
7891 : bool
7892 115199 : heap_tuple_needs_eventual_freeze(HeapTupleHeader tuple)
7893 : {
7894 : TransactionId xid;
7895 :
7896 : /*
7897 : * If xmin is a normal transaction ID, this tuple is definitely not
7898 : * frozen.
7899 : */
7900 115199 : xid = HeapTupleHeaderGetXmin(tuple);
7901 115199 : if (TransactionIdIsNormal(xid))
7902 2386 : return true;
7903 :
7904 : /*
7905 : * If xmax is a valid xact or multixact, this tuple is also not frozen.
7906 : */
7907 112813 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7908 : {
7909 : MultiXactId multi;
7910 :
7911 0 : multi = HeapTupleHeaderGetRawXmax(tuple);
7912 0 : if (MultiXactIdIsValid(multi))
7913 0 : return true;
7914 : }
7915 : else
7916 : {
7917 112813 : xid = HeapTupleHeaderGetRawXmax(tuple);
7918 112813 : if (TransactionIdIsNormal(xid))
7919 7 : return true;
7920 : }
7921 :
7922 112806 : if (tuple->t_infomask & HEAP_MOVED)
7923 : {
7924 0 : xid = HeapTupleHeaderGetXvac(tuple);
7925 0 : if (TransactionIdIsNormal(xid))
7926 0 : return true;
7927 : }
7928 :
7929 112806 : return false;
7930 : }
7931 :
7932 : /*
7933 : * heap_tuple_should_freeze
7934 : *
7935 : * Return value indicates if heap_prepare_freeze_tuple sibling function would
7936 : * (or should) force freezing of the heap page that contains caller's tuple.
7937 : * Tuple header XIDs/MXIDs < FreezeLimit/MultiXactCutoff trigger freezing.
7938 : * This includes (xmin, xmax, xvac) fields, as well as MultiXact member XIDs.
7939 : *
7940 : * The *NoFreezePageRelfrozenXid and *NoFreezePageRelminMxid input/output
7941 : * arguments help VACUUM track the oldest extant XID/MXID remaining in rel.
7942 : * Our working assumption is that caller won't decide to freeze this tuple.
7943 : * It's up to caller to only ratchet back its own top-level trackers after the
7944 : * point that it fully commits to not freezing the tuple/page in question.
7945 : */
7946 : bool
7947 9382299 : heap_tuple_should_freeze(HeapTupleHeader tuple,
7948 : const struct VacuumCutoffs *cutoffs,
7949 : TransactionId *NoFreezePageRelfrozenXid,
7950 : MultiXactId *NoFreezePageRelminMxid)
7951 : {
7952 : TransactionId xid;
7953 : MultiXactId multi;
7954 9382299 : bool freeze = false;
7955 :
7956 : /* First deal with xmin */
7957 9382299 : xid = HeapTupleHeaderGetXmin(tuple);
7958 9382299 : if (TransactionIdIsNormal(xid))
7959 : {
7960 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7961 1696601 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7962 22794 : *NoFreezePageRelfrozenXid = xid;
7963 1696601 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7964 21233 : freeze = true;
7965 : }
7966 :
7967 : /* Now deal with xmax */
7968 9382299 : xid = InvalidTransactionId;
7969 9382299 : multi = InvalidMultiXactId;
7970 9382299 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7971 2 : multi = HeapTupleHeaderGetRawXmax(tuple);
7972 : else
7973 9382297 : xid = HeapTupleHeaderGetRawXmax(tuple);
7974 :
7975 9382299 : if (TransactionIdIsNormal(xid))
7976 : {
7977 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7978 : /* xmax is a non-permanent XID */
7979 7933700 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7980 2 : *NoFreezePageRelfrozenXid = xid;
7981 7933700 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7982 32 : freeze = true;
7983 : }
7984 1448599 : else if (!MultiXactIdIsValid(multi))
7985 : {
7986 : /* xmax is a permanent XID or invalid MultiXactId/XID */
7987 : }
7988 2 : else if (HEAP_LOCKED_UPGRADED(tuple->t_infomask))
7989 : {
7990 : /* xmax is a pg_upgrade'd MultiXact, which can't have updater XID */
7991 0 : if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
7992 0 : *NoFreezePageRelminMxid = multi;
7993 : /* heap_prepare_freeze_tuple always freezes pg_upgrade'd xmax */
7994 0 : freeze = true;
7995 : }
7996 : else
7997 : {
7998 : /* xmax is a MultiXactId that may have an updater XID */
7999 : MultiXactMember *members;
8000 : int nmembers;
8001 :
8002 : Assert(MultiXactIdPrecedesOrEquals(cutoffs->relminmxid, multi));
8003 2 : if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
8004 2 : *NoFreezePageRelminMxid = multi;
8005 2 : if (MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff))
8006 2 : freeze = true;
8007 :
8008 : /* need to check whether any member of the mxact is old */
8009 2 : nmembers = GetMultiXactIdMembers(multi, &members, false,
8010 2 : HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask));
8011 :
8012 5 : for (int i = 0; i < nmembers; i++)
8013 : {
8014 3 : xid = members[i].xid;
8015 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
8016 3 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
8017 0 : *NoFreezePageRelfrozenXid = xid;
8018 3 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
8019 0 : freeze = true;
8020 : }
8021 2 : if (nmembers > 0)
8022 1 : pfree(members);
8023 : }
8024 :
8025 9382299 : if (tuple->t_infomask & HEAP_MOVED)
8026 : {
8027 0 : xid = HeapTupleHeaderGetXvac(tuple);
8028 0 : if (TransactionIdIsNormal(xid))
8029 : {
8030 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
8031 0 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
8032 0 : *NoFreezePageRelfrozenXid = xid;
8033 : /* heap_prepare_freeze_tuple forces xvac freezing */
8034 0 : freeze = true;
8035 : }
8036 : }
8037 :
8038 9382299 : return freeze;
8039 : }
8040 :
8041 : /*
8042 : * Maintain snapshotConflictHorizon for caller by ratcheting forward its value
8043 : * using any committed XIDs contained in 'tuple', an obsolescent heap tuple
8044 : * that caller is in the process of physically removing, e.g. via HOT pruning
8045 : * or index deletion.
8046 : *
8047 : * Caller must initialize its value to InvalidTransactionId, which is
8048 : * generally interpreted as "definitely no need for a recovery conflict".
8049 : * Final value must reflect all heap tuples that caller will physically remove
8050 : * (or remove TID references to) via its ongoing pruning/deletion operation.
8051 : * ResolveRecoveryConflictWithSnapshot() is passed the final value (taken from
8052 : * caller's WAL record) by REDO routine when it replays caller's operation.
8053 : */
8054 : void
8055 1577506 : HeapTupleHeaderAdvanceConflictHorizon(HeapTupleHeader tuple,
8056 : TransactionId *snapshotConflictHorizon)
8057 : {
8058 1577506 : TransactionId xmin = HeapTupleHeaderGetXmin(tuple);
8059 1577506 : TransactionId xmax = HeapTupleHeaderGetUpdateXid(tuple);
8060 1577506 : TransactionId xvac = HeapTupleHeaderGetXvac(tuple);
8061 :
8062 1577506 : if (tuple->t_infomask & HEAP_MOVED)
8063 : {
8064 0 : if (TransactionIdPrecedes(*snapshotConflictHorizon, xvac))
8065 0 : *snapshotConflictHorizon = xvac;
8066 : }
8067 :
8068 : /*
8069 : * Ignore tuples inserted by an aborted transaction or if the tuple was
8070 : * updated/deleted by the inserting transaction.
8071 : *
8072 : * Look for a committed hint bit, or if no xmin bit is set, check clog.
8073 : */
8074 1577506 : if (HeapTupleHeaderXminCommitted(tuple) ||
8075 101155 : (!HeapTupleHeaderXminInvalid(tuple) && TransactionIdDidCommit(xmin)))
8076 : {
8077 2810279 : if (xmax != xmin &&
8078 1315418 : TransactionIdFollows(xmax, *snapshotConflictHorizon))
8079 99267 : *snapshotConflictHorizon = xmax;
8080 : }
8081 1577506 : }
8082 :
8083 : #ifdef USE_PREFETCH
8084 : /*
8085 : * Helper function for heap_index_delete_tuples. Issues prefetch requests for
8086 : * prefetch_count buffers. The prefetch_state keeps track of all the buffers
8087 : * we can prefetch, and which have already been prefetched; each call to this
8088 : * function picks up where the previous call left off.
8089 : *
8090 : * Note: we expect the deltids array to be sorted in an order that groups TIDs
8091 : * by heap block, with all TIDs for each block appearing together in exactly
8092 : * one group.
