Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * heapam.c
4 : * heap access method code
5 : *
6 : * Portions Copyright (c) 1996-2025, 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/procarray.h"
51 : #include "utils/datum.h"
52 : #include "utils/injection_point.h"
53 : #include "utils/inval.h"
54 : #include "utils/spccache.h"
55 : #include "utils/syscache.h"
56 :
57 :
58 : static HeapTuple heap_prepare_insert(Relation relation, HeapTuple tup,
59 : TransactionId xid, CommandId cid, int options);
60 : static XLogRecPtr log_heap_update(Relation reln, Buffer oldbuf,
61 : Buffer newbuf, HeapTuple oldtup,
62 : HeapTuple newtup, HeapTuple old_key_tuple,
63 : bool all_visible_cleared, bool new_all_visible_cleared);
64 : #ifdef USE_ASSERT_CHECKING
65 : static void check_lock_if_inplace_updateable_rel(Relation relation,
66 : const ItemPointerData *otid,
67 : HeapTuple newtup);
68 : static void check_inplace_rel_lock(HeapTuple oldtup);
69 : #endif
70 : static Bitmapset *HeapDetermineColumnsInfo(Relation relation,
71 : Bitmapset *interesting_cols,
72 : Bitmapset *external_cols,
73 : HeapTuple oldtup, HeapTuple newtup,
74 : bool *has_external);
75 : static bool heap_acquire_tuplock(Relation relation, const ItemPointerData *tid,
76 : LockTupleMode mode, LockWaitPolicy wait_policy,
77 : bool *have_tuple_lock);
78 : static inline BlockNumber heapgettup_advance_block(HeapScanDesc scan,
79 : BlockNumber block,
80 : ScanDirection dir);
81 : static pg_noinline BlockNumber heapgettup_initial_block(HeapScanDesc scan,
82 : ScanDirection dir);
83 : static void compute_new_xmax_infomask(TransactionId xmax, uint16 old_infomask,
84 : uint16 old_infomask2, TransactionId add_to_xmax,
85 : LockTupleMode mode, bool is_update,
86 : TransactionId *result_xmax, uint16 *result_infomask,
87 : uint16 *result_infomask2);
88 : static TM_Result heap_lock_updated_tuple(Relation rel, HeapTuple tuple,
89 : const ItemPointerData *ctid, TransactionId xid,
90 : LockTupleMode mode);
91 : static void GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
92 : uint16 *new_infomask2);
93 : static TransactionId MultiXactIdGetUpdateXid(TransactionId xmax,
94 : uint16 t_infomask);
95 : static bool DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
96 : LockTupleMode lockmode, bool *current_is_member);
97 : static void MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
98 : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
99 : int *remaining);
100 : static bool ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status,
101 : uint16 infomask, Relation rel, int *remaining,
102 : bool logLockFailure);
103 : static void index_delete_sort(TM_IndexDeleteOp *delstate);
104 : static int bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate);
105 : static XLogRecPtr log_heap_new_cid(Relation relation, HeapTuple tup);
106 : static HeapTuple ExtractReplicaIdentity(Relation relation, HeapTuple tp, bool key_required,
107 : bool *copy);
108 :
109 :
110 : /*
111 : * Each tuple lock mode has a corresponding heavyweight lock, and one or two
112 : * corresponding MultiXactStatuses (one to merely lock tuples, another one to
113 : * update them). This table (and the macros below) helps us determine the
114 : * heavyweight lock mode and MultiXactStatus values to use for any particular
115 : * tuple lock strength.
116 : *
117 : * These interact with InplaceUpdateTupleLock, an alias for ExclusiveLock.
118 : *
119 : * Don't look at lockstatus/updstatus directly! Use get_mxact_status_for_lock
120 : * instead.
121 : */
122 : static const struct
123 : {
124 : LOCKMODE hwlock;
125 : int lockstatus;
126 : int updstatus;
127 : }
128 :
129 : tupleLockExtraInfo[MaxLockTupleMode + 1] =
130 : {
131 : { /* LockTupleKeyShare */
132 : AccessShareLock,
133 : MultiXactStatusForKeyShare,
134 : -1 /* KeyShare does not allow updating tuples */
135 : },
136 : { /* LockTupleShare */
137 : RowShareLock,
138 : MultiXactStatusForShare,
139 : -1 /* Share does not allow updating tuples */
140 : },
141 : { /* LockTupleNoKeyExclusive */
142 : ExclusiveLock,
143 : MultiXactStatusForNoKeyUpdate,
144 : MultiXactStatusNoKeyUpdate
145 : },
146 : { /* LockTupleExclusive */
147 : AccessExclusiveLock,
148 : MultiXactStatusForUpdate,
149 : MultiXactStatusUpdate
150 : }
151 : };
152 :
153 : /* Get the LOCKMODE for a given MultiXactStatus */
154 : #define LOCKMODE_from_mxstatus(status) \
155 : (tupleLockExtraInfo[TUPLOCK_from_mxstatus((status))].hwlock)
156 :
157 : /*
158 : * Acquire heavyweight locks on tuples, using a LockTupleMode strength value.
159 : * This is more readable than having every caller translate it to lock.h's
160 : * LOCKMODE.
161 : */
162 : #define LockTupleTuplock(rel, tup, mode) \
163 : LockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
164 : #define UnlockTupleTuplock(rel, tup, mode) \
165 : UnlockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
166 : #define ConditionalLockTupleTuplock(rel, tup, mode, log) \
167 : ConditionalLockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock, (log))
168 :
169 : #ifdef USE_PREFETCH
170 : /*
171 : * heap_index_delete_tuples and index_delete_prefetch_buffer use this
172 : * structure to coordinate prefetching activity
173 : */
174 : typedef struct
175 : {
176 : BlockNumber cur_hblkno;
177 : int next_item;
178 : int ndeltids;
179 : TM_IndexDelete *deltids;
180 : } IndexDeletePrefetchState;
181 : #endif
182 :
183 : /* heap_index_delete_tuples bottom-up index deletion costing constants */
184 : #define BOTTOMUP_MAX_NBLOCKS 6
185 : #define BOTTOMUP_TOLERANCE_NBLOCKS 3
186 :
187 : /*
188 : * heap_index_delete_tuples uses this when determining which heap blocks it
189 : * must visit to help its bottom-up index deletion caller
190 : */
191 : typedef struct IndexDeleteCounts
192 : {
193 : int16 npromisingtids; /* Number of "promising" TIDs in group */
194 : int16 ntids; /* Number of TIDs in group */
195 : int16 ifirsttid; /* Offset to group's first deltid */
196 : } IndexDeleteCounts;
197 :
198 : /*
199 : * This table maps tuple lock strength values for each particular
200 : * MultiXactStatus value.
201 : */
202 : static const int MultiXactStatusLock[MaxMultiXactStatus + 1] =
203 : {
204 : LockTupleKeyShare, /* ForKeyShare */
205 : LockTupleShare, /* ForShare */
206 : LockTupleNoKeyExclusive, /* ForNoKeyUpdate */
207 : LockTupleExclusive, /* ForUpdate */
208 : LockTupleNoKeyExclusive, /* NoKeyUpdate */
209 : LockTupleExclusive /* Update */
210 : };
211 :
212 : /* Get the LockTupleMode for a given MultiXactStatus */
213 : #define TUPLOCK_from_mxstatus(status) \
214 : (MultiXactStatusLock[(status)])
215 :
216 : /*
217 : * Check that we have a valid snapshot if we might need TOAST access.
218 : */
219 : static inline void
220 21030770 : AssertHasSnapshotForToast(Relation rel)
221 : {
222 : #ifdef USE_ASSERT_CHECKING
223 :
224 : /* bootstrap mode in particular breaks this rule */
225 : if (!IsNormalProcessingMode())
226 : return;
227 :
228 : /* if the relation doesn't have a TOAST table, we are good */
229 : if (!OidIsValid(rel->rd_rel->reltoastrelid))
230 : return;
231 :
232 : Assert(HaveRegisteredOrActiveSnapshot());
233 :
234 : #endif /* USE_ASSERT_CHECKING */
235 21030770 : }
236 :
237 : /* ----------------------------------------------------------------
238 : * heap support routines
239 : * ----------------------------------------------------------------
240 : */
241 :
242 : /*
243 : * Streaming read API callback for parallel sequential scans. Returns the next
244 : * block the caller wants from the read stream or InvalidBlockNumber when done.
245 : */
246 : static BlockNumber
247 202546 : heap_scan_stream_read_next_parallel(ReadStream *stream,
248 : void *callback_private_data,
249 : void *per_buffer_data)
250 : {
251 202546 : HeapScanDesc scan = (HeapScanDesc) callback_private_data;
252 :
253 : Assert(ScanDirectionIsForward(scan->rs_dir));
254 : Assert(scan->rs_base.rs_parallel);
255 :
256 202546 : if (unlikely(!scan->rs_inited))
257 : {
258 : /* parallel scan */
259 3154 : table_block_parallelscan_startblock_init(scan->rs_base.rs_rd,
260 3154 : scan->rs_parallelworkerdata,
261 3154 : (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel,
262 : scan->rs_startblock,
263 : scan->rs_numblocks);
264 :
265 : /* may return InvalidBlockNumber if there are no more blocks */
266 6308 : scan->rs_prefetch_block = table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
267 3154 : scan->rs_parallelworkerdata,
268 3154 : (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel);
269 3154 : scan->rs_inited = true;
270 : }
271 : else
272 : {
273 199392 : scan->rs_prefetch_block = table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
274 199392 : scan->rs_parallelworkerdata, (ParallelBlockTableScanDesc)
275 199392 : scan->rs_base.rs_parallel);
276 : }
277 :
278 202546 : return scan->rs_prefetch_block;
279 : }
280 :
281 : /*
282 : * Streaming read API callback for serial sequential and TID range scans.
283 : * Returns the next block the caller wants from the read stream or
284 : * InvalidBlockNumber when done.
285 : */
286 : static BlockNumber
287 7561364 : heap_scan_stream_read_next_serial(ReadStream *stream,
288 : void *callback_private_data,
289 : void *per_buffer_data)
290 : {
291 7561364 : HeapScanDesc scan = (HeapScanDesc) callback_private_data;
292 :
293 7561364 : if (unlikely(!scan->rs_inited))
294 : {
295 1959338 : scan->rs_prefetch_block = heapgettup_initial_block(scan, scan->rs_dir);
296 1959338 : scan->rs_inited = true;
297 : }
298 : else
299 5602026 : scan->rs_prefetch_block = heapgettup_advance_block(scan,
300 : scan->rs_prefetch_block,
301 : scan->rs_dir);
302 :
303 7561364 : return scan->rs_prefetch_block;
304 : }
305 :
306 : /*
307 : * Read stream API callback for bitmap heap scans.
308 : * Returns the next block the caller wants from the read stream or
309 : * InvalidBlockNumber when done.
310 : */
311 : static BlockNumber
312 421696 : bitmapheap_stream_read_next(ReadStream *pgsr, void *private_data,
313 : void *per_buffer_data)
314 : {
315 421696 : TBMIterateResult *tbmres = per_buffer_data;
316 421696 : BitmapHeapScanDesc bscan = (BitmapHeapScanDesc) private_data;
317 421696 : HeapScanDesc hscan = (HeapScanDesc) bscan;
318 421696 : TableScanDesc sscan = &hscan->rs_base;
319 :
320 : for (;;)
321 : {
322 421696 : CHECK_FOR_INTERRUPTS();
323 :
324 : /* no more entries in the bitmap */
325 421696 : if (!tbm_iterate(&sscan->st.rs_tbmiterator, tbmres))
326 20550 : return InvalidBlockNumber;
327 :
328 : /*
329 : * Ignore any claimed entries past what we think is the end of the
330 : * relation. It may have been extended after the start of our scan (we
331 : * only hold an AccessShareLock, and it could be inserts from this
332 : * backend). We don't take this optimization in SERIALIZABLE
333 : * isolation though, as we need to examine all invisible tuples
334 : * reachable by the index.
335 : */
336 401146 : if (!IsolationIsSerializable() &&
337 400928 : tbmres->blockno >= hscan->rs_nblocks)
338 0 : continue;
339 :
340 401146 : return tbmres->blockno;
341 : }
342 :
343 : /* not reachable */
344 : Assert(false);
345 : }
346 :
347 : /* ----------------
348 : * initscan - scan code common to heap_beginscan and heap_rescan
349 : * ----------------
350 : */
351 : static void
352 2002806 : initscan(HeapScanDesc scan, ScanKey key, bool keep_startblock)
353 : {
354 2002806 : ParallelBlockTableScanDesc bpscan = NULL;
355 : bool allow_strat;
356 : bool allow_sync;
357 :
358 : /*
359 : * Determine the number of blocks we have to scan.
360 : *
361 : * It is sufficient to do this once at scan start, since any tuples added
362 : * while the scan is in progress will be invisible to my snapshot anyway.
363 : * (That is not true when using a non-MVCC snapshot. However, we couldn't
364 : * guarantee to return tuples added after scan start anyway, since they
365 : * might go into pages we already scanned. To guarantee consistent
366 : * results for a non-MVCC snapshot, the caller must hold some higher-level
367 : * lock that ensures the interesting tuple(s) won't change.)
368 : */
369 2002806 : if (scan->rs_base.rs_parallel != NULL)
370 : {
371 4338 : bpscan = (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel;
372 4338 : scan->rs_nblocks = bpscan->phs_nblocks;
373 : }
374 : else
375 1998468 : scan->rs_nblocks = RelationGetNumberOfBlocks(scan->rs_base.rs_rd);
376 :
377 : /*
378 : * If the table is large relative to NBuffers, use a bulk-read access
379 : * strategy and enable synchronized scanning (see syncscan.c). Although
380 : * the thresholds for these features could be different, we make them the
381 : * same so that there are only two behaviors to tune rather than four.
382 : * (However, some callers need to be able to disable one or both of these
383 : * behaviors, independently of the size of the table; also there is a GUC
384 : * variable that can disable synchronized scanning.)
385 : *
386 : * Note that table_block_parallelscan_initialize has a very similar test;
387 : * if you change this, consider changing that one, too.
388 : */
389 2002802 : if (!RelationUsesLocalBuffers(scan->rs_base.rs_rd) &&
390 1988202 : scan->rs_nblocks > NBuffers / 4)
391 : {
392 28420 : allow_strat = (scan->rs_base.rs_flags & SO_ALLOW_STRAT) != 0;
393 28420 : allow_sync = (scan->rs_base.rs_flags & SO_ALLOW_SYNC) != 0;
394 : }
395 : else
396 1974382 : allow_strat = allow_sync = false;
397 :
398 2002802 : if (allow_strat)
399 : {
400 : /* During a rescan, keep the previous strategy object. */
401 25772 : if (scan->rs_strategy == NULL)
402 25430 : scan->rs_strategy = GetAccessStrategy(BAS_BULKREAD);
403 : }
404 : else
405 : {
406 1977030 : if (scan->rs_strategy != NULL)
407 0 : FreeAccessStrategy(scan->rs_strategy);
408 1977030 : scan->rs_strategy = NULL;
409 : }
410 :
411 2002802 : if (scan->rs_base.rs_parallel != NULL)
412 : {
413 : /* For parallel scan, believe whatever ParallelTableScanDesc says. */
414 4338 : if (scan->rs_base.rs_parallel->phs_syncscan)
415 4 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
416 : else
417 4334 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
418 :
419 : /*
420 : * If not rescanning, initialize the startblock. Finding the actual
421 : * start location is done in table_block_parallelscan_startblock_init,
422 : * based on whether an alternative start location has been set with
423 : * heap_setscanlimits, or using the syncscan location, when syncscan
424 : * is enabled.
425 : */
426 4338 : if (!keep_startblock)
427 4110 : scan->rs_startblock = InvalidBlockNumber;
428 : }
429 : else
430 : {
431 1998464 : if (keep_startblock)
432 : {
433 : /*
434 : * When rescanning, we want to keep the previous startblock
435 : * setting, so that rewinding a cursor doesn't generate surprising
436 : * results. Reset the active syncscan setting, though.
437 : */
438 1232100 : if (allow_sync && synchronize_seqscans)
439 100 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
440 : else
441 1232000 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
442 : }
443 766364 : else if (allow_sync && synchronize_seqscans)
444 : {
445 144 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
446 144 : scan->rs_startblock = ss_get_location(scan->rs_base.rs_rd, scan->rs_nblocks);
447 : }
448 : else
449 : {
450 766220 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
451 766220 : scan->rs_startblock = 0;
452 : }
453 : }
454 :
455 2002802 : scan->rs_numblocks = InvalidBlockNumber;
456 2002802 : scan->rs_inited = false;
457 2002802 : scan->rs_ctup.t_data = NULL;
458 2002802 : ItemPointerSetInvalid(&scan->rs_ctup.t_self);
459 2002802 : scan->rs_cbuf = InvalidBuffer;
460 2002802 : scan->rs_cblock = InvalidBlockNumber;
461 2002802 : scan->rs_ntuples = 0;
462 2002802 : scan->rs_cindex = 0;
463 :
464 : /*
465 : * Initialize to ForwardScanDirection because it is most common and
466 : * because heap scans go forward before going backward (e.g. CURSORs).
467 : */
468 2002802 : scan->rs_dir = ForwardScanDirection;
469 2002802 : scan->rs_prefetch_block = InvalidBlockNumber;
470 :
471 : /* page-at-a-time fields are always invalid when not rs_inited */
472 :
473 : /*
474 : * copy the scan key, if appropriate
475 : */
476 2002802 : if (key != NULL && scan->rs_base.rs_nkeys > 0)
477 442620 : memcpy(scan->rs_base.rs_key, key, scan->rs_base.rs_nkeys * sizeof(ScanKeyData));
478 :
479 : /*
480 : * Currently, we only have a stats counter for sequential heap scans (but
481 : * e.g for bitmap scans the underlying bitmap index scans will be counted,
482 : * and for sample scans we update stats for tuple fetches).
483 : */
484 2002802 : if (scan->rs_base.rs_flags & SO_TYPE_SEQSCAN)
485 1961734 : pgstat_count_heap_scan(scan->rs_base.rs_rd);
486 2002802 : }
487 :
488 : /*
489 : * heap_setscanlimits - restrict range of a heapscan
490 : *
491 : * startBlk is the page to start at
492 : * numBlks is number of pages to scan (InvalidBlockNumber means "all")
493 : */
494 : void
495 5802 : heap_setscanlimits(TableScanDesc sscan, BlockNumber startBlk, BlockNumber numBlks)
496 : {
497 5802 : HeapScanDesc scan = (HeapScanDesc) sscan;
498 :
499 : Assert(!scan->rs_inited); /* else too late to change */
500 : /* else rs_startblock is significant */
501 : Assert(!(scan->rs_base.rs_flags & SO_ALLOW_SYNC));
502 :
503 : /* Check startBlk is valid (but allow case of zero blocks...) */
504 : Assert(startBlk == 0 || startBlk < scan->rs_nblocks);
505 :
506 5802 : scan->rs_startblock = startBlk;
507 5802 : scan->rs_numblocks = numBlks;
508 5802 : }
509 :
510 : /*
511 : * Per-tuple loop for heap_prepare_pagescan(). Pulled out so it can be called
512 : * multiple times, with constant arguments for all_visible,
513 : * check_serializable.
514 : */
515 : pg_attribute_always_inline
516 : static int
517 5576570 : page_collect_tuples(HeapScanDesc scan, Snapshot snapshot,
518 : Page page, Buffer buffer,
519 : BlockNumber block, int lines,
520 : bool all_visible, bool check_serializable)
521 : {
522 5576570 : int ntup = 0;
523 : OffsetNumber lineoff;
524 :
525 282063236 : for (lineoff = FirstOffsetNumber; lineoff <= lines; lineoff++)
526 : {
527 276486682 : ItemId lpp = PageGetItemId(page, lineoff);
528 : HeapTupleData loctup;
529 : bool valid;
530 :
531 276486682 : if (!ItemIdIsNormal(lpp))
532 58941702 : continue;
533 :
534 217544980 : loctup.t_data = (HeapTupleHeader) PageGetItem(page, lpp);
535 217544980 : loctup.t_len = ItemIdGetLength(lpp);
536 217544980 : loctup.t_tableOid = RelationGetRelid(scan->rs_base.rs_rd);
537 217544980 : ItemPointerSet(&(loctup.t_self), block, lineoff);
538 :
539 217544980 : if (all_visible)
540 85851128 : valid = true;
541 : else
542 131693852 : valid = HeapTupleSatisfiesVisibility(&loctup, snapshot, buffer);
543 :
544 217544980 : if (check_serializable)
545 2818 : HeapCheckForSerializableConflictOut(valid, scan->rs_base.rs_rd,
546 : &loctup, buffer, snapshot);
547 :
548 217544964 : if (valid)
549 : {
550 200812564 : scan->rs_vistuples[ntup] = lineoff;
551 200812564 : ntup++;
552 : }
553 : }
554 :
555 : Assert(ntup <= MaxHeapTuplesPerPage);
556 :
557 5576554 : return ntup;
558 : }
559 :
560 : /*
561 : * heap_prepare_pagescan - Prepare current scan page to be scanned in pagemode
562 : *
563 : * Preparation currently consists of 1. prune the scan's rs_cbuf page, and 2.
564 : * fill the rs_vistuples[] array with the OffsetNumbers of visible tuples.
565 : */
566 : void
567 5576570 : heap_prepare_pagescan(TableScanDesc sscan)
568 : {
569 5576570 : HeapScanDesc scan = (HeapScanDesc) sscan;
570 5576570 : Buffer buffer = scan->rs_cbuf;
571 5576570 : BlockNumber block = scan->rs_cblock;
572 : Snapshot snapshot;
573 : Page page;
574 : int lines;
575 : bool all_visible;
576 : bool check_serializable;
577 :
578 : Assert(BufferGetBlockNumber(buffer) == block);
579 :
580 : /* ensure we're not accidentally being used when not in pagemode */
581 : Assert(scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE);
582 5576570 : snapshot = scan->rs_base.rs_snapshot;
583 :
584 : /*
585 : * Prune and repair fragmentation for the whole page, if possible.
586 : */
587 5576570 : heap_page_prune_opt(scan->rs_base.rs_rd, buffer);
588 :
589 : /*
590 : * We must hold share lock on the buffer content while examining tuple
591 : * visibility. Afterwards, however, the tuples we have found to be
592 : * visible are guaranteed good as long as we hold the buffer pin.
593 : */
594 5576570 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
595 :
596 5576570 : page = BufferGetPage(buffer);
597 5576570 : lines = PageGetMaxOffsetNumber(page);
598 :
599 : /*
600 : * If the all-visible flag indicates that all tuples on the page are
601 : * visible to everyone, we can skip the per-tuple visibility tests.
602 : *
603 : * Note: In hot standby, a tuple that's already visible to all
604 : * transactions on the primary might still be invisible to a read-only
605 : * transaction in the standby. We partly handle this problem by tracking
606 : * the minimum xmin of visible tuples as the cut-off XID while marking a
607 : * page all-visible on the primary and WAL log that along with the
608 : * visibility map SET operation. In hot standby, we wait for (or abort)
609 : * all transactions that can potentially may not see one or more tuples on
610 : * the page. That's how index-only scans work fine in hot standby. A
611 : * crucial difference between index-only scans and heap scans is that the
612 : * index-only scan completely relies on the visibility map where as heap
613 : * scan looks at the page-level PD_ALL_VISIBLE flag. We are not sure if
614 : * the page-level flag can be trusted in the same way, because it might
615 : * get propagated somehow without being explicitly WAL-logged, e.g. via a
616 : * full page write. Until we can prove that beyond doubt, let's check each
617 : * tuple for visibility the hard way.
618 : */
619 5576570 : all_visible = PageIsAllVisible(page) && !snapshot->takenDuringRecovery;
620 : check_serializable =
621 5576570 : CheckForSerializableConflictOutNeeded(scan->rs_base.rs_rd, snapshot);
622 :
623 : /*
624 : * We call page_collect_tuples() with constant arguments, to get the
625 : * compiler to constant fold the constant arguments. Separate calls with
626 : * constant arguments, rather than variables, are needed on several
627 : * compilers to actually perform constant folding.
628 : */
629 5576570 : if (likely(all_visible))
630 : {
631 1986304 : if (likely(!check_serializable))
632 1986304 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
633 : block, lines, true, false);
634 : else
635 0 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
636 : block, lines, true, true);
637 : }
638 : else
639 : {
640 3590266 : if (likely(!check_serializable))
641 3589020 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
642 : block, lines, false, false);
643 : else
644 1246 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
645 : block, lines, false, true);
646 : }
647 :
648 5576554 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
649 5576554 : }
650 :
651 : /*
652 : * heap_fetch_next_buffer - read and pin the next block from MAIN_FORKNUM.
653 : *
654 : * Read the next block of the scan relation from the read stream and save it
655 : * in the scan descriptor. It is already pinned.
656 : */
657 : static inline void
658 7392580 : heap_fetch_next_buffer(HeapScanDesc scan, ScanDirection dir)
659 : {
660 : Assert(scan->rs_read_stream);
661 :
662 : /* release previous scan buffer, if any */
663 7392580 : if (BufferIsValid(scan->rs_cbuf))
664 : {
665 5430088 : ReleaseBuffer(scan->rs_cbuf);
666 5430088 : scan->rs_cbuf = InvalidBuffer;
667 : }
668 :
669 : /*
670 : * Be sure to check for interrupts at least once per page. Checks at
671 : * higher code levels won't be able to stop a seqscan that encounters many
672 : * pages' worth of consecutive dead tuples.
673 : */
674 7392580 : CHECK_FOR_INTERRUPTS();
675 :
676 : /*
677 : * If the scan direction is changing, reset the prefetch block to the
678 : * current block. Otherwise, we will incorrectly prefetch the blocks
679 : * between the prefetch block and the current block again before
680 : * prefetching blocks in the new, correct scan direction.
681 : */
682 7392580 : if (unlikely(scan->rs_dir != dir))
683 : {
684 154 : scan->rs_prefetch_block = scan->rs_cblock;
685 154 : read_stream_reset(scan->rs_read_stream);
686 : }
687 :
688 7392580 : scan->rs_dir = dir;
689 :
690 7392580 : scan->rs_cbuf = read_stream_next_buffer(scan->rs_read_stream, NULL);
691 7392530 : if (BufferIsValid(scan->rs_cbuf))
692 5758184 : scan->rs_cblock = BufferGetBlockNumber(scan->rs_cbuf);
693 7392530 : }
694 :
695 : /*
696 : * heapgettup_initial_block - return the first BlockNumber to scan
697 : *
698 : * Returns InvalidBlockNumber when there are no blocks to scan. This can
699 : * occur with empty tables and in parallel scans when parallel workers get all
700 : * of the pages before we can get a chance to get our first page.
701 : */
702 : static pg_noinline BlockNumber
703 1959338 : heapgettup_initial_block(HeapScanDesc scan, ScanDirection dir)
704 : {
705 : Assert(!scan->rs_inited);
706 : Assert(scan->rs_base.rs_parallel == NULL);
707 :
708 : /* When there are no pages to scan, return InvalidBlockNumber */
709 1959338 : if (scan->rs_nblocks == 0 || scan->rs_numblocks == 0)
710 993204 : return InvalidBlockNumber;
711 :
712 966134 : if (ScanDirectionIsForward(dir))
713 : {
714 966070 : return scan->rs_startblock;
715 : }
716 : else
717 : {
718 : /*
719 : * Disable reporting to syncscan logic in a backwards scan; it's not
720 : * very likely anyone else is doing the same thing at the same time,
721 : * and much more likely that we'll just bollix things for forward
722 : * scanners.
723 : */
724 64 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
725 :
726 : /*
727 : * Start from last page of the scan. Ensure we take into account
728 : * rs_numblocks if it's been adjusted by heap_setscanlimits().
729 : */
730 64 : if (scan->rs_numblocks != InvalidBlockNumber)
731 6 : return (scan->rs_startblock + scan->rs_numblocks - 1) % scan->rs_nblocks;
732 :
733 58 : if (scan->rs_startblock > 0)
734 0 : return scan->rs_startblock - 1;
735 :
736 58 : return scan->rs_nblocks - 1;
737 : }
738 : }
739 :
740 :
741 : /*
742 : * heapgettup_start_page - helper function for heapgettup()
743 : *
744 : * Return the next page to scan based on the scan->rs_cbuf and set *linesleft
745 : * to the number of tuples on this page. Also set *lineoff to the first
746 : * offset to scan with forward scans getting the first offset and backward
747 : * getting the final offset on the page.
748 : */
749 : static Page
750 190166 : heapgettup_start_page(HeapScanDesc scan, ScanDirection dir, int *linesleft,
751 : OffsetNumber *lineoff)
752 : {
753 : Page page;
754 :
755 : Assert(scan->rs_inited);
756 : Assert(BufferIsValid(scan->rs_cbuf));
757 :
758 : /* Caller is responsible for ensuring buffer is locked if needed */
759 190166 : page = BufferGetPage(scan->rs_cbuf);
760 :
761 190166 : *linesleft = PageGetMaxOffsetNumber(page) - FirstOffsetNumber + 1;
762 :
763 190166 : if (ScanDirectionIsForward(dir))
764 190166 : *lineoff = FirstOffsetNumber;
765 : else
766 0 : *lineoff = (OffsetNumber) (*linesleft);
767 :
768 : /* lineoff now references the physically previous or next tid */
769 190166 : return page;
770 : }
771 :
772 :
773 : /*
774 : * heapgettup_continue_page - helper function for heapgettup()
775 : *
776 : * Return the next page to scan based on the scan->rs_cbuf and set *linesleft
777 : * to the number of tuples left to scan on this page. Also set *lineoff to
778 : * the next offset to scan according to the ScanDirection in 'dir'.
779 : */
780 : static inline Page
781 15767264 : heapgettup_continue_page(HeapScanDesc scan, ScanDirection dir, int *linesleft,
782 : OffsetNumber *lineoff)
783 : {
784 : Page page;
785 :
786 : Assert(scan->rs_inited);
787 : Assert(BufferIsValid(scan->rs_cbuf));
788 :
789 : /* Caller is responsible for ensuring buffer is locked if needed */
790 15767264 : page = BufferGetPage(scan->rs_cbuf);
791 :
792 15767264 : if (ScanDirectionIsForward(dir))
793 : {
794 15767264 : *lineoff = OffsetNumberNext(scan->rs_coffset);
795 15767264 : *linesleft = PageGetMaxOffsetNumber(page) - (*lineoff) + 1;
796 : }
797 : else
798 : {
799 : /*
800 : * The previous returned tuple may have been vacuumed since the
801 : * previous scan when we use a non-MVCC snapshot, so we must
802 : * re-establish the lineoff <= PageGetMaxOffsetNumber(page) invariant
803 : */
804 0 : *lineoff = Min(PageGetMaxOffsetNumber(page), OffsetNumberPrev(scan->rs_coffset));
805 0 : *linesleft = *lineoff;
806 : }
807 :
808 : /* lineoff now references the physically previous or next tid */
809 15767264 : return page;
810 : }
811 :
812 : /*
813 : * heapgettup_advance_block - helper for heap_fetch_next_buffer()
814 : *
815 : * Given the current block number, the scan direction, and various information
816 : * contained in the scan descriptor, calculate the BlockNumber to scan next
817 : * and return it. If there are no further blocks to scan, return
818 : * InvalidBlockNumber to indicate this fact to the caller.