8093 : */
8094 : static void
8095 20691 : index_delete_prefetch_buffer(Relation rel,
8096 : IndexDeletePrefetchState *prefetch_state,
8097 : int prefetch_count)
8098 : {
8099 20691 : BlockNumber cur_hblkno = prefetch_state->cur_hblkno;
8100 20691 : int count = 0;
8101 : int i;
8102 20691 : int ndeltids = prefetch_state->ndeltids;
8103 20691 : TM_IndexDelete *deltids = prefetch_state->deltids;
8104 :
8105 20691 : for (i = prefetch_state->next_item;
8106 720490 : i < ndeltids && count < prefetch_count;
8107 699799 : i++)
8108 : {
8109 699799 : ItemPointer htid = &deltids[i].tid;
8110 :
8111 1393483 : if (cur_hblkno == InvalidBlockNumber ||
8112 693684 : ItemPointerGetBlockNumber(htid) != cur_hblkno)
8113 : {
8114 18760 : cur_hblkno = ItemPointerGetBlockNumber(htid);
8115 18760 : PrefetchBuffer(rel, MAIN_FORKNUM, cur_hblkno);
8116 18760 : count++;
8117 : }
8118 : }
8119 :
8120 : /*
8121 : * Save the prefetch position so that next time we can continue from that
8122 : * position.
8123 : */
8124 20691 : prefetch_state->next_item = i;
8125 20691 : prefetch_state->cur_hblkno = cur_hblkno;
8126 20691 : }
8127 : #endif
8128 :
8129 : /*
8130 : * Helper function for heap_index_delete_tuples. Checks for index corruption
8131 : * involving an invalid TID in index AM caller's index page.
8132 : *
8133 : * This is an ideal place for these checks. The index AM must hold a buffer
8134 : * lock on the index page containing the TIDs we examine here, so we don't
8135 : * have to worry about concurrent VACUUMs at all. We can be sure that the
8136 : * index is corrupt when htid points directly to an LP_UNUSED item or
8137 : * heap-only tuple, which is not the case during standard index scans.
8138 : */
8139 : static inline void
8140 584260 : index_delete_check_htid(TM_IndexDeleteOp *delstate,
8141 : Page page, OffsetNumber maxoff,
8142 : const ItemPointerData *htid, TM_IndexStatus *istatus)
8143 : {
8144 584260 : OffsetNumber indexpagehoffnum = ItemPointerGetOffsetNumber(htid);
8145 : ItemId iid;
8146 :
8147 : Assert(OffsetNumberIsValid(istatus->idxoffnum));
8148 :
8149 584260 : if (unlikely(indexpagehoffnum > maxoff))
8150 0 : ereport(ERROR,
8151 : (errcode(ERRCODE_INDEX_CORRUPTED),
8152 : 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\"",
8153 : ItemPointerGetBlockNumber(htid),
8154 : indexpagehoffnum,
8155 : istatus->idxoffnum, delstate->iblknum,
8156 : RelationGetRelationName(delstate->irel))));
8157 :
8158 584260 : iid = PageGetItemId(page, indexpagehoffnum);
8159 584260 : if (unlikely(!ItemIdIsUsed(iid)))
8160 0 : ereport(ERROR,
8161 : (errcode(ERRCODE_INDEX_CORRUPTED),
8162 : errmsg_internal("heap tid from index tuple (%u,%u) points to unused heap page item at offset %u of block %u in index \"%s\"",
8163 : ItemPointerGetBlockNumber(htid),
8164 : indexpagehoffnum,
8165 : istatus->idxoffnum, delstate->iblknum,
8166 : RelationGetRelationName(delstate->irel))));
8167 :
8168 584260 : if (ItemIdHasStorage(iid))
8169 : {
8170 : HeapTupleHeader htup;
8171 :
8172 : Assert(ItemIdIsNormal(iid));
8173 344100 : htup = (HeapTupleHeader) PageGetItem(page, iid);
8174 :
8175 344100 : if (unlikely(HeapTupleHeaderIsHeapOnly(htup)))
8176 0 : ereport(ERROR,
8177 : (errcode(ERRCODE_INDEX_CORRUPTED),
8178 : errmsg_internal("heap tid from index tuple (%u,%u) points to heap-only tuple at offset %u of block %u in index \"%s\"",
8179 : ItemPointerGetBlockNumber(htid),
8180 : indexpagehoffnum,
8181 : istatus->idxoffnum, delstate->iblknum,
8182 : RelationGetRelationName(delstate->irel))));
8183 : }
8184 584260 : }
8185 :
8186 : /*
8187 : * heapam implementation of tableam's index_delete_tuples interface.
8188 : *
8189 : * This helper function is called by index AMs during index tuple deletion.
8190 : * See tableam header comments for an explanation of the interface implemented
8191 : * here and a general theory of operation. Note that each call here is either
8192 : * a simple index deletion call, or a bottom-up index deletion call.
8193 : *
8194 : * It's possible for this to generate a fair amount of I/O, since we may be
8195 : * deleting hundreds of tuples from a single index block. To amortize that
8196 : * cost to some degree, this uses prefetching and combines repeat accesses to
8197 : * the same heap block.
8198 : */
8199 : TransactionId
8200 6115 : heap_index_delete_tuples(Relation rel, TM_IndexDeleteOp *delstate)
8201 : {
8202 : /* Initial assumption is that earlier pruning took care of conflict */
8203 6115 : TransactionId snapshotConflictHorizon = InvalidTransactionId;
8204 6115 : BlockNumber blkno = InvalidBlockNumber;
8205 6115 : Buffer buf = InvalidBuffer;
8206 6115 : Page page = NULL;
8207 6115 : OffsetNumber maxoff = InvalidOffsetNumber;
8208 : TransactionId priorXmax;
8209 : #ifdef USE_PREFETCH
8210 : IndexDeletePrefetchState prefetch_state;
8211 : int prefetch_distance;
8212 : #endif
8213 : SnapshotData SnapshotNonVacuumable;
8214 6115 : int finalndeltids = 0,
8215 6115 : nblocksaccessed = 0;
8216 :
8217 : /* State that's only used in bottom-up index deletion case */
8218 6115 : int nblocksfavorable = 0;
8219 6115 : int curtargetfreespace = delstate->bottomupfreespace,
8220 6115 : lastfreespace = 0,
8221 6115 : actualfreespace = 0;
8222 6115 : bool bottomup_final_block = false;
8223 :
8224 6115 : InitNonVacuumableSnapshot(SnapshotNonVacuumable, GlobalVisTestFor(rel));
8225 :
8226 : /* Sort caller's deltids array by TID for further processing */
8227 6115 : index_delete_sort(delstate);
8228 :
8229 : /*
8230 : * Bottom-up case: resort deltids array in an order attuned to where the
8231 : * greatest number of promising TIDs are to be found, and determine how
8232 : * many blocks from the start of sorted array should be considered
8233 : * favorable. This will also shrink the deltids array in order to
8234 : * eliminate completely unfavorable blocks up front.
8235 : */
8236 6115 : if (delstate->bottomup)
8237 1996 : nblocksfavorable = bottomup_sort_and_shrink(delstate);
8238 :
8239 : #ifdef USE_PREFETCH
8240 : /* Initialize prefetch state. */
8241 6115 : prefetch_state.cur_hblkno = InvalidBlockNumber;
8242 6115 : prefetch_state.next_item = 0;
8243 6115 : prefetch_state.ndeltids = delstate->ndeltids;
8244 6115 : prefetch_state.deltids = delstate->deltids;
8245 :
8246 : /*
8247 : * Determine the prefetch distance that we will attempt to maintain.
8248 : *
8249 : * Since the caller holds a buffer lock somewhere in rel, we'd better make
8250 : * sure that isn't a catalog relation before we call code that does
8251 : * syscache lookups, to avoid risk of deadlock.
8252 : */
8253 6115 : if (IsCatalogRelation(rel))
8254 4490 : prefetch_distance = maintenance_io_concurrency;
8255 : else
8256 : prefetch_distance =
8257 1625 : get_tablespace_maintenance_io_concurrency(rel->rd_rel->reltablespace);
8258 :
8259 : /* Cap initial prefetch distance for bottom-up deletion caller */
8260 6115 : if (delstate->bottomup)
8261 : {
8262 : Assert(nblocksfavorable >= 1);
8263 : Assert(nblocksfavorable <= BOTTOMUP_MAX_NBLOCKS);
8264 1996 : prefetch_distance = Min(prefetch_distance, nblocksfavorable);
8265 : }
8266 :
8267 : /* Start prefetching. */
8268 6115 : index_delete_prefetch_buffer(rel, &prefetch_state, prefetch_distance);
8269 : #endif
8270 :
8271 : /* Iterate over deltids, determine which to delete, check their horizon */
8272 : Assert(delstate->ndeltids > 0);
8273 590375 : for (int i = 0; i < delstate->ndeltids; i++)
8274 : {
8275 586256 : TM_IndexDelete *ideltid = &delstate->deltids[i];
8276 586256 : TM_IndexStatus *istatus = delstate->status + ideltid->id;
8277 586256 : ItemPointer htid = &ideltid->tid;
8278 : OffsetNumber offnum;
8279 :
8280 : /*
8281 : * Read buffer, and perform required extra steps each time a new block
8282 : * is encountered. Avoid refetching if it's the same block as the one
8283 : * from the last htid.