819 : *
820 : * This should not be called to determine the initial block number -- only for
821 : * subsequent blocks.
822 : *
823 : * This also adjusts rs_numblocks when a limit has been imposed by
824 : * heap_setscanlimits().
825 : */
826 : static inline BlockNumber
827 5602026 : heapgettup_advance_block(HeapScanDesc scan, BlockNumber block, ScanDirection dir)
828 : {
829 : Assert(scan->rs_base.rs_parallel == NULL);
830 :
831 5602026 : if (likely(ScanDirectionIsForward(dir)))
832 : {
833 5601908 : block++;
834 :
835 : /* wrap back to the start of the heap */
836 5601908 : if (block >= scan->rs_nblocks)
837 763234 : block = 0;
838 :
839 : /*
840 : * Report our new scan position for synchronization purposes. We don't
841 : * do that when moving backwards, however. That would just mess up any
842 : * other forward-moving scanners.
843 : *
844 : * Note: we do this before checking for end of scan so that the final
845 : * state of the position hint is back at the start of the rel. That's
846 : * not strictly necessary, but otherwise when you run the same query
847 : * multiple times the starting position would shift a little bit
848 : * backwards on every invocation, which is confusing. We don't
849 : * guarantee any specific ordering in general, though.
850 : */
851 5601908 : if (scan->rs_base.rs_flags & SO_ALLOW_SYNC)
852 22530 : ss_report_location(scan->rs_base.rs_rd, block);
853 :
854 : /* we're done if we're back at where we started */
855 5601908 : if (block == scan->rs_startblock)
856 763152 : return InvalidBlockNumber;
857 :
858 : /* check if the limit imposed by heap_setscanlimits() is met */
859 4838756 : if (scan->rs_numblocks != InvalidBlockNumber)
860 : {
861 4968 : if (--scan->rs_numblocks == 0)
862 3092 : return InvalidBlockNumber;
863 : }
864 :
865 4835664 : return block;
866 : }
867 : else
868 : {
869 : /* we're done if the last block is the start position */
870 118 : if (block == scan->rs_startblock)
871 118 : return InvalidBlockNumber;
872 :
873 : /* check if the limit imposed by heap_setscanlimits() is met */
874 0 : if (scan->rs_numblocks != InvalidBlockNumber)
875 : {
876 0 : if (--scan->rs_numblocks == 0)
877 0 : return InvalidBlockNumber;
878 : }
879 :
880 : /* wrap to the end of the heap when the last page was page 0 */
881 0 : if (block == 0)
882 0 : block = scan->rs_nblocks;
883 :
884 0 : block--;
885 :
886 0 : return block;
887 : }
888 : }
889 :
890 : /* ----------------
891 : * heapgettup - fetch next heap tuple
892 : *
893 : * Initialize the scan if not already done; then advance to the next
894 : * tuple as indicated by "dir"; return the next tuple in scan->rs_ctup,
895 : * or set scan->rs_ctup.t_data = NULL if no more tuples.
896 : *
897 : * Note: the reason nkeys/key are passed separately, even though they are
898 : * kept in the scan descriptor, is that the caller may not want us to check
899 : * the scankeys.
900 : *
901 : * Note: when we fall off the end of the scan in either direction, we
902 : * reset rs_inited. This means that a further request with the same
903 : * scan direction will restart the scan, which is a bit odd, but a
904 : * request with the opposite scan direction will start a fresh scan
905 : * in the proper direction. The latter is required behavior for cursors,
906 : * while the former case is generally undefined behavior in Postgres
907 : * so we don't care too much.
908 : * ----------------
909 : */
910 : static void
911 15808304 : heapgettup(HeapScanDesc scan,
912 : ScanDirection dir,
913 : int nkeys,
914 : ScanKey key)
915 : {
916 15808304 : HeapTuple tuple = &(scan->rs_ctup);
917 : Page page;
918 : OffsetNumber lineoff;
919 : int linesleft;
920 :
921 15808304 : if (likely(scan->rs_inited))
922 : {
923 : /* continue from previously returned page/tuple */
924 15767264 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
925 15767264 : page = heapgettup_continue_page(scan, dir, &linesleft, &lineoff);
926 15767264 : goto continue_page;
927 : }
928 :
929 : /*
930 : * advance the scan until we find a qualifying tuple or run out of stuff
931 : * to scan
932 : */
933 : while (true)
934 : {
935 230906 : heap_fetch_next_buffer(scan, dir);
936 :
937 : /* did we run out of blocks to scan? */
938 230906 : if (!BufferIsValid(scan->rs_cbuf))
939 40740 : break;
940 :
941 : Assert(BufferGetBlockNumber(scan->rs_cbuf) == scan->rs_cblock);
942 :
943 190166 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
944 190166 : page = heapgettup_start_page(scan, dir, &linesleft, &lineoff);
945 15957430 : continue_page:
946 :
947 : /*
948 : * Only continue scanning the page while we have lines left.
949 : *
950 : * Note that this protects us from accessing line pointers past
951 : * PageGetMaxOffsetNumber(); both for forward scans when we resume the
952 : * table scan, and for when we start scanning a new page.
953 : */
954 16041156 : for (; linesleft > 0; linesleft--, lineoff += dir)
955 : {
956 : bool visible;
957 15851290 : ItemId lpp = PageGetItemId(page, lineoff);
958 :
959 15851290 : if (!ItemIdIsNormal(lpp))
960 73264 : continue;
961 :
962 15778026 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
963 15778026 : tuple->t_len = ItemIdGetLength(lpp);
964 15778026 : ItemPointerSet(&(tuple->t_self), scan->rs_cblock, lineoff);
965 :
966 15778026 : visible = HeapTupleSatisfiesVisibility(tuple,
967 : scan->rs_base.rs_snapshot,
968 : scan->rs_cbuf);
969 :
970 15778026 : HeapCheckForSerializableConflictOut(visible, scan->rs_base.rs_rd,
971 : tuple, scan->rs_cbuf,
972 : scan->rs_base.rs_snapshot);
973 :
974 : /* skip tuples not visible to this snapshot */
975 15778026 : if (!visible)
976 10462 : continue;
977 :
978 : /* skip any tuples that don't match the scan key */
979 15767564 : if (key != NULL &&
980 0 : !HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
981 : nkeys, key))
982 0 : continue;
983 :
984 15767564 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
985 15767564 : scan->rs_coffset = lineoff;
986 15767564 : return;
987 : }
988 :
989 : /*
990 : * if we get here, it means we've exhausted the items on this page and
991 : * it's time to move to the next.
992 : */
993 189866 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
994 : }
995 :
996 : /* end of scan */
997 40740 : if (BufferIsValid(scan->rs_cbuf))
998 0 : ReleaseBuffer(scan->rs_cbuf);
999 :
1000 40740 : scan->rs_cbuf = InvalidBuffer;
1001 40740 : scan->rs_cblock = InvalidBlockNumber;
1002 40740 : scan->rs_prefetch_block = InvalidBlockNumber;
1003 40740 : tuple->t_data = NULL;
1004 40740 : scan->rs_inited = false;
1005 : }
1006 :
1007 : /* ----------------
1008 : * heapgettup_pagemode - fetch next heap tuple in page-at-a-time mode
1009 : *
1010 : * Same API as heapgettup, but used in page-at-a-time mode
1011 : *
1012 : * The internal logic is much the same as heapgettup's too, but there are some
1013 : * differences: we do not take the buffer content lock (that only needs to
1014 : * happen inside heap_prepare_pagescan), and we iterate through just the
1015 : * tuples listed in rs_vistuples[] rather than all tuples on the page. Notice
1016 : * that lineindex is 0-based, where the corresponding loop variable lineoff in
1017 : * heapgettup is 1-based.
1018 : * ----------------
1019 : */
1020 : static void
1021 99417478 : heapgettup_pagemode(HeapScanDesc scan,
1022 : ScanDirection dir,
1023 : int nkeys,
1024 : ScanKey key)
1025 : {
1026 99417478 : HeapTuple tuple = &(scan->rs_ctup);
1027 : Page page;
1028 : uint32 lineindex;
1029 : uint32 linesleft;
1030 :
1031 99417478 : if (likely(scan->rs_inited))
1032 : {
1033 : /* continue from previously returned page/tuple */
1034 97496026 : page = BufferGetPage(scan->rs_cbuf);
1035 :
1036 97496026 : lineindex = scan->rs_cindex + dir;
1037 97496026 : if (ScanDirectionIsForward(dir))
1038 97495368 : linesleft = scan->rs_ntuples - lineindex;
1039 : else
1040 658 : linesleft = scan->rs_cindex;
1041 : /* lineindex now references the next or previous visible tid */
1042 :
1043 97496026 : goto continue_page;
1044 : }
1045 :
1046 : /*
1047 : * advance the scan until we find a qualifying tuple or run out of stuff
1048 : * to scan
1049 : */
1050 : while (true)
1051 : {
1052 7161674 : heap_fetch_next_buffer(scan, dir);
1053 :
1054 : /* did we run out of blocks to scan? */
1055 7161624 : if (!BufferIsValid(scan->rs_cbuf))
1056 1593606 : break;
1057 :
1058 : Assert(BufferGetBlockNumber(scan->rs_cbuf) == scan->rs_cblock);
1059 :
1060 : /* prune the page and determine visible tuple offsets */
1061 5568018 : heap_prepare_pagescan((TableScanDesc) scan);
1062 5568002 : page = BufferGetPage(scan->rs_cbuf);
1063 5568002 : linesleft = scan->rs_ntuples;
1064 5568002 : lineindex = ScanDirectionIsForward(dir) ? 0 : linesleft - 1;
1065 :
1066 : /* block is the same for all tuples, set it once outside the loop */
1067 5568002 : ItemPointerSetBlockNumber(&tuple->t_self, scan->rs_cblock);
1068 :
1069 : /* lineindex now references the next or previous visible tid */
1070 103064028 : continue_page:
1071 :
1072 199603018 : for (; linesleft > 0; linesleft--, lineindex += dir)
1073 : {
1074 : ItemId lpp;
1075 : OffsetNumber lineoff;
1076 :
1077 : Assert(lineindex <= scan->rs_ntuples);
1078 194362796 : lineoff = scan->rs_vistuples[lineindex];
1079 194362796 : lpp = PageGetItemId(page, lineoff);
1080 : Assert(ItemIdIsNormal(lpp));
1081 :
1082 194362796 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
1083 194362796 : tuple->t_len = ItemIdGetLength(lpp);
1084 194362796 : ItemPointerSetOffsetNumber(&tuple->t_self, lineoff);
1085 :
1086 : /* skip any tuples that don't match the scan key */
1087 194362796 : if (key != NULL &&
1088 97343896 : !HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
1089 : nkeys, key))
1090 96538990 : continue;
1091 :
1092 97823806 : scan->rs_cindex = lineindex;
1093 97823806 : return;
1094 : }
1095 : }
1096 :
1097 : /* end of scan */
1098 1593606 : if (BufferIsValid(scan->rs_cbuf))
1099 0 : ReleaseBuffer(scan->rs_cbuf);
1100 1593606 : scan->rs_cbuf = InvalidBuffer;
1101 1593606 : scan->rs_cblock = InvalidBlockNumber;
1102 1593606 : scan->rs_prefetch_block = InvalidBlockNumber;
1103 1593606 : tuple->t_data = NULL;
1104 1593606 : scan->rs_inited = false;
1105 : }
1106 :
1107 :
1108 : /* ----------------------------------------------------------------
1109 : * heap access method interface
1110 : * ----------------------------------------------------------------
1111 : */
1112 :
1113 :
1114 : TableScanDesc
1115 770478 : heap_beginscan(Relation relation, Snapshot snapshot,
1116 : int nkeys, ScanKey key,
1117 : ParallelTableScanDesc parallel_scan,
1118 : uint32 flags)
1119 : {
1120 : HeapScanDesc scan;
1121 :
1122 : /*
1123 : * increment relation ref count while scanning relation
1124 : *
1125 : * This is just to make really sure the relcache entry won't go away while
1126 : * the scan has a pointer to it. Caller should be holding the rel open
1127 : * anyway, so this is redundant in all normal scenarios...
1128 : */
1129 770478 : RelationIncrementReferenceCount(relation);
1130 :
1131 : /*
1132 : * allocate and initialize scan descriptor
1133 : */
1134 770478 : if (flags & SO_TYPE_BITMAPSCAN)
1135 : {
1136 16376 : BitmapHeapScanDesc bscan = palloc(sizeof(BitmapHeapScanDescData));
1137 :
1138 : /*
1139 : * Bitmap Heap scans do not have any fields that a normal Heap Scan
1140 : * does not have, so no special initializations required here.
1141 : */
1142 16376 : scan = (HeapScanDesc) bscan;
1143 : }
1144 : else
1145 754102 : scan = (HeapScanDesc) palloc(sizeof(HeapScanDescData));
1146 :
1147 770478 : scan->rs_base.rs_rd = relation;
1148 770478 : scan->rs_base.rs_snapshot = snapshot;
1149 770478 : scan->rs_base.rs_nkeys = nkeys;
1150 770478 : scan->rs_base.rs_flags = flags;
1151 770478 : scan->rs_base.rs_parallel = parallel_scan;
1152 770478 : scan->rs_strategy = NULL; /* set in initscan */
1153 770478 : scan->rs_cbuf = InvalidBuffer;
1154 :
1155 : /*
1156 : * Disable page-at-a-time mode if it's not a MVCC-safe snapshot.
1157 : */
1158 770478 : if (!(snapshot && IsMVCCSnapshot(snapshot)))
1159 58368 : scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
1160 :
1161 : /* Check that a historic snapshot is not used for non-catalog tables */
1162 770478 : if (snapshot &&
1163 753154 : IsHistoricMVCCSnapshot(snapshot) &&
1164 1338 : !RelationIsAccessibleInLogicalDecoding(relation))
1165 : {
1166 0 : ereport(ERROR,
1167 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
1168 : errmsg("cannot query non-catalog table \"%s\" during logical decoding",
1169 : RelationGetRelationName(relation))));
1170 : }
1171 :
1172 : /*
1173 : * For seqscan and sample scans in a serializable transaction, acquire a
1174 : * predicate lock on the entire relation. This is required not only to
1175 : * lock all the matching tuples, but also to conflict with new insertions
1176 : * into the table. In an indexscan, we take page locks on the index pages
1177 : * covering the range specified in the scan qual, but in a heap scan there
1178 : * is nothing more fine-grained to lock. A bitmap scan is a different
1179 : * story, there we have already scanned the index and locked the index
1180 : * pages covering the predicate. But in that case we still have to lock
1181 : * any matching heap tuples. For sample scan we could optimize the locking
1182 : * to be at least page-level granularity, but we'd need to add per-tuple
1183 : * locking for that.
1184 : */
1185 770478 : if (scan->rs_base.rs_flags & (SO_TYPE_SEQSCAN | SO_TYPE_SAMPLESCAN))
1186 : {
1187 : /*
1188 : * Ensure a missing snapshot is noticed reliably, even if the
1189 : * isolation mode means predicate locking isn't performed (and
1190 : * therefore the snapshot isn't used here).
1191 : */
1192 : Assert(snapshot);
1193 734012 : PredicateLockRelation(relation, snapshot);
1194 : }
1195 :
1196 : /* we only need to set this up once */
1197 770478 : scan->rs_ctup.t_tableOid = RelationGetRelid(relation);
1198 :
1199 : /*
1200 : * Allocate memory to keep track of page allocation for parallel workers
1201 : * when doing a parallel scan.
1202 : */
1203 770478 : if (parallel_scan != NULL)
1204 4110 : scan->rs_parallelworkerdata = palloc(sizeof(ParallelBlockTableScanWorkerData));
1205 : else
1206 766368 : scan->rs_parallelworkerdata = NULL;
1207 :
1208 : /*
1209 : * we do this here instead of in initscan() because heap_rescan also calls
1210 : * initscan() and we don't want to allocate memory again
1211 : */
1212 770478 : if (nkeys > 0)
1213 442620 : scan->rs_base.rs_key = (ScanKey) palloc(sizeof(ScanKeyData) * nkeys);
1214 : else
1215 327858 : scan->rs_base.rs_key = NULL;
1216 :
1217 770478 : initscan(scan, key, false);
1218 :
1219 770474 : scan->rs_read_stream = NULL;
1220 :
1221 : /*
1222 : * Set up a read stream for sequential scans and TID range scans. This
1223 : * should be done after initscan() because initscan() allocates the
1224 : * BufferAccessStrategy object passed to the read stream API.
1225 : */
1226 770474 : if (scan->rs_base.rs_flags & SO_TYPE_SEQSCAN ||
1227 36612 : scan->rs_base.rs_flags & SO_TYPE_TIDRANGESCAN)
1228 735842 : {
1229 : ReadStreamBlockNumberCB cb;
1230 :
1231 735842 : if (scan->rs_base.rs_parallel)
1232 4110 : cb = heap_scan_stream_read_next_parallel;
1233 : else
1234 731732 : cb = heap_scan_stream_read_next_serial;
1235 :
1236 : /* ---
1237 : * It is safe to use batchmode as the only locks taken by `cb`
1238 : * are never taken while waiting for IO:
1239 : * - SyncScanLock is used in the non-parallel case
1240 : * - in the parallel case, only spinlocks and atomics are used
1241 : * ---
1242 : */
1243 735842 : scan->rs_read_stream = read_stream_begin_relation(READ_STREAM_SEQUENTIAL |
1244 : READ_STREAM_USE_BATCHING,
1245 : scan->rs_strategy,
1246 : scan->rs_base.rs_rd,
1247 : MAIN_FORKNUM,
1248 : cb,
1249 : scan,
1250 : 0);
1251 : }
1252 34632 : else if (scan->rs_base.rs_flags & SO_TYPE_BITMAPSCAN)
1253 : {
1254 16376 : scan->rs_read_stream = read_stream_begin_relation(READ_STREAM_DEFAULT |
1255 : READ_STREAM_USE_BATCHING,
1256 : scan->rs_strategy,
1257 : scan->rs_base.rs_rd,
1258 : MAIN_FORKNUM,
1259 : bitmapheap_stream_read_next,
1260 : scan,
1261 : sizeof(TBMIterateResult));
1262 : }
1263 :
1264 :
1265 770474 : return (TableScanDesc) scan;
1266 : }
1267 :
1268 : void
1269 1232328 : heap_rescan(TableScanDesc sscan, ScanKey key, bool set_params,
1270 : bool allow_strat, bool allow_sync, bool allow_pagemode)
1271 : {
1272 1232328 : HeapScanDesc scan = (HeapScanDesc) sscan;
1273 :
1274 1232328 : if (set_params)
1275 : {
1276 30 : if (allow_strat)
1277 30 : scan->rs_base.rs_flags |= SO_ALLOW_STRAT;
1278 : else
1279 0 : scan->rs_base.rs_flags &= ~SO_ALLOW_STRAT;
1280 :
1281 30 : if (allow_sync)
1282 12 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
1283 : else
1284 18 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
1285 :
1286 30 : if (allow_pagemode && scan->rs_base.rs_snapshot &&
1287 30 : IsMVCCSnapshot(scan->rs_base.rs_snapshot))
1288 30 : scan->rs_base.rs_flags |= SO_ALLOW_PAGEMODE;
1289 : else
1290 0 : scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
1291 : }
1292 :
1293 : /*
1294 : * unpin scan buffers
1295 : */
1296 1232328 : if (BufferIsValid(scan->rs_cbuf))
1297 : {
1298 3264 : ReleaseBuffer(scan->rs_cbuf);
1299 3264 : scan->rs_cbuf = InvalidBuffer;
1300 : }
1301 :
1302 : /*
1303 : * SO_TYPE_BITMAPSCAN would be cleaned up here, but it does not hold any
1304 : * additional data vs a normal HeapScan
1305 : */
1306 :
1307 : /*
1308 : * The read stream is reset on rescan. This must be done before
1309 : * initscan(), as some state referred to by read_stream_reset() is reset
1310 : * in initscan().
1311 : */
1312 1232328 : if (scan->rs_read_stream)
1313 1232292 : read_stream_reset(scan->rs_read_stream);
1314 :
1315 : /*
1316 : * reinitialize scan descriptor
1317 : */
1318 1232328 : initscan(scan, key, true);
1319 1232328 : }
1320 :
1321 : void
1322 765742 : heap_endscan(TableScanDesc sscan)
1323 : {
1324 765742 : HeapScanDesc scan = (HeapScanDesc) sscan;
1325 :
1326 : /* Note: no locking manipulations needed */
1327 :
1328 : /*
1329 : * unpin scan buffers
1330 : */
1331 765742 : if (BufferIsValid(scan->rs_cbuf))
1332 321572 : ReleaseBuffer(scan->rs_cbuf);
1333 :
1334 : /*
1335 : * Must free the read stream before freeing the BufferAccessStrategy.
1336 : */
1337 765742 : if (scan->rs_read_stream)
1338 747592 : read_stream_end(scan->rs_read_stream);
1339 :
1340 : /*
1341 : * decrement relation reference count and free scan descriptor storage
1342 : */
1343 765742 : RelationDecrementReferenceCount(scan->rs_base.rs_rd);
1344 :
1345 765742 : if (scan->rs_base.rs_key)
1346 442566 : pfree(scan->rs_base.rs_key);
1347 :
1348 765742 : if (scan->rs_strategy != NULL)
1349 25412 : FreeAccessStrategy(scan->rs_strategy);
1350 :
1351 765742 : if (scan->rs_parallelworkerdata != NULL)
1352 4110 : pfree(scan->rs_parallelworkerdata);
1353 :
1354 765742 : if (scan->rs_base.rs_flags & SO_TEMP_SNAPSHOT)
1355 78824 : UnregisterSnapshot(scan->rs_base.rs_snapshot);
1356 :
1357 765742 : pfree(scan);
1358 765742 : }
1359 :
1360 : HeapTuple
1361 19938624 : heap_getnext(TableScanDesc sscan, ScanDirection direction)
1362 : {
1363 19938624 : HeapScanDesc scan = (HeapScanDesc) sscan;
1364 :
1365 : /*
1366 : * This is still widely used directly, without going through table AM, so
1367 : * add a safety check. It's possible we should, at a later point,
1368 : * downgrade this to an assert. The reason for checking the AM routine,
1369 : * rather than the AM oid, is that this allows to write regression tests
1370 : * that create another AM reusing the heap handler.
1371 : */
1372 19938624 : if (unlikely(sscan->rs_rd->rd_tableam != GetHeapamTableAmRoutine()))
1373 0 : ereport(ERROR,
1374 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1375 : errmsg_internal("only heap AM is supported")));
1376 :
1377 : /*
1378 : * We don't expect direct calls to heap_getnext with valid CheckXidAlive
1379 : * for catalog or regular tables. See detailed comments in xact.c where
1380 : * these variables are declared. Normally we have such a check at tableam
1381 : * level API but this is called from many places so we need to ensure it
1382 : * here.
1383 : */
1384 19938624 : if (unlikely(TransactionIdIsValid(CheckXidAlive) && !bsysscan))
1385 0 : elog(ERROR, "unexpected heap_getnext call during logical decoding");
1386 :
1387 : /* Note: no locking manipulations needed */
1388 :
1389 19938624 : if (scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE)
1390 5106388 : heapgettup_pagemode(scan, direction,
1391 5106388 : scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1392 : else
1393 14832236 : heapgettup(scan, direction,
1394 14832236 : scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1395 :
1396 19938624 : if (scan->rs_ctup.t_data == NULL)
1397 131546 : return NULL;
1398 :
1399 : /*
1400 : * if we get here it means we have a new current scan tuple, so point to
1401 : * the proper return buffer and return the tuple.
1402 : */
1403 :
1404 19807078 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1405 :
1406 19807078 : return &scan->rs_ctup;
1407 : }
1408 :
1409 : bool
1410 95275700 : heap_getnextslot(TableScanDesc sscan, ScanDirection direction, TupleTableSlot *slot)
1411 : {
1412 95275700 : HeapScanDesc scan = (HeapScanDesc) sscan;
1413 :
1414 : /* Note: no locking manipulations needed */
1415 :
1416 95275700 : if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1417 94299632 : heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1418 : else
1419 976068 : heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1420 :
1421 95275650 : if (scan->rs_ctup.t_data == NULL)
1422 : {
1423 1502592 : ExecClearTuple(slot);
1424 1502592 : return false;
1425 : }
1426 :
1427 : /*
1428 : * if we get here it means we have a new current scan tuple, so point to
1429 : * the proper return buffer and return the tuple.
1430 : */
1431 :
1432 93773058 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1433 :
1434 93773058 : ExecStoreBufferHeapTuple(&scan->rs_ctup, slot,
1435 : scan->rs_cbuf);
1436 93773058 : return true;
1437 : }
1438 :
1439 : void
1440 2070 : heap_set_tidrange(TableScanDesc sscan, ItemPointer mintid,
1441 : ItemPointer maxtid)
1442 : {
1443 2070 : HeapScanDesc scan = (HeapScanDesc) sscan;
1444 : BlockNumber startBlk;
1445 : BlockNumber numBlks;
1446 : ItemPointerData highestItem;
1447 : ItemPointerData lowestItem;
1448 :
1449 : /*
1450 : * For relations without any pages, we can simply leave the TID range
1451 : * unset. There will be no tuples to scan, therefore no tuples outside
1452 : * the given TID range.
1453 : */
1454 2070 : if (scan->rs_nblocks == 0)
1455 48 : return;
1456 :
1457 : /*
1458 : * Set up some ItemPointers which point to the first and last possible
1459 : * tuples in the heap.
1460 : */
1461 2058 : ItemPointerSet(&highestItem, scan->rs_nblocks - 1, MaxOffsetNumber);
1462 2058 : ItemPointerSet(&lowestItem, 0, FirstOffsetNumber);
1463 :
1464 : /*
1465 : * If the given maximum TID is below the highest possible TID in the
1466 : * relation, then restrict the range to that, otherwise we scan to the end
1467 : * of the relation.
1468 : */
1469 2058 : if (ItemPointerCompare(maxtid, &highestItem) < 0)
1470 260 : ItemPointerCopy(maxtid, &highestItem);
1471 :
1472 : /*
1473 : * If the given minimum TID is above the lowest possible TID in the
1474 : * relation, then restrict the range to only scan for TIDs above that.
1475 : */
1476 2058 : if (ItemPointerCompare(mintid, &lowestItem) > 0)
1477 1822 : ItemPointerCopy(mintid, &lowestItem);
1478 :
1479 : /*
1480 : * Check for an empty range and protect from would be negative results
1481 : * from the numBlks calculation below.
1482 : */
1483 2058 : if (ItemPointerCompare(&highestItem, &lowestItem) < 0)
1484 : {
1485 : /* Set an empty range of blocks to scan */
1486 36 : heap_setscanlimits(sscan, 0, 0);
1487 36 : return;
1488 : }
1489 :
1490 : /*
1491 : * Calculate the first block and the number of blocks we must scan. We
1492 : * could be more aggressive here and perform some more validation to try
1493 : * and further narrow the scope of blocks to scan by checking if the
1494 : * lowestItem has an offset above MaxOffsetNumber. In this case, we could
1495 : * advance startBlk by one. Likewise, if highestItem has an offset of 0
1496 : * we could scan one fewer blocks. However, such an optimization does not
1497 : * seem worth troubling over, currently.
1498 : */
1499 2022 : startBlk = ItemPointerGetBlockNumberNoCheck(&lowestItem);
1500 :
1501 2022 : numBlks = ItemPointerGetBlockNumberNoCheck(&highestItem) -
1502 2022 : ItemPointerGetBlockNumberNoCheck(&lowestItem) + 1;
1503 :
1504 : /* Set the start block and number of blocks to scan */
1505 2022 : heap_setscanlimits(sscan, startBlk, numBlks);
1506 :
1507 : /* Finally, set the TID range in sscan */
1508 2022 : ItemPointerCopy(&lowestItem, &sscan->st.tidrange.rs_mintid);
1509 2022 : ItemPointerCopy(&highestItem, &sscan->st.tidrange.rs_maxtid);
1510 : }
1511 :
1512 : bool
1513 11272 : heap_getnextslot_tidrange(TableScanDesc sscan, ScanDirection direction,
1514 : TupleTableSlot *slot)
1515 : {
1516 11272 : HeapScanDesc scan = (HeapScanDesc) sscan;
1517 11272 : ItemPointer mintid = &sscan->st.tidrange.rs_mintid;
1518 11272 : ItemPointer maxtid = &sscan->st.tidrange.rs_maxtid;
1519 :
1520 : /* Note: no locking manipulations needed */
1521 : for (;;)
1522 : {
1523 11458 : if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1524 11458 : heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1525 : else
1526 0 : heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1527 :
1528 11442 : if (scan->rs_ctup.t_data == NULL)
1529 : {
1530 208 : ExecClearTuple(slot);
1531 208 : return false;
1532 : }
1533 :
1534 : /*
1535 : * heap_set_tidrange will have used heap_setscanlimits to limit the
1536 : * range of pages we scan to only ones that can contain the TID range
1537 : * we're scanning for. Here we must filter out any tuples from these
1538 : * pages that are outside of that range.
1539 : */
1540 11234 : if (ItemPointerCompare(&scan->rs_ctup.t_self, mintid) < 0)
1541 : {
1542 186 : ExecClearTuple(slot);
1543 :
1544 : /*
1545 : * When scanning backwards, the TIDs will be in descending order.
1546 : * Future tuples in this direction will be lower still, so we can
1547 : * just return false to indicate there will be no more tuples.
1548 : */
1549 186 : if (ScanDirectionIsBackward(direction))
1550 0 : return false;
1551 :
1552 186 : continue;
1553 : }
1554 :
1555 : /*
1556 : * Likewise for the final page, we must filter out TIDs greater than
1557 : * maxtid.
1558 : */
1559 11048 : if (ItemPointerCompare(&scan->rs_ctup.t_self, maxtid) > 0)
1560 : {
1561 112 : ExecClearTuple(slot);
1562 :
1563 : /*
1564 : * When scanning forward, the TIDs will be in ascending order.
1565 : * Future tuples in this direction will be higher still, so we can
1566 : * just return false to indicate there will be no more tuples.
1567 : */
1568 112 : if (ScanDirectionIsForward(direction))
1569 112 : return false;
1570 0 : continue;
1571 : }
1572 :
1573 10936 : break;
1574 : }
1575 :
1576 : /*
1577 : * if we get here it means we have a new current scan tuple, so point to
1578 : * the proper return buffer and return the tuple.
1579 : */
1580 10936 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1581 :
1582 10936 : ExecStoreBufferHeapTuple(&scan->rs_ctup, slot, scan->rs_cbuf);
1583 10936 : return true;
1584 : }
1585 :
1586 : /*
1587 : * heap_fetch - retrieve tuple with given tid
1588 : *
1589 : * On entry, tuple->t_self is the TID to fetch. We pin the buffer holding
1590 : * the tuple, fill in the remaining fields of *tuple, and check the tuple
1591 : * against the specified snapshot.