8284 : */
8285 1166397 : if (blkno == InvalidBlockNumber ||
8286 580141 : ItemPointerGetBlockNumber(htid) != blkno)
8287 : {
8288 : /*
8289 : * Consider giving up early for bottom-up index deletion caller
8290 : * first. (Only prefetch next-next block afterwards, when it
8291 : * becomes clear that we're at least going to access the next
8292 : * block in line.)
8293 : *
8294 : * Sometimes the first block frees so much space for bottom-up
8295 : * caller that the deletion process can end without accessing any
8296 : * more blocks. It is usually necessary to access 2 or 3 blocks
8297 : * per bottom-up deletion operation, though.
8298 : */
8299 16572 : if (delstate->bottomup)
8300 : {
8301 : /*
8302 : * We often allow caller to delete a few additional items
8303 : * whose entries we reached after the point that space target
8304 : * from caller was satisfied. The cost of accessing the page
8305 : * was already paid at that point, so it made sense to finish
8306 : * it off. When that happened, we finalize everything here
8307 : * (by finishing off the whole bottom-up deletion operation
8308 : * without needlessly paying the cost of accessing any more
8309 : * blocks).
8310 : */
8311 4478 : if (bottomup_final_block)
8312 153 : break;
8313 :
8314 : /*
8315 : * Give up when we didn't enable our caller to free any
8316 : * additional space as a result of processing the page that we
8317 : * just finished up with. This rule is the main way in which
8318 : * we keep the cost of bottom-up deletion under control.
8319 : */
8320 4325 : if (nblocksaccessed >= 1 && actualfreespace == lastfreespace)
8321 1843 : break;
8322 2482 : lastfreespace = actualfreespace; /* for next time */
8323 :
8324 : /*
8325 : * Deletion operation (which is bottom-up) will definitely
8326 : * access the next block in line. Prepare for that now.
8327 : *
8328 : * Decay target free space so that we don't hang on for too
8329 : * long with a marginal case. (Space target is only truly
8330 : * helpful when it allows us to recognize that we don't need
8331 : * to access more than 1 or 2 blocks to satisfy caller due to
8332 : * agreeable workload characteristics.)
8333 : *
8334 : * We are a bit more patient when we encounter contiguous
8335 : * blocks, though: these are treated as favorable blocks. The
8336 : * decay process is only applied when the next block in line
8337 : * is not a favorable/contiguous block. This is not an
8338 : * exception to the general rule; we still insist on finding
8339 : * at least one deletable item per block accessed. See
8340 : * bottomup_nblocksfavorable() for full details of the theory
8341 : * behind favorable blocks and heap block locality in general.
8342 : *
8343 : * Note: The first block in line is always treated as a
8344 : * favorable block, so the earliest possible point that the
8345 : * decay can be applied is just before we access the second
8346 : * block in line. The Assert() verifies this for us.
8347 : */
8348 : Assert(nblocksaccessed > 0 || nblocksfavorable > 0);
8349 2482 : if (nblocksfavorable > 0)
8350 2234 : nblocksfavorable--;
8351 : else
8352 248 : curtargetfreespace /= 2;
8353 : }
8354 :
8355 : /* release old buffer */
8356 14576 : if (BufferIsValid(buf))
8357 8461 : UnlockReleaseBuffer(buf);
8358 :
8359 14576 : blkno = ItemPointerGetBlockNumber(htid);
8360 14576 : buf = ReadBuffer(rel, blkno);
8361 14576 : nblocksaccessed++;
8362 : Assert(!delstate->bottomup ||
8363 : nblocksaccessed <= BOTTOMUP_MAX_NBLOCKS);
8364 :
8365 : #ifdef USE_PREFETCH
8366 :
8367 : /*
8368 : * To maintain the prefetch distance, prefetch one more page for
8369 : * each page we read.
8370 : */
8371 14576 : index_delete_prefetch_buffer(rel, &prefetch_state, 1);
8372 : #endif
8373 :
8374 14576 : LockBuffer(buf, BUFFER_LOCK_SHARE);
8375 :
8376 14576 : page = BufferGetPage(buf);
8377 14576 : maxoff = PageGetMaxOffsetNumber(page);
8378 : }
8379 :
8380 : /*
8381 : * In passing, detect index corruption involving an index page with a
8382 : * TID that points to a location in the heap that couldn't possibly be
8383 : * correct. We only do this with actual TIDs from caller's index page
8384 : * (not items reached by traversing through a HOT chain).
8385 : */
8386 584260 : index_delete_check_htid(delstate, page, maxoff, htid, istatus);
8387 :
8388 584260 : if (istatus->knowndeletable)
8389 : Assert(!delstate->bottomup && !istatus->promising);
8390 : else
8391 : {
8392 444411 : ItemPointerData tmp = *htid;
8393 : HeapTupleData heapTuple;
8394 :
8395 : /* Are any tuples from this HOT chain non-vacuumable? */
8396 444411 : if (heap_hot_search_buffer(&tmp, rel, buf, &SnapshotNonVacuumable,
8397 : &heapTuple, NULL, true))
8398 256993 : continue; /* can't delete entry */
8399 :
8400 : /* Caller will delete, since whole HOT chain is vacuumable */
8401 187418 : istatus->knowndeletable = true;
8402 :
8403 : /* Maintain index free space info for bottom-up deletion case */
8404 187418 : if (delstate->bottomup)
8405 : {
8406 : Assert(istatus->freespace > 0);
8407 9203 : actualfreespace += istatus->freespace;
8408 9203 : if (actualfreespace >= curtargetfreespace)
8409 2163 : bottomup_final_block = true;
8410 : }
8411 : }
8412 :
8413 : /*
8414 : * Maintain snapshotConflictHorizon value for deletion operation as a
8415 : * whole by advancing current value using heap tuple headers. This is
8416 : * loosely based on the logic for pruning a HOT chain.
8417 : */
8418 327267 : offnum = ItemPointerGetOffsetNumber(htid);
8419 327267 : priorXmax = InvalidTransactionId; /* cannot check first XMIN */
8420 : for (;;)
8421 21392 : {
8422 : ItemId lp;
8423 : HeapTupleHeader htup;
8424 :
8425 : /* Sanity check (pure paranoia) */
8426 348659 : if (offnum < FirstOffsetNumber)
8427 0 : break;
8428 :
8429 : /*
8430 : * An offset past the end of page's line pointer array is possible
8431 : * when the array was truncated
8432 : */
8433 348659 : if (offnum > maxoff)
8434 0 : break;
8435 :
8436 348659 : lp = PageGetItemId(page, offnum);
8437 348659 : if (ItemIdIsRedirected(lp))
8438 : {
8439 9563 : offnum = ItemIdGetRedirect(lp);
8440 9563 : continue;
8441 : }
8442 :
8443 : /*
8444 : * We'll often encounter LP_DEAD line pointers (especially with an
8445 : * entry marked knowndeletable by our caller up front). No heap
8446 : * tuple headers get examined for an htid that leads us to an
8447 : * LP_DEAD item. This is okay because the earlier pruning
8448 : * operation that made the line pointer LP_DEAD in the first place
8449 : * must have considered the original tuple header as part of
8450 : * generating its own snapshotConflictHorizon value.
8451 : *
8452 : * Relying on XLOG_HEAP2_PRUNE_VACUUM_SCAN records like this is
8453 : * the same strategy that index vacuuming uses in all cases. Index
8454 : * VACUUM WAL records don't even have a snapshotConflictHorizon
8455 : * field of their own for this reason.
8456 : */
8457 339096 : if (!ItemIdIsNormal(lp))
8458 214265 : break;
8459 :
8460 124831 : htup = (HeapTupleHeader) PageGetItem(page, lp);
8461 :
8462 : /*
8463 : * Check the tuple XMIN against prior XMAX, if any
8464 : */
8465 136660 : if (TransactionIdIsValid(priorXmax) &&
8466 11829 : !TransactionIdEquals(HeapTupleHeaderGetXmin(htup), priorXmax))
8467 0 : break;
8468 :
8469 124831 : HeapTupleHeaderAdvanceConflictHorizon(htup,
8470 : &snapshotConflictHorizon);
8471 :
8472 : /*
8473 : * If the tuple is not HOT-updated, then we are at the end of this
8474 : * HOT-chain. No need to visit later tuples from the same update
8475 : * chain (they get their own index entries) -- just move on to
8476 : * next htid from index AM caller.