1592 : *
1593 : * If successful (tuple found and passes snapshot time qual), then *userbuf
1594 : * is set to the buffer holding the tuple and true is returned. The caller
1595 : * must unpin the buffer when done with the tuple.
1596 : *
1597 : * If the tuple is not found (ie, item number references a deleted slot),
1598 : * then tuple->t_data is set to NULL, *userbuf is set to InvalidBuffer,
1599 : * and false is returned.
1600 : *
1601 : * If the tuple is found but fails the time qual check, then the behavior
1602 : * depends on the keep_buf parameter. If keep_buf is false, the results
1603 : * are the same as for the tuple-not-found case. If keep_buf is true,
1604 : * then tuple->t_data and *userbuf are returned as for the success case,
1605 : * and again the caller must unpin the buffer; but false is returned.
1606 : *
1607 : * heap_fetch does not follow HOT chains: only the exact TID requested will
1608 : * be fetched.
1609 : *
1610 : * It is somewhat inconsistent that we ereport() on invalid block number but
1611 : * return false on invalid item number. There are a couple of reasons though.
1612 : * One is that the caller can relatively easily check the block number for
1613 : * validity, but cannot check the item number without reading the page
1614 : * himself. Another is that when we are following a t_ctid link, we can be
1615 : * reasonably confident that the page number is valid (since VACUUM shouldn't
1616 : * truncate off the destination page without having killed the referencing
1617 : * tuple first), but the item number might well not be good.
1618 : */
1619 : bool
1620 356134 : heap_fetch(Relation relation,
1621 : Snapshot snapshot,
1622 : HeapTuple tuple,
1623 : Buffer *userbuf,
1624 : bool keep_buf)
1625 : {
1626 356134 : ItemPointer tid = &(tuple->t_self);
1627 : ItemId lp;
1628 : Buffer buffer;
1629 : Page page;
1630 : OffsetNumber offnum;
1631 : bool valid;
1632 :
1633 : /*
1634 : * Fetch and pin the appropriate page of the relation.
1635 : */
1636 356134 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
1637 :
1638 : /*
1639 : * Need share lock on buffer to examine tuple commit status.
1640 : */
1641 356118 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
1642 356118 : page = BufferGetPage(buffer);
1643 :
1644 : /*
1645 : * We'd better check for out-of-range offnum in case of VACUUM since the
1646 : * TID was obtained.
1647 : */
1648 356118 : offnum = ItemPointerGetOffsetNumber(tid);
1649 356118 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1650 : {
1651 6 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1652 6 : ReleaseBuffer(buffer);
1653 6 : *userbuf = InvalidBuffer;
1654 6 : tuple->t_data = NULL;
1655 6 : return false;
1656 : }
1657 :
1658 : /*
1659 : * get the item line pointer corresponding to the requested tid
1660 : */
1661 356112 : lp = PageGetItemId(page, offnum);
1662 :
1663 : /*
1664 : * Must check for deleted tuple.
1665 : */
1666 356112 : if (!ItemIdIsNormal(lp))
1667 : {
1668 708 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1669 708 : ReleaseBuffer(buffer);
1670 708 : *userbuf = InvalidBuffer;
1671 708 : tuple->t_data = NULL;
1672 708 : return false;
1673 : }
1674 :
1675 : /*
1676 : * fill in *tuple fields
1677 : */
1678 355404 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
1679 355404 : tuple->t_len = ItemIdGetLength(lp);
1680 355404 : tuple->t_tableOid = RelationGetRelid(relation);
1681 :
1682 : /*
1683 : * check tuple visibility, then release lock
1684 : */
1685 355404 : valid = HeapTupleSatisfiesVisibility(tuple, snapshot, buffer);
1686 :
1687 355404 : if (valid)
1688 355294 : PredicateLockTID(relation, &(tuple->t_self), snapshot,
1689 355294 : HeapTupleHeaderGetXmin(tuple->t_data));
1690 :
1691 355404 : HeapCheckForSerializableConflictOut(valid, relation, tuple, buffer, snapshot);
1692 :
1693 355404 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1694 :
1695 355404 : if (valid)
1696 : {
1697 : /*
1698 : * All checks passed, so return the tuple as valid. Caller is now
1699 : * responsible for releasing the buffer.
1700 : */
1701 355294 : *userbuf = buffer;
1702 :
1703 355294 : return true;
1704 : }
1705 :
1706 : /* Tuple failed time qual, but maybe caller wants to see it anyway. */
1707 110 : if (keep_buf)
1708 68 : *userbuf = buffer;
1709 : else
1710 : {
1711 42 : ReleaseBuffer(buffer);
1712 42 : *userbuf = InvalidBuffer;
1713 42 : tuple->t_data = NULL;
1714 : }
1715 :
1716 110 : return false;
1717 : }
1718 :
1719 : /*
1720 : * heap_hot_search_buffer - search HOT chain for tuple satisfying snapshot
1721 : *
1722 : * On entry, *tid is the TID of a tuple (either a simple tuple, or the root
1723 : * of a HOT chain), and buffer is the buffer holding this tuple. We search
1724 : * for the first chain member satisfying the given snapshot. If one is
1725 : * found, we update *tid to reference that tuple's offset number, and
1726 : * return true. If no match, return false without modifying *tid.
1727 : *
1728 : * heapTuple is a caller-supplied buffer. When a match is found, we return
1729 : * the tuple here, in addition to updating *tid. If no match is found, the
1730 : * contents of this buffer on return are undefined.
1731 : *
1732 : * If all_dead is not NULL, we check non-visible tuples to see if they are
1733 : * globally dead; *all_dead is set true if all members of the HOT chain
1734 : * are vacuumable, false if not.
1735 : *
1736 : * Unlike heap_fetch, the caller must already have pin and (at least) share
1737 : * lock on the buffer; it is still pinned/locked at exit.
1738 : */
1739 : bool
1740 44738660 : heap_hot_search_buffer(ItemPointer tid, Relation relation, Buffer buffer,
1741 : Snapshot snapshot, HeapTuple heapTuple,
1742 : bool *all_dead, bool first_call)
1743 : {
1744 44738660 : Page page = BufferGetPage(buffer);
1745 44738660 : TransactionId prev_xmax = InvalidTransactionId;
1746 : BlockNumber blkno;
1747 : OffsetNumber offnum;
1748 : bool at_chain_start;
1749 : bool valid;
1750 : bool skip;
1751 44738660 : GlobalVisState *vistest = NULL;
1752 :
1753 : /* If this is not the first call, previous call returned a (live!) tuple */
1754 44738660 : if (all_dead)
1755 38231144 : *all_dead = first_call;
1756 :
1757 44738660 : blkno = ItemPointerGetBlockNumber(tid);
1758 44738660 : offnum = ItemPointerGetOffsetNumber(tid);
1759 44738660 : at_chain_start = first_call;
1760 44738660 : skip = !first_call;
1761 :
1762 : /* XXX: we should assert that a snapshot is pushed or registered */
1763 : Assert(TransactionIdIsValid(RecentXmin));
1764 : Assert(BufferGetBlockNumber(buffer) == blkno);
1765 :
1766 : /* Scan through possible multiple members of HOT-chain */
1767 : for (;;)
1768 2569974 : {
1769 : ItemId lp;
1770 :
1771 : /* check for bogus TID */
1772 47308634 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1773 : break;
1774 :
1775 47308634 : lp = PageGetItemId(page, offnum);
1776 :
1777 : /* check for unused, dead, or redirected items */
1778 47308634 : if (!ItemIdIsNormal(lp))
1779 : {
1780 : /* We should only see a redirect at start of chain */
1781 1726726 : if (ItemIdIsRedirected(lp) && at_chain_start)
1782 : {
1783 : /* Follow the redirect */
1784 971768 : offnum = ItemIdGetRedirect(lp);
1785 971768 : at_chain_start = false;
1786 971768 : continue;
1787 : }
1788 : /* else must be end of chain */
1789 754958 : break;
1790 : }
1791 :
1792 : /*
1793 : * Update heapTuple to point to the element of the HOT chain we're
1794 : * currently investigating. Having t_self set correctly is important
1795 : * because the SSI checks and the *Satisfies routine for historical
1796 : * MVCC snapshots need the correct tid to decide about the visibility.
1797 : */
1798 45581908 : heapTuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
1799 45581908 : heapTuple->t_len = ItemIdGetLength(lp);
1800 45581908 : heapTuple->t_tableOid = RelationGetRelid(relation);
1801 45581908 : ItemPointerSet(&heapTuple->t_self, blkno, offnum);
1802 :
1803 : /*
1804 : * Shouldn't see a HEAP_ONLY tuple at chain start.
1805 : */
1806 45581908 : if (at_chain_start && HeapTupleIsHeapOnly(heapTuple))
1807 0 : break;
1808 :
1809 : /*
1810 : * The xmin should match the previous xmax value, else chain is
1811 : * broken.
1812 : */
1813 47180114 : if (TransactionIdIsValid(prev_xmax) &&
1814 1598206 : !TransactionIdEquals(prev_xmax,
1815 : HeapTupleHeaderGetXmin(heapTuple->t_data)))
1816 0 : break;
1817 :
1818 : /*
1819 : * When first_call is true (and thus, skip is initially false) we'll
1820 : * return the first tuple we find. But on later passes, heapTuple
1821 : * will initially be pointing to the tuple we returned last time.
1822 : * Returning it again would be incorrect (and would loop forever), so
1823 : * we skip it and return the next match we find.
1824 : */
1825 45581908 : if (!skip)
1826 : {
1827 : /* If it's visible per the snapshot, we must return it */
1828 45410086 : valid = HeapTupleSatisfiesVisibility(heapTuple, snapshot, buffer);
1829 45410086 : HeapCheckForSerializableConflictOut(valid, relation, heapTuple,
1830 : buffer, snapshot);
1831 :
1832 45410076 : if (valid)
1833 : {
1834 31543946 : ItemPointerSetOffsetNumber(tid, offnum);
1835 31543946 : PredicateLockTID(relation, &heapTuple->t_self, snapshot,
1836 31543946 : HeapTupleHeaderGetXmin(heapTuple->t_data));
1837 31543946 : if (all_dead)
1838 25613370 : *all_dead = false;
1839 31543946 : return true;
1840 : }
1841 : }
1842 14037952 : skip = false;
1843 :
1844 : /*
1845 : * If we can't see it, maybe no one else can either. At caller
1846 : * request, check whether all chain members are dead to all
1847 : * transactions.
1848 : *
1849 : * Note: if you change the criterion here for what is "dead", fix the
1850 : * planner's get_actual_variable_range() function to match.
1851 : */
1852 14037952 : if (all_dead && *all_dead)
1853 : {
1854 12879382 : if (!vistest)
1855 12645194 : vistest = GlobalVisTestFor(relation);
1856 :
1857 12879382 : if (!HeapTupleIsSurelyDead(heapTuple, vistest))
1858 12177192 : *all_dead = false;
1859 : }
1860 :
1861 : /*
1862 : * Check to see if HOT chain continues past this tuple; if so fetch
1863 : * the next offnum and loop around.
1864 : */
1865 14037952 : if (HeapTupleIsHotUpdated(heapTuple))
1866 : {
1867 : Assert(ItemPointerGetBlockNumber(&heapTuple->t_data->t_ctid) ==
1868 : blkno);
1869 1598206 : offnum = ItemPointerGetOffsetNumber(&heapTuple->t_data->t_ctid);
1870 1598206 : at_chain_start = false;
1871 1598206 : prev_xmax = HeapTupleHeaderGetUpdateXid(heapTuple->t_data);
1872 : }
1873 : else
1874 12439746 : break; /* end of chain */
1875 : }
1876 :
1877 13194704 : return false;
1878 : }
1879 :
1880 : /*
1881 : * heap_get_latest_tid - get the latest tid of a specified tuple
1882 : *
1883 : * Actually, this gets the latest version that is visible according to the
1884 : * scan's snapshot. Create a scan using SnapshotDirty to get the very latest,
1885 : * possibly uncommitted version.
1886 : *
1887 : * *tid is both an input and an output parameter: it is updated to
1888 : * show the latest version of the row. Note that it will not be changed
1889 : * if no version of the row passes the snapshot test.
1890 : */
1891 : void
1892 300 : heap_get_latest_tid(TableScanDesc sscan,
1893 : ItemPointer tid)
1894 : {
1895 300 : Relation relation = sscan->rs_rd;
1896 300 : Snapshot snapshot = sscan->rs_snapshot;
1897 : ItemPointerData ctid;
1898 : TransactionId priorXmax;
1899 :
1900 : /*
1901 : * table_tuple_get_latest_tid() verified that the passed in tid is valid.
1902 : * Assume that t_ctid links are valid however - there shouldn't be invalid
1903 : * ones in the table.
1904 : */
1905 : Assert(ItemPointerIsValid(tid));
1906 :
1907 : /*
1908 : * Loop to chase down t_ctid links. At top of loop, ctid is the tuple we
1909 : * need to examine, and *tid is the TID we will return if ctid turns out
1910 : * to be bogus.
1911 : *
1912 : * Note that we will loop until we reach the end of the t_ctid chain.
1913 : * Depending on the snapshot passed, there might be at most one visible
1914 : * version of the row, but we don't try to optimize for that.
1915 : */
1916 300 : ctid = *tid;
1917 300 : priorXmax = InvalidTransactionId; /* cannot check first XMIN */
1918 : for (;;)
1919 90 : {
1920 : Buffer buffer;
1921 : Page page;
1922 : OffsetNumber offnum;
1923 : ItemId lp;
1924 : HeapTupleData tp;
1925 : bool valid;
1926 :
1927 : /*
1928 : * Read, pin, and lock the page.
1929 : */
1930 390 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(&ctid));
1931 390 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
1932 390 : page = BufferGetPage(buffer);
1933 :
1934 : /*
1935 : * Check for bogus item number. This is not treated as an error
1936 : * condition because it can happen while following a t_ctid link. We
1937 : * just assume that the prior tid is OK and return it unchanged.
1938 : */
1939 390 : offnum = ItemPointerGetOffsetNumber(&ctid);
1940 390 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1941 : {
1942 0 : UnlockReleaseBuffer(buffer);
1943 0 : break;
1944 : }
1945 390 : lp = PageGetItemId(page, offnum);
1946 390 : if (!ItemIdIsNormal(lp))
1947 : {
1948 0 : UnlockReleaseBuffer(buffer);
1949 0 : break;
1950 : }
1951 :
1952 : /* OK to access the tuple */
1953 390 : tp.t_self = ctid;
1954 390 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
1955 390 : tp.t_len = ItemIdGetLength(lp);
1956 390 : tp.t_tableOid = RelationGetRelid(relation);
1957 :
1958 : /*
1959 : * After following a t_ctid link, we might arrive at an unrelated
1960 : * tuple. Check for XMIN match.
1961 : */
1962 480 : if (TransactionIdIsValid(priorXmax) &&
1963 90 : !TransactionIdEquals(priorXmax, HeapTupleHeaderGetXmin(tp.t_data)))
1964 : {
1965 0 : UnlockReleaseBuffer(buffer);
1966 0 : break;
1967 : }
1968 :
1969 : /*
1970 : * Check tuple visibility; if visible, set it as the new result
1971 : * candidate.
1972 : */
1973 390 : valid = HeapTupleSatisfiesVisibility(&tp, snapshot, buffer);
1974 390 : HeapCheckForSerializableConflictOut(valid, relation, &tp, buffer, snapshot);
1975 390 : if (valid)
1976 276 : *tid = ctid;
1977 :
1978 : /*
1979 : * If there's a valid t_ctid link, follow it, else we're done.
1980 : */
1981 552 : if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
1982 276 : HeapTupleHeaderIsOnlyLocked(tp.t_data) ||
1983 228 : HeapTupleHeaderIndicatesMovedPartitions(tp.t_data) ||
1984 114 : ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
1985 : {
1986 300 : UnlockReleaseBuffer(buffer);
1987 300 : break;
1988 : }
1989 :
1990 90 : ctid = tp.t_data->t_ctid;
1991 90 : priorXmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
1992 90 : UnlockReleaseBuffer(buffer);
1993 : } /* end of loop */
1994 300 : }
1995 :
1996 :
1997 : /*
1998 : * UpdateXmaxHintBits - update tuple hint bits after xmax transaction ends
1999 : *
2000 : * This is called after we have waited for the XMAX transaction to terminate.
2001 : * If the transaction aborted, we guarantee the XMAX_INVALID hint bit will
2002 : * be set on exit. If the transaction committed, we set the XMAX_COMMITTED
2003 : * hint bit if possible --- but beware that that may not yet be possible,
2004 : * if the transaction committed asynchronously.
2005 : *
2006 : * Note that if the transaction was a locker only, we set HEAP_XMAX_INVALID
2007 : * even if it commits.
2008 : *
2009 : * Hence callers should look only at XMAX_INVALID.
2010 : *
2011 : * Note this is not allowed for tuples whose xmax is a multixact.
2012 : */
2013 : static void
2014 438 : UpdateXmaxHintBits(HeapTupleHeader tuple, Buffer buffer, TransactionId xid)
2015 : {
2016 : Assert(TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple), xid));
2017 : Assert(!(tuple->t_infomask & HEAP_XMAX_IS_MULTI));
2018 :
2019 438 : if (!(tuple->t_infomask & (HEAP_XMAX_COMMITTED | HEAP_XMAX_INVALID)))
2020 : {
2021 786 : if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask) &&
2022 348 : TransactionIdDidCommit(xid))
2023 294 : HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_COMMITTED,
2024 : xid);
2025 : else
2026 144 : HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_INVALID,
2027 : InvalidTransactionId);
2028 : }
2029 438 : }
2030 :
2031 :
2032 : /*
2033 : * GetBulkInsertState - prepare status object for a bulk insert
2034 : */
2035 : BulkInsertState
2036 4784 : GetBulkInsertState(void)
2037 : {
2038 : BulkInsertState bistate;
2039 :
2040 4784 : bistate = (BulkInsertState) palloc(sizeof(BulkInsertStateData));
2041 4784 : bistate->strategy = GetAccessStrategy(BAS_BULKWRITE);
2042 4784 : bistate->current_buf = InvalidBuffer;
2043 4784 : bistate->next_free = InvalidBlockNumber;
2044 4784 : bistate->last_free = InvalidBlockNumber;
2045 4784 : bistate->already_extended_by = 0;
2046 4784 : return bistate;
2047 : }
2048 :
2049 : /*
2050 : * FreeBulkInsertState - clean up after finishing a bulk insert
2051 : */
2052 : void
2053 4496 : FreeBulkInsertState(BulkInsertState bistate)
2054 : {
2055 4496 : if (bistate->current_buf != InvalidBuffer)
2056 3612 : ReleaseBuffer(bistate->current_buf);
2057 4496 : FreeAccessStrategy(bistate->strategy);
2058 4496 : pfree(bistate);
2059 4496 : }
2060 :
2061 : /*
2062 : * ReleaseBulkInsertStatePin - release a buffer currently held in bistate
2063 : */
2064 : void
2065 161516 : ReleaseBulkInsertStatePin(BulkInsertState bistate)
2066 : {
2067 161516 : if (bistate->current_buf != InvalidBuffer)
2068 60042 : ReleaseBuffer(bistate->current_buf);
2069 161516 : bistate->current_buf = InvalidBuffer;
2070 :
2071 : /*
2072 : * Despite the name, we also reset bulk relation extension state.
2073 : * Otherwise we can end up erroring out due to looking for free space in
2074 : * ->next_free of one partition, even though ->next_free was set when
2075 : * extending another partition. It could obviously also be bad for
2076 : * efficiency to look at existing blocks at offsets from another
2077 : * partition, even if we don't error out.
2078 : */
2079 161516 : bistate->next_free = InvalidBlockNumber;
2080 161516 : bistate->last_free = InvalidBlockNumber;
2081 161516 : }
2082 :
2083 :
2084 : /*
2085 : * heap_insert - insert tuple into a heap
2086 : *
2087 : * The new tuple is stamped with current transaction ID and the specified
2088 : * command ID.
2089 : *
2090 : * See table_tuple_insert for comments about most of the input flags, except
2091 : * that this routine directly takes a tuple rather than a slot.
2092 : *
2093 : * There's corresponding HEAP_INSERT_ options to all the TABLE_INSERT_
2094 : * options, and there additionally is HEAP_INSERT_SPECULATIVE which is used to
2095 : * implement table_tuple_insert_speculative().
2096 : *
2097 : * On return the header fields of *tup are updated to match the stored tuple;
2098 : * in particular tup->t_self receives the actual TID where the tuple was
2099 : * stored. But note that any toasting of fields within the tuple data is NOT
2100 : * reflected into *tup.
2101 : */
2102 : void
2103 16701978 : heap_insert(Relation relation, HeapTuple tup, CommandId cid,
2104 : int options, BulkInsertState bistate)
2105 : {
2106 16701978 : TransactionId xid = GetCurrentTransactionId();
2107 : HeapTuple heaptup;
2108 : Buffer buffer;
2109 16701964 : Buffer vmbuffer = InvalidBuffer;
2110 16701964 : bool all_visible_cleared = false;
2111 :
2112 : /* Cheap, simplistic check that the tuple matches the rel's rowtype. */
2113 : Assert(HeapTupleHeaderGetNatts(tup->t_data) <=
2114 : RelationGetNumberOfAttributes(relation));
2115 :
2116 16701964 : AssertHasSnapshotForToast(relation);
2117 :
2118 : /*
2119 : * Fill in tuple header fields and toast the tuple if necessary.
2120 : *
2121 : * Note: below this point, heaptup is the data we actually intend to store
2122 : * into the relation; tup is the caller's original untoasted data.
2123 : */
2124 16701964 : heaptup = heap_prepare_insert(relation, tup, xid, cid, options);
2125 :
2126 : /*
2127 : * Find buffer to insert this tuple into. If the page is all visible,
2128 : * this will also pin the requisite visibility map page.
2129 : */
2130 16701964 : buffer = RelationGetBufferForTuple(relation, heaptup->t_len,
2131 : InvalidBuffer, options, bistate,
2132 : &vmbuffer, NULL,
2133 : 0);
2134 :
2135 : /*
2136 : * We're about to do the actual insert -- but check for conflict first, to
2137 : * avoid possibly having to roll back work we've just done.
2138 : *
2139 : * This is safe without a recheck as long as there is no possibility of
2140 : * another process scanning the page between this check and the insert
2141 : * being visible to the scan (i.e., an exclusive buffer content lock is
2142 : * continuously held from this point until the tuple insert is visible).
2143 : *
2144 : * For a heap insert, we only need to check for table-level SSI locks. Our
2145 : * new tuple can't possibly conflict with existing tuple locks, and heap
2146 : * page locks are only consolidated versions of tuple locks; they do not
2147 : * lock "gaps" as index page locks do. So we don't need to specify a
2148 : * buffer when making the call, which makes for a faster check.
2149 : */
2150 16701964 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2151 :
2152 : /* NO EREPORT(ERROR) from here till changes are logged */
2153 16701940 : START_CRIT_SECTION();
2154 :
2155 16701940 : RelationPutHeapTuple(relation, buffer, heaptup,
2156 16701940 : (options & HEAP_INSERT_SPECULATIVE) != 0);
2157 :
2158 16701940 : if (PageIsAllVisible(BufferGetPage(buffer)))
2159 : {
2160 14766 : all_visible_cleared = true;
2161 14766 : PageClearAllVisible(BufferGetPage(buffer));
2162 14766 : visibilitymap_clear(relation,
2163 14766 : ItemPointerGetBlockNumber(&(heaptup->t_self)),
2164 : vmbuffer, VISIBILITYMAP_VALID_BITS);
2165 : }
2166 :
2167 : /*
2168 : * XXX Should we set PageSetPrunable on this page ?
2169 : *
2170 : * The inserting transaction may eventually abort thus making this tuple
2171 : * DEAD and hence available for pruning. Though we don't want to optimize
2172 : * for aborts, if no other tuple in this page is UPDATEd/DELETEd, the
2173 : * aborted tuple will never be pruned until next vacuum is triggered.
2174 : *
2175 : * If you do add PageSetPrunable here, add it in heap_xlog_insert too.
2176 : */
2177 :
2178 16701940 : MarkBufferDirty(buffer);
2179 :
2180 : /* XLOG stuff */
2181 16701940 : if (RelationNeedsWAL(relation))
2182 : {
2183 : xl_heap_insert xlrec;
2184 : xl_heap_header xlhdr;
2185 : XLogRecPtr recptr;
2186 14189356 : Page page = BufferGetPage(buffer);
2187 14189356 : uint8 info = XLOG_HEAP_INSERT;
2188 14189356 : int bufflags = 0;
2189 :
2190 : /*
2191 : * If this is a catalog, we need to transmit combo CIDs to properly
2192 : * decode, so log that as well.
2193 : */
2194 14189356 : if (RelationIsAccessibleInLogicalDecoding(relation))
2195 6850 : log_heap_new_cid(relation, heaptup);
2196 :
2197 : /*
2198 : * If this is the single and first tuple on page, we can reinit the
2199 : * page instead of restoring the whole thing. Set flag, and hide
2200 : * buffer references from XLogInsert.
2201 : */
2202 14372704 : if (ItemPointerGetOffsetNumber(&(heaptup->t_self)) == FirstOffsetNumber &&
2203 183348 : PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
2204 : {
2205 181590 : info |= XLOG_HEAP_INIT_PAGE;
2206 181590 : bufflags |= REGBUF_WILL_INIT;
2207 : }
2208 :
2209 14189356 : xlrec.offnum = ItemPointerGetOffsetNumber(&heaptup->t_self);
2210 14189356 : xlrec.flags = 0;
2211 14189356 : if (all_visible_cleared)
2212 14760 : xlrec.flags |= XLH_INSERT_ALL_VISIBLE_CLEARED;
2213 14189356 : if (options & HEAP_INSERT_SPECULATIVE)
2214 4140 : xlrec.flags |= XLH_INSERT_IS_SPECULATIVE;
2215 : Assert(ItemPointerGetBlockNumber(&heaptup->t_self) == BufferGetBlockNumber(buffer));
2216 :
2217 : /*
2218 : * For logical decoding, we need the tuple even if we're doing a full
2219 : * page write, so make sure it's included even if we take a full-page
2220 : * image. (XXX We could alternatively store a pointer into the FPW).
2221 : */
2222 14189356 : if (RelationIsLogicallyLogged(relation) &&
2223 501044 : !(options & HEAP_INSERT_NO_LOGICAL))
2224 : {
2225 500990 : xlrec.flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
2226 500990 : bufflags |= REGBUF_KEEP_DATA;
2227 :
2228 500990 : if (IsToastRelation(relation))
2229 3572 : xlrec.flags |= XLH_INSERT_ON_TOAST_RELATION;
2230 : }
2231 :
2232 14189356 : XLogBeginInsert();
2233 14189356 : XLogRegisterData(&xlrec, SizeOfHeapInsert);
2234 :
2235 14189356 : xlhdr.t_infomask2 = heaptup->t_data->t_infomask2;
2236 14189356 : xlhdr.t_infomask = heaptup->t_data->t_infomask;
2237 14189356 : xlhdr.t_hoff = heaptup->t_data->t_hoff;
2238 :
2239 : /*
2240 : * note we mark xlhdr as belonging to buffer; if XLogInsert decides to
2241 : * write the whole page to the xlog, we don't need to store
2242 : * xl_heap_header in the xlog.
2243 : */
2244 14189356 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD | bufflags);
2245 14189356 : XLogRegisterBufData(0, &xlhdr, SizeOfHeapHeader);
2246 : /* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
2247 14189356 : XLogRegisterBufData(0,
2248 14189356 : (char *) heaptup->t_data + SizeofHeapTupleHeader,
2249 14189356 : heaptup->t_len - SizeofHeapTupleHeader);
2250 :
2251 : /* filtering by origin on a row level is much more efficient */
2252 14189356 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
2253 :
2254 14189356 : recptr = XLogInsert(RM_HEAP_ID, info);
2255 :
2256 14189356 : PageSetLSN(page, recptr);
2257 : }
2258 :
2259 16701940 : END_CRIT_SECTION();
2260 :
2261 16701940 : UnlockReleaseBuffer(buffer);
2262 16701940 : if (vmbuffer != InvalidBuffer)
2263 15328 : ReleaseBuffer(vmbuffer);
2264 :
2265 : /*
2266 : * If tuple is cacheable, mark it for invalidation from the caches in case
2267 : * we abort. Note it is OK to do this after releasing the buffer, because
2268 : * the heaptup data structure is all in local memory, not in the shared
2269 : * buffer.
2270 : */
2271 16701940 : CacheInvalidateHeapTuple(relation, heaptup, NULL);
2272 :
2273 : /* Note: speculative insertions are counted too, even if aborted later */
2274 16701940 : pgstat_count_heap_insert(relation, 1);
2275 :
2276 : /*
2277 : * If heaptup is a private copy, release it. Don't forget to copy t_self
2278 : * back to the caller's image, too.
2279 : */
2280 16701940 : if (heaptup != tup)
2281 : {
2282 36448 : tup->t_self = heaptup->t_self;
2283 36448 : heap_freetuple(heaptup);
2284 : }
2285 16701940 : }
2286 :
2287 : /*
2288 : * Subroutine for heap_insert(). Prepares a tuple for insertion. This sets the
2289 : * tuple header fields and toasts the tuple if necessary. Returns a toasted
2290 : * version of the tuple if it was toasted, or the original tuple if not. Note
2291 : * that in any case, the header fields are also set in the original tuple.
2292 : */
2293 : static HeapTuple
2294 19657896 : heap_prepare_insert(Relation relation, HeapTuple tup, TransactionId xid,
2295 : CommandId cid, int options)
2296 : {
2297 : /*
2298 : * To allow parallel inserts, we need to ensure that they are safe to be
2299 : * performed in workers. We have the infrastructure to allow parallel
2300 : * inserts in general except for the cases where inserts generate a new
2301 : * CommandId (eg. inserts into a table having a foreign key column).
2302 : */
2303 19657896 : if (IsParallelWorker())
2304 0 : ereport(ERROR,
2305 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
2306 : errmsg("cannot insert tuples in a parallel worker")));
2307 :
2308 19657896 : tup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
2309 19657896 : tup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
2310 19657896 : tup->t_data->t_infomask |= HEAP_XMAX_INVALID;
2311 19657896 : HeapTupleHeaderSetXmin(tup->t_data, xid);
2312 19657896 : if (options & HEAP_INSERT_FROZEN)
2313 204176 : HeapTupleHeaderSetXminFrozen(tup->t_data);
2314 :
2315 19657896 : HeapTupleHeaderSetCmin(tup->t_data, cid);
2316 19657896 : HeapTupleHeaderSetXmax(tup->t_data, 0); /* for cleanliness */
2317 19657896 : tup->t_tableOid = RelationGetRelid(relation);
2318 :
2319 : /*
2320 : * If the new tuple is too big for storage or contains already toasted
2321 : * out-of-line attributes from some other relation, invoke the toaster.