8477 : */
8478 124831 : if (!HeapTupleHeaderIsHotUpdated(htup))
8479 113002 : break;
8480 :
8481 : /* Advance to next HOT chain member */
8482 : Assert(ItemPointerGetBlockNumber(&htup->t_ctid) == blkno);
8483 11829 : offnum = ItemPointerGetOffsetNumber(&htup->t_ctid);
8484 11829 : priorXmax = HeapTupleHeaderGetUpdateXid(htup);
8485 : }
8486 :
8487 : /* Enable further/final shrinking of deltids for caller */
8488 327267 : finalndeltids = i + 1;
8489 : }
8490 :
8491 6115 : UnlockReleaseBuffer(buf);
8492 :
8493 : /*
8494 : * Shrink deltids array to exclude non-deletable entries at the end. This
8495 : * is not just a minor optimization. Final deltids array size might be
8496 : * zero for a bottom-up caller. Index AM is explicitly allowed to rely on
8497 : * ndeltids being zero in all cases with zero total deletable entries.
8498 : */
8499 : Assert(finalndeltids > 0 || delstate->bottomup);
8500 6115 : delstate->ndeltids = finalndeltids;
8501 :
8502 6115 : return snapshotConflictHorizon;
8503 : }
8504 :
8505 : /*
8506 : * Specialized inlineable comparison function for index_delete_sort()
8507 : */
8508 : static inline int
8509 13263564 : index_delete_sort_cmp(TM_IndexDelete *deltid1, TM_IndexDelete *deltid2)
8510 : {
8511 13263564 : ItemPointer tid1 = &deltid1->tid;
8512 13263564 : ItemPointer tid2 = &deltid2->tid;
8513 :
8514 : {
8515 13263564 : BlockNumber blk1 = ItemPointerGetBlockNumber(tid1);
8516 13263564 : BlockNumber blk2 = ItemPointerGetBlockNumber(tid2);
8517 :
8518 13263564 : if (blk1 != blk2)
8519 5353857 : return (blk1 < blk2) ? -1 : 1;
8520 : }
8521 : {
8522 7909707 : OffsetNumber pos1 = ItemPointerGetOffsetNumber(tid1);
8523 7909707 : OffsetNumber pos2 = ItemPointerGetOffsetNumber(tid2);
8524 :
8525 7909707 : if (pos1 != pos2)
8526 7909707 : return (pos1 < pos2) ? -1 : 1;
8527 : }
8528 :
8529 : Assert(false);
8530 :
8531 0 : return 0;
8532 : }
8533 :
8534 : /*
8535 : * Sort deltids array from delstate by TID. This prepares it for further
8536 : * processing by heap_index_delete_tuples().
8537 : *
8538 : * This operation becomes a noticeable consumer of CPU cycles with some
8539 : * workloads, so we go to the trouble of specialization/micro optimization.
8540 : * We use shellsort for this because it's easy to specialize, compiles to
8541 : * relatively few instructions, and is adaptive to presorted inputs/subsets
8542 : * (which are typical here).
8543 : */
8544 : static void
8545 6115 : index_delete_sort(TM_IndexDeleteOp *delstate)
8546 : {
8547 6115 : TM_IndexDelete *deltids = delstate->deltids;
8548 6115 : int ndeltids = delstate->ndeltids;
8549 :
8550 : /*
8551 : * Shellsort gap sequence (taken from Sedgewick-Incerpi paper).
8552 : *
8553 : * This implementation is fast with array sizes up to ~4500. This covers
8554 : * all supported BLCKSZ values.
8555 : */
8556 6115 : const int gaps[9] = {1968, 861, 336, 112, 48, 21, 7, 3, 1};
8557 :
8558 : /* Think carefully before changing anything here -- keep swaps cheap */
8559 : StaticAssertDecl(sizeof(TM_IndexDelete) <= 8,
8560 : "element size exceeds 8 bytes");
8561 :
8562 61150 : for (int g = 0; g < lengthof(gaps); g++)
8563 : {
8564 7919900 : for (int hi = gaps[g], i = hi; i < ndeltids; i++)
8565 : {
8566 7864865 : TM_IndexDelete d = deltids[i];
8567 7864865 : int j = i;
8568 :
8569 13653850 : while (j >= hi && index_delete_sort_cmp(&deltids[j - hi], &d) >= 0)
8570 : {
8571 5788985 : deltids[j] = deltids[j - hi];
8572 5788985 : j -= hi;
8573 : }
8574 7864865 : deltids[j] = d;
8575 : }
8576 : }
8577 6115 : }
8578 :
8579 : /*
8580 : * Returns how many blocks should be considered favorable/contiguous for a
8581 : * bottom-up index deletion pass. This is a number of heap blocks that starts
8582 : * from and includes the first block in line.
8583 : *
8584 : * There is always at least one favorable block during bottom-up index
8585 : * deletion. In the worst case (i.e. with totally random heap blocks) the
8586 : * first block in line (the only favorable block) can be thought of as a
8587 : * degenerate array of contiguous blocks that consists of a single block.
8588 : * heap_index_delete_tuples() will expect this.
8589 : *
8590 : * Caller passes blockgroups, a description of the final order that deltids
8591 : * will be sorted in for heap_index_delete_tuples() bottom-up index deletion
8592 : * processing. Note that deltids need not actually be sorted just yet (caller
8593 : * only passes deltids to us so that we can interpret blockgroups).
8594 : *
8595 : * You might guess that the existence of contiguous blocks cannot matter much,
8596 : * since in general the main factor that determines which blocks we visit is
8597 : * the number of promising TIDs, which is a fixed hint from the index AM.
8598 : * We're not really targeting the general case, though -- the actual goal is
8599 : * to adapt our behavior to a wide variety of naturally occurring conditions.
8600 : * The effects of most of the heuristics we apply are only noticeable in the
8601 : * aggregate, over time and across many _related_ bottom-up index deletion
8602 : * passes.
8603 : *
8604 : * Deeming certain blocks favorable allows heapam to recognize and adapt to
8605 : * workloads where heap blocks visited during bottom-up index deletion can be
8606 : * accessed contiguously, in the sense that each newly visited block is the
8607 : * neighbor of the block that bottom-up deletion just finished processing (or
8608 : * close enough to it). It will likely be cheaper to access more favorable
8609 : * blocks sooner rather than later (e.g. in this pass, not across a series of
8610 : * related bottom-up passes). Either way it is probably only a matter of time
8611 : * (or a matter of further correlated version churn) before all blocks that
8612 : * appear together as a single large batch of favorable blocks get accessed by
8613 : * _some_ bottom-up pass. Large batches of favorable blocks tend to either
8614 : * appear almost constantly or not even once (it all depends on per-index
8615 : * workload characteristics).
8616 : *
8617 : * Note that the blockgroups sort order applies a power-of-two bucketing
8618 : * scheme that creates opportunities for contiguous groups of blocks to get
8619 : * batched together, at least with workloads that are naturally amenable to
8620 : * being driven by heap block locality. This doesn't just enhance the spatial
8621 : * locality of bottom-up heap block processing in the obvious way. It also
8622 : * enables temporal locality of access, since sorting by heap block number
8623 : * naturally tends to make the bottom-up processing order deterministic.
8624 : *
8625 : * Consider the following example to get a sense of how temporal locality
8626 : * might matter: There is a heap relation with several indexes, each of which
8627 : * is low to medium cardinality. It is subject to constant non-HOT updates.
8628 : * The updates are skewed (in one part of the primary key, perhaps). None of
8629 : * the indexes are logically modified by the UPDATE statements (if they were
8630 : * then bottom-up index deletion would not be triggered in the first place).
8631 : * Naturally, each new round of index tuples (for each heap tuple that gets a
8632 : * heap_update() call) will have the same heap TID in each and every index.
8633 : * Since these indexes are low cardinality and never get logically modified,
8634 : * heapam processing during bottom-up deletion passes will access heap blocks
8635 : * in approximately sequential order. Temporal locality of access occurs due
8636 : * to bottom-up deletion passes behaving very similarly across each of the
8637 : * indexes at any given moment. This keeps the number of buffer misses needed
8638 : * to visit heap blocks to a minimum.
8639 : */
8640 : static int
8641 1996 : bottomup_nblocksfavorable(IndexDeleteCounts *blockgroups, int nblockgroups,
8642 : TM_IndexDelete *deltids)
8643 : {
8644 1996 : int64 lastblock = -1;
8645 1996 : int nblocksfavorable = 0;
8646 :
8647 : Assert(nblockgroups >= 1);
8648 : Assert(nblockgroups <= BOTTOMUP_MAX_NBLOCKS);
8649 :
8650 : /*
8651 : * We tolerate heap blocks that will be accessed only slightly out of
8652 : * physical order. Small blips occur when a pair of almost-contiguous
8653 : * blocks happen to fall into different buckets (perhaps due only to a
8654 : * small difference in npromisingtids that the bucketing scheme didn't
8655 : * quite manage to ignore). We effectively ignore these blips by applying
8656 : * a small tolerance. The precise tolerance we use is a little arbitrary,
8657 : * but it works well enough in practice.
8658 : */
8659 6552 : for (int b = 0; b < nblockgroups; b++)
8660 : {
8661 6232 : IndexDeleteCounts *group = blockgroups + b;
8662 6232 : TM_IndexDelete *firstdtid = deltids + group->ifirsttid;
8663 6232 : BlockNumber block = ItemPointerGetBlockNumber(&firstdtid->tid);
8664 :
8665 6232 : if (lastblock != -1 &&
8666 4236 : ((int64) block < lastblock - BOTTOMUP_TOLERANCE_NBLOCKS ||
8667 3752 : (int64) block > lastblock + BOTTOMUP_TOLERANCE_NBLOCKS))
8668 : break;
8669 :
8670 4556 : nblocksfavorable++;
8671 4556 : lastblock = block;
8672 : }
8673 :
8674 : /* Always indicate that there is at least 1 favorable block */
8675 : Assert(nblocksfavorable >= 1);
8676 :
8677 1996 : return nblocksfavorable;
8678 : }
8679 :
8680 : /*
8681 : * qsort comparison function for bottomup_sort_and_shrink()
8682 : */
8683 : static int
8684 195201 : bottomup_sort_and_shrink_cmp(const void *arg1, const void *arg2)
8685 : {
8686 195201 : const IndexDeleteCounts *group1 = (const IndexDeleteCounts *) arg1;
8687 195201 : const IndexDeleteCounts *group2 = (const IndexDeleteCounts *) arg2;
8688 :
8689 : /*
8690 : * Most significant field is npromisingtids (which we invert the order of
8691 : * so as to sort in desc order).
8692 : *
8693 : * Caller should have already normalized npromisingtids fields into
8694 : * power-of-two values (buckets).
8695 : */
8696 195201 : if (group1->npromisingtids > group2->npromisingtids)
8697 9378 : return -1;
8698 185823 : if (group1->npromisingtids < group2->npromisingtids)
8699 10757 : return 1;
8700 :
8701 : /*
8702 : * Tiebreak: desc ntids sort order.
8703 : *
8704 : * We cannot expect power-of-two values for ntids fields. We should
8705 : * behave as if they were already rounded up for us instead.
8706 : */
8707 175066 : if (group1->ntids != group2->ntids)
8708 : {
8709 125640 : uint32 ntids1 = pg_nextpower2_32((uint32) group1->ntids);
8710 125640 : uint32 ntids2 = pg_nextpower2_32((uint32) group2->ntids);
8711 :
8712 125640 : if (ntids1 > ntids2)
8713 18650 : return -1;
8714 106990 : if (ntids1 < ntids2)
8715 23170 : return 1;
8716 : }
8717 :
8718 : /*
8719 : * Tiebreak: asc offset-into-deltids-for-block (offset to first TID for
8720 : * block in deltids array) order.
8721 : *
8722 : * This is equivalent to sorting in ascending heap block number order
8723 : * (among otherwise equal subsets of the array). This approach allows us
8724 : * to avoid accessing the out-of-line TID. (We rely on the assumption
8725 : * that the deltids array was sorted in ascending heap TID order when
8726 : * these offsets to the first TID from each heap block group were formed.)
8727 : */
8728 133246 : if (group1->ifirsttid > group2->ifirsttid)
8729 64585 : return 1;
8730 68661 : if (group1->ifirsttid < group2->ifirsttid)
8731 68661 : return -1;
8732 :
8733 0 : pg_unreachable();
8734 :
8735 : return 0;
8736 : }
8737 :
8738 : /*
8739 : * heap_index_delete_tuples() helper function for bottom-up deletion callers.
8740 : *
8741 : * Sorts deltids array in the order needed for useful processing by bottom-up
8742 : * deletion. The array should already be sorted in TID order when we're
8743 : * called. The sort process groups heap TIDs from deltids into heap block
8744 : * groupings. Earlier/more-promising groups/blocks are usually those that are
8745 : * known to have the most "promising" TIDs.
8746 : *
8747 : * Sets new size of deltids array (ndeltids) in state. deltids will only have
8748 : * TIDs from the BOTTOMUP_MAX_NBLOCKS most promising heap blocks when we
8749 : * return. This often means that deltids will be shrunk to a small fraction
8750 : * of its original size (we eliminate many heap blocks from consideration for
8751 : * caller up front).
8752 : *
8753 : * Returns the number of "favorable" blocks. See bottomup_nblocksfavorable()
8754 : * for a definition and full details.
8755 : */
8756 : static int
8757 1996 : bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate)
8758 : {
8759 : IndexDeleteCounts *blockgroups;
8760 : TM_IndexDelete *reordereddeltids;
8761 1996 : BlockNumber curblock = InvalidBlockNumber;
8762 1996 : int nblockgroups = 0;
8763 1996 : int ncopied = 0;
8764 1996 : int nblocksfavorable = 0;
8765 :
8766 : Assert(delstate->bottomup);
8767 : Assert(delstate->ndeltids > 0);
8768 :
8769 : /* Calculate per-heap-block count of TIDs */
8770 1996 : blockgroups = palloc_array(IndexDeleteCounts, delstate->ndeltids);
8771 965163 : for (int i = 0; i < delstate->ndeltids; i++)
8772 : {
8773 963167 : TM_IndexDelete *ideltid = &delstate->deltids[i];
8774 963167 : TM_IndexStatus *istatus = delstate->status + ideltid->id;
8775 963167 : ItemPointer htid = &ideltid->tid;
8776 963167 : bool promising = istatus->promising;
8777 :
8778 963167 : if (curblock != ItemPointerGetBlockNumber(htid))
8779 : {
8780 : /* New block group */
8781 38305 : nblockgroups++;
8782 :
8783 : Assert(curblock < ItemPointerGetBlockNumber(htid) ||
8784 : !BlockNumberIsValid(curblock));
8785 :
8786 38305 : curblock = ItemPointerGetBlockNumber(htid);
8787 38305 : blockgroups[nblockgroups - 1].ifirsttid = i;
8788 38305 : blockgroups[nblockgroups - 1].ntids = 1;
8789 38305 : blockgroups[nblockgroups - 1].npromisingtids = 0;
8790 : }
8791 : else
8792 : {
8793 924862 : blockgroups[nblockgroups - 1].ntids++;
8794 : }
8795 :
8796 963167 : if (promising)
8797 122554 : blockgroups[nblockgroups - 1].npromisingtids++;
8798 : }
8799 :
8800 : /*
8801 : * We're about ready to sort block groups to determine the optimal order
8802 : * for visiting heap blocks. But before we do, round the number of
8803 : * promising tuples for each block group up to the next power-of-two,
8804 : * unless it is very low (less than 4), in which case we round up to 4.
8805 : * npromisingtids is far too noisy to trust when choosing between a pair
8806 : * of block groups that both have very low values.
8807 : *
8808 : * This scheme divides heap blocks/block groups into buckets. Each bucket
8809 : * contains blocks that have _approximately_ the same number of promising
8810 : * TIDs as each other. The goal is to ignore relatively small differences
8811 : * in the total number of promising entries, so that the whole process can
8812 : * give a little weight to heapam factors (like heap block locality)
8813 : * instead. This isn't a trade-off, really -- we have nothing to lose. It
8814 : * would be foolish to interpret small differences in npromisingtids
8815 : * values as anything more than noise.
8816 : *
8817 : * We tiebreak on nhtids when sorting block group subsets that have the
8818 : * same npromisingtids, but this has the same issues as npromisingtids,
8819 : * and so nhtids is subject to the same power-of-two bucketing scheme. The
8820 : * only reason that we don't fix nhtids in the same way here too is that
8821 : * we'll need accurate nhtids values after the sort. We handle nhtids
8822 : * bucketization dynamically instead (in the sort comparator).
8823 : *
8824 : * See bottomup_nblocksfavorable() for a full explanation of when and how
8825 : * heap locality/favorable blocks can significantly influence when and how
8826 : * heap blocks are accessed.