2322 : */
2323 19657896 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
2324 62194 : relation->rd_rel->relkind != RELKIND_MATVIEW)
2325 : {
2326 : /* toast table entries should never be recursively toasted */
2327 : Assert(!HeapTupleHasExternal(tup));
2328 62098 : return tup;
2329 : }
2330 19595798 : else if (HeapTupleHasExternal(tup) || tup->t_len > TOAST_TUPLE_THRESHOLD)
2331 36536 : return heap_toast_insert_or_update(relation, tup, NULL, options);
2332 : else
2333 19559262 : return tup;
2334 : }
2335 :
2336 : /*
2337 : * Helper for heap_multi_insert() that computes the number of entire pages
2338 : * that inserting the remaining heaptuples requires. Used to determine how
2339 : * much the relation needs to be extended by.
2340 : */
2341 : static int
2342 733938 : heap_multi_insert_pages(HeapTuple *heaptuples, int done, int ntuples, Size saveFreeSpace)
2343 : {
2344 733938 : size_t page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
2345 733938 : int npages = 1;
2346 :
2347 4874590 : for (int i = done; i < ntuples; i++)
2348 : {
2349 4140652 : size_t tup_sz = sizeof(ItemIdData) + MAXALIGN(heaptuples[i]->t_len);
2350 :
2351 4140652 : if (page_avail < tup_sz)
2352 : {
2353 31092 : npages++;
2354 31092 : page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
2355 : }
2356 4140652 : page_avail -= tup_sz;
2357 : }
2358 :
2359 733938 : return npages;
2360 : }
2361 :
2362 : /*
2363 : * heap_multi_insert - insert multiple tuples into a heap
2364 : *
2365 : * This is like heap_insert(), but inserts multiple tuples in one operation.
2366 : * That's faster than calling heap_insert() in a loop, because when multiple
2367 : * tuples can be inserted on a single page, we can write just a single WAL
2368 : * record covering all of them, and only need to lock/unlock the page once.
2369 : *
2370 : * Note: this leaks memory into the current memory context. You can create a
2371 : * temporary context before calling this, if that's a problem.
2372 : */
2373 : void
2374 721108 : heap_multi_insert(Relation relation, TupleTableSlot **slots, int ntuples,
2375 : CommandId cid, int options, BulkInsertState bistate)
2376 : {
2377 721108 : TransactionId xid = GetCurrentTransactionId();
2378 : HeapTuple *heaptuples;
2379 : int i;
2380 : int ndone;
2381 : PGAlignedBlock scratch;
2382 : Page page;
2383 721108 : Buffer vmbuffer = InvalidBuffer;
2384 : bool needwal;
2385 : Size saveFreeSpace;
2386 721108 : bool need_tuple_data = RelationIsLogicallyLogged(relation);
2387 721108 : bool need_cids = RelationIsAccessibleInLogicalDecoding(relation);
2388 721108 : bool starting_with_empty_page = false;
2389 721108 : int npages = 0;
2390 721108 : int npages_used = 0;
2391 :
2392 : /* currently not needed (thus unsupported) for heap_multi_insert() */
2393 : Assert(!(options & HEAP_INSERT_NO_LOGICAL));
2394 :
2395 721108 : AssertHasSnapshotForToast(relation);
2396 :
2397 721108 : needwal = RelationNeedsWAL(relation);
2398 721108 : saveFreeSpace = RelationGetTargetPageFreeSpace(relation,
2399 : HEAP_DEFAULT_FILLFACTOR);
2400 :
2401 : /* Toast and set header data in all the slots */
2402 721108 : heaptuples = palloc(ntuples * sizeof(HeapTuple));
2403 3677040 : for (i = 0; i < ntuples; i++)
2404 : {
2405 : HeapTuple tuple;
2406 :
2407 2955932 : tuple = ExecFetchSlotHeapTuple(slots[i], true, NULL);
2408 2955932 : slots[i]->tts_tableOid = RelationGetRelid(relation);
2409 2955932 : tuple->t_tableOid = slots[i]->tts_tableOid;
2410 2955932 : heaptuples[i] = heap_prepare_insert(relation, tuple, xid, cid,
2411 : options);
2412 : }
2413 :
2414 : /*
2415 : * We're about to do the actual inserts -- but check for conflict first,
2416 : * to minimize the possibility of having to roll back work we've just
2417 : * done.
2418 : *
2419 : * A check here does not definitively prevent a serialization anomaly;
2420 : * that check MUST be done at least past the point of acquiring an
2421 : * exclusive buffer content lock on every buffer that will be affected,
2422 : * and MAY be done after all inserts are reflected in the buffers and
2423 : * those locks are released; otherwise there is a race condition. Since
2424 : * multiple buffers can be locked and unlocked in the loop below, and it
2425 : * would not be feasible to identify and lock all of those buffers before
2426 : * the loop, we must do a final check at the end.
2427 : *
2428 : * The check here could be omitted with no loss of correctness; it is
2429 : * present strictly as an optimization.
2430 : *
2431 : * For heap inserts, we only need to check for table-level SSI locks. Our
2432 : * new tuples can't possibly conflict with existing tuple locks, and heap
2433 : * page locks are only consolidated versions of tuple locks; they do not
2434 : * lock "gaps" as index page locks do. So we don't need to specify a
2435 : * buffer when making the call, which makes for a faster check.
2436 : */
2437 721108 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2438 :
2439 721108 : ndone = 0;
2440 1471348 : while (ndone < ntuples)
2441 : {
2442 : Buffer buffer;
2443 750240 : bool all_visible_cleared = false;
2444 750240 : bool all_frozen_set = false;
2445 : int nthispage;
2446 :
2447 750240 : CHECK_FOR_INTERRUPTS();
2448 :
2449 : /*
2450 : * Compute number of pages needed to fit the to-be-inserted tuples in
2451 : * the worst case. This will be used to determine how much to extend
2452 : * the relation by in RelationGetBufferForTuple(), if needed. If we
2453 : * filled a prior page from scratch, we can just update our last
2454 : * computation, but if we started with a partially filled page,
2455 : * recompute from scratch, the number of potentially required pages
2456 : * can vary due to tuples needing to fit onto the page, page headers
2457 : * etc.
2458 : */
2459 750240 : if (ndone == 0 || !starting_with_empty_page)
2460 : {
2461 733938 : npages = heap_multi_insert_pages(heaptuples, ndone, ntuples,
2462 : saveFreeSpace);
2463 733938 : npages_used = 0;
2464 : }
2465 : else
2466 16302 : npages_used++;
2467 :
2468 : /*
2469 : * Find buffer where at least the next tuple will fit. If the page is
2470 : * all-visible, this will also pin the requisite visibility map page.
2471 : *
2472 : * Also pin visibility map page if COPY FREEZE inserts tuples into an
2473 : * empty page. See all_frozen_set below.
2474 : */
2475 750240 : buffer = RelationGetBufferForTuple(relation, heaptuples[ndone]->t_len,
2476 : InvalidBuffer, options, bistate,
2477 : &vmbuffer, NULL,
2478 : npages - npages_used);
2479 750240 : page = BufferGetPage(buffer);
2480 :
2481 750240 : starting_with_empty_page = PageGetMaxOffsetNumber(page) == 0;
2482 :
2483 750240 : if (starting_with_empty_page && (options & HEAP_INSERT_FROZEN))
2484 : {
2485 3322 : all_frozen_set = true;
2486 : /* Lock the vmbuffer before entering the critical section */
2487 3322 : LockBuffer(vmbuffer, BUFFER_LOCK_EXCLUSIVE);
2488 : }
2489 :
2490 : /* NO EREPORT(ERROR) from here till changes are logged */
2491 750240 : START_CRIT_SECTION();
2492 :
2493 : /*
2494 : * RelationGetBufferForTuple has ensured that the first tuple fits.
2495 : * Put that on the page, and then as many other tuples as fit.
2496 : */
2497 750240 : RelationPutHeapTuple(relation, buffer, heaptuples[ndone], false);
2498 :
2499 : /*
2500 : * For logical decoding we need combo CIDs to properly decode the
2501 : * catalog.
2502 : */
2503 750240 : if (needwal && need_cids)
2504 10100 : log_heap_new_cid(relation, heaptuples[ndone]);
2505 :
2506 2955932 : for (nthispage = 1; ndone + nthispage < ntuples; nthispage++)
2507 : {
2508 2234824 : HeapTuple heaptup = heaptuples[ndone + nthispage];
2509 :
2510 2234824 : if (PageGetHeapFreeSpace(page) < MAXALIGN(heaptup->t_len) + saveFreeSpace)
2511 29132 : break;
2512 :
2513 2205692 : RelationPutHeapTuple(relation, buffer, heaptup, false);
2514 :
2515 : /*
2516 : * For logical decoding we need combo CIDs to properly decode the
2517 : * catalog.
2518 : */
2519 2205692 : if (needwal && need_cids)
2520 9454 : log_heap_new_cid(relation, heaptup);
2521 : }
2522 :
2523 : /*
2524 : * If the page is all visible, need to clear that, unless we're only
2525 : * going to add further frozen rows to it.
2526 : *
2527 : * If we're only adding already frozen rows to a previously empty
2528 : * page, mark it as all-frozen and update the visibility map. We're
2529 : * already holding a pin on the vmbuffer.
2530 : */
2531 750240 : if (PageIsAllVisible(page) && !(options & HEAP_INSERT_FROZEN))
2532 : {
2533 5626 : all_visible_cleared = true;
2534 5626 : PageClearAllVisible(page);
2535 5626 : visibilitymap_clear(relation,
2536 : BufferGetBlockNumber(buffer),
2537 : vmbuffer, VISIBILITYMAP_VALID_BITS);
2538 : }
2539 744614 : else if (all_frozen_set)
2540 : {
2541 3322 : PageSetAllVisible(page);
2542 3322 : visibilitymap_set_vmbits(BufferGetBlockNumber(buffer),
2543 : vmbuffer,
2544 : VISIBILITYMAP_ALL_VISIBLE |
2545 : VISIBILITYMAP_ALL_FROZEN,
2546 : relation->rd_locator);
2547 : }
2548 :
2549 : /*
2550 : * XXX Should we set PageSetPrunable on this page ? See heap_insert()
2551 : */
2552 :
2553 750240 : MarkBufferDirty(buffer);
2554 :
2555 : /* XLOG stuff */
2556 750240 : if (needwal)
2557 : {
2558 : XLogRecPtr recptr;
2559 : xl_heap_multi_insert *xlrec;
2560 742584 : uint8 info = XLOG_HEAP2_MULTI_INSERT;
2561 : char *tupledata;
2562 : int totaldatalen;
2563 742584 : char *scratchptr = scratch.data;
2564 : bool init;
2565 742584 : int bufflags = 0;
2566 :
2567 : /*
2568 : * If the page was previously empty, we can reinit the page
2569 : * instead of restoring the whole thing.
2570 : */
2571 742584 : init = starting_with_empty_page;
2572 :
2573 : /* allocate xl_heap_multi_insert struct from the scratch area */
2574 742584 : xlrec = (xl_heap_multi_insert *) scratchptr;
2575 742584 : scratchptr += SizeOfHeapMultiInsert;
2576 :
2577 : /*
2578 : * Allocate offsets array. Unless we're reinitializing the page,
2579 : * in that case the tuples are stored in order starting at
2580 : * FirstOffsetNumber and we don't need to store the offsets
2581 : * explicitly.
2582 : */
2583 742584 : if (!init)
2584 716164 : scratchptr += nthispage * sizeof(OffsetNumber);
2585 :
2586 : /* the rest of the scratch space is used for tuple data */
2587 742584 : tupledata = scratchptr;
2588 :
2589 : /* check that the mutually exclusive flags are not both set */
2590 : Assert(!(all_visible_cleared && all_frozen_set));
2591 :
2592 742584 : xlrec->flags = 0;
2593 742584 : if (all_visible_cleared)
2594 5626 : xlrec->flags = XLH_INSERT_ALL_VISIBLE_CLEARED;
2595 :
2596 : /*
2597 : * We don't have to worry about including a conflict xid in the
2598 : * WAL record, as HEAP_INSERT_FROZEN intentionally violates
2599 : * visibility rules.
2600 : */
2601 742584 : if (all_frozen_set)
2602 26 : xlrec->flags = XLH_INSERT_ALL_FROZEN_SET;
2603 :
2604 742584 : xlrec->ntuples = nthispage;
2605 :
2606 : /*
2607 : * Write out an xl_multi_insert_tuple and the tuple data itself
2608 : * for each tuple.
2609 : */
2610 3287686 : for (i = 0; i < nthispage; i++)
2611 : {
2612 2545102 : HeapTuple heaptup = heaptuples[ndone + i];
2613 : xl_multi_insert_tuple *tuphdr;
2614 : int datalen;
2615 :
2616 2545102 : if (!init)
2617 1496636 : xlrec->offsets[i] = ItemPointerGetOffsetNumber(&heaptup->t_self);
2618 : /* xl_multi_insert_tuple needs two-byte alignment. */
2619 2545102 : tuphdr = (xl_multi_insert_tuple *) SHORTALIGN(scratchptr);
2620 2545102 : scratchptr = ((char *) tuphdr) + SizeOfMultiInsertTuple;
2621 :
2622 2545102 : tuphdr->t_infomask2 = heaptup->t_data->t_infomask2;
2623 2545102 : tuphdr->t_infomask = heaptup->t_data->t_infomask;
2624 2545102 : tuphdr->t_hoff = heaptup->t_data->t_hoff;
2625 :
2626 : /* write bitmap [+ padding] [+ oid] + data */
2627 2545102 : datalen = heaptup->t_len - SizeofHeapTupleHeader;
2628 2545102 : memcpy(scratchptr,
2629 2545102 : (char *) heaptup->t_data + SizeofHeapTupleHeader,
2630 : datalen);
2631 2545102 : tuphdr->datalen = datalen;
2632 2545102 : scratchptr += datalen;
2633 : }
2634 742584 : totaldatalen = scratchptr - tupledata;
2635 : Assert((scratchptr - scratch.data) < BLCKSZ);
2636 :
2637 742584 : if (need_tuple_data)
2638 144 : xlrec->flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
2639 :
2640 : /*
2641 : * Signal that this is the last xl_heap_multi_insert record
2642 : * emitted by this call to heap_multi_insert(). Needed for logical
2643 : * decoding so it knows when to cleanup temporary data.
2644 : */
2645 742584 : if (ndone + nthispage == ntuples)
2646 720270 : xlrec->flags |= XLH_INSERT_LAST_IN_MULTI;
2647 :
2648 742584 : if (init)
2649 : {
2650 26420 : info |= XLOG_HEAP_INIT_PAGE;
2651 26420 : bufflags |= REGBUF_WILL_INIT;
2652 : }
2653 :
2654 : /*
2655 : * If we're doing logical decoding, include the new tuple data
2656 : * even if we take a full-page image of the page.
2657 : */
2658 742584 : if (need_tuple_data)
2659 144 : bufflags |= REGBUF_KEEP_DATA;
2660 :
2661 742584 : XLogBeginInsert();
2662 742584 : XLogRegisterData(xlrec, tupledata - scratch.data);
2663 742584 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD | bufflags);
2664 742584 : if (all_frozen_set)
2665 26 : XLogRegisterBuffer(1, vmbuffer, 0);
2666 :
2667 742584 : XLogRegisterBufData(0, tupledata, totaldatalen);
2668 :
2669 : /* filtering by origin on a row level is much more efficient */
2670 742584 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
2671 :
2672 742584 : recptr = XLogInsert(RM_HEAP2_ID, info);
2673 :
2674 742584 : PageSetLSN(page, recptr);
2675 742584 : if (all_frozen_set)
2676 : {
2677 : Assert(BufferIsDirty(vmbuffer));
2678 26 : PageSetLSN(BufferGetPage(vmbuffer), recptr);
2679 : }
2680 : }
2681 :
2682 750240 : END_CRIT_SECTION();
2683 :
2684 750240 : if (all_frozen_set)
2685 3322 : LockBuffer(vmbuffer, BUFFER_LOCK_UNLOCK);
2686 :
2687 750240 : UnlockReleaseBuffer(buffer);
2688 750240 : ndone += nthispage;
2689 :
2690 : /*
2691 : * NB: Only release vmbuffer after inserting all tuples - it's fairly
2692 : * likely that we'll insert into subsequent heap pages that are likely
2693 : * to use the same vm page.
2694 : */
2695 : }
2696 :
2697 : /* We're done with inserting all tuples, so release the last vmbuffer. */
2698 721108 : if (vmbuffer != InvalidBuffer)
2699 5812 : ReleaseBuffer(vmbuffer);
2700 :
2701 : /*
2702 : * We're done with the actual inserts. Check for conflicts again, to
2703 : * ensure that all rw-conflicts in to these inserts are detected. Without
2704 : * this final check, a sequential scan of the heap may have locked the
2705 : * table after the "before" check, missing one opportunity to detect the
2706 : * conflict, and then scanned the table before the new tuples were there,
2707 : * missing the other chance to detect the conflict.
2708 : *
2709 : * For heap inserts, we only need to check for table-level SSI locks. Our
2710 : * new tuples can't possibly conflict with existing tuple locks, and heap
2711 : * page locks are only consolidated versions of tuple locks; they do not
2712 : * lock "gaps" as index page locks do. So we don't need to specify a
2713 : * buffer when making the call.
2714 : */
2715 721108 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2716 :
2717 : /*
2718 : * If tuples are cacheable, mark them for invalidation from the caches in
2719 : * case we abort. Note it is OK to do this after releasing the buffer,
2720 : * because the heaptuples data structure is all in local memory, not in
2721 : * the shared buffer.
2722 : */
2723 721108 : if (IsCatalogRelation(relation))
2724 : {
2725 2471490 : for (i = 0; i < ntuples; i++)
2726 1752812 : CacheInvalidateHeapTuple(relation, heaptuples[i], NULL);
2727 : }
2728 :
2729 : /* copy t_self fields back to the caller's slots */
2730 3677040 : for (i = 0; i < ntuples; i++)
2731 2955932 : slots[i]->tts_tid = heaptuples[i]->t_self;
2732 :
2733 721108 : pgstat_count_heap_insert(relation, ntuples);
2734 721108 : }
2735 :
2736 : /*
2737 : * simple_heap_insert - insert a tuple
2738 : *
2739 : * Currently, this routine differs from heap_insert only in supplying
2740 : * a default command ID and not allowing access to the speedup options.
2741 : *
2742 : * This should be used rather than using heap_insert directly in most places
2743 : * where we are modifying system catalogs.
2744 : */
2745 : void
2746 1803290 : simple_heap_insert(Relation relation, HeapTuple tup)
2747 : {
2748 1803290 : heap_insert(relation, tup, GetCurrentCommandId(true), 0, NULL);
2749 1803290 : }
2750 :
2751 : /*
2752 : * Given infomask/infomask2, compute the bits that must be saved in the
2753 : * "infobits" field of xl_heap_delete, xl_heap_update, xl_heap_lock,
2754 : * xl_heap_lock_updated WAL records.
2755 : *
2756 : * See fix_infomask_from_infobits.
2757 : */
2758 : static uint8
2759 3903834 : compute_infobits(uint16 infomask, uint16 infomask2)
2760 : {
2761 : return
2762 3903834 : ((infomask & HEAP_XMAX_IS_MULTI) != 0 ? XLHL_XMAX_IS_MULTI : 0) |
2763 3903834 : ((infomask & HEAP_XMAX_LOCK_ONLY) != 0 ? XLHL_XMAX_LOCK_ONLY : 0) |
2764 3903834 : ((infomask & HEAP_XMAX_EXCL_LOCK) != 0 ? XLHL_XMAX_EXCL_LOCK : 0) |
2765 : /* note we ignore HEAP_XMAX_SHR_LOCK here */
2766 7807668 : ((infomask & HEAP_XMAX_KEYSHR_LOCK) != 0 ? XLHL_XMAX_KEYSHR_LOCK : 0) |
2767 : ((infomask2 & HEAP_KEYS_UPDATED) != 0 ?
2768 3903834 : XLHL_KEYS_UPDATED : 0);
2769 : }
2770 :
2771 : /*
2772 : * Given two versions of the same t_infomask for a tuple, compare them and
2773 : * return whether the relevant status for a tuple Xmax has changed. This is
2774 : * used after a buffer lock has been released and reacquired: we want to ensure
2775 : * that the tuple state continues to be the same it was when we previously
2776 : * examined it.
2777 : *
2778 : * Note the Xmax field itself must be compared separately.
2779 : */
2780 : static inline bool
2781 10748 : xmax_infomask_changed(uint16 new_infomask, uint16 old_infomask)
2782 : {
2783 10748 : const uint16 interesting =
2784 : HEAP_XMAX_IS_MULTI | HEAP_XMAX_LOCK_ONLY | HEAP_LOCK_MASK;
2785 :
2786 10748 : if ((new_infomask & interesting) != (old_infomask & interesting))
2787 28 : return true;
2788 :
2789 10720 : return false;
2790 : }
2791 :
2792 : /*
2793 : * heap_delete - delete a tuple
2794 : *
2795 : * See table_tuple_delete() for an explanation of the parameters, except that
2796 : * this routine directly takes a tuple rather than a slot.
2797 : *
2798 : * In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
2799 : * t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
2800 : * only for TM_SelfModified, since we cannot obtain cmax from a combo CID
2801 : * generated by another transaction).
2802 : */
2803 : TM_Result
2804 3000022 : heap_delete(Relation relation, const ItemPointerData *tid,
2805 : CommandId cid, Snapshot crosscheck, bool wait,
2806 : TM_FailureData *tmfd, bool changingPart)
2807 : {
2808 : TM_Result result;
2809 3000022 : TransactionId xid = GetCurrentTransactionId();
2810 : ItemId lp;
2811 : HeapTupleData tp;
2812 : Page page;
2813 : BlockNumber block;
2814 : Buffer buffer;
2815 3000022 : Buffer vmbuffer = InvalidBuffer;
2816 : TransactionId new_xmax;
2817 : uint16 new_infomask,
2818 : new_infomask2;
2819 3000022 : bool have_tuple_lock = false;
2820 : bool iscombo;
2821 3000022 : bool all_visible_cleared = false;
2822 3000022 : HeapTuple old_key_tuple = NULL; /* replica identity of the tuple */
2823 3000022 : bool old_key_copied = false;
2824 :
2825 : Assert(ItemPointerIsValid(tid));
2826 :
2827 3000022 : AssertHasSnapshotForToast(relation);
2828 :
2829 : /*
2830 : * Forbid this during a parallel operation, lest it allocate a combo CID.
2831 : * Other workers might need that combo CID for visibility checks, and we
2832 : * have no provision for broadcasting it to them.
2833 : */
2834 3000022 : if (IsInParallelMode())
2835 0 : ereport(ERROR,
2836 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
2837 : errmsg("cannot delete tuples during a parallel operation")));
2838 :
2839 3000022 : block = ItemPointerGetBlockNumber(tid);
2840 3000022 : buffer = ReadBuffer(relation, block);
2841 3000022 : page = BufferGetPage(buffer);
2842 :
2843 : /*
2844 : * Before locking the buffer, pin the visibility map page if it appears to
2845 : * be necessary. Since we haven't got the lock yet, someone else might be
2846 : * in the middle of changing this, so we'll need to recheck after we have
2847 : * the lock.
2848 : */
2849 3000022 : if (PageIsAllVisible(page))
2850 374 : visibilitymap_pin(relation, block, &vmbuffer);
2851 :
2852 3000022 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2853 :
2854 3000022 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
2855 : Assert(ItemIdIsNormal(lp));
2856 :
2857 3000022 : tp.t_tableOid = RelationGetRelid(relation);
2858 3000022 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
2859 3000022 : tp.t_len = ItemIdGetLength(lp);
2860 3000022 : tp.t_self = *tid;
2861 :
2862 2 : l1:
2863 :
2864 : /*
2865 : * If we didn't pin the visibility map page and the page has become all
2866 : * visible while we were busy locking the buffer, we'll have to unlock and
2867 : * re-lock, to avoid holding the buffer lock across an I/O. That's a bit
2868 : * unfortunate, but hopefully shouldn't happen often.
2869 : */
2870 3000024 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
2871 : {
2872 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2873 0 : visibilitymap_pin(relation, block, &vmbuffer);
2874 0 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2875 : }
2876 :
2877 3000024 : result = HeapTupleSatisfiesUpdate(&tp, cid, buffer);
2878 :
2879 3000024 : if (result == TM_Invisible)
2880 : {
2881 0 : UnlockReleaseBuffer(buffer);
2882 0 : ereport(ERROR,
2883 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2884 : errmsg("attempted to delete invisible tuple")));
2885 : }
2886 3000024 : else if (result == TM_BeingModified && wait)
2887 : {
2888 : TransactionId xwait;
2889 : uint16 infomask;
2890 :
2891 : /* must copy state data before unlocking buffer */
2892 81120 : xwait = HeapTupleHeaderGetRawXmax(tp.t_data);
2893 81120 : infomask = tp.t_data->t_infomask;
2894 :
2895 : /*
2896 : * Sleep until concurrent transaction ends -- except when there's a
2897 : * single locker and it's our own transaction. Note we don't care
2898 : * which lock mode the locker has, because we need the strongest one.
2899 : *
2900 : * Before sleeping, we need to acquire tuple lock to establish our
2901 : * priority for the tuple (see heap_lock_tuple). LockTuple will
2902 : * release us when we are next-in-line for the tuple.
2903 : *
2904 : * If we are forced to "start over" below, we keep the tuple lock;
2905 : * this arranges that we stay at the head of the line while rechecking
2906 : * tuple state.
2907 : */
2908 81120 : if (infomask & HEAP_XMAX_IS_MULTI)
2909 : {
2910 16 : bool current_is_member = false;
2911 :
2912 16 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
2913 : LockTupleExclusive, ¤t_is_member))
2914 : {
2915 16 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2916 :
2917 : /*
2918 : * Acquire the lock, if necessary (but skip it when we're
2919 : * requesting a lock and already have one; avoids deadlock).
2920 : */
2921 16 : if (!current_is_member)
2922 12 : heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
2923 : LockWaitBlock, &have_tuple_lock);
2924 :
2925 : /* wait for multixact */
2926 16 : MultiXactIdWait((MultiXactId) xwait, MultiXactStatusUpdate, infomask,
2927 : relation, &(tp.t_self), XLTW_Delete,
2928 : NULL);
2929 16 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2930 :
2931 : /*
2932 : * If xwait had just locked the tuple then some other xact
2933 : * could update this tuple before we get to this point. Check
2934 : * for xmax change, and start over if so.
2935 : *
2936 : * We also must start over if we didn't pin the VM page, and
2937 : * the page has become all visible.
2938 : */
2939 32 : if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
2940 32 : xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
2941 16 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
2942 : xwait))
2943 0 : goto l1;
2944 : }
2945 :
2946 : /*
2947 : * You might think the multixact is necessarily done here, but not
2948 : * so: it could have surviving members, namely our own xact or
2949 : * other subxacts of this backend. It is legal for us to delete
2950 : * the tuple in either case, however (the latter case is
2951 : * essentially a situation of upgrading our former shared lock to
2952 : * exclusive). We don't bother changing the on-disk hint bits
2953 : * since we are about to overwrite the xmax altogether.
2954 : */
2955 : }
2956 81104 : else if (!TransactionIdIsCurrentTransactionId(xwait))
2957 : {
2958 : /*
2959 : * Wait for regular transaction to end; but first, acquire tuple
2960 : * lock.
2961 : */
2962 104 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2963 104 : heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
2964 : LockWaitBlock, &have_tuple_lock);
2965 104 : XactLockTableWait(xwait, relation, &(tp.t_self), XLTW_Delete);
2966 96 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2967 :
2968 : /*
2969 : * xwait is done, but if xwait had just locked the tuple then some
2970 : * other xact could update this tuple before we get to this point.
2971 : * Check for xmax change, and start over if so.
2972 : *
2973 : * We also must start over if we didn't pin the VM page, and the
2974 : * page has become all visible.
2975 : */
2976 192 : if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
2977 190 : xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
2978 94 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
2979 : xwait))
2980 2 : goto l1;
2981 :
2982 : /* Otherwise check if it committed or aborted */
2983 94 : UpdateXmaxHintBits(tp.t_data, buffer, xwait);
2984 : }
2985 :
2986 : /*
2987 : * We may overwrite if previous xmax aborted, or if it committed but
2988 : * only locked the tuple without updating it.
2989 : */
2990 162180 : if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
2991 81132 : HEAP_XMAX_IS_LOCKED_ONLY(tp.t_data->t_infomask) ||
2992 62 : HeapTupleHeaderIsOnlyLocked(tp.t_data))
2993 81056 : result = TM_Ok;
2994 54 : else if (!ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
2995 46 : result = TM_Updated;
2996 : else
2997 8 : result = TM_Deleted;
2998 : }
2999 :
3000 : /* sanity check the result HeapTupleSatisfiesUpdate() and the logic above */
3001 : if (result != TM_Ok)
3002 : {
3003 : Assert(result == TM_SelfModified ||
3004 : result == TM_Updated ||
3005 : result == TM_Deleted ||
3006 : result == TM_BeingModified);
3007 : Assert(!(tp.t_data->t_infomask & HEAP_XMAX_INVALID));
3008 : Assert(result != TM_Updated ||
3009 : !ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid));
3010 : }
3011 :
3012 3000014 : if (crosscheck != InvalidSnapshot && result == TM_Ok)
3013 : {
3014 : /* Perform additional check for transaction-snapshot mode RI updates */
3015 2 : if (!HeapTupleSatisfiesVisibility(&tp, crosscheck, buffer))
3016 2 : result = TM_Updated;
3017 : }
3018 :
3019 3000014 : if (result != TM_Ok)
3020 : {
3021 124 : tmfd->ctid = tp.t_data->t_ctid;
3022 124 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
3023 124 : if (result == TM_SelfModified)
3024 42 : tmfd->cmax = HeapTupleHeaderGetCmax(tp.t_data);
3025 : else
3026 82 : tmfd->cmax = InvalidCommandId;
3027 124 : UnlockReleaseBuffer(buffer);
3028 124 : if (have_tuple_lock)
3029 54 : UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
3030 124 : if (vmbuffer != InvalidBuffer)
3031 0 : ReleaseBuffer(vmbuffer);
3032 124 : return result;
3033 : }
3034 :
3035 : /*
3036 : * We're about to do the actual delete -- check for conflict first, to
3037 : * avoid possibly having to roll back work we've just done.
3038 : *
3039 : * This is safe without a recheck as long as there is no possibility of
3040 : * another process scanning the page between this check and the delete
3041 : * being visible to the scan (i.e., an exclusive buffer content lock is
3042 : * continuously held from this point until the tuple delete is visible).
3043 : */
3044 2999890 : CheckForSerializableConflictIn(relation, tid, BufferGetBlockNumber(buffer));
3045 :
3046 : /* replace cid with a combo CID if necessary */
3047 2999862 : HeapTupleHeaderAdjustCmax(tp.t_data, &cid, &iscombo);
3048 :
3049 : /*
3050 : * Compute replica identity tuple before entering the critical section so
3051 : * we don't PANIC upon a memory allocation failure.
3052 : */
3053 2999862 : old_key_tuple = ExtractReplicaIdentity(relation, &tp, true, &old_key_copied);
3054 :
3055 : /*
3056 : * If this is the first possibly-multixact-able operation in the current
3057 : * transaction, set my per-backend OldestMemberMXactId setting. We can be
3058 : * certain that the transaction will never become a member of any older
3059 : * MultiXactIds than that. (We have to do this even if we end up just
3060 : * using our own TransactionId below, since some other backend could
3061 : * incorporate our XID into a MultiXact immediately afterwards.)
3062 : */
3063 2999862 : MultiXactIdSetOldestMember();
3064 :
3065 2999862 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(tp.t_data),
3066 2999862 : tp.t_data->t_infomask, tp.t_data->t_infomask2,
3067 : xid, LockTupleExclusive, true,
3068 : &new_xmax, &new_infomask, &new_infomask2);
3069 :
3070 2999862 : START_CRIT_SECTION();
3071 :
3072 : /*
3073 : * If this transaction commits, the tuple will become DEAD sooner or
3074 : * later. Set flag that this page is a candidate for pruning once our xid
3075 : * falls below the OldestXmin horizon. If the transaction finally aborts,
3076 : * the subsequent page pruning will be a no-op and the hint will be
3077 : * cleared.
3078 : */
3079 2999862 : PageSetPrunable(page, xid);
3080 :
3081 2999862 : if (PageIsAllVisible(page))
3082 : {
3083 374 : all_visible_cleared = true;
3084 374 : PageClearAllVisible(page);
3085 374 : visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
3086 : vmbuffer, VISIBILITYMAP_VALID_BITS);
3087 : }
3088 :
3089 : /* store transaction information of xact deleting the tuple */
3090 2999862 : tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
3091 2999862 : tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
3092 2999862 : tp.t_data->t_infomask |= new_infomask;
3093 2999862 : tp.t_data->t_infomask2 |= new_infomask2;
3094 2999862 : HeapTupleHeaderClearHotUpdated(tp.t_data);
3095 2999862 : HeapTupleHeaderSetXmax(tp.t_data, new_xmax);
3096 2999862 : HeapTupleHeaderSetCmax(tp.t_data, cid, iscombo);
3097 : /* Make sure there is no forward chain link in t_ctid */
3098 2999862 : tp.t_data->t_ctid = tp.t_self;
3099 :
3100 : /* Signal that this is actually a move into another partition */
3101 2999862 : if (changingPart)
3102 978 : HeapTupleHeaderSetMovedPartitions(tp.t_data);
3103 :
3104 2999862 : MarkBufferDirty(buffer);
3105 :
3106 : /*
3107 : * XLOG stuff
3108 : *
3109 : * NB: heap_abort_speculative() uses the same xlog record and replay
3110 : * routines.
3111 : */
3112 2999862 : if (RelationNeedsWAL(relation))
3113 : {
3114 : xl_heap_delete xlrec;
3115 : xl_heap_header xlhdr;
3116 : XLogRecPtr recptr;
3117 :
3118 : /*
3119 : * For logical decode we need combo CIDs to properly decode the
3120 : * catalog
3121 : */
3122 2874624 : if (RelationIsAccessibleInLogicalDecoding(relation))
3123 12544 : log_heap_new_cid(relation, &tp);
3124 :
3125 2874624 : xlrec.flags = 0;
3126 2874624 : if (all_visible_cleared)
3127 374 : xlrec.flags |= XLH_DELETE_ALL_VISIBLE_CLEARED;
3128 2874624 : if (changingPart)
3129 978 : xlrec.flags |= XLH_DELETE_IS_PARTITION_MOVE;
3130 5749248 : xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
3131 2874624 : tp.t_data->t_infomask2);
3132 2874624 : xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
3133 2874624 : xlrec.xmax = new_xmax;
3134 :
3135 2874624 : if (old_key_tuple != NULL)
3136 : {
3137 94040 : if (relation->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
3138 264 : xlrec.flags |= XLH_DELETE_CONTAINS_OLD_TUPLE;
3139 : else
3140 93776 : xlrec.flags |= XLH_DELETE_CONTAINS_OLD_KEY;
3141 : }
3142 :
3143 2874624 : XLogBeginInsert();
3144 2874624 : XLogRegisterData(&xlrec, SizeOfHeapDelete);
3145 :
3146 2874624 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
3147 :
3148 : /*
3149 : * Log replica identity of the deleted tuple if there is one
3150 : */
3151 2874624 : if (old_key_tuple != NULL)
3152 : {
3153 94040 : xlhdr.t_infomask2 = old_key_tuple->t_data->t_infomask2;
3154 94040 : xlhdr.t_infomask = old_key_tuple->t_data->t_infomask;
3155 94040 : xlhdr.t_hoff = old_key_tuple->t_data->t_hoff;
3156 :
3157 94040 : XLogRegisterData(&xlhdr, SizeOfHeapHeader);
3158 94040 : XLogRegisterData((char *) old_key_tuple->t_data
3159 : + SizeofHeapTupleHeader,
3160 94040 : old_key_tuple->t_len
3161 : - SizeofHeapTupleHeader);
3162 : }
3163 :
3164 : /* filtering by origin on a row level is much more efficient */
3165 2874624 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
3166 :
3167 2874624 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
3168 :
3169 2874624 : PageSetLSN(page, recptr);
3170 : }
3171 :
3172 2999862 : END_CRIT_SECTION();
3173 :
3174 2999862 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3175 :
3176 2999862 : if (vmbuffer != InvalidBuffer)
3177 374 : ReleaseBuffer(vmbuffer);
3178 :
3179 : /*
3180 : * If the tuple has toasted out-of-line attributes, we need to delete
3181 : * those items too. We have to do this before releasing the buffer
3182 : * because we need to look at the contents of the tuple, but it's OK to
3183 : * release the content lock on the buffer first.
3184 : */
3185 2999862 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
3186 5114 : relation->rd_rel->relkind != RELKIND_MATVIEW)
3187 : {
3188 : /* toast table entries should never be recursively toasted */
3189 : Assert(!HeapTupleHasExternal(&tp));
3190 : }
3191 2994768 : else if (HeapTupleHasExternal(&tp))
3192 598 : heap_toast_delete(relation, &tp, false);
3193 :
3194 : /*
3195 : * Mark tuple for invalidation from system caches at next command
3196 : * boundary. We have to do this before releasing the buffer because we
3197 : * need to look at the contents of the tuple.
3198 : */
3199 2999862 : CacheInvalidateHeapTuple(relation, &tp, NULL);
3200 :
3201 : /* Now we can release the buffer */
3202 2999862 : ReleaseBuffer(buffer);
3203 :
3204 : /*
3205 : * Release the lmgr tuple lock, if we had it.
3206 : */
3207 2999862 : if (have_tuple_lock)
3208 52 : UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
3209 :
3210 2999862 : pgstat_count_heap_delete(relation);
3211 :
3212 2999862 : if (old_key_tuple != NULL && old_key_copied)
3213 93778 : heap_freetuple(old_key_tuple);
3214 :
3215 2999862 : return TM_Ok;
3216 : }
3217 :
3218 : /*
3219 : * simple_heap_delete - delete a tuple
3220 : *
3221 : * This routine may be used to delete a tuple when concurrent updates of
3222 : * the target tuple are not expected (for example, because we have a lock
3223 : * on the relation associated with the tuple). Any failure is reported
3224 : * via ereport().
3225 : */
3226 : void
3227 1268078 : simple_heap_delete(Relation relation, const ItemPointerData *tid)
3228 : {
3229 : TM_Result result;
3230 : TM_FailureData tmfd;
3231 :
3232 1268078 : result = heap_delete(relation, tid,
3233 : GetCurrentCommandId(true), InvalidSnapshot,
3234 : true /* wait for commit */ ,
3235 : &tmfd, false /* changingPart */ );
3236 1268078 : switch (result)
3237 : {
3238 0 : case TM_SelfModified:
3239 : /* Tuple was already updated in current command? */
3240 0 : elog(ERROR, "tuple already updated by self");
3241 : break;
3242 :
3243 1268078 : case TM_Ok:
3244 : /* done successfully */
3245 1268078 : break;
3246 :
3247 0 : case TM_Updated:
3248 0 : elog(ERROR, "tuple concurrently updated");
3249 : break;
3250 :
3251 0 : case TM_Deleted:
3252 0 : elog(ERROR, "tuple concurrently deleted");
3253 : break;
3254 :
3255 0 : default:
3256 0 : elog(ERROR, "unrecognized heap_delete status: %u", result);
3257 : break;
3258 : }
3259 1268078 : }
3260 :
3261 : /*
3262 : * heap_update - replace a tuple
3263 : *
3264 : * See table_tuple_update() for an explanation of the parameters, except that
3265 : * this routine directly takes a tuple rather than a slot.
3266 : *
3267 : * In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
3268 : * t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
3269 : * only for TM_SelfModified, since we cannot obtain cmax from a combo CID
3270 : * generated by another transaction).
3271 : */
3272 : TM_Result
3273 607676 : heap_update(Relation relation, const ItemPointerData *otid, HeapTuple newtup,
3274 : CommandId cid, Snapshot crosscheck, bool wait,
3275 : TM_FailureData *tmfd, LockTupleMode *lockmode,
3276 : TU_UpdateIndexes *update_indexes)
3277 : {
3278 : TM_Result result;
3279 607676 : TransactionId xid = GetCurrentTransactionId();
3280 : Bitmapset *hot_attrs;
3281 : Bitmapset *sum_attrs;
3282 : Bitmapset *key_attrs;
3283 : Bitmapset *id_attrs;
3284 : Bitmapset *interesting_attrs;
3285 : Bitmapset *modified_attrs;
3286 : ItemId lp;
3287 : HeapTupleData oldtup;
3288 : HeapTuple heaptup;
3289 607676 : HeapTuple old_key_tuple = NULL;
3290 607676 : bool old_key_copied = false;
3291 : Page page;
3292 : BlockNumber block;
3293 : MultiXactStatus mxact_status;
3294 : Buffer buffer,
3295 : newbuf,
3296 607676 : vmbuffer = InvalidBuffer,
3297 607676 : vmbuffer_new = InvalidBuffer;
3298 : bool need_toast;
3299 : Size newtupsize,
3300 : pagefree;
3301 607676 : bool have_tuple_lock = false;
3302 : bool iscombo;
3303 607676 : bool use_hot_update = false;
3304 607676 : bool summarized_update = false;
3305 : bool key_intact;
3306 607676 : bool all_visible_cleared = false;
3307 607676 : bool all_visible_cleared_new = false;
3308 : bool checked_lockers;
3309 : bool locker_remains;
3310 607676 : bool id_has_external = false;
3311 : TransactionId xmax_new_tuple,
3312 : xmax_old_tuple;
3313 : uint16 infomask_old_tuple,
3314 : infomask2_old_tuple,
3315 : infomask_new_tuple,
3316 : infomask2_new_tuple;
3317 :
3318 : Assert(ItemPointerIsValid(otid));
3319 :
3320 : /* Cheap, simplistic check that the tuple matches the rel's rowtype. */
3321 : Assert(HeapTupleHeaderGetNatts(newtup->t_data) <=
3322 : RelationGetNumberOfAttributes(relation));
3323 :
3324 607676 : AssertHasSnapshotForToast(relation);
3325 :
3326 : /*
3327 : * Forbid this during a parallel operation, lest it allocate a combo CID.
3328 : * Other workers might need that combo CID for visibility checks, and we
3329 : * have no provision for broadcasting it to them.
3330 : */
3331 607676 : if (IsInParallelMode())
3332 0 : ereport(ERROR,
3333 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
3334 : errmsg("cannot update tuples during a parallel operation")));
3335 :
3336 : #ifdef USE_ASSERT_CHECKING
3337 : check_lock_if_inplace_updateable_rel(relation, otid, newtup);
3338 : #endif
3339 :
3340 : /*
3341 : * Fetch the list of attributes to be checked for various operations.
3342 : *
3343 : * For HOT considerations, this is wasted effort if we fail to update or
3344 : * have to put the new tuple on a different page. But we must compute the
3345 : * list before obtaining buffer lock --- in the worst case, if we are
3346 : * doing an update on one of the relevant system catalogs, we could
3347 : * deadlock if we try to fetch the list later. In any case, the relcache
3348 : * caches the data so this is usually pretty cheap.
3349 : *
3350 : * We also need columns used by the replica identity and columns that are
3351 : * considered the "key" of rows in the table.
3352 : *
3353 : * Note that we get copies of each bitmap, so we need not worry about
3354 : * relcache flush happening midway through.
3355 : */
3356 607676 : hot_attrs = RelationGetIndexAttrBitmap(relation,
3357 : INDEX_ATTR_BITMAP_HOT_BLOCKING);
3358 607676 : sum_attrs = RelationGetIndexAttrBitmap(relation,
3359 : INDEX_ATTR_BITMAP_SUMMARIZED);
3360 607676 : key_attrs = RelationGetIndexAttrBitmap(relation, INDEX_ATTR_BITMAP_KEY);
3361 607676 : id_attrs = RelationGetIndexAttrBitmap(relation,
3362 : INDEX_ATTR_BITMAP_IDENTITY_KEY);
3363 607676 : interesting_attrs = NULL;
3364 607676 : interesting_attrs = bms_add_members(interesting_attrs, hot_attrs);
3365 607676 : interesting_attrs = bms_add_members(interesting_attrs, sum_attrs);
3366 607676 : interesting_attrs = bms_add_members(interesting_attrs, key_attrs);
3367 607676 : interesting_attrs = bms_add_members(interesting_attrs, id_attrs);
3368 :
3369 607676 : block = ItemPointerGetBlockNumber(otid);
3370 607676 : INJECTION_POINT("heap_update-before-pin", NULL);
3371 607676 : buffer = ReadBuffer(relation, block);
3372 607676 : page = BufferGetPage(buffer);
3373 :
3374 : /*
3375 : * Before locking the buffer, pin the visibility map page if it appears to
3376 : * be necessary. Since we haven't got the lock yet, someone else might be
3377 : * in the middle of changing this, so we'll need to recheck after we have
3378 : * the lock.
3379 : */
3380 607676 : if (PageIsAllVisible(page))
3381 2844 : visibilitymap_pin(relation, block, &vmbuffer);
3382 :
3383 607676 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3384 :
3385 607676 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(otid));
3386 :
3387 : /*
3388 : * Usually, a buffer pin and/or snapshot blocks pruning of otid, ensuring
3389 : * we see LP_NORMAL here. When the otid origin is a syscache, we may have
3390 : * neither a pin nor a snapshot. Hence, we may see other LP_ states, each
3391 : * of which indicates concurrent pruning.
3392 : *
3393 : * Failing with TM_Updated would be most accurate. However, unlike other
3394 : * TM_Updated scenarios, we don't know the successor ctid in LP_UNUSED and
3395 : * LP_DEAD cases. While the distinction between TM_Updated and TM_Deleted
3396 : * does matter to SQL statements UPDATE and MERGE, those SQL statements
3397 : * hold a snapshot that ensures LP_NORMAL. Hence, the choice between
3398 : * TM_Updated and TM_Deleted affects only the wording of error messages.
3399 : * Settle on TM_Deleted, for two reasons. First, it avoids complicating
3400 : * the specification of when tmfd->ctid is valid. Second, it creates
3401 : * error log evidence that we took this branch.
3402 : *
3403 : * Since it's possible to see LP_UNUSED at otid, it's also possible to see
3404 : * LP_NORMAL for a tuple that replaced LP_UNUSED. If it's a tuple for an
3405 : * unrelated row, we'll fail with "duplicate key value violates unique".
3406 : * XXX if otid is the live, newer version of the newtup row, we'll discard
3407 : * changes originating in versions of this catalog row after the version
3408 : * the caller got from syscache. See syscache-update-pruned.spec.
3409 : */
3410 607676 : if (!ItemIdIsNormal(lp))
3411 : {
3412 : Assert(RelationSupportsSysCache(RelationGetRelid(relation)));
3413 :
3414 2 : UnlockReleaseBuffer(buffer);
3415 : Assert(!have_tuple_lock);
3416 2 : if (vmbuffer != InvalidBuffer)
3417 2 : ReleaseBuffer(vmbuffer);
3418 2 : tmfd->ctid = *otid;
3419 2 : tmfd->xmax = InvalidTransactionId;
3420 2 : tmfd->cmax = InvalidCommandId;
3421 2 : *update_indexes = TU_None;
3422 :
3423 2 : bms_free(hot_attrs);
3424 2 : bms_free(sum_attrs);
3425 2 : bms_free(key_attrs);
3426 2 : bms_free(id_attrs);
3427 : /* modified_attrs not yet initialized */
3428 2 : bms_free(interesting_attrs);
3429 2 : return TM_Deleted;
3430 : }
3431 :
3432 : /*
3433 : * Fill in enough data in oldtup for HeapDetermineColumnsInfo to work
3434 : * properly.
3435 : */
3436 607674 : oldtup.t_tableOid = RelationGetRelid(relation);
3437 607674 : oldtup.t_data = (HeapTupleHeader) PageGetItem(page, lp);
3438 607674 : oldtup.t_len = ItemIdGetLength(lp);
3439 607674 : oldtup.t_self = *otid;
3440 :
3441 : /* the new tuple is ready, except for this: */
3442 607674 : newtup->t_tableOid = RelationGetRelid(relation);
3443 :
3444 : /*
3445 : * Determine columns modified by the update. Additionally, identify
3446 : * whether any of the unmodified replica identity key attributes in the
3447 : * old tuple is externally stored or not. This is required because for
3448 : * such attributes the flattened value won't be WAL logged as part of the
3449 : * new tuple so we must include it as part of the old_key_tuple. See
3450 : * ExtractReplicaIdentity.
3451 : */
3452 607674 : modified_attrs = HeapDetermineColumnsInfo(relation, interesting_attrs,
3453 : id_attrs, &oldtup,
3454 : newtup, &id_has_external);
3455 :
3456 : /*
3457 : * If we're not updating any "key" column, we can grab a weaker lock type.
3458 : * This allows for more concurrency when we are running simultaneously
3459 : * with foreign key checks.
3460 : *
3461 : * Note that if a column gets detoasted while executing the update, but
3462 : * the value ends up being the same, this test will fail and we will use
3463 : * the stronger lock. This is acceptable; the important case to optimize
3464 : * is updates that don't manipulate key columns, not those that
3465 : * serendipitously arrive at the same key values.
3466 : */
3467 607674 : if (!bms_overlap(modified_attrs, key_attrs))
3468 : {
3469 599282 : *lockmode = LockTupleNoKeyExclusive;
3470 599282 : mxact_status = MultiXactStatusNoKeyUpdate;
3471 599282 : key_intact = true;
3472 :
3473 : /*
3474 : * If this is the first possibly-multixact-able operation in the
3475 : * current transaction, set my per-backend OldestMemberMXactId
3476 : * setting. We can be certain that the transaction will never become a
3477 : * member of any older MultiXactIds than that. (We have to do this
3478 : * even if we end up just using our own TransactionId below, since
3479 : * some other backend could incorporate our XID into a MultiXact
3480 : * immediately afterwards.)
3481 : */
3482 599282 : MultiXactIdSetOldestMember();
3483 : }
3484 : else
3485 : {
3486 8392 : *lockmode = LockTupleExclusive;
3487 8392 : mxact_status = MultiXactStatusUpdate;
3488 8392 : key_intact = false;
3489 : }
3490 :
3491 : /*
3492 : * Note: beyond this point, use oldtup not otid to refer to old tuple.
3493 : * otid may very well point at newtup->t_self, which we will overwrite
3494 : * with the new tuple's location, so there's great risk of confusion if we
3495 : * use otid anymore.
3496 : */
3497 :
3498 607674 : l2:
3499 607676 : checked_lockers = false;
3500 607676 : locker_remains = false;
3501 607676 : result = HeapTupleSatisfiesUpdate(&oldtup, cid, buffer);
3502 :
3503 : /* see below about the "no wait" case */
3504 : Assert(result != TM_BeingModified || wait);
3505 :
3506 607676 : if (result == TM_Invisible)
3507 : {
3508 0 : UnlockReleaseBuffer(buffer);
3509 0 : ereport(ERROR,
3510 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3511 : errmsg("attempted to update invisible tuple")));
3512 : }
3513 607676 : else if (result == TM_BeingModified && wait)
3514 : {
3515 : TransactionId xwait;
3516 : uint16 infomask;
3517 71968 : bool can_continue = false;
3518 :
3519 : /*
3520 : * XXX note that we don't consider the "no wait" case here. This
3521 : * isn't a problem currently because no caller uses that case, but it
3522 : * should be fixed if such a caller is introduced. It wasn't a
3523 : * problem previously because this code would always wait, but now
3524 : * that some tuple locks do not conflict with one of the lock modes we
3525 : * use, it is possible that this case is interesting to handle
3526 : * specially.
3527 : *
3528 : * This may cause failures with third-party code that calls
3529 : * heap_update directly.
3530 : */
3531 :
3532 : /* must copy state data before unlocking buffer */
3533 71968 : xwait = HeapTupleHeaderGetRawXmax(oldtup.t_data);
3534 71968 : infomask = oldtup.t_data->t_infomask;
3535 :
3536 : /*
3537 : * Now we have to do something about the existing locker. If it's a
3538 : * multi, sleep on it; we might be awakened before it is completely
3539 : * gone (or even not sleep at all in some cases); we need to preserve
3540 : * it as locker, unless it is gone completely.
3541 : *
3542 : * If it's not a multi, we need to check for sleeping conditions
3543 : * before actually going to sleep. If the update doesn't conflict
3544 : * with the locks, we just continue without sleeping (but making sure
3545 : * it is preserved).
3546 : *
3547 : * Before sleeping, we need to acquire tuple lock to establish our
3548 : * priority for the tuple (see heap_lock_tuple). LockTuple will
3549 : * release us when we are next-in-line for the tuple. Note we must
3550 : * not acquire the tuple lock until we're sure we're going to sleep;
3551 : * otherwise we're open for race conditions with other transactions
3552 : * holding the tuple lock which sleep on us.
3553 : *
3554 : * If we are forced to "start over" below, we keep the tuple lock;
3555 : * this arranges that we stay at the head of the line while rechecking
3556 : * tuple state.
3557 : */
3558 71968 : if (infomask & HEAP_XMAX_IS_MULTI)
3559 : {
3560 : TransactionId update_xact;
3561 : int remain;
3562 120 : bool current_is_member = false;
3563 :
3564 120 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
3565 : *lockmode, ¤t_is_member))
3566 : {
3567 16 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3568 :
3569 : /*
3570 : * Acquire the lock, if necessary (but skip it when we're
3571 : * requesting a lock and already have one; avoids deadlock).
3572 : */
3573 16 : if (!current_is_member)
3574 0 : heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3575 : LockWaitBlock, &have_tuple_lock);
3576 :
3577 : /* wait for multixact */
3578 16 : MultiXactIdWait((MultiXactId) xwait, mxact_status, infomask,
3579 : relation, &oldtup.t_self, XLTW_Update,
3580 : &remain);
3581 16 : checked_lockers = true;
3582 16 : locker_remains = remain != 0;
3583 16 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3584 :
3585 : /*
3586 : * If xwait had just locked the tuple then some other xact
3587 : * could update this tuple before we get to this point. Check
3588 : * for xmax change, and start over if so.
3589 : */
3590 16 : if (xmax_infomask_changed(oldtup.t_data->t_infomask,
3591 16 : infomask) ||
3592 16 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3593 : xwait))
3594 0 : goto l2;
3595 : }
3596 :
3597 : /*
3598 : * Note that the multixact may not be done by now. It could have
3599 : * surviving members; our own xact or other subxacts of this
3600 : * backend, and also any other concurrent transaction that locked
3601 : * the tuple with LockTupleKeyShare if we only got
3602 : * LockTupleNoKeyExclusive. If this is the case, we have to be
3603 : * careful to mark the updated tuple with the surviving members in
3604 : * Xmax.
3605 : *
3606 : * Note that there could have been another update in the
3607 : * MultiXact. In that case, we need to check whether it committed
3608 : * or aborted. If it aborted we are safe to update it again;
3609 : * otherwise there is an update conflict, and we have to return
3610 : * TableTuple{Deleted, Updated} below.
3611 : *
3612 : * In the LockTupleExclusive case, we still need to preserve the
3613 : * surviving members: those would include the tuple locks we had
3614 : * before this one, which are important to keep in case this
3615 : * subxact aborts.
3616 : */
3617 120 : if (!HEAP_XMAX_IS_LOCKED_ONLY(oldtup.t_data->t_infomask))
3618 16 : update_xact = HeapTupleGetUpdateXid(oldtup.t_data);
3619 : else
3620 104 : update_xact = InvalidTransactionId;
3621 :
3622 : /*
3623 : * There was no UPDATE in the MultiXact; or it aborted. No
3624 : * TransactionIdIsInProgress() call needed here, since we called
3625 : * MultiXactIdWait() above.
3626 : */
3627 136 : if (!TransactionIdIsValid(update_xact) ||
3628 16 : TransactionIdDidAbort(update_xact))
3629 106 : can_continue = true;
3630 : }
3631 71848 : else if (TransactionIdIsCurrentTransactionId(xwait))
3632 : {
3633 : /*
3634 : * The only locker is ourselves; we can avoid grabbing the tuple
3635 : * lock here, but must preserve our locking information.
3636 : */
3637 71634 : checked_lockers = true;
3638 71634 : locker_remains = true;
3639 71634 : can_continue = true;
3640 : }
3641 214 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) && key_intact)
3642 : {
3643 : /*
3644 : * If it's just a key-share locker, and we're not changing the key
3645 : * columns, we don't need to wait for it to end; but we need to
3646 : * preserve it as locker.
3647 : */
3648 58 : checked_lockers = true;
3649 58 : locker_remains = true;
3650 58 : can_continue = true;
3651 : }
3652 : else
3653 : {
3654 : /*
3655 : * Wait for regular transaction to end; but first, acquire tuple
3656 : * lock.
3657 : */
3658 156 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3659 156 : heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3660 : LockWaitBlock, &have_tuple_lock);
3661 156 : XactLockTableWait(xwait, relation, &oldtup.t_self,
3662 : XLTW_Update);
3663 156 : checked_lockers = true;
3664 156 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3665 :
3666 : /*
3667 : * xwait is done, but if xwait had just locked the tuple then some
3668 : * other xact could update this tuple before we get to this point.
3669 : * Check for xmax change, and start over if so.
3670 : */
3671 310 : if (xmax_infomask_changed(oldtup.t_data->t_infomask, infomask) ||
3672 154 : !TransactionIdEquals(xwait,
3673 : HeapTupleHeaderGetRawXmax(oldtup.t_data)))
3674 2 : goto l2;
3675 :
3676 : /* Otherwise check if it committed or aborted */
3677 154 : UpdateXmaxHintBits(oldtup.t_data, buffer, xwait);
3678 154 : if (oldtup.t_data->t_infomask & HEAP_XMAX_INVALID)
3679 44 : can_continue = true;
3680 : }
3681 :
3682 71966 : if (can_continue)
3683 71842 : result = TM_Ok;
3684 124 : else if (!ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid))
3685 114 : result = TM_Updated;
3686 : else
3687 10 : result = TM_Deleted;
3688 : }
3689 :
3690 : /* Sanity check the result HeapTupleSatisfiesUpdate() and the logic above */
3691 : if (result != TM_Ok)
3692 : {
3693 : Assert(result == TM_SelfModified ||
3694 : result == TM_Updated ||
3695 : result == TM_Deleted ||
3696 : result == TM_BeingModified);
3697 : Assert(!(oldtup.t_data->t_infomask & HEAP_XMAX_INVALID));
3698 : Assert(result != TM_Updated ||
3699 : !ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid));
3700 : }
3701 :
3702 607674 : if (crosscheck != InvalidSnapshot && result == TM_Ok)
3703 : {
3704 : /* Perform additional check for transaction-snapshot mode RI updates */
3705 2 : if (!HeapTupleSatisfiesVisibility(&oldtup, crosscheck, buffer))
3706 2 : result = TM_Updated;
3707 : }
3708 :
3709 607674 : if (result != TM_Ok)
3710 : {
3711 320 : tmfd->ctid = oldtup.t_data->t_ctid;
3712 320 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(oldtup.t_data);
3713 320 : if (result == TM_SelfModified)
3714 104 : tmfd->cmax = HeapTupleHeaderGetCmax(oldtup.t_data);
3715 : else
3716 216 : tmfd->cmax = InvalidCommandId;
3717 320 : UnlockReleaseBuffer(buffer);
3718 320 : if (have_tuple_lock)
3719 110 : UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
3720 320 : if (vmbuffer != InvalidBuffer)
3721 0 : ReleaseBuffer(vmbuffer);
3722 320 : *update_indexes = TU_None;
3723 :
3724 320 : bms_free(hot_attrs);
3725 320 : bms_free(sum_attrs);
3726 320 : bms_free(key_attrs);
3727 320 : bms_free(id_attrs);
3728 320 : bms_free(modified_attrs);
3729 320 : bms_free(interesting_attrs);
3730 320 : return result;
3731 : }
3732 :
3733 : /*
3734 : * If we didn't pin the visibility map page and the page has become all
3735 : * visible while we were busy locking the buffer, or during some
3736 : * subsequent window during which we had it unlocked, we'll have to unlock
3737 : * and re-lock, to avoid holding the buffer lock across an I/O. That's a
3738 : * bit unfortunate, especially since we'll now have to recheck whether the
3739 : * tuple has been locked or updated under us, but hopefully it won't
3740 : * happen very often.
3741 : */
3742 607354 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
3743 : {
3744 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3745 0 : visibilitymap_pin(relation, block, &vmbuffer);
3746 0 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3747 0 : goto l2;
3748 : }
3749 :
3750 : /* Fill in transaction status data */
3751 :
3752 : /*
3753 : * If the tuple we're updating is locked, we need to preserve the locking
3754 : * info in the old tuple's Xmax. Prepare a new Xmax value for this.
3755 : */
3756 607354 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3757 607354 : oldtup.t_data->t_infomask,
3758 607354 : oldtup.t_data->t_infomask2,
3759 : xid, *lockmode, true,
3760 : &xmax_old_tuple, &infomask_old_tuple,
3761 : &infomask2_old_tuple);
3762 :
3763 : /*
3764 : * And also prepare an Xmax value for the new copy of the tuple. If there
3765 : * was no xmax previously, or there was one but all lockers are now gone,
3766 : * then use InvalidTransactionId; otherwise, get the xmax from the old
3767 : * tuple. (In rare cases that might also be InvalidTransactionId and yet
3768 : * not have the HEAP_XMAX_INVALID bit set; that's fine.)