8827 : */
8828 40301 : for (int b = 0; b < nblockgroups; b++)
8829 : {
8830 38305 : IndexDeleteCounts *group = blockgroups + b;
8831 :
8832 : /* Better off falling back on nhtids with low npromisingtids */
8833 38305 : if (group->npromisingtids <= 4)
8834 32728 : group->npromisingtids = 4;
8835 : else
8836 5577 : group->npromisingtids =
8837 5577 : pg_nextpower2_32((uint32) group->npromisingtids);
8838 : }
8839 :
8840 : /* Sort groups and rearrange caller's deltids array */
8841 1996 : qsort(blockgroups, nblockgroups, sizeof(IndexDeleteCounts),
8842 : bottomup_sort_and_shrink_cmp);
8843 1996 : reordereddeltids = palloc(delstate->ndeltids * sizeof(TM_IndexDelete));
8844 :
8845 1996 : nblockgroups = Min(BOTTOMUP_MAX_NBLOCKS, nblockgroups);
8846 : /* Determine number of favorable blocks at the start of final deltids */
8847 1996 : nblocksfavorable = bottomup_nblocksfavorable(blockgroups, nblockgroups,
8848 : delstate->deltids);
8849 :
8850 13362 : for (int b = 0; b < nblockgroups; b++)
8851 : {
8852 11366 : IndexDeleteCounts *group = blockgroups + b;
8853 11366 : TM_IndexDelete *firstdtid = delstate->deltids + group->ifirsttid;
8854 :
8855 11366 : memcpy(reordereddeltids + ncopied, firstdtid,
8856 11366 : sizeof(TM_IndexDelete) * group->ntids);
8857 11366 : ncopied += group->ntids;
8858 : }
8859 :
8860 : /* Copy final grouped and sorted TIDs back into start of caller's array */
8861 1996 : memcpy(delstate->deltids, reordereddeltids,
8862 : sizeof(TM_IndexDelete) * ncopied);
8863 1996 : delstate->ndeltids = ncopied;
8864 :
8865 1996 : pfree(reordereddeltids);
8866 1996 : pfree(blockgroups);
8867 :
8868 1996 : return nblocksfavorable;
8869 : }
8870 :
8871 : /*
8872 : * Perform XLogInsert for a heap-visible operation. 'block' is the block
8873 : * being marked all-visible, and vm_buffer is the buffer containing the
8874 : * corresponding visibility map block. Both should have already been modified
8875 : * and dirtied.
8876 : *
8877 : * snapshotConflictHorizon comes from the largest xmin on the page being
8878 : * marked all-visible. REDO routine uses it to generate recovery conflicts.
8879 : *
8880 : * If checksums or wal_log_hints are enabled, we may also generate a full-page
8881 : * image of heap_buffer. Otherwise, we optimize away the FPI (by specifying
8882 : * REGBUF_NO_IMAGE for the heap buffer), in which case the caller should *not*
8883 : * update the heap page's LSN.
8884 : */
8885 : XLogRecPtr
8886 34686 : log_heap_visible(Relation rel, Buffer heap_buffer, Buffer vm_buffer,
8887 : TransactionId snapshotConflictHorizon, uint8 vmflags)
8888 : {
8889 : xl_heap_visible xlrec;
8890 : XLogRecPtr recptr;
8891 : uint8 flags;
8892 :
8893 : Assert(BufferIsValid(heap_buffer));
8894 : Assert(BufferIsValid(vm_buffer));
8895 :
8896 34686 : xlrec.snapshotConflictHorizon = snapshotConflictHorizon;
8897 34686 : xlrec.flags = vmflags;
8898 34686 : if (RelationIsAccessibleInLogicalDecoding(rel))
8899 56 : xlrec.flags |= VISIBILITYMAP_XLOG_CATALOG_REL;
8900 34686 : XLogBeginInsert();
8901 34686 : XLogRegisterData(&xlrec, SizeOfHeapVisible);
8902 :
8903 34686 : XLogRegisterBuffer(0, vm_buffer, 0);
8904 :
8905 34686 : flags = REGBUF_STANDARD;
8906 34686 : if (!XLogHintBitIsNeeded())
8907 3135 : flags |= REGBUF_NO_IMAGE;
8908 34686 : XLogRegisterBuffer(1, heap_buffer, flags);
8909 :
8910 34686 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_VISIBLE);
8911 :
8912 34686 : return recptr;
8913 : }
8914 :
8915 : /*
8916 : * Perform XLogInsert for a heap-update operation. Caller must already
8917 : * have modified the buffer(s) and marked them dirty.
8918 : */
8919 : static XLogRecPtr
8920 299869 : log_heap_update(Relation reln, Buffer oldbuf,
8921 : Buffer newbuf, HeapTuple oldtup, HeapTuple newtup,
8922 : HeapTuple old_key_tuple,
8923 : bool all_visible_cleared, bool new_all_visible_cleared)
8924 : {
8925 : xl_heap_update xlrec;
8926 : xl_heap_header xlhdr;
8927 : xl_heap_header xlhdr_idx;
8928 : uint8 info;
8929 : uint16 prefix_suffix[2];
8930 299869 : uint16 prefixlen = 0,
8931 299869 : suffixlen = 0;
8932 : XLogRecPtr recptr;
8933 299869 : Page page = BufferGetPage(newbuf);
8934 299869 : bool need_tuple_data = RelationIsLogicallyLogged(reln);
8935 : bool init;
8936 : int bufflags;
8937 :
8938 : /* Caller should not call me on a non-WAL-logged relation */
8939 : Assert(RelationNeedsWAL(reln));
8940 :
8941 299869 : XLogBeginInsert();
8942 :
8943 299869 : if (HeapTupleIsHeapOnly(newtup))
8944 148330 : info = XLOG_HEAP_HOT_UPDATE;
8945 : else
8946 151539 : info = XLOG_HEAP_UPDATE;
8947 :
8948 : /*
8949 : * If the old and new tuple are on the same page, we only need to log the
8950 : * parts of the new tuple that were changed. That saves on the amount of
8951 : * WAL we need to write. Currently, we just count any unchanged bytes in
8952 : * the beginning and end of the tuple. That's quick to check, and
8953 : * perfectly covers the common case that only one field is updated.
8954 : *
8955 : * We could do this even if the old and new tuple are on different pages,
8956 : * but only if we don't make a full-page image of the old page, which is
8957 : * difficult to know in advance. Also, if the old tuple is corrupt for
8958 : * some reason, it would allow the corruption to propagate the new page,
8959 : * so it seems best to avoid. Under the general assumption that most
8960 : * updates tend to create the new tuple version on the same page, there
8961 : * isn't much to be gained by doing this across pages anyway.
8962 : *
8963 : * Skip this if we're taking a full-page image of the new page, as we
8964 : * don't include the new tuple in the WAL record in that case. Also
8965 : * disable if effective_wal_level='logical', as logical decoding needs to
8966 : * be able to read the new tuple in whole from the WAL record alone.
8967 : */
8968 299869 : if (oldbuf == newbuf && !need_tuple_data &&
8969 148887 : !XLogCheckBufferNeedsBackup(newbuf))
8970 : {
8971 148257 : char *oldp = (char *) oldtup->t_data + oldtup->t_data->t_hoff;
8972 148257 : char *newp = (char *) newtup->t_data + newtup->t_data->t_hoff;
8973 148257 : int oldlen = oldtup->t_len - oldtup->t_data->t_hoff;
8974 148257 : int newlen = newtup->t_len - newtup->t_data->t_hoff;
8975 :
8976 : /* Check for common prefix between old and new tuple */
8977 12382038 : for (prefixlen = 0; prefixlen < Min(oldlen, newlen); prefixlen++)
8978 : {
8979 12355787 : if (newp[prefixlen] != oldp[prefixlen])
8980 122006 : break;
8981 : }
8982 :
8983 : /*
8984 : * Storing the length of the prefix takes 2 bytes, so we need to save
8985 : * at least 3 bytes or there's no point.
8986 : */
8987 148257 : if (prefixlen < 3)
8988 23119 : prefixlen = 0;
8989 :
8990 : /* Same for suffix */
8991 4849559 : for (suffixlen = 0; suffixlen < Min(oldlen, newlen) - prefixlen; suffixlen++)
8992 : {
8993 4823052 : if (newp[newlen - suffixlen - 1] != oldp[oldlen - suffixlen - 1])
8994 121750 : break;
8995 : }
8996 148257 : if (suffixlen < 3)
8997 37861 : suffixlen = 0;
8998 : }
8999 :
9000 : /* Prepare main WAL data chain */
9001 299869 : xlrec.flags = 0;
9002 299869 : if (all_visible_cleared)
9003 1745 : xlrec.flags |= XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED;
9004 299869 : if (new_all_visible_cleared)
9005 712 : xlrec.flags |= XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED;
9006 299869 : if (prefixlen > 0)
9007 125138 : xlrec.flags |= XLH_UPDATE_PREFIX_FROM_OLD;
9008 299869 : if (suffixlen > 0)
9009 110396 : xlrec.flags |= XLH_UPDATE_SUFFIX_FROM_OLD;
9010 299869 : if (need_tuple_data)
9011 : {
9012 47026 : xlrec.flags |= XLH_UPDATE_CONTAINS_NEW_TUPLE;
9013 47026 : if (old_key_tuple)
9014 : {
9015 146 : if (reln->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
9016 65 : xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_TUPLE;
9017 : else
9018 81 : xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_KEY;
9019 : }
9020 : }
9021 :
9022 : /* If new tuple is the single and first tuple on page... */
9023 303307 : if (ItemPointerGetOffsetNumber(&(newtup->t_self)) == FirstOffsetNumber &&
9024 3438 : PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
9025 : {
9026 3213 : info |= XLOG_HEAP_INIT_PAGE;
9027 3213 : init = true;
9028 : }
9029 : else
9030 296656 : init = false;
9031 :
9032 : /* Prepare WAL data for the old page */
9033 299869 : xlrec.old_offnum = ItemPointerGetOffsetNumber(&oldtup->t_self);
9034 299869 : xlrec.old_xmax = HeapTupleHeaderGetRawXmax(oldtup->t_data);
9035 599738 : xlrec.old_infobits_set = compute_infobits(oldtup->t_data->t_infomask,
9036 299869 : oldtup->t_data->t_infomask2);
9037 :
9038 : /* Prepare WAL data for the new page */
9039 299869 : xlrec.new_offnum = ItemPointerGetOffsetNumber(&newtup->t_self);
9040 299869 : xlrec.new_xmax = HeapTupleHeaderGetRawXmax(newtup->t_data);
9041 :
9042 299869 : bufflags = REGBUF_STANDARD;
9043 299869 : if (init)
9044 3213 : bufflags |= REGBUF_WILL_INIT;
9045 299869 : if (need_tuple_data)
9046 47026 : bufflags |= REGBUF_KEEP_DATA;
9047 :
9048 299869 : XLogRegisterBuffer(0, newbuf, bufflags);
9049 299869 : if (oldbuf != newbuf)
9050 139036 : XLogRegisterBuffer(1, oldbuf, REGBUF_STANDARD);
9051 :
9052 299869 : XLogRegisterData(&xlrec, SizeOfHeapUpdate);
9053 :
9054 : /*
9055 : * Prepare WAL data for the new tuple.
9056 : */
9057 299869 : if (prefixlen > 0 || suffixlen > 0)
9058 : {
9059 147042 : if (prefixlen > 0 && suffixlen > 0)
9060 : {
9061 88492 : prefix_suffix[0] = prefixlen;
9062 88492 : prefix_suffix[1] = suffixlen;
9063 88492 : XLogRegisterBufData(0, &prefix_suffix, sizeof(uint16) * 2);
9064 : }
9065 58550 : else if (prefixlen > 0)
9066 : {
9067 36646 : XLogRegisterBufData(0, &prefixlen, sizeof(uint16));
9068 : }
9069 : else
9070 : {
9071 21904 : XLogRegisterBufData(0, &suffixlen, sizeof(uint16));
9072 : }
9073 : }
9074 :
9075 299869 : xlhdr.t_infomask2 = newtup->t_data->t_infomask2;
9076 299869 : xlhdr.t_infomask = newtup->t_data->t_infomask;
9077 299869 : xlhdr.t_hoff = newtup->t_data->t_hoff;
9078 : Assert(SizeofHeapTupleHeader + prefixlen + suffixlen <= newtup->t_len);
9079 :
9080 : /*
9081 : * PG73FORMAT: write bitmap [+ padding] [+ oid] + data
9082 : *
9083 : * The 'data' doesn't include the common prefix or suffix.
9084 : */
9085 299869 : XLogRegisterBufData(0, &xlhdr, SizeOfHeapHeader);
9086 299869 : if (prefixlen == 0)
9087 : {
9088 174731 : XLogRegisterBufData(0,
9089 174731 : (char *) newtup->t_data + SizeofHeapTupleHeader,
9090 174731 : newtup->t_len - SizeofHeapTupleHeader - suffixlen);
9091 : }
9092 : else
9093 : {
9094 : /*
9095 : * Have to write the null bitmap and data after the common prefix as
9096 : * two separate rdata entries.
9097 : */
9098 : /* bitmap [+ padding] [+ oid] */
9099 125138 : if (newtup->t_data->t_hoff - SizeofHeapTupleHeader > 0)
9100 : {
9101 125138 : XLogRegisterBufData(0,
9102 125138 : (char *) newtup->t_data + SizeofHeapTupleHeader,
9103 125138 : newtup->t_data->t_hoff - SizeofHeapTupleHeader);
9104 : }
9105 :
9106 : /* data after common prefix */
9107 125138 : XLogRegisterBufData(0,
9108 125138 : (char *) newtup->t_data + newtup->t_data->t_hoff + prefixlen,
9109 125138 : newtup->t_len - newtup->t_data->t_hoff - prefixlen - suffixlen);
9110 : }
9111 :
9112 : /* We need to log a tuple identity */
9113 299869 : if (need_tuple_data && old_key_tuple)
9114 : {
9115 : /* don't really need this, but its more comfy to decode */
9116 146 : xlhdr_idx.t_infomask2 = old_key_tuple->t_data->t_infomask2;
9117 146 : xlhdr_idx.t_infomask = old_key_tuple->t_data->t_infomask;
9118 146 : xlhdr_idx.t_hoff = old_key_tuple->t_data->t_hoff;
9119 :
9120 146 : XLogRegisterData(&xlhdr_idx, SizeOfHeapHeader);
9121 :
9122 : /* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
9123 146 : XLogRegisterData((char *) old_key_tuple->t_data + SizeofHeapTupleHeader,
9124 146 : old_key_tuple->t_len - SizeofHeapTupleHeader);
9125 : }
9126 :
9127 : /* filtering by origin on a row level is much more efficient */
9128 299869 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
9129 :
9130 299869 : recptr = XLogInsert(RM_HEAP_ID, info);
9131 :
9132 299869 : return recptr;
9133 : }
9134 :
9135 : /*
9136 : * Perform XLogInsert of an XLOG_HEAP2_NEW_CID record
9137 : *
9138 : * This is only used when effective_wal_level is logical, and only for
9139 : * catalog tuples.
9140 : */
9141 : static XLogRecPtr
9142 24667 : log_heap_new_cid(Relation relation, HeapTuple tup)
9143 : {
9144 : xl_heap_new_cid xlrec;
9145 :
9146 : XLogRecPtr recptr;
9147 24667 : HeapTupleHeader hdr = tup->t_data;
9148 :
9149 : Assert(ItemPointerIsValid(&tup->t_self));
9150 : Assert(tup->t_tableOid != InvalidOid);
9151 :
9152 24667 : xlrec.top_xid = GetTopTransactionId();
9153 24667 : xlrec.target_locator = relation->rd_locator;
9154 24667 : xlrec.target_tid = tup->t_self;
9155 :
9156 : /*
9157 : * If the tuple got inserted & deleted in the same TX we definitely have a
9158 : * combo CID, set cmin and cmax.
9159 : */
9160 24667 : if (hdr->t_infomask & HEAP_COMBOCID)
9161 : {
9162 : Assert(!(hdr->t_infomask & HEAP_XMAX_INVALID));
9163 : Assert(!HeapTupleHeaderXminInvalid(hdr));
9164 2024 : xlrec.cmin = HeapTupleHeaderGetCmin(hdr);
9165 2024 : xlrec.cmax = HeapTupleHeaderGetCmax(hdr);
9166 2024 : xlrec.combocid = HeapTupleHeaderGetRawCommandId(hdr);
9167 : }
9168 : /* No combo CID, so only cmin or cmax can be set by this TX */
9169 : else
9170 : {
9171 : /*
9172 : * Tuple inserted.
9173 : *
9174 : * We need to check for LOCK ONLY because multixacts might be
9175 : * transferred to the new tuple in case of FOR KEY SHARE updates in
9176 : * which case there will be an xmax, although the tuple just got
9177 : * inserted.
9178 : */
9179 29483 : if (hdr->t_infomask & HEAP_XMAX_INVALID ||
9180 6840 : HEAP_XMAX_IS_LOCKED_ONLY(hdr->t_infomask))
9181 : {
9182 15804 : xlrec.cmin = HeapTupleHeaderGetRawCommandId(hdr);
9183 15804 : xlrec.cmax = InvalidCommandId;
9184 : }
9185 : /* Tuple from a different tx updated or deleted. */
9186 : else
9187 : {
9188 6839 : xlrec.cmin = InvalidCommandId;
9189 6839 : xlrec.cmax = HeapTupleHeaderGetRawCommandId(hdr);
9190 : }
9191 22643 : xlrec.combocid = InvalidCommandId;
9192 : }
9193 :
9194 : /*
9195 : * Note that we don't need to register the buffer here, because this
9196 : * operation does not modify the page. The insert/update/delete that
9197 : * called us certainly did, but that's WAL-logged separately.