3769 : */
3770 679152 : if ((oldtup.t_data->t_infomask & HEAP_XMAX_INVALID) ||
3771 143596 : HEAP_LOCKED_UPGRADED(oldtup.t_data->t_infomask) ||
3772 71694 : (checked_lockers && !locker_remains))
3773 535556 : xmax_new_tuple = InvalidTransactionId;
3774 : else
3775 71798 : xmax_new_tuple = HeapTupleHeaderGetRawXmax(oldtup.t_data);
3776 :
3777 607354 : if (!TransactionIdIsValid(xmax_new_tuple))
3778 : {
3779 535556 : infomask_new_tuple = HEAP_XMAX_INVALID;
3780 535556 : infomask2_new_tuple = 0;
3781 : }
3782 : else
3783 : {
3784 : /*
3785 : * If we found a valid Xmax for the new tuple, then the infomask bits
3786 : * to use on the new tuple depend on what was there on the old one.
3787 : * Note that since we're doing an update, the only possibility is that
3788 : * the lockers had FOR KEY SHARE lock.
3789 : */
3790 71798 : if (oldtup.t_data->t_infomask & HEAP_XMAX_IS_MULTI)
3791 : {
3792 106 : GetMultiXactIdHintBits(xmax_new_tuple, &infomask_new_tuple,
3793 : &infomask2_new_tuple);
3794 : }
3795 : else
3796 : {
3797 71692 : infomask_new_tuple = HEAP_XMAX_KEYSHR_LOCK | HEAP_XMAX_LOCK_ONLY;
3798 71692 : infomask2_new_tuple = 0;
3799 : }
3800 : }
3801 :
3802 : /*
3803 : * Prepare the new tuple with the appropriate initial values of Xmin and
3804 : * Xmax, as well as initial infomask bits as computed above.
3805 : */
3806 607354 : newtup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
3807 607354 : newtup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
3808 607354 : HeapTupleHeaderSetXmin(newtup->t_data, xid);
3809 607354 : HeapTupleHeaderSetCmin(newtup->t_data, cid);
3810 607354 : newtup->t_data->t_infomask |= HEAP_UPDATED | infomask_new_tuple;
3811 607354 : newtup->t_data->t_infomask2 |= infomask2_new_tuple;
3812 607354 : HeapTupleHeaderSetXmax(newtup->t_data, xmax_new_tuple);
3813 :
3814 : /*
3815 : * Replace cid with a combo CID if necessary. Note that we already put
3816 : * the plain cid into the new tuple.
3817 : */
3818 607354 : HeapTupleHeaderAdjustCmax(oldtup.t_data, &cid, &iscombo);
3819 :
3820 : /*
3821 : * If the toaster needs to be activated, OR if the new tuple will not fit
3822 : * on the same page as the old, then we need to release the content lock
3823 : * (but not the pin!) on the old tuple's buffer while we are off doing
3824 : * TOAST and/or table-file-extension work. We must mark the old tuple to
3825 : * show that it's locked, else other processes may try to update it
3826 : * themselves.
3827 : *
3828 : * We need to invoke the toaster if there are already any out-of-line
3829 : * toasted values present, or if the new tuple is over-threshold.
3830 : */
3831 607354 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
3832 0 : relation->rd_rel->relkind != RELKIND_MATVIEW)
3833 : {
3834 : /* toast table entries should never be recursively toasted */
3835 : Assert(!HeapTupleHasExternal(&oldtup));
3836 : Assert(!HeapTupleHasExternal(newtup));
3837 0 : need_toast = false;
3838 : }
3839 : else
3840 1821338 : need_toast = (HeapTupleHasExternal(&oldtup) ||
3841 1213984 : HeapTupleHasExternal(newtup) ||
3842 606582 : newtup->t_len > TOAST_TUPLE_THRESHOLD);
3843 :
3844 607354 : pagefree = PageGetHeapFreeSpace(page);
3845 :
3846 607354 : newtupsize = MAXALIGN(newtup->t_len);
3847 :
3848 607354 : if (need_toast || newtupsize > pagefree)
3849 298890 : {
3850 : TransactionId xmax_lock_old_tuple;
3851 : uint16 infomask_lock_old_tuple,
3852 : infomask2_lock_old_tuple;
3853 298890 : bool cleared_all_frozen = false;
3854 :
3855 : /*
3856 : * To prevent concurrent sessions from updating the tuple, we have to
3857 : * temporarily mark it locked, while we release the page-level lock.
3858 : *
3859 : * To satisfy the rule that any xid potentially appearing in a buffer
3860 : * written out to disk, we unfortunately have to WAL log this
3861 : * temporary modification. We can reuse xl_heap_lock for this
3862 : * purpose. If we crash/error before following through with the
3863 : * actual update, xmax will be of an aborted transaction, allowing
3864 : * other sessions to proceed.
3865 : */
3866 :
3867 : /*
3868 : * Compute xmax / infomask appropriate for locking the tuple. This has
3869 : * to be done separately from the combo that's going to be used for
3870 : * updating, because the potentially created multixact would otherwise
3871 : * be wrong.
3872 : */
3873 298890 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3874 298890 : oldtup.t_data->t_infomask,
3875 298890 : oldtup.t_data->t_infomask2,
3876 : xid, *lockmode, false,
3877 : &xmax_lock_old_tuple, &infomask_lock_old_tuple,
3878 : &infomask2_lock_old_tuple);
3879 :
3880 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(infomask_lock_old_tuple));
3881 :
3882 298890 : START_CRIT_SECTION();
3883 :
3884 : /* Clear obsolete visibility flags ... */
3885 298890 : oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
3886 298890 : oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
3887 298890 : HeapTupleClearHotUpdated(&oldtup);
3888 : /* ... and store info about transaction updating this tuple */
3889 : Assert(TransactionIdIsValid(xmax_lock_old_tuple));
3890 298890 : HeapTupleHeaderSetXmax(oldtup.t_data, xmax_lock_old_tuple);
3891 298890 : oldtup.t_data->t_infomask |= infomask_lock_old_tuple;
3892 298890 : oldtup.t_data->t_infomask2 |= infomask2_lock_old_tuple;
3893 298890 : HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
3894 :
3895 : /* temporarily make it look not-updated, but locked */
3896 298890 : oldtup.t_data->t_ctid = oldtup.t_self;
3897 :
3898 : /*
3899 : * Clear all-frozen bit on visibility map if needed. We could
3900 : * immediately reset ALL_VISIBLE, but given that the WAL logging
3901 : * overhead would be unchanged, that doesn't seem necessarily
3902 : * worthwhile.
3903 : */
3904 300592 : if (PageIsAllVisible(page) &&
3905 1702 : visibilitymap_clear(relation, block, vmbuffer,
3906 : VISIBILITYMAP_ALL_FROZEN))
3907 1368 : cleared_all_frozen = true;
3908 :
3909 298890 : MarkBufferDirty(buffer);
3910 :
3911 298890 : if (RelationNeedsWAL(relation))
3912 : {
3913 : xl_heap_lock xlrec;
3914 : XLogRecPtr recptr;
3915 :
3916 278630 : XLogBeginInsert();
3917 278630 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
3918 :
3919 278630 : xlrec.offnum = ItemPointerGetOffsetNumber(&oldtup.t_self);
3920 278630 : xlrec.xmax = xmax_lock_old_tuple;
3921 557260 : xlrec.infobits_set = compute_infobits(oldtup.t_data->t_infomask,
3922 278630 : oldtup.t_data->t_infomask2);
3923 278630 : xlrec.flags =
3924 278630 : cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
3925 278630 : XLogRegisterData(&xlrec, SizeOfHeapLock);
3926 278630 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
3927 278630 : PageSetLSN(page, recptr);
3928 : }
3929 :
3930 298890 : END_CRIT_SECTION();
3931 :
3932 298890 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3933 :
3934 : /*
3935 : * Let the toaster do its thing, if needed.
3936 : *
3937 : * Note: below this point, heaptup is the data we actually intend to
3938 : * store into the relation; newtup is the caller's original untoasted
3939 : * data.
3940 : */
3941 298890 : if (need_toast)
3942 : {
3943 : /* Note we always use WAL and FSM during updates */
3944 3220 : heaptup = heap_toast_insert_or_update(relation, newtup, &oldtup, 0);
3945 3220 : newtupsize = MAXALIGN(heaptup->t_len);
3946 : }
3947 : else
3948 295670 : heaptup = newtup;
3949 :
3950 : /*
3951 : * Now, do we need a new page for the tuple, or not? This is a bit
3952 : * tricky since someone else could have added tuples to the page while
3953 : * we weren't looking. We have to recheck the available space after
3954 : * reacquiring the buffer lock. But don't bother to do that if the
3955 : * former amount of free space is still not enough; it's unlikely
3956 : * there's more free now than before.
3957 : *
3958 : * What's more, if we need to get a new page, we will need to acquire
3959 : * buffer locks on both old and new pages. To avoid deadlock against
3960 : * some other backend trying to get the same two locks in the other
3961 : * order, we must be consistent about the order we get the locks in.
3962 : * We use the rule "lock the lower-numbered page of the relation
3963 : * first". To implement this, we must do RelationGetBufferForTuple
3964 : * while not holding the lock on the old page, and we must rely on it
3965 : * to get the locks on both pages in the correct order.
3966 : *
3967 : * Another consideration is that we need visibility map page pin(s) if
3968 : * we will have to clear the all-visible flag on either page. If we
3969 : * call RelationGetBufferForTuple, we rely on it to acquire any such
3970 : * pins; but if we don't, we have to handle that here. Hence we need
3971 : * a loop.
3972 : */
3973 : for (;;)
3974 : {
3975 298890 : if (newtupsize > pagefree)
3976 : {
3977 : /* It doesn't fit, must use RelationGetBufferForTuple. */
3978 297792 : newbuf = RelationGetBufferForTuple(relation, heaptup->t_len,
3979 : buffer, 0, NULL,
3980 : &vmbuffer_new, &vmbuffer,
3981 : 0);
3982 : /* We're all done. */
3983 297792 : break;
3984 : }
3985 : /* Acquire VM page pin if needed and we don't have it. */
3986 1098 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
3987 0 : visibilitymap_pin(relation, block, &vmbuffer);
3988 : /* Re-acquire the lock on the old tuple's page. */
3989 1098 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3990 : /* Re-check using the up-to-date free space */
3991 1098 : pagefree = PageGetHeapFreeSpace(page);
3992 1098 : if (newtupsize > pagefree ||
3993 1098 : (vmbuffer == InvalidBuffer && PageIsAllVisible(page)))
3994 : {
3995 : /*
3996 : * Rats, it doesn't fit anymore, or somebody just now set the
3997 : * all-visible flag. We must now unlock and loop to avoid
3998 : * deadlock. Fortunately, this path should seldom be taken.
3999 : */
4000 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
4001 : }
4002 : else
4003 : {
4004 : /* We're all done. */
4005 1098 : newbuf = buffer;
4006 1098 : break;
4007 : }
4008 : }
4009 : }
4010 : else
4011 : {
4012 : /* No TOAST work needed, and it'll fit on same page */
4013 308464 : newbuf = buffer;
4014 308464 : heaptup = newtup;
4015 : }
4016 :
4017 : /*
4018 : * We're about to do the actual update -- check for conflict first, to
4019 : * avoid possibly having to roll back work we've just done.
4020 : *
4021 : * This is safe without a recheck as long as there is no possibility of
4022 : * another process scanning the pages between this check and the update
4023 : * being visible to the scan (i.e., exclusive buffer content lock(s) are
4024 : * continuously held from this point until the tuple update is visible).
4025 : *
4026 : * For the new tuple the only check needed is at the relation level, but
4027 : * since both tuples are in the same relation and the check for oldtup
4028 : * will include checking the relation level, there is no benefit to a
4029 : * separate check for the new tuple.
4030 : */
4031 607354 : CheckForSerializableConflictIn(relation, &oldtup.t_self,
4032 : BufferGetBlockNumber(buffer));
4033 :
4034 : /*
4035 : * At this point newbuf and buffer are both pinned and locked, and newbuf
4036 : * has enough space for the new tuple. If they are the same buffer, only
4037 : * one pin is held.
4038 : */
4039 :
4040 607330 : if (newbuf == buffer)
4041 : {
4042 : /*
4043 : * Since the new tuple is going into the same page, we might be able
4044 : * to do a HOT update. Check if any of the index columns have been
4045 : * changed.
4046 : */
4047 309538 : if (!bms_overlap(modified_attrs, hot_attrs))
4048 : {
4049 285154 : use_hot_update = true;
4050 :
4051 : /*
4052 : * If none of the columns that are used in hot-blocking indexes
4053 : * were updated, we can apply HOT, but we do still need to check
4054 : * if we need to update the summarizing indexes, and update those
4055 : * indexes if the columns were updated, or we may fail to detect
4056 : * e.g. value bound changes in BRIN minmax indexes.
4057 : */
4058 285154 : if (bms_overlap(modified_attrs, sum_attrs))
4059 3282 : summarized_update = true;
4060 : }
4061 : }
4062 : else
4063 : {
4064 : /* Set a hint that the old page could use prune/defrag */
4065 297792 : PageSetFull(page);
4066 : }
4067 :
4068 : /*
4069 : * Compute replica identity tuple before entering the critical section so
4070 : * we don't PANIC upon a memory allocation failure.
4071 : * ExtractReplicaIdentity() will return NULL if nothing needs to be
4072 : * logged. Pass old key required as true only if the replica identity key
4073 : * columns are modified or it has external data.
4074 : */
4075 607330 : old_key_tuple = ExtractReplicaIdentity(relation, &oldtup,
4076 607330 : bms_overlap(modified_attrs, id_attrs) ||
4077 : id_has_external,
4078 : &old_key_copied);
4079 :
4080 : /* NO EREPORT(ERROR) from here till changes are logged */
4081 607330 : START_CRIT_SECTION();
4082 :
4083 : /*
4084 : * If this transaction commits, the old tuple will become DEAD sooner or
4085 : * later. Set flag that this page is a candidate for pruning once our xid
4086 : * falls below the OldestXmin horizon. If the transaction finally aborts,
4087 : * the subsequent page pruning will be a no-op and the hint will be
4088 : * cleared.
4089 : *
4090 : * XXX Should we set hint on newbuf as well? If the transaction aborts,
4091 : * there would be a prunable tuple in the newbuf; but for now we choose
4092 : * not to optimize for aborts. Note that heap_xlog_update must be kept in
4093 : * sync if this decision changes.
4094 : */
4095 607330 : PageSetPrunable(page, xid);
4096 :
4097 607330 : if (use_hot_update)
4098 : {
4099 : /* Mark the old tuple as HOT-updated */
4100 285154 : HeapTupleSetHotUpdated(&oldtup);
4101 : /* And mark the new tuple as heap-only */
4102 285154 : HeapTupleSetHeapOnly(heaptup);
4103 : /* Mark the caller's copy too, in case different from heaptup */
4104 285154 : HeapTupleSetHeapOnly(newtup);
4105 : }
4106 : else
4107 : {
4108 : /* Make sure tuples are correctly marked as not-HOT */
4109 322176 : HeapTupleClearHotUpdated(&oldtup);
4110 322176 : HeapTupleClearHeapOnly(heaptup);
4111 322176 : HeapTupleClearHeapOnly(newtup);
4112 : }
4113 :
4114 607330 : RelationPutHeapTuple(relation, newbuf, heaptup, false); /* insert new tuple */
4115 :
4116 :
4117 : /* Clear obsolete visibility flags, possibly set by ourselves above... */
4118 607330 : oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
4119 607330 : oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
4120 : /* ... and store info about transaction updating this tuple */
4121 : Assert(TransactionIdIsValid(xmax_old_tuple));
4122 607330 : HeapTupleHeaderSetXmax(oldtup.t_data, xmax_old_tuple);
4123 607330 : oldtup.t_data->t_infomask |= infomask_old_tuple;
4124 607330 : oldtup.t_data->t_infomask2 |= infomask2_old_tuple;
4125 607330 : HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
4126 :
4127 : /* record address of new tuple in t_ctid of old one */
4128 607330 : oldtup.t_data->t_ctid = heaptup->t_self;
4129 :
4130 : /* clear PD_ALL_VISIBLE flags, reset all visibilitymap bits */
4131 607330 : if (PageIsAllVisible(BufferGetPage(buffer)))
4132 : {
4133 2842 : all_visible_cleared = true;
4134 2842 : PageClearAllVisible(BufferGetPage(buffer));
4135 2842 : visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
4136 : vmbuffer, VISIBILITYMAP_VALID_BITS);
4137 : }
4138 607330 : if (newbuf != buffer && PageIsAllVisible(BufferGetPage(newbuf)))
4139 : {
4140 1660 : all_visible_cleared_new = true;
4141 1660 : PageClearAllVisible(BufferGetPage(newbuf));
4142 1660 : visibilitymap_clear(relation, BufferGetBlockNumber(newbuf),
4143 : vmbuffer_new, VISIBILITYMAP_VALID_BITS);
4144 : }
4145 :
4146 607330 : if (newbuf != buffer)
4147 297792 : MarkBufferDirty(newbuf);
4148 607330 : MarkBufferDirty(buffer);
4149 :
4150 : /* XLOG stuff */
4151 607330 : if (RelationNeedsWAL(relation))
4152 : {
4153 : XLogRecPtr recptr;
4154 :
4155 : /*
4156 : * For logical decoding we need combo CIDs to properly decode the
4157 : * catalog.
4158 : */
4159 584584 : if (RelationIsAccessibleInLogicalDecoding(relation))
4160 : {
4161 5114 : log_heap_new_cid(relation, &oldtup);
4162 5114 : log_heap_new_cid(relation, heaptup);
4163 : }
4164 :
4165 584584 : recptr = log_heap_update(relation, buffer,
4166 : newbuf, &oldtup, heaptup,
4167 : old_key_tuple,
4168 : all_visible_cleared,
4169 : all_visible_cleared_new);
4170 584584 : if (newbuf != buffer)
4171 : {
4172 277544 : PageSetLSN(BufferGetPage(newbuf), recptr);
4173 : }
4174 584584 : PageSetLSN(BufferGetPage(buffer), recptr);
4175 : }
4176 :
4177 607330 : END_CRIT_SECTION();
4178 :
4179 607330 : if (newbuf != buffer)
4180 297792 : LockBuffer(newbuf, BUFFER_LOCK_UNLOCK);
4181 607330 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
4182 :
4183 : /*
4184 : * Mark old tuple for invalidation from system caches at next command
4185 : * boundary, and mark the new tuple for invalidation in case we abort. We
4186 : * have to do this before releasing the buffer because oldtup is in the
4187 : * buffer. (heaptup is all in local memory, but it's necessary to process
4188 : * both tuple versions in one call to inval.c so we can avoid redundant
4189 : * sinval messages.)
4190 : */
4191 607330 : CacheInvalidateHeapTuple(relation, &oldtup, heaptup);
4192 :
4193 : /* Now we can release the buffer(s) */
4194 607330 : if (newbuf != buffer)
4195 297792 : ReleaseBuffer(newbuf);
4196 607330 : ReleaseBuffer(buffer);
4197 607330 : if (BufferIsValid(vmbuffer_new))
4198 1660 : ReleaseBuffer(vmbuffer_new);
4199 607330 : if (BufferIsValid(vmbuffer))
4200 2842 : ReleaseBuffer(vmbuffer);
4201 :
4202 : /*
4203 : * Release the lmgr tuple lock, if we had it.
4204 : */
4205 607330 : if (have_tuple_lock)
4206 44 : UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
4207 :
4208 607330 : pgstat_count_heap_update(relation, use_hot_update, newbuf != buffer);
4209 :
4210 : /*
4211 : * If heaptup is a private copy, release it. Don't forget to copy t_self
4212 : * back to the caller's image, too.
4213 : */
4214 607330 : if (heaptup != newtup)
4215 : {
4216 3124 : newtup->t_self = heaptup->t_self;
4217 3124 : heap_freetuple(heaptup);
4218 : }
4219 :
4220 : /*
4221 : * If it is a HOT update, the update may still need to update summarized
4222 : * indexes, lest we fail to update those summaries and get incorrect
4223 : * results (for example, minmax bounds of the block may change with this
4224 : * update).
4225 : */
4226 607330 : if (use_hot_update)
4227 : {
4228 285154 : if (summarized_update)
4229 3282 : *update_indexes = TU_Summarizing;
4230 : else
4231 281872 : *update_indexes = TU_None;
4232 : }
4233 : else
4234 322176 : *update_indexes = TU_All;
4235 :
4236 607330 : if (old_key_tuple != NULL && old_key_copied)
4237 168 : heap_freetuple(old_key_tuple);
4238 :
4239 607330 : bms_free(hot_attrs);
4240 607330 : bms_free(sum_attrs);
4241 607330 : bms_free(key_attrs);
4242 607330 : bms_free(id_attrs);
4243 607330 : bms_free(modified_attrs);
4244 607330 : bms_free(interesting_attrs);
4245 :
4246 607330 : return TM_Ok;
4247 : }
4248 :
4249 : #ifdef USE_ASSERT_CHECKING
4250 : /*
4251 : * Confirm adequate lock held during heap_update(), per rules from
4252 : * README.tuplock section "Locking to write inplace-updated tables".
4253 : */
4254 : static void
4255 : check_lock_if_inplace_updateable_rel(Relation relation,
4256 : const ItemPointerData *otid,
4257 : HeapTuple newtup)
4258 : {
4259 : /* LOCKTAG_TUPLE acceptable for any catalog */
4260 : switch (RelationGetRelid(relation))
4261 : {
4262 : case RelationRelationId:
4263 : case DatabaseRelationId:
4264 : {
4265 : LOCKTAG tuptag;
4266 :
4267 : SET_LOCKTAG_TUPLE(tuptag,
4268 : relation->rd_lockInfo.lockRelId.dbId,
4269 : relation->rd_lockInfo.lockRelId.relId,
4270 : ItemPointerGetBlockNumber(otid),
4271 : ItemPointerGetOffsetNumber(otid));
4272 : if (LockHeldByMe(&tuptag, InplaceUpdateTupleLock, false))
4273 : return;
4274 : }
4275 : break;
4276 : default:
4277 : Assert(!IsInplaceUpdateRelation(relation));
4278 : return;
4279 : }
4280 :
4281 : switch (RelationGetRelid(relation))
4282 : {
4283 : case RelationRelationId:
4284 : {
4285 : /* LOCKTAG_TUPLE or LOCKTAG_RELATION ok */
4286 : Form_pg_class classForm = (Form_pg_class) GETSTRUCT(newtup);
4287 : Oid relid = classForm->oid;
4288 : Oid dbid;
4289 : LOCKTAG tag;
4290 :
4291 : if (IsSharedRelation(relid))
4292 : dbid = InvalidOid;
4293 : else
4294 : dbid = MyDatabaseId;
4295 :
4296 : if (classForm->relkind == RELKIND_INDEX)
4297 : {
4298 : Relation irel = index_open(relid, AccessShareLock);
4299 :
4300 : SET_LOCKTAG_RELATION(tag, dbid, irel->rd_index->indrelid);
4301 : index_close(irel, AccessShareLock);
4302 : }
4303 : else
4304 : SET_LOCKTAG_RELATION(tag, dbid, relid);
4305 :
4306 : if (!LockHeldByMe(&tag, ShareUpdateExclusiveLock, false) &&
4307 : !LockHeldByMe(&tag, ShareRowExclusiveLock, true))
4308 : elog(WARNING,
4309 : "missing lock for relation \"%s\" (OID %u, relkind %c) @ TID (%u,%u)",
4310 : NameStr(classForm->relname),
4311 : relid,
4312 : classForm->relkind,
4313 : ItemPointerGetBlockNumber(otid),
4314 : ItemPointerGetOffsetNumber(otid));
4315 : }
4316 : break;
4317 : case DatabaseRelationId:
4318 : {
4319 : /* LOCKTAG_TUPLE required */
4320 : Form_pg_database dbForm = (Form_pg_database) GETSTRUCT(newtup);
4321 :
4322 : elog(WARNING,
4323 : "missing lock on database \"%s\" (OID %u) @ TID (%u,%u)",
4324 : NameStr(dbForm->datname),
4325 : dbForm->oid,
4326 : ItemPointerGetBlockNumber(otid),
4327 : ItemPointerGetOffsetNumber(otid));
4328 : }
4329 : break;
4330 : }
4331 : }
4332 :
4333 : /*
4334 : * Confirm adequate relation lock held, per rules from README.tuplock section
4335 : * "Locking to write inplace-updated tables".
4336 : */
4337 : static void
4338 : check_inplace_rel_lock(HeapTuple oldtup)
4339 : {
4340 : Form_pg_class classForm = (Form_pg_class) GETSTRUCT(oldtup);
4341 : Oid relid = classForm->oid;
4342 : Oid dbid;
4343 : LOCKTAG tag;
4344 :
4345 : if (IsSharedRelation(relid))
4346 : dbid = InvalidOid;
4347 : else
4348 : dbid = MyDatabaseId;
4349 :
4350 : if (classForm->relkind == RELKIND_INDEX)
4351 : {
4352 : Relation irel = index_open(relid, AccessShareLock);
4353 :
4354 : SET_LOCKTAG_RELATION(tag, dbid, irel->rd_index->indrelid);
4355 : index_close(irel, AccessShareLock);
4356 : }
4357 : else
4358 : SET_LOCKTAG_RELATION(tag, dbid, relid);
4359 :
4360 : if (!LockHeldByMe(&tag, ShareUpdateExclusiveLock, true))
4361 : elog(WARNING,
4362 : "missing lock for relation \"%s\" (OID %u, relkind %c) @ TID (%u,%u)",
4363 : NameStr(classForm->relname),
4364 : relid,
4365 : classForm->relkind,
4366 : ItemPointerGetBlockNumber(&oldtup->t_self),
4367 : ItemPointerGetOffsetNumber(&oldtup->t_self));
4368 : }
4369 : #endif
4370 :
4371 : /*
4372 : * Check if the specified attribute's values are the same. Subroutine for
4373 : * HeapDetermineColumnsInfo.
4374 : */
4375 : static bool
4376 1465070 : heap_attr_equals(TupleDesc tupdesc, int attrnum, Datum value1, Datum value2,
4377 : bool isnull1, bool isnull2)
4378 : {
4379 : /*
4380 : * If one value is NULL and other is not, then they are certainly not
4381 : * equal
4382 : */
4383 1465070 : if (isnull1 != isnull2)
4384 90 : return false;
4385 :
4386 : /*
4387 : * If both are NULL, they can be considered equal.
4388 : */
4389 1464980 : if (isnull1)
4390 9982 : return true;
4391 :
4392 : /*
4393 : * We do simple binary comparison of the two datums. This may be overly
4394 : * strict because there can be multiple binary representations for the
4395 : * same logical value. But we should be OK as long as there are no false
4396 : * positives. Using a type-specific equality operator is messy because
4397 : * there could be multiple notions of equality in different operator
4398 : * classes; furthermore, we cannot safely invoke user-defined functions
4399 : * while holding exclusive buffer lock.
4400 : */
4401 1454998 : if (attrnum <= 0)
4402 : {
4403 : /* The only allowed system columns are OIDs, so do this */
4404 0 : return (DatumGetObjectId(value1) == DatumGetObjectId(value2));
4405 : }
4406 : else
4407 : {
4408 : CompactAttribute *att;
4409 :
4410 : Assert(attrnum <= tupdesc->natts);
4411 1454998 : att = TupleDescCompactAttr(tupdesc, attrnum - 1);
4412 1454998 : return datumIsEqual(value1, value2, att->attbyval, att->attlen);
4413 : }
4414 : }
4415 :
4416 : /*
4417 : * Check which columns are being updated.
4418 : *
4419 : * Given an updated tuple, determine (and return into the output bitmapset),
4420 : * from those listed as interesting, the set of columns that changed.
4421 : *
4422 : * has_external indicates if any of the unmodified attributes (from those
4423 : * listed as interesting) of the old tuple is a member of external_cols and is
4424 : * stored externally.
4425 : */
4426 : static Bitmapset *
4427 607674 : HeapDetermineColumnsInfo(Relation relation,
4428 : Bitmapset *interesting_cols,
4429 : Bitmapset *external_cols,
4430 : HeapTuple oldtup, HeapTuple newtup,
4431 : bool *has_external)
4432 : {
4433 : int attidx;
4434 607674 : Bitmapset *modified = NULL;
4435 607674 : TupleDesc tupdesc = RelationGetDescr(relation);
4436 :
4437 607674 : attidx = -1;
4438 2072744 : while ((attidx = bms_next_member(interesting_cols, attidx)) >= 0)
4439 : {
4440 : /* attidx is zero-based, attrnum is the normal attribute number */
4441 1465070 : AttrNumber attrnum = attidx + FirstLowInvalidHeapAttributeNumber;
4442 : Datum value1,
4443 : value2;
4444 : bool isnull1,
4445 : isnull2;
4446 :
4447 : /*
4448 : * If it's a whole-tuple reference, say "not equal". It's not really
4449 : * worth supporting this case, since it could only succeed after a
4450 : * no-op update, which is hardly a case worth optimizing for.
4451 : */
4452 1465070 : if (attrnum == 0)
4453 : {
4454 0 : modified = bms_add_member(modified, attidx);
4455 1400714 : continue;
4456 : }
4457 :
4458 : /*
4459 : * Likewise, automatically say "not equal" for any system attribute
4460 : * other than tableOID; we cannot expect these to be consistent in a
4461 : * HOT chain, or even to be set correctly yet in the new tuple.
4462 : */
4463 1465070 : if (attrnum < 0)
4464 : {
4465 0 : if (attrnum != TableOidAttributeNumber)
4466 : {
4467 0 : modified = bms_add_member(modified, attidx);
4468 0 : continue;
4469 : }
4470 : }
4471 :
4472 : /*
4473 : * Extract the corresponding values. XXX this is pretty inefficient
4474 : * if there are many indexed columns. Should we do a single
4475 : * heap_deform_tuple call on each tuple, instead? But that doesn't
4476 : * work for system columns ...
4477 : */
4478 1465070 : value1 = heap_getattr(oldtup, attrnum, tupdesc, &isnull1);
4479 1465070 : value2 = heap_getattr(newtup, attrnum, tupdesc, &isnull2);
4480 :
4481 1465070 : if (!heap_attr_equals(tupdesc, attrnum, value1,
4482 : value2, isnull1, isnull2))
4483 : {
4484 53952 : modified = bms_add_member(modified, attidx);
4485 53952 : continue;
4486 : }
4487 :
4488 : /*
4489 : * No need to check attributes that can't be stored externally. Note
4490 : * that system attributes can't be stored externally.
4491 : */
4492 1411118 : if (attrnum < 0 || isnull1 ||
4493 1401136 : TupleDescCompactAttr(tupdesc, attrnum - 1)->attlen != -1)
4494 1346762 : continue;
4495 :
4496 : /*
4497 : * Check if the old tuple's attribute is stored externally and is a
4498 : * member of external_cols.
4499 : */
4500 64366 : if (VARATT_IS_EXTERNAL((struct varlena *) DatumGetPointer(value1)) &&
4501 10 : bms_is_member(attidx, external_cols))
4502 4 : *has_external = true;
4503 : }
4504 :
4505 607674 : return modified;
4506 : }
4507 :
4508 : /*
4509 : * simple_heap_update - replace a tuple
4510 : *
4511 : * This routine may be used to update a tuple when concurrent updates of
4512 : * the target tuple are not expected (for example, because we have a lock
4513 : * on the relation associated with the tuple). Any failure is reported
4514 : * via ereport().