9198 : */
9199 24667 : XLogBeginInsert();
9200 24667 : XLogRegisterData(&xlrec, SizeOfHeapNewCid);
9201 :
9202 : /* will be looked at irrespective of origin */
9203 :
9204 24667 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_NEW_CID);
9205 :
9206 24667 : return recptr;
9207 : }
9208 :
9209 : /*
9210 : * Build a heap tuple representing the configured REPLICA IDENTITY to represent
9211 : * the old tuple in an UPDATE or DELETE.
9212 : *
9213 : * Returns NULL if there's no need to log an identity or if there's no suitable
9214 : * key defined.
9215 : *
9216 : * Pass key_required true if any replica identity columns changed value, or if
9217 : * any of them have any external data. Delete must always pass true.
9218 : *
9219 : * *copy is set to true if the returned tuple is a modified copy rather than
9220 : * the same tuple that was passed in.
9221 : */
9222 : static HeapTuple
9223 1839126 : ExtractReplicaIdentity(Relation relation, HeapTuple tp, bool key_required,
9224 : bool *copy)
9225 : {
9226 1839126 : TupleDesc desc = RelationGetDescr(relation);
9227 1839126 : char replident = relation->rd_rel->relreplident;
9228 : Bitmapset *idattrs;
9229 : HeapTuple key_tuple;
9230 : bool nulls[MaxHeapAttributeNumber];
9231 : Datum values[MaxHeapAttributeNumber];
9232 :
9233 1839126 : *copy = false;
9234 :
9235 1839126 : if (!RelationIsLogicallyLogged(relation))
9236 1738828 : return NULL;
9237 :
9238 100298 : if (replident == REPLICA_IDENTITY_NOTHING)
9239 231 : return NULL;
9240 :
9241 100067 : if (replident == REPLICA_IDENTITY_FULL)
9242 : {
9243 : /*
9244 : * When logging the entire old tuple, it very well could contain
9245 : * toasted columns. If so, force them to be inlined.
9246 : */
9247 197 : if (HeapTupleHasExternal(tp))
9248 : {
9249 4 : *copy = true;
9250 4 : tp = toast_flatten_tuple(tp, desc);
9251 : }
9252 197 : return tp;
9253 : }
9254 :
9255 : /* if the key isn't required and we're only logging the key, we're done */
9256 99870 : if (!key_required)
9257 46880 : return NULL;
9258 :
9259 : /* find out the replica identity columns */
9260 52990 : idattrs = RelationGetIndexAttrBitmap(relation,
9261 : INDEX_ATTR_BITMAP_IDENTITY_KEY);
9262 :
9263 : /*
9264 : * If there's no defined replica identity columns, treat as !key_required.
9265 : * (This case should not be reachable from heap_update, since that should
9266 : * calculate key_required accurately. But heap_delete just passes
9267 : * constant true for key_required, so we can hit this case in deletes.)
9268 : */
9269 52990 : if (bms_is_empty(idattrs))
9270 6021 : return NULL;
9271 :
9272 : /*
9273 : * Construct a new tuple containing only the replica identity columns,
9274 : * with nulls elsewhere. While we're at it, assert that the replica
9275 : * identity columns aren't null.
9276 : */
9277 46969 : heap_deform_tuple(tp, desc, values, nulls);
9278 :
9279 150901 : for (int i = 0; i < desc->natts; i++)
9280 : {
9281 103932 : if (bms_is_member(i + 1 - FirstLowInvalidHeapAttributeNumber,
9282 : idattrs))
9283 : Assert(!nulls[i]);
9284 : else
9285 56951 : nulls[i] = true;
9286 : }
9287 :
9288 46969 : key_tuple = heap_form_tuple(desc, values, nulls);
9289 46969 : *copy = true;
9290 :
9291 46969 : bms_free(idattrs);
9292 :
9293 : /*
9294 : * If the tuple, which by here only contains indexed columns, still has
9295 : * toasted columns, force them to be inlined. This is somewhat unlikely
9296 : * since there's limits on the size of indexed columns, so we don't
9297 : * duplicate toast_flatten_tuple()s functionality in the above loop over
9298 : * the indexed columns, even if it would be more efficient.
9299 : */
9300 46969 : if (HeapTupleHasExternal(key_tuple))
9301 : {
9302 4 : HeapTuple oldtup = key_tuple;
9303 :
9304 4 : key_tuple = toast_flatten_tuple(oldtup, desc);
9305 4 : heap_freetuple(oldtup);
9306 : }
9307 :
9308 46969 : return key_tuple;
9309 : }
9310 :
9311 : /*
9312 : * HeapCheckForSerializableConflictOut
9313 : * We are reading a tuple. If it's not visible, there may be a
9314 : * rw-conflict out with the inserter. Otherwise, if it is visible to us
9315 : * but has been deleted, there may be a rw-conflict out with the deleter.
9316 : *
9317 : * We will determine the top level xid of the writing transaction with which
9318 : * we may be in conflict, and ask CheckForSerializableConflictOut() to check
9319 : * for overlap with our own transaction.
9320 : *
9321 : * This function should be called just about anywhere in heapam.c where a
9322 : * tuple has been read. The caller must hold at least a shared lock on the
9323 : * buffer, because this function might set hint bits on the tuple. There is
9324 : * currently no known reason to call this function from an index AM.
9325 : */
9326 : void
9327 32450506 : HeapCheckForSerializableConflictOut(bool visible, Relation relation,
9328 : HeapTuple tuple, Buffer buffer,
9329 : Snapshot snapshot)
9330 : {
9331 : TransactionId xid;
9332 : HTSV_Result htsvResult;
9333 :
9334 32450506 : if (!CheckForSerializableConflictOutNeeded(relation, snapshot))
9335 32425132 : return;
9336 :
9337 : /*
9338 : * Check to see whether the tuple has been written to by a concurrent
9339 : * transaction, either to create it not visible to us, or to delete it
9340 : * while it is visible to us. The "visible" bool indicates whether the
9341 : * tuple is visible to us, while HeapTupleSatisfiesVacuum checks what else
9342 : * is going on with it.
9343 : *
9344 : * In the event of a concurrently inserted tuple that also happens to have
9345 : * been concurrently updated (by a separate transaction), the xmin of the
9346 : * tuple will be used -- not the updater's xid.
9347 : */
9348 25374 : htsvResult = HeapTupleSatisfiesVacuum(tuple, TransactionXmin, buffer);
9349 25374 : switch (htsvResult)
9350 : {
9351 24561 : case HEAPTUPLE_LIVE:
9352 24561 : if (visible)
9353 24548 : return;
9354 13 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9355 13 : break;
9356 361 : case HEAPTUPLE_RECENTLY_DEAD:
9357 : case HEAPTUPLE_DELETE_IN_PROGRESS:
9358 361 : if (visible)
9359 285 : xid = HeapTupleHeaderGetUpdateXid(tuple->t_data);
9360 : else
9361 76 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9362 :
9363 361 : if (TransactionIdPrecedes(xid, TransactionXmin))
9364 : {
9365 : /* This is like the HEAPTUPLE_DEAD case */
9366 : Assert(!visible);
9367 70 : return;
9368 : }
9369 291 : break;
9370 328 : case HEAPTUPLE_INSERT_IN_PROGRESS:
9371 328 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9372 328 : break;
9373 124 : case HEAPTUPLE_DEAD:
9374 : Assert(!visible);
9375 124 : return;
9376 0 : default:
9377 :
9378 : /*
9379 : * The only way to get to this default clause is if a new value is
9380 : * added to the enum type without adding it to this switch
9381 : * statement. That's a bug, so elog.
9382 : */
9383 0 : elog(ERROR, "unrecognized return value from HeapTupleSatisfiesVacuum: %u", htsvResult);
9384 :
9385 : /*
9386 : * In spite of having all enum values covered and calling elog on
9387 : * this default, some compilers think this is a code path which
9388 : * allows xid to be used below without initialization. Silence
9389 : * that warning.
9390 : */
9391 : xid = InvalidTransactionId;
9392 : }
9393 :
9394 : Assert(TransactionIdIsValid(xid));
9395 : Assert(TransactionIdFollowsOrEquals(xid, TransactionXmin));
9396 :
9397 : /*
9398 : * Find top level xid. Bail out if xid is too early to be a conflict, or
9399 : * if it's our own xid.
9400 : */
9401 632 : if (TransactionIdEquals(xid, GetTopTransactionIdIfAny()))
9402 64 : return;
9403 568 : xid = SubTransGetTopmostTransaction(xid);
9404 568 : if (TransactionIdPrecedes(xid, TransactionXmin))
9405 0 : return;
9406 :
9407 568 : CheckForSerializableConflictOut(relation, xid, snapshot);
9408 : }
|