4515 : */
4516 : void
4517 220294 : simple_heap_update(Relation relation, const ItemPointerData *otid, HeapTuple tup,
4518 : TU_UpdateIndexes *update_indexes)
4519 : {
4520 : TM_Result result;
4521 : TM_FailureData tmfd;
4522 : LockTupleMode lockmode;
4523 :
4524 220294 : result = heap_update(relation, otid, tup,
4525 : GetCurrentCommandId(true), InvalidSnapshot,
4526 : true /* wait for commit */ ,
4527 : &tmfd, &lockmode, update_indexes);
4528 220294 : switch (result)
4529 : {
4530 0 : case TM_SelfModified:
4531 : /* Tuple was already updated in current command? */
4532 0 : elog(ERROR, "tuple already updated by self");
4533 : break;
4534 :
4535 220292 : case TM_Ok:
4536 : /* done successfully */
4537 220292 : break;
4538 :
4539 0 : case TM_Updated:
4540 0 : elog(ERROR, "tuple concurrently updated");
4541 : break;
4542 :
4543 2 : case TM_Deleted:
4544 2 : elog(ERROR, "tuple concurrently deleted");
4545 : break;
4546 :
4547 0 : default:
4548 0 : elog(ERROR, "unrecognized heap_update status: %u", result);
4549 : break;
4550 : }
4551 220292 : }
4552 :
4553 :
4554 : /*
4555 : * Return the MultiXactStatus corresponding to the given tuple lock mode.
4556 : */
4557 : static MultiXactStatus
4558 2410 : get_mxact_status_for_lock(LockTupleMode mode, bool is_update)
4559 : {
4560 : int retval;
4561 :
4562 2410 : if (is_update)
4563 192 : retval = tupleLockExtraInfo[mode].updstatus;
4564 : else
4565 2218 : retval = tupleLockExtraInfo[mode].lockstatus;
4566 :
4567 2410 : if (retval == -1)
4568 0 : elog(ERROR, "invalid lock tuple mode %d/%s", mode,
4569 : is_update ? "true" : "false");
4570 :
4571 2410 : return (MultiXactStatus) retval;
4572 : }
4573 :
4574 : /*
4575 : * heap_lock_tuple - lock a tuple in shared or exclusive mode
4576 : *
4577 : * Note that this acquires a buffer pin, which the caller must release.
4578 : *
4579 : * Input parameters:
4580 : * relation: relation containing tuple (caller must hold suitable lock)
4581 : * cid: current command ID (used for visibility test, and stored into
4582 : * tuple's cmax if lock is successful)
4583 : * mode: indicates if shared or exclusive tuple lock is desired
4584 : * wait_policy: what to do if tuple lock is not available
4585 : * follow_updates: if true, follow the update chain to also lock descendant
4586 : * tuples.
4587 : *
4588 : * Output parameters:
4589 : * *tuple: all fields filled in
4590 : * *buffer: set to buffer holding tuple (pinned but not locked at exit)
4591 : * *tmfd: filled in failure cases (see below)
4592 : *
4593 : * Function results are the same as the ones for table_tuple_lock().
4594 : *
4595 : * In the failure cases other than TM_Invisible, the routine fills
4596 : * *tmfd with the tuple's t_ctid, t_xmax (resolving a possible MultiXact,
4597 : * if necessary), and t_cmax (the last only for TM_SelfModified,
4598 : * since we cannot obtain cmax from a combo CID generated by another
4599 : * transaction).
4600 : * See comments for struct TM_FailureData for additional info.
4601 : *
4602 : * See README.tuplock for a thorough explanation of this mechanism.
4603 : */
4604 : TM_Result
4605 170130 : heap_lock_tuple(Relation relation, HeapTuple tuple,
4606 : CommandId cid, LockTupleMode mode, LockWaitPolicy wait_policy,
4607 : bool follow_updates,
4608 : Buffer *buffer, TM_FailureData *tmfd)
4609 : {
4610 : TM_Result result;
4611 170130 : ItemPointer tid = &(tuple->t_self);
4612 : ItemId lp;
4613 : Page page;
4614 170130 : Buffer vmbuffer = InvalidBuffer;
4615 : BlockNumber block;
4616 : TransactionId xid,
4617 : xmax;
4618 : uint16 old_infomask,
4619 : new_infomask,
4620 : new_infomask2;
4621 170130 : bool first_time = true;
4622 170130 : bool skip_tuple_lock = false;
4623 170130 : bool have_tuple_lock = false;
4624 170130 : bool cleared_all_frozen = false;
4625 :
4626 170130 : *buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
4627 170130 : block = ItemPointerGetBlockNumber(tid);
4628 :
4629 : /*
4630 : * Before locking the buffer, pin the visibility map page if it appears to
4631 : * be necessary. Since we haven't got the lock yet, someone else might be
4632 : * in the middle of changing this, so we'll need to recheck after we have
4633 : * the lock.
4634 : */
4635 170130 : if (PageIsAllVisible(BufferGetPage(*buffer)))
4636 3332 : visibilitymap_pin(relation, block, &vmbuffer);
4637 :
4638 170130 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4639 :
4640 170130 : page = BufferGetPage(*buffer);
4641 170130 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
4642 : Assert(ItemIdIsNormal(lp));
4643 :
4644 170130 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
4645 170130 : tuple->t_len = ItemIdGetLength(lp);
4646 170130 : tuple->t_tableOid = RelationGetRelid(relation);
4647 :
4648 28 : l3:
4649 170158 : result = HeapTupleSatisfiesUpdate(tuple, cid, *buffer);
4650 :
4651 170158 : if (result == TM_Invisible)
4652 : {
4653 : /*
4654 : * This is possible, but only when locking a tuple for ON CONFLICT
4655 : * UPDATE. We return this value here rather than throwing an error in
4656 : * order to give that case the opportunity to throw a more specific
4657 : * error.
4658 : */
4659 24 : result = TM_Invisible;
4660 24 : goto out_locked;
4661 : }
4662 170134 : else if (result == TM_BeingModified ||
4663 154296 : result == TM_Updated ||
4664 : result == TM_Deleted)
4665 : {
4666 : TransactionId xwait;
4667 : uint16 infomask;
4668 : uint16 infomask2;
4669 : bool require_sleep;
4670 : ItemPointerData t_ctid;
4671 :
4672 : /* must copy state data before unlocking buffer */
4673 15840 : xwait = HeapTupleHeaderGetRawXmax(tuple->t_data);
4674 15840 : infomask = tuple->t_data->t_infomask;
4675 15840 : infomask2 = tuple->t_data->t_infomask2;
4676 15840 : ItemPointerCopy(&tuple->t_data->t_ctid, &t_ctid);
4677 :
4678 15840 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
4679 :
4680 : /*
4681 : * If any subtransaction of the current top transaction already holds
4682 : * a lock as strong as or stronger than what we're requesting, we
4683 : * effectively hold the desired lock already. We *must* succeed
4684 : * without trying to take the tuple lock, else we will deadlock
4685 : * against anyone wanting to acquire a stronger lock.
4686 : *
4687 : * Note we only do this the first time we loop on the HTSU result;
4688 : * there is no point in testing in subsequent passes, because
4689 : * evidently our own transaction cannot have acquired a new lock after
4690 : * the first time we checked.
4691 : */
4692 15840 : if (first_time)
4693 : {
4694 15822 : first_time = false;
4695 :
4696 15822 : if (infomask & HEAP_XMAX_IS_MULTI)
4697 : {
4698 : int i;
4699 : int nmembers;
4700 : MultiXactMember *members;
4701 :
4702 : /*
4703 : * We don't need to allow old multixacts here; if that had
4704 : * been the case, HeapTupleSatisfiesUpdate would have returned
4705 : * MayBeUpdated and we wouldn't be here.
4706 : */
4707 : nmembers =
4708 160 : GetMultiXactIdMembers(xwait, &members, false,
4709 160 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
4710 :
4711 474 : for (i = 0; i < nmembers; i++)
4712 : {
4713 : /* only consider members of our own transaction */
4714 342 : if (!TransactionIdIsCurrentTransactionId(members[i].xid))
4715 244 : continue;
4716 :
4717 98 : if (TUPLOCK_from_mxstatus(members[i].status) >= mode)
4718 : {
4719 28 : pfree(members);
4720 28 : result = TM_Ok;
4721 28 : goto out_unlocked;
4722 : }
4723 : else
4724 : {
4725 : /*
4726 : * Disable acquisition of the heavyweight tuple lock.
4727 : * Otherwise, when promoting a weaker lock, we might
4728 : * deadlock with another locker that has acquired the
4729 : * heavyweight tuple lock and is waiting for our
4730 : * transaction to finish.
4731 : *
4732 : * Note that in this case we still need to wait for
4733 : * the multixact if required, to avoid acquiring
4734 : * conflicting locks.
4735 : */
4736 70 : skip_tuple_lock = true;
4737 : }
4738 : }
4739 :
4740 132 : if (members)
4741 132 : pfree(members);
4742 : }
4743 15662 : else if (TransactionIdIsCurrentTransactionId(xwait))
4744 : {
4745 13134 : switch (mode)
4746 : {
4747 332 : case LockTupleKeyShare:
4748 : Assert(HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) ||
4749 : HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4750 : HEAP_XMAX_IS_EXCL_LOCKED(infomask));
4751 332 : result = TM_Ok;
4752 332 : goto out_unlocked;
4753 48 : case LockTupleShare:
4754 60 : if (HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4755 12 : HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4756 : {
4757 36 : result = TM_Ok;
4758 36 : goto out_unlocked;
4759 : }
4760 12 : break;
4761 144 : case LockTupleNoKeyExclusive:
4762 144 : if (HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4763 : {
4764 120 : result = TM_Ok;
4765 120 : goto out_unlocked;
4766 : }
4767 24 : break;
4768 12610 : case LockTupleExclusive:
4769 12610 : if (HEAP_XMAX_IS_EXCL_LOCKED(infomask) &&
4770 2530 : infomask2 & HEAP_KEYS_UPDATED)
4771 : {
4772 2488 : result = TM_Ok;
4773 2488 : goto out_unlocked;
4774 : }
4775 10122 : break;
4776 : }
4777 : }
4778 : }
4779 :
4780 : /*
4781 : * Initially assume that we will have to wait for the locking
4782 : * transaction(s) to finish. We check various cases below in which
4783 : * this can be turned off.
4784 : */
4785 12836 : require_sleep = true;
4786 12836 : if (mode == LockTupleKeyShare)
4787 : {
4788 : /*
4789 : * If we're requesting KeyShare, and there's no update present, we
4790 : * don't need to wait. Even if there is an update, we can still
4791 : * continue if the key hasn't been modified.
4792 : *
4793 : * However, if there are updates, we need to walk the update chain
4794 : * to mark future versions of the row as locked, too. That way,
4795 : * if somebody deletes that future version, we're protected
4796 : * against the key going away. This locking of future versions
4797 : * could block momentarily, if a concurrent transaction is
4798 : * deleting a key; or it could return a value to the effect that
4799 : * the transaction deleting the key has already committed. So we
4800 : * do this before re-locking the buffer; otherwise this would be
4801 : * prone to deadlocks.
4802 : *
4803 : * Note that the TID we're locking was grabbed before we unlocked
4804 : * the buffer. For it to change while we're not looking, the
4805 : * other properties we're testing for below after re-locking the
4806 : * buffer would also change, in which case we would restart this
4807 : * loop above.
4808 : */
4809 1190 : if (!(infomask2 & HEAP_KEYS_UPDATED))
4810 : {
4811 : bool updated;
4812 :
4813 1104 : updated = !HEAP_XMAX_IS_LOCKED_ONLY(infomask);
4814 :
4815 : /*
4816 : * If there are updates, follow the update chain; bail out if
4817 : * that cannot be done.
4818 : */
4819 1104 : if (follow_updates && updated)
4820 : {
4821 : TM_Result res;
4822 :
4823 100 : res = heap_lock_updated_tuple(relation, tuple, &t_ctid,
4824 : GetCurrentTransactionId(),
4825 : mode);
4826 100 : if (res != TM_Ok)
4827 : {
4828 12 : result = res;
4829 : /* recovery code expects to have buffer lock held */
4830 12 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4831 388 : goto failed;
4832 : }
4833 : }
4834 :
4835 1092 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4836 :
4837 : /*
4838 : * Make sure it's still an appropriate lock, else start over.
4839 : * Also, if it wasn't updated before we released the lock, but
4840 : * is updated now, we start over too; the reason is that we
4841 : * now need to follow the update chain to lock the new
4842 : * versions.
4843 : */
4844 1092 : if (!HeapTupleHeaderIsOnlyLocked(tuple->t_data) &&
4845 86 : ((tuple->t_data->t_infomask2 & HEAP_KEYS_UPDATED) ||
4846 86 : !updated))
4847 28 : goto l3;
4848 :
4849 : /* Things look okay, so we can skip sleeping */
4850 1092 : require_sleep = false;
4851 :
4852 : /*
4853 : * Note we allow Xmax to change here; other updaters/lockers
4854 : * could have modified it before we grabbed the buffer lock.
4855 : * However, this is not a problem, because with the recheck we
4856 : * just did we ensure that they still don't conflict with the
4857 : * lock we want.
4858 : */
4859 : }
4860 : }
4861 11646 : else if (mode == LockTupleShare)
4862 : {
4863 : /*
4864 : * If we're requesting Share, we can similarly avoid sleeping if
4865 : * there's no update and no exclusive lock present.
4866 : */
4867 886 : if (HEAP_XMAX_IS_LOCKED_ONLY(infomask) &&
4868 886 : !HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4869 : {
4870 874 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4871 :
4872 : /*
4873 : * Make sure it's still an appropriate lock, else start over.
4874 : * See above about allowing xmax to change.
4875 : */
4876 1748 : if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
4877 874 : HEAP_XMAX_IS_EXCL_LOCKED(tuple->t_data->t_infomask))
4878 0 : goto l3;
4879 874 : require_sleep = false;
4880 : }
4881 : }
4882 10760 : else if (mode == LockTupleNoKeyExclusive)
4883 : {
4884 : /*
4885 : * If we're requesting NoKeyExclusive, we might also be able to
4886 : * avoid sleeping; just ensure that there no conflicting lock
4887 : * already acquired.
4888 : */
4889 328 : if (infomask & HEAP_XMAX_IS_MULTI)
4890 : {
4891 52 : if (!DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
4892 : mode, NULL))
4893 : {
4894 : /*
4895 : * No conflict, but if the xmax changed under us in the
4896 : * meantime, start over.
4897 : */
4898 26 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4899 52 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4900 26 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4901 : xwait))
4902 0 : goto l3;
4903 :
4904 : /* otherwise, we're good */
4905 26 : require_sleep = false;
4906 : }
4907 : }
4908 276 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask))
4909 : {
4910 36 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4911 :
4912 : /* if the xmax changed in the meantime, start over */
4913 72 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4914 36 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4915 : xwait))
4916 0 : goto l3;
4917 : /* otherwise, we're good */
4918 36 : require_sleep = false;
4919 : }
4920 : }
4921 :
4922 : /*
4923 : * As a check independent from those above, we can also avoid sleeping
4924 : * if the current transaction is the sole locker of the tuple. Note
4925 : * that the strength of the lock already held is irrelevant; this is
4926 : * not about recording the lock in Xmax (which will be done regardless
4927 : * of this optimization, below). Also, note that the cases where we
4928 : * hold a lock stronger than we are requesting are already handled
4929 : * above by not doing anything.
4930 : *
4931 : * Note we only deal with the non-multixact case here; MultiXactIdWait
4932 : * is well equipped to deal with this situation on its own.
4933 : */
4934 23538 : if (require_sleep && !(infomask & HEAP_XMAX_IS_MULTI) &&
4935 10714 : TransactionIdIsCurrentTransactionId(xwait))
4936 : {
4937 : /* ... but if the xmax changed in the meantime, start over */
4938 10122 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4939 20244 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4940 10122 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4941 : xwait))
4942 0 : goto l3;
4943 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask));
4944 10122 : require_sleep = false;
4945 : }
4946 :
4947 : /*
4948 : * Time to sleep on the other transaction/multixact, if necessary.
4949 : *
4950 : * If the other transaction is an update/delete that's already
4951 : * committed, then sleeping cannot possibly do any good: if we're
4952 : * required to sleep, get out to raise an error instead.
4953 : *
4954 : * By here, we either have already acquired the buffer exclusive lock,
4955 : * or we must wait for the locking transaction or multixact; so below
4956 : * we ensure that we grab buffer lock after the sleep.
4957 : */
4958 12824 : if (require_sleep && (result == TM_Updated || result == TM_Deleted))
4959 : {
4960 300 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4961 300 : goto failed;
4962 : }
4963 12524 : else if (require_sleep)
4964 : {
4965 : /*
4966 : * Acquire tuple lock to establish our priority for the tuple, or
4967 : * die trying. LockTuple will release us when we are next-in-line
4968 : * for the tuple. We must do this even if we are share-locking,
4969 : * but not if we already have a weaker lock on the tuple.
4970 : *
4971 : * If we are forced to "start over" below, we keep the tuple lock;
4972 : * this arranges that we stay at the head of the line while
4973 : * rechecking tuple state.
4974 : */
4975 374 : if (!skip_tuple_lock &&
4976 342 : !heap_acquire_tuplock(relation, tid, mode, wait_policy,
4977 : &have_tuple_lock))
4978 : {
4979 : /*
4980 : * This can only happen if wait_policy is Skip and the lock
4981 : * couldn't be obtained.
4982 : */
4983 2 : result = TM_WouldBlock;
4984 : /* recovery code expects to have buffer lock held */
4985 2 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4986 2 : goto failed;
4987 : }
4988 :
4989 370 : if (infomask & HEAP_XMAX_IS_MULTI)
4990 : {
4991 80 : MultiXactStatus status = get_mxact_status_for_lock(mode, false);
4992 :
4993 : /* We only ever lock tuples, never update them */
4994 80 : if (status >= MultiXactStatusNoKeyUpdate)
4995 0 : elog(ERROR, "invalid lock mode in heap_lock_tuple");
4996 :
4997 : /* wait for multixact to end, or die trying */
4998 80 : switch (wait_policy)
4999 : {
5000 72 : case LockWaitBlock:
5001 72 : MultiXactIdWait((MultiXactId) xwait, status, infomask,
5002 72 : relation, &tuple->t_self, XLTW_Lock, NULL);
5003 72 : break;
5004 4 : case LockWaitSkip:
5005 4 : if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
5006 : status, infomask, relation,
5007 : NULL, false))
5008 : {
5009 4 : result = TM_WouldBlock;
5010 : /* recovery code expects to have buffer lock held */
5011 4 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5012 4 : goto failed;
5013 : }
5014 0 : break;
5015 4 : case LockWaitError:
5016 4 : if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
5017 : status, infomask, relation,
5018 : NULL, log_lock_failures))
5019 4 : ereport(ERROR,
5020 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5021 : errmsg("could not obtain lock on row in relation \"%s\"",
5022 : RelationGetRelationName(relation))));
5023 :
5024 0 : break;
5025 : }
5026 :
5027 : /*
5028 : * Of course, the multixact might not be done here: if we're
5029 : * requesting a light lock mode, other transactions with light
5030 : * locks could still be alive, as well as locks owned by our
5031 : * own xact or other subxacts of this backend. We need to
5032 : * preserve the surviving MultiXact members. Note that it
5033 : * isn't absolutely necessary in the latter case, but doing so
5034 : * is simpler.
5035 : */
5036 : }
5037 : else
5038 : {
5039 : /* wait for regular transaction to end, or die trying */
5040 290 : switch (wait_policy)
5041 : {
5042 212 : case LockWaitBlock:
5043 212 : XactLockTableWait(xwait, relation, &tuple->t_self,
5044 : XLTW_Lock);
5045 212 : break;
5046 66 : case LockWaitSkip:
5047 66 : if (!ConditionalXactLockTableWait(xwait, false))
5048 : {
5049 66 : result = TM_WouldBlock;
5050 : /* recovery code expects to have buffer lock held */
5051 66 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5052 66 : goto failed;
5053 : }
5054 0 : break;
5055 12 : case LockWaitError:
5056 12 : if (!ConditionalXactLockTableWait(xwait, log_lock_failures))
5057 12 : ereport(ERROR,
5058 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5059 : errmsg("could not obtain lock on row in relation \"%s\"",
5060 : RelationGetRelationName(relation))));
5061 0 : break;
5062 : }
5063 : }
5064 :
5065 : /* if there are updates, follow the update chain */
5066 284 : if (follow_updates && !HEAP_XMAX_IS_LOCKED_ONLY(infomask))
5067 : {
5068 : TM_Result res;
5069 :
5070 112 : res = heap_lock_updated_tuple(relation, tuple, &t_ctid,
5071 : GetCurrentTransactionId(),
5072 : mode);
5073 112 : if (res != TM_Ok)
5074 : {
5075 4 : result = res;
5076 : /* recovery code expects to have buffer lock held */
5077 4 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5078 4 : goto failed;
5079 : }
5080 : }
5081 :
5082 280 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5083 :
5084 : /*
5085 : * xwait is done, but if xwait had just locked the tuple then some
5086 : * other xact could update this tuple before we get to this point.
5087 : * Check for xmax change, and start over if so.
5088 : */
5089 536 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
5090 256 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
5091 : xwait))
5092 28 : goto l3;
5093 :
5094 252 : if (!(infomask & HEAP_XMAX_IS_MULTI))
5095 : {
5096 : /*
5097 : * Otherwise check if it committed or aborted. Note we cannot
5098 : * be here if the tuple was only locked by somebody who didn't
5099 : * conflict with us; that would have been handled above. So
5100 : * that transaction must necessarily be gone by now. But
5101 : * don't check for this in the multixact case, because some
5102 : * locker transactions might still be running.
5103 : */
5104 190 : UpdateXmaxHintBits(tuple->t_data, *buffer, xwait);
5105 : }
5106 : }
5107 :
5108 : /* By here, we're certain that we hold buffer exclusive lock again */
5109 :
5110 : /*
5111 : * We may lock if previous xmax aborted, or if it committed but only
5112 : * locked the tuple without updating it; or if we didn't have to wait
5113 : * at all for whatever reason.
5114 : */
5115 12402 : if (!require_sleep ||
5116 444 : (tuple->t_data->t_infomask & HEAP_XMAX_INVALID) ||
5117 354 : HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
5118 162 : HeapTupleHeaderIsOnlyLocked(tuple->t_data))
5119 12252 : result = TM_Ok;
5120 150 : else if (!ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid))
5121 114 : result = TM_Updated;
5122 : else
5123 36 : result = TM_Deleted;
5124 : }
5125 :
5126 154294 : failed:
5127 167084 : if (result != TM_Ok)
5128 : {
5129 : Assert(result == TM_SelfModified || result == TM_Updated ||
5130 : result == TM_Deleted || result == TM_WouldBlock);
5131 :
5132 : /*
5133 : * When locking a tuple under LockWaitSkip semantics and we fail with
5134 : * TM_WouldBlock above, it's possible for concurrent transactions to
5135 : * release the lock and set HEAP_XMAX_INVALID in the meantime. So
5136 : * this assert is slightly different from the equivalent one in
5137 : * heap_delete and heap_update.
5138 : */
5139 : Assert((result == TM_WouldBlock) ||
5140 : !(tuple->t_data->t_infomask & HEAP_XMAX_INVALID));
5141 : Assert(result != TM_Updated ||
5142 : !ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid));
5143 550 : tmfd->ctid = tuple->t_data->t_ctid;
5144 550 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(tuple->t_data);
5145 550 : if (result == TM_SelfModified)
5146 12 : tmfd->cmax = HeapTupleHeaderGetCmax(tuple->t_data);
5147 : else
5148 538 : tmfd->cmax = InvalidCommandId;
5149 550 : goto out_locked;
5150 : }
5151 :
5152 : /*
5153 : * If we didn't pin the visibility map page and the page has become all
5154 : * visible while we were busy locking the buffer, or during some
5155 : * subsequent window during which we had it unlocked, we'll have to unlock
5156 : * and re-lock, to avoid holding the buffer lock across I/O. That's a bit
5157 : * unfortunate, especially since we'll now have to recheck whether the
5158 : * tuple has been locked or updated under us, but hopefully it won't
5159 : * happen very often.
5160 : */
5161 166534 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
5162 : {
5163 0 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
5164 0 : visibilitymap_pin(relation, block, &vmbuffer);
5165 0 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5166 0 : goto l3;
5167 : }
5168 :
5169 166534 : xmax = HeapTupleHeaderGetRawXmax(tuple->t_data);
5170 166534 : old_infomask = tuple->t_data->t_infomask;
5171 :
5172 : /*
5173 : * If this is the first possibly-multixact-able operation in the current
5174 : * transaction, set my per-backend OldestMemberMXactId setting. We can be
5175 : * certain that the transaction will never become a member of any older
5176 : * MultiXactIds than that. (We have to do this even if we end up just
5177 : * using our own TransactionId below, since some other backend could
5178 : * incorporate our XID into a MultiXact immediately afterwards.)
5179 : */
5180 166534 : MultiXactIdSetOldestMember();
5181 :
5182 : /*
5183 : * Compute the new xmax and infomask to store into the tuple. Note we do
5184 : * not modify the tuple just yet, because that would leave it in the wrong
5185 : * state if multixact.c elogs.
5186 : */
5187 166534 : compute_new_xmax_infomask(xmax, old_infomask, tuple->t_data->t_infomask2,
5188 : GetCurrentTransactionId(), mode, false,
5189 : &xid, &new_infomask, &new_infomask2);
5190 :
5191 166534 : START_CRIT_SECTION();
5192 :
5193 : /*
5194 : * Store transaction information of xact locking the tuple.
5195 : *
5196 : * Note: Cmax is meaningless in this context, so don't set it; this avoids
5197 : * possibly generating a useless combo CID. Moreover, if we're locking a
5198 : * previously updated tuple, it's important to preserve the Cmax.
5199 : *
5200 : * Also reset the HOT UPDATE bit, but only if there's no update; otherwise
5201 : * we would break the HOT chain.
5202 : */
5203 166534 : tuple->t_data->t_infomask &= ~HEAP_XMAX_BITS;
5204 166534 : tuple->t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
5205 166534 : tuple->t_data->t_infomask |= new_infomask;
5206 166534 : tuple->t_data->t_infomask2 |= new_infomask2;
5207 166534 : if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
5208 166456 : HeapTupleHeaderClearHotUpdated(tuple->t_data);
5209 166534 : HeapTupleHeaderSetXmax(tuple->t_data, xid);
5210 :
5211 : /*
5212 : * Make sure there is no forward chain link in t_ctid. Note that in the
5213 : * cases where the tuple has been updated, we must not overwrite t_ctid,
5214 : * because it was set by the updater. Moreover, if the tuple has been
5215 : * updated, we need to follow the update chain to lock the new versions of
5216 : * the tuple as well.
5217 : */
5218 166534 : if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
5219 166456 : tuple->t_data->t_ctid = *tid;
5220 :
5221 : /* Clear only the all-frozen bit on visibility map if needed */
5222 169866 : if (PageIsAllVisible(page) &&
5223 3332 : visibilitymap_clear(relation, block, vmbuffer,
5224 : VISIBILITYMAP_ALL_FROZEN))
5225 28 : cleared_all_frozen = true;
5226 :
5227 :
5228 166534 : MarkBufferDirty(*buffer);
5229 :
5230 : /*
5231 : * XLOG stuff. You might think that we don't need an XLOG record because
5232 : * there is no state change worth restoring after a crash. You would be
5233 : * wrong however: we have just written either a TransactionId or a
5234 : * MultiXactId that may never have been seen on disk before, and we need
5235 : * to make sure that there are XLOG entries covering those ID numbers.
5236 : * Else the same IDs might be re-used after a crash, which would be
5237 : * disastrous if this page made it to disk before the crash. Essentially
5238 : * we have to enforce the WAL log-before-data rule even in this case.
5239 : * (Also, in a PITR log-shipping or 2PC environment, we have to have XLOG
5240 : * entries for everything anyway.)
5241 : */
5242 166534 : if (RelationNeedsWAL(relation))
5243 : {
5244 : xl_heap_lock xlrec;
5245 : XLogRecPtr recptr;
5246 :
5247 165830 : XLogBeginInsert();
5248 165830 : XLogRegisterBuffer(0, *buffer, REGBUF_STANDARD);
5249 :
5250 165830 : xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
5251 165830 : xlrec.xmax = xid;
5252 331660 : xlrec.infobits_set = compute_infobits(new_infomask,
5253 165830 : tuple->t_data->t_infomask2);
5254 165830 : xlrec.flags = cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
5255 165830 : XLogRegisterData(&xlrec, SizeOfHeapLock);
5256 :
5257 : /* we don't decode row locks atm, so no need to log the origin */
5258 :
5259 165830 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
5260 :
5261 165830 : PageSetLSN(page, recptr);
5262 : }
5263 :
5264 166534 : END_CRIT_SECTION();
5265 :
5266 166534 : result = TM_Ok;
5267 :
5268 167108 : out_locked:
5269 167108 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
5270 :
5271 170112 : out_unlocked:
5272 170112 : if (BufferIsValid(vmbuffer))
5273 3332 : ReleaseBuffer(vmbuffer);
5274 :
5275 : /*
5276 : * Don't update the visibility map here. Locking a tuple doesn't change
5277 : * visibility info.
5278 : */
5279 :
5280 : /*
5281 : * Now that we have successfully marked the tuple as locked, we can
5282 : * release the lmgr tuple lock, if we had it.
5283 : */
5284 170112 : if (have_tuple_lock)
5285 312 : UnlockTupleTuplock(relation, tid, mode);
5286 :
5287 170112 : return result;
5288 : }
5289 :
5290 : /*
5291 : * Acquire heavyweight lock on the given tuple, in preparation for acquiring
5292 : * its normal, Xmax-based tuple lock.
5293 : *
5294 : * have_tuple_lock is an input and output parameter: on input, it indicates
5295 : * whether the lock has previously been acquired (and this function does
5296 : * nothing in that case). If this function returns success, have_tuple_lock
5297 : * has been flipped to true.
5298 : *
5299 : * Returns false if it was unable to obtain the lock; this can only happen if
5300 : * wait_policy is Skip.
5301 : */
5302 : static bool
5303 614 : heap_acquire_tuplock(Relation relation, const ItemPointerData *tid, LockTupleMode mode,
5304 : LockWaitPolicy wait_policy, bool *have_tuple_lock)
5305 : {
5306 614 : if (*have_tuple_lock)
5307 18 : return true;
5308 :
5309 596 : switch (wait_policy)
5310 : {
5311 514 : case LockWaitBlock:
5312 514 : LockTupleTuplock(relation, tid, mode);
5313 514 : break;
5314 :
5315 68 : case LockWaitSkip:
5316 68 : if (!ConditionalLockTupleTuplock(relation, tid, mode, false))
5317 2 : return false;
5318 66 : break;
5319 :
5320 14 : case LockWaitError:
5321 14 : if (!ConditionalLockTupleTuplock(relation, tid, mode, log_lock_failures))
5322 2 : ereport(ERROR,
5323 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5324 : errmsg("could not obtain lock on row in relation \"%s\"",
5325 : RelationGetRelationName(relation))));
5326 12 : break;
5327 : }
5328 592 : *have_tuple_lock = true;
5329 :
5330 592 : return true;
5331 : }
5332 :
5333 : /*
5334 : * Given an original set of Xmax and infomask, and a transaction (identified by
5335 : * add_to_xmax) acquiring a new lock of some mode, compute the new Xmax and
5336 : * corresponding infomasks to use on the tuple.
5337 : *
5338 : * Note that this might have side effects such as creating a new MultiXactId.
5339 : *
5340 : * Most callers will have called HeapTupleSatisfiesUpdate before this function;
5341 : * that will have set the HEAP_XMAX_INVALID bit if the xmax was a MultiXactId
5342 : * but it was not running anymore. There is a race condition, which is that the
5343 : * MultiXactId may have finished since then, but that uncommon case is handled
5344 : * either here, or within MultiXactIdExpand.
5345 : *
5346 : * There is a similar race condition possible when the old xmax was a regular
5347 : * TransactionId. We test TransactionIdIsInProgress again just to narrow the
5348 : * window, but it's still possible to end up creating an unnecessary
5349 : * MultiXactId. Fortunately this is harmless.
5350 : */
5351 : static void
5352 4072786 : compute_new_xmax_infomask(TransactionId xmax, uint16 old_infomask,
5353 : uint16 old_infomask2, TransactionId add_to_xmax,
5354 : LockTupleMode mode, bool is_update,
5355 : TransactionId *result_xmax, uint16 *result_infomask,
5356 : uint16 *result_infomask2)
5357 : {
5358 : TransactionId new_xmax;
5359 : uint16 new_infomask,
5360 : new_infomask2;
5361 :
5362 : Assert(TransactionIdIsCurrentTransactionId(add_to_xmax));
5363 :
5364 208068 : l5:
5365 4280854 : new_infomask = 0;
5366 4280854 : new_infomask2 = 0;
5367 4280854 : if (old_infomask & HEAP_XMAX_INVALID)
5368 : {
5369 : /*
5370 : * No previous locker; we just insert our own TransactionId.
5371 : *
5372 : * Note that it's critical that this case be the first one checked,
5373 : * because there are several blocks below that come back to this one
5374 : * to implement certain optimizations; old_infomask might contain
5375 : * other dirty bits in those cases, but we don't really care.
5376 : */
5377 4070520 : if (is_update)
5378 : {
5379 3607024 : new_xmax = add_to_xmax;
5380 3607024 : if (mode == LockTupleExclusive)
5381 3072328 : new_infomask2 |= HEAP_KEYS_UPDATED;
5382 : }
5383 : else
5384 : {
5385 463496 : new_infomask |= HEAP_XMAX_LOCK_ONLY;
5386 463496 : switch (mode)
5387 : {
5388 5154 : case LockTupleKeyShare:
5389 5154 : new_xmax = add_to_xmax;
5390 5154 : new_infomask |= HEAP_XMAX_KEYSHR_LOCK;
5391 5154 : break;
5392 1488 : case LockTupleShare:
5393 1488 : new_xmax = add_to_xmax;
5394 1488 : new_infomask |= HEAP_XMAX_SHR_LOCK;
5395 1488 : break;
5396 265474 : case LockTupleNoKeyExclusive:
5397 265474 : new_xmax = add_to_xmax;
5398 265474 : new_infomask |= HEAP_XMAX_EXCL_LOCK;
5399 265474 : break;
5400 191380 : case LockTupleExclusive:
5401 191380 : new_xmax = add_to_xmax;
5402 191380 : new_infomask |= HEAP_XMAX_EXCL_LOCK;
5403 191380 : new_infomask2 |= HEAP_KEYS_UPDATED;
5404 191380 : break;
5405 0 : default:
5406 0 : new_xmax = InvalidTransactionId; /* silence compiler */
5407 0 : elog(ERROR, "invalid lock mode");
5408 : }
5409 : }
5410 : }
5411 210334 : else if (old_infomask & HEAP_XMAX_IS_MULTI)
5412 : {
5413 : MultiXactStatus new_status;
5414 :
5415 : /*
5416 : * Currently we don't allow XMAX_COMMITTED to be set for multis, so
5417 : * cross-check.
5418 : */
5419 : Assert(!(old_infomask & HEAP_XMAX_COMMITTED));
5420 :
5421 : /*
5422 : * A multixact together with LOCK_ONLY set but neither lock bit set
5423 : * (i.e. a pg_upgraded share locked tuple) cannot possibly be running
5424 : * anymore. This check is critical for databases upgraded by
5425 : * pg_upgrade; both MultiXactIdIsRunning and MultiXactIdExpand assume
5426 : * that such multis are never passed.
5427 : */
5428 232 : if (HEAP_LOCKED_UPGRADED(old_infomask))
5429 : {
5430 0 : old_infomask &= ~HEAP_XMAX_IS_MULTI;
5431 0 : old_infomask |= HEAP_XMAX_INVALID;
5432 0 : goto l5;
5433 : }
5434 :
5435 : /*
5436 : * If the XMAX is already a MultiXactId, then we need to expand it to
5437 : * include add_to_xmax; but if all the members were lockers and are
5438 : * all gone, we can do away with the IS_MULTI bit and just set
5439 : * add_to_xmax as the only locker/updater. If all lockers are gone
5440 : * and we have an updater that aborted, we can also do without a
5441 : * multi.
5442 : *
5443 : * The cost of doing GetMultiXactIdMembers would be paid by
5444 : * MultiXactIdExpand if we weren't to do this, so this check is not
5445 : * incurring extra work anyhow.
5446 : */
5447 232 : if (!MultiXactIdIsRunning(xmax, HEAP_XMAX_IS_LOCKED_ONLY(old_infomask)))
5448 : {
5449 46 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) ||
5450 16 : !TransactionIdDidCommit(MultiXactIdGetUpdateXid(xmax,
5451 : old_infomask)))
5452 : {
5453 : /*
5454 : * Reset these bits and restart; otherwise fall through to
5455 : * create a new multi below.
5456 : */
5457 46 : old_infomask &= ~HEAP_XMAX_IS_MULTI;
5458 46 : old_infomask |= HEAP_XMAX_INVALID;
5459 46 : goto l5;
5460 : }
5461 : }
5462 :
5463 186 : new_status = get_mxact_status_for_lock(mode, is_update);
5464 :
5465 186 : new_xmax = MultiXactIdExpand((MultiXactId) xmax, add_to_xmax,
5466 : new_status);
5467 186 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5468 : }
5469 210102 : else if (old_infomask & HEAP_XMAX_COMMITTED)
5470 : {
5471 : /*
5472 : * It's a committed update, so we need to preserve him as updater of
5473 : * the tuple.
5474 : */
5475 : MultiXactStatus status;
5476 : MultiXactStatus new_status;
5477 :
5478 26 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5479 0 : status = MultiXactStatusUpdate;
5480 : else
5481 26 : status = MultiXactStatusNoKeyUpdate;
5482 :
5483 26 : new_status = get_mxact_status_for_lock(mode, is_update);
5484 :
5485 : /*
5486 : * since it's not running, it's obviously impossible for the old
5487 : * updater to be identical to the current one, so we need not check
5488 : * for that case as we do in the block above.
5489 : */
5490 26 : new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5491 26 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5492 : }
5493 210076 : else if (TransactionIdIsInProgress(xmax))
5494 : {
5495 : /*
5496 : * If the XMAX is a valid, in-progress TransactionId, then we need to
5497 : * create a new MultiXactId that includes both the old locker or
5498 : * updater and our own TransactionId.
5499 : */
5500 : MultiXactStatus new_status;
5501 : MultiXactStatus old_status;
5502 : LockTupleMode old_mode;
5503 :
5504 210058 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
5505 : {
5506 210006 : if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
5507 11254 : old_status = MultiXactStatusForKeyShare;
5508 198752 : else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
5509 866 : old_status = MultiXactStatusForShare;
5510 197886 : else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
5511 : {
5512 197886 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5513 185544 : old_status = MultiXactStatusForUpdate;
5514 : else
5515 12342 : old_status = MultiXactStatusForNoKeyUpdate;
5516 : }
5517 : else
5518 : {
5519 : /*
5520 : * LOCK_ONLY can be present alone only when a page has been
5521 : * upgraded by pg_upgrade. But in that case,
5522 : * TransactionIdIsInProgress() should have returned false. We
5523 : * assume it's no longer locked in this case.
5524 : */
5525 0 : elog(WARNING, "LOCK_ONLY found for Xid in progress %u", xmax);
5526 0 : old_infomask |= HEAP_XMAX_INVALID;
5527 0 : old_infomask &= ~HEAP_XMAX_LOCK_ONLY;
5528 0 : goto l5;
5529 : }
5530 : }
5531 : else
5532 : {
5533 : /* it's an update, but which kind? */
5534 52 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5535 0 : old_status = MultiXactStatusUpdate;
5536 : else
5537 52 : old_status = MultiXactStatusNoKeyUpdate;
5538 : }
5539 :
5540 210058 : old_mode = TUPLOCK_from_mxstatus(old_status);
5541 :
5542 : /*
5543 : * If the lock to be acquired is for the same TransactionId as the
5544 : * existing lock, there's an optimization possible: consider only the
5545 : * strongest of both locks as the only one present, and restart.
5546 : */
5547 210058 : if (xmax == add_to_xmax)
5548 : {
5549 : /*
5550 : * Note that it's not possible for the original tuple to be
5551 : * updated: we wouldn't be here because the tuple would have been
5552 : * invisible and we wouldn't try to update it. As a subtlety,
5553 : * this code can also run when traversing an update chain to lock
5554 : * future versions of a tuple. But we wouldn't be here either,
5555 : * because the add_to_xmax would be different from the original
5556 : * updater.
5557 : */
5558 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
5559 :
5560 : /* acquire the strongest of both */
5561 208006 : if (mode < old_mode)
5562 104414 : mode = old_mode;
5563 : /* mustn't touch is_update */
5564 :
5565 208006 : old_infomask |= HEAP_XMAX_INVALID;
5566 208006 : goto l5;
5567 : }
5568 :
5569 : /* otherwise, just fall back to creating a new multixact */
5570 2052 : new_status = get_mxact_status_for_lock(mode, is_update);
5571 2052 : new_xmax = MultiXactIdCreate(xmax, old_status,
5572 : add_to_xmax, new_status);
5573 2052 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5574 : }
5575 28 : else if (!HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) &&
5576 10 : TransactionIdDidCommit(xmax))
5577 2 : {
5578 : /*
5579 : * It's a committed update, so we gotta preserve him as updater of the
5580 : * tuple.
5581 : */
5582 : MultiXactStatus status;
5583 : MultiXactStatus new_status;
5584 :
5585 2 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5586 0 : status = MultiXactStatusUpdate;
5587 : else
5588 2 : status = MultiXactStatusNoKeyUpdate;
5589 :
5590 2 : new_status = get_mxact_status_for_lock(mode, is_update);
5591 :
5592 : /*
5593 : * since it's not running, it's obviously impossible for the old
5594 : * updater to be identical to the current one, so we need not check
5595 : * for that case as we do in the block above.
5596 : */
5597 2 : new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5598 2 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5599 : }
5600 : else
5601 : {
5602 : /*
5603 : * Can get here iff the locking/updating transaction was running when
5604 : * the infomask was extracted from the tuple, but finished before
5605 : * TransactionIdIsInProgress got to run. Deal with it as if there was
5606 : * no locker at all in the first place.
5607 : */
5608 16 : old_infomask |= HEAP_XMAX_INVALID;
5609 16 : goto l5;
5610 : }
5611 :
5612 4072786 : *result_infomask = new_infomask;
5613 4072786 : *result_infomask2 = new_infomask2;
5614 4072786 : *result_xmax = new_xmax;
5615 4072786 : }
5616 :
5617 : /*
5618 : * Subroutine for heap_lock_updated_tuple_rec.
5619 : *
5620 : * Given a hypothetical multixact status held by the transaction identified
5621 : * with the given xid, does the current transaction need to wait, fail, or can
5622 : * it continue if it wanted to acquire a lock of the given mode? "needwait"
5623 : * is set to true if waiting is necessary; if it can continue, then TM_Ok is
5624 : * returned. If the lock is already held by the current transaction, return
5625 : * TM_SelfModified. In case of a conflict with another transaction, a
5626 : * different HeapTupleSatisfiesUpdate return code is returned.
5627 : *
5628 : * The held status is said to be hypothetical because it might correspond to a
5629 : * lock held by a single Xid, i.e. not a real MultiXactId; we express it this
5630 : * way for simplicity of API.
5631 : */
5632 : static TM_Result
5633 64 : test_lockmode_for_conflict(MultiXactStatus status, TransactionId xid,
5634 : LockTupleMode mode, HeapTuple tup,
5635 : bool *needwait)
5636 : {
5637 : MultiXactStatus wantedstatus;
5638 :
5639 64 : *needwait = false;
5640 64 : wantedstatus = get_mxact_status_for_lock(mode, false);
5641 :
5642 : /*
5643 : * Note: we *must* check TransactionIdIsInProgress before
5644 : * TransactionIdDidAbort/Commit; see comment at top of heapam_visibility.c
5645 : * for an explanation.
5646 : */
5647 64 : if (TransactionIdIsCurrentTransactionId(xid))
5648 : {
5649 : /*
5650 : * The tuple has already been locked by our own transaction. This is
5651 : * very rare but can happen if multiple transactions are trying to
5652 : * lock an ancient version of the same tuple.
5653 : */
5654 0 : return TM_SelfModified;
5655 : }
5656 64 : else if (TransactionIdIsInProgress(xid))
5657 : {
5658 : /*
5659 : * If the locking transaction is running, what we do depends on
5660 : * whether the lock modes conflict: if they do, then we must wait for
5661 : * it to finish; otherwise we can fall through to lock this tuple
5662 : * version without waiting.
5663 : */
5664 32 : if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
5665 32 : LOCKMODE_from_mxstatus(wantedstatus)))
5666 : {
5667 16 : *needwait = true;
5668 : }
5669 :
5670 : /*
5671 : * If we set needwait above, then this value doesn't matter;
5672 : * otherwise, this value signals to caller that it's okay to proceed.
5673 : */
5674 32 : return TM_Ok;
5675 : }
5676 32 : else if (TransactionIdDidAbort(xid))
5677 6 : return TM_Ok;
5678 26 : else if (TransactionIdDidCommit(xid))
5679 : {
5680 : /*
5681 : * The other transaction committed. If it was only a locker, then the
5682 : * lock is completely gone now and we can return success; but if it
5683 : * was an update, then what we do depends on whether the two lock
5684 : * modes conflict. If they conflict, then we must report error to
5685 : * caller. But if they don't, we can fall through to allow the current
5686 : * transaction to lock the tuple.
5687 : *
5688 : * Note: the reason we worry about ISUPDATE here is because as soon as
5689 : * a transaction ends, all its locks are gone and meaningless, and
5690 : * thus we can ignore them; whereas its updates persist. In the
5691 : * TransactionIdIsInProgress case, above, we don't need to check
5692 : * because we know the lock is still "alive" and thus a conflict needs
5693 : * always be checked.
5694 : */
5695 26 : if (!ISUPDATE_from_mxstatus(status))
5696 8 : return TM_Ok;
5697 :
5698 18 : if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
5699 18 : LOCKMODE_from_mxstatus(wantedstatus)))
5700 : {
5701 : /* bummer */
5702 16 : if (!ItemPointerEquals(&tup->t_self, &tup->t_data->t_ctid))
5703 12 : return TM_Updated;
5704 : else
5705 4 : return TM_Deleted;
5706 : }
5707 :
5708 2 : return TM_Ok;
5709 : }
5710 :
5711 : /* Not in progress, not aborted, not committed -- must have crashed */
5712 0 : return TM_Ok;
5713 : }
5714 :
5715 :
5716 : /*
5717 : * Recursive part of heap_lock_updated_tuple
5718 : *
5719 : * Fetch the tuple pointed to by tid in rel, and mark it as locked by the given
5720 : * xid with the given mode; if this tuple is updated, recurse to lock the new
5721 : * version as well.
5722 : */
5723 : static TM_Result
5724 182 : heap_lock_updated_tuple_rec(Relation rel, const ItemPointerData *tid, TransactionId xid,
5725 : LockTupleMode mode)
5726 : {
5727 : TM_Result result;
5728 : ItemPointerData tupid;
5729 : HeapTupleData mytup;
5730 : Buffer buf;
5731 : uint16 new_infomask,
5732 : new_infomask2,
5733 : old_infomask,
5734 : old_infomask2;
5735 : TransactionId xmax,
5736 : new_xmax;
5737 182 : TransactionId priorXmax = InvalidTransactionId;
5738 182 : bool cleared_all_frozen = false;
5739 : bool pinned_desired_page;
5740 182 : Buffer vmbuffer = InvalidBuffer;
5741 : BlockNumber block;
5742 :
5743 182 : ItemPointerCopy(tid, &tupid);
5744 :
5745 : for (;;)
5746 : {
5747 188 : new_infomask = 0;
5748 188 : new_xmax = InvalidTransactionId;
5749 188 : block = ItemPointerGetBlockNumber(&tupid);
5750 188 : ItemPointerCopy(&tupid, &(mytup.t_self));
5751 :
5752 188 : if (!heap_fetch(rel, SnapshotAny, &mytup, &buf, false))
5753 : {
5754 : /*
5755 : * if we fail to find the updated version of the tuple, it's
5756 : * because it was vacuumed/pruned away after its creator
5757 : * transaction aborted. So behave as if we got to the end of the
5758 : * chain, and there's no further tuple to lock: return success to
5759 : * caller.
5760 : */
5761 0 : result = TM_Ok;
5762 0 : goto out_unlocked;
5763 : }
5764 :
5765 188 : l4:
5766 204 : CHECK_FOR_INTERRUPTS();
5767 :
5768 : /*
5769 : * Before locking the buffer, pin the visibility map page if it
5770 : * appears to be necessary. Since we haven't got the lock yet,
5771 : * someone else might be in the middle of changing this, so we'll need
5772 : * to recheck after we have the lock.
5773 : */
5774 204 : if (PageIsAllVisible(BufferGetPage(buf)))
5775 : {
5776 0 : visibilitymap_pin(rel, block, &vmbuffer);
5777 0 : pinned_desired_page = true;
5778 : }
5779 : else
5780 204 : pinned_desired_page = false;
5781 :
5782 204 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
5783 :
5784 : /*
5785 : * If we didn't pin the visibility map page and the page has become
5786 : * all visible while we were busy locking the buffer, we'll have to
5787 : * unlock and re-lock, to avoid holding the buffer lock across I/O.
5788 : * That's a bit unfortunate, but hopefully shouldn't happen often.
5789 : *
5790 : * Note: in some paths through this function, we will reach here
5791 : * holding a pin on a vm page that may or may not be the one matching
5792 : * this page. If this page isn't all-visible, we won't use the vm
5793 : * page, but we hold onto such a pin till the end of the function.
5794 : */
5795 204 : if (!pinned_desired_page && PageIsAllVisible(BufferGetPage(buf)))
5796 : {
5797 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5798 0 : visibilitymap_pin(rel, block, &vmbuffer);
5799 0 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
5800 : }
5801 :
5802 : /*
5803 : * Check the tuple XMIN against prior XMAX, if any. If we reached the
5804 : * end of the chain, we're done, so return success.
5805 : */
5806 210 : if (TransactionIdIsValid(priorXmax) &&
5807 6 : !TransactionIdEquals(HeapTupleHeaderGetXmin(mytup.t_data),
5808 : priorXmax))
5809 : {
5810 0 : result = TM_Ok;
5811 0 : goto out_locked;
5812 : }
5813 :
5814 : /*
5815 : * Also check Xmin: if this tuple was created by an aborted
5816 : * (sub)transaction, then we already locked the last live one in the
5817 : * chain, thus we're done, so return success.
5818 : */
5819 204 : if (TransactionIdDidAbort(HeapTupleHeaderGetXmin(mytup.t_data)))
5820 : {
5821 26 : result = TM_Ok;
5822 26 : goto out_locked;
5823 : }
5824 :
5825 178 : old_infomask = mytup.t_data->t_infomask;
5826 178 : old_infomask2 = mytup.t_data->t_infomask2;
5827 178 : xmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
5828 :
5829 : /*
5830 : * If this tuple version has been updated or locked by some concurrent
5831 : * transaction(s), what we do depends on whether our lock mode
5832 : * conflicts with what those other transactions hold, and also on the
5833 : * status of them.
5834 : */
5835 178 : if (!(old_infomask & HEAP_XMAX_INVALID))
5836 : {
5837 : TransactionId rawxmax;
5838 : bool needwait;
5839 :
5840 60 : rawxmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
5841 60 : if (old_infomask & HEAP_XMAX_IS_MULTI)
5842 : {
5843 : int nmembers;
5844 : int i;
5845 : MultiXactMember *members;
5846 :
5847 : /*
5848 : * We don't need a test for pg_upgrade'd tuples: this is only
5849 : * applied to tuples after the first in an update chain. Said
5850 : * first tuple in the chain may well be locked-in-9.2-and-
5851 : * pg_upgraded, but that one was already locked by our caller,
5852 : * not us; and any subsequent ones cannot be because our
5853 : * caller must necessarily have obtained a snapshot later than
5854 : * the pg_upgrade itself.
5855 : */
5856 : Assert(!HEAP_LOCKED_UPGRADED(mytup.t_data->t_infomask));
5857 :
5858 2 : nmembers = GetMultiXactIdMembers(rawxmax, &members, false,
5859 2 : HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
5860 8 : for (i = 0; i < nmembers; i++)
5861 : {
5862 6 : result = test_lockmode_for_conflict(members[i].status,
5863 6 : members[i].xid,
5864 : mode,
5865 : &mytup,
5866 : &needwait);
5867 :
5868 : /*
5869 : * If the tuple was already locked by ourselves in a
5870 : * previous iteration of this (say heap_lock_tuple was
5871 : * forced to restart the locking loop because of a change
5872 : * in xmax), then we hold the lock already on this tuple
5873 : * version and we don't need to do anything; and this is
5874 : * not an error condition either. We just need to skip
5875 : * this tuple and continue locking the next version in the
5876 : * update chain.
5877 : */
5878 6 : if (result == TM_SelfModified)
5879 : {
5880 0 : pfree(members);
5881 0 : goto next;
5882 : }
5883 :
5884 6 : if (needwait)
5885 : {
5886 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5887 0 : XactLockTableWait(members[i].xid, rel,
5888 : &mytup.t_self,
5889 : XLTW_LockUpdated);
5890 0 : pfree(members);
5891 0 : goto l4;
5892 : }
5893 6 : if (result != TM_Ok)
5894 : {
5895 0 : pfree(members);
5896 0 : goto out_locked;
5897 : }
5898 : }
5899 2 : if (members)
5900 2 : pfree(members);
5901 : }
5902 : else
5903 : {
5904 : MultiXactStatus status;
5905 :
5906 : /*
5907 : * For a non-multi Xmax, we first need to compute the
5908 : * corresponding MultiXactStatus by using the infomask bits.
5909 : */
5910 58 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
5911 : {
5912 20 : if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
5913 20 : status = MultiXactStatusForKeyShare;
5914 0 : else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
5915 0 : status = MultiXactStatusForShare;
5916 0 : else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
5917 : {
5918 0 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5919 0 : status = MultiXactStatusForUpdate;
5920 : else
5921 0 : status = MultiXactStatusForNoKeyUpdate;
5922 : }
5923 : else
5924 : {
5925 : /*
5926 : * LOCK_ONLY present alone (a pg_upgraded tuple marked
5927 : * as share-locked in the old cluster) shouldn't be
5928 : * seen in the middle of an update chain.
5929 : */
5930 0 : elog(ERROR, "invalid lock status in tuple");
5931 : }
5932 : }
5933 : else
5934 : {
5935 : /* it's an update, but which kind? */
5936 38 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5937 28 : status = MultiXactStatusUpdate;
5938 : else
5939 10 : status = MultiXactStatusNoKeyUpdate;
5940 : }
5941 :
5942 58 : result = test_lockmode_for_conflict(status, rawxmax, mode,
5943 : &mytup, &needwait);
5944 :
5945 : /*
5946 : * If the tuple was already locked by ourselves in a previous
5947 : * iteration of this (say heap_lock_tuple was forced to
5948 : * restart the locking loop because of a change in xmax), then
5949 : * we hold the lock already on this tuple version and we don't
5950 : * need to do anything; and this is not an error condition
5951 : * either. We just need to skip this tuple and continue
5952 : * locking the next version in the update chain.
5953 : */
5954 58 : if (result == TM_SelfModified)
5955 0 : goto next;
5956 :
5957 58 : if (needwait)
5958 : {
5959 16 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5960 16 : XactLockTableWait(rawxmax, rel, &mytup.t_self,
5961 : XLTW_LockUpdated);
5962 16 : goto l4;
5963 : }
5964 42 : if (result != TM_Ok)
5965 : {
5966 16 : goto out_locked;
5967 : }
5968 : }
5969 : }
5970 :
5971 : /* compute the new Xmax and infomask values for the tuple ... */
5972 146 : compute_new_xmax_infomask(xmax, old_infomask, mytup.t_data->t_infomask2,
5973 : xid, mode, false,
5974 : &new_xmax, &new_infomask, &new_infomask2);
5975 :
5976 146 : if (PageIsAllVisible(BufferGetPage(buf)) &&
5977 0 : visibilitymap_clear(rel, block, vmbuffer,
5978 : VISIBILITYMAP_ALL_FROZEN))
5979 0 : cleared_all_frozen = true;
5980 :
5981 146 : START_CRIT_SECTION();
5982 :
5983 : /* ... and set them */
5984 146 : HeapTupleHeaderSetXmax(mytup.t_data, new_xmax);
5985 146 : mytup.t_data->t_infomask &= ~HEAP_XMAX_BITS;
5986 146 : mytup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
5987 146 : mytup.t_data->t_infomask |= new_infomask;
5988 146 : mytup.t_data->t_infomask2 |= new_infomask2;
5989 :
5990 146 : MarkBufferDirty(buf);
5991 :
5992 : /* XLOG stuff */
5993 146 : if (RelationNeedsWAL(rel))
5994 : {
5995 : xl_heap_lock_updated xlrec;
5996 : XLogRecPtr recptr;
5997 146 : Page page = BufferGetPage(buf);
5998 :
5999 146 : XLogBeginInsert();
6000 146 : XLogRegisterBuffer(0, buf, REGBUF_STANDARD);
6001 :
6002 146 : xlrec.offnum = ItemPointerGetOffsetNumber(&mytup.t_self);
6003 146 : xlrec.xmax = new_xmax;
6004 146 : xlrec.infobits_set = compute_infobits(new_infomask, new_infomask2);
6005 146 : xlrec.flags =
6006 146 : cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
6007 :
6008 146 : XLogRegisterData(&xlrec, SizeOfHeapLockUpdated);
6009 :
6010 146 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_LOCK_UPDATED);
6011 :
6012 146 : PageSetLSN(page, recptr);
6013 : }
6014 :
6015 146 : END_CRIT_SECTION();
6016 :
6017 146 : next:
6018 : /* if we find the end of update chain, we're done. */
6019 292 : if (mytup.t_data->t_infomask & HEAP_XMAX_INVALID ||
6020 292 : HeapTupleHeaderIndicatesMovedPartitions(mytup.t_data) ||
6021 154 : ItemPointerEquals(&mytup.t_self, &mytup.t_data->t_ctid) ||
6022 8 : HeapTupleHeaderIsOnlyLocked(mytup.t_data))
6023 : {
6024 140 : result = TM_Ok;
6025 140 : goto out_locked;
6026 : }
6027 :
6028 : /* tail recursion */
6029 6 : priorXmax = HeapTupleHeaderGetUpdateXid(mytup.t_data);
6030 6 : ItemPointerCopy(&(mytup.t_data->t_ctid), &tupid);
6031 6 : UnlockReleaseBuffer(buf);
6032 : }
6033 :
6034 : result = TM_Ok;
6035 :
6036 182 : out_locked:
6037 182 : UnlockReleaseBuffer(buf);
6038 :
6039 182 : out_unlocked:
6040 182 : if (vmbuffer != InvalidBuffer)
6041 0 : ReleaseBuffer(vmbuffer);
6042 :
6043 182 : return result;
6044 : }
6045 :
6046 : /*
6047 : * heap_lock_updated_tuple
6048 : * Follow update chain when locking an updated tuple, acquiring locks (row
6049 : * marks) on the updated versions.
6050 : *
6051 : * The initial tuple is assumed to be already locked.
6052 : *
6053 : * This function doesn't check visibility, it just unconditionally marks the
6054 : * tuple(s) as locked. If any tuple in the updated chain is being deleted
6055 : * concurrently (or updated with the key being modified), sleep until the
6056 : * transaction doing it is finished.
6057 : *
6058 : * Note that we don't acquire heavyweight tuple locks on the tuples we walk
6059 : * when we have to wait for other transactions to release them, as opposed to
6060 : * what heap_lock_tuple does. The reason is that having more than one
6061 : * transaction walking the chain is probably uncommon enough that risk of
6062 : * starvation is not likely: one of the preconditions for being here is that
6063 : * the snapshot in use predates the update that created this tuple (because we
6064 : * started at an earlier version of the tuple), but at the same time such a
6065 : * transaction cannot be using repeatable read or serializable isolation
6066 : * levels, because that would lead to a serializability failure.
6067 : */
6068 : static TM_Result
6069 212 : heap_lock_updated_tuple(Relation rel, HeapTuple tuple, const ItemPointerData *ctid,
6070 : TransactionId xid, LockTupleMode mode)
6071 : {
6072 : /*
6073 : * If the tuple has not been updated, or has moved into another partition
6074 : * (effectively a delete) stop here.
6075 : */
6076 212 : if (!HeapTupleHeaderIndicatesMovedPartitions(tuple->t_data) &&
6077 208 : !ItemPointerEquals(&tuple->t_self, ctid))
6078 : {
6079 : /*
6080 : * If this is the first possibly-multixact-able operation in the
6081 : * current transaction, set my per-backend OldestMemberMXactId
6082 : * setting. We can be certain that the transaction will never become a
6083 : * member of any older MultiXactIds than that. (We have to do this
6084 : * even if we end up just using our own TransactionId below, since
6085 : * some other backend could incorporate our XID into a MultiXact
6086 : * immediately afterwards.)
6087 : */
6088 182 : MultiXactIdSetOldestMember();
6089 :
6090 182 : return heap_lock_updated_tuple_rec(rel, ctid, xid, mode);
6091 : }
6092 :
6093 : /* nothing to lock */
6094 30 : return TM_Ok;
6095 : }
6096 :
6097 : /*
6098 : * heap_finish_speculative - mark speculative insertion as successful
6099 : *
6100 : * To successfully finish a speculative insertion we have to clear speculative
6101 : * token from tuple. To do so the t_ctid field, which will contain a
6102 : * speculative token value, is modified in place to point to the tuple itself,
6103 : * which is characteristic of a newly inserted ordinary tuple.
6104 : *
6105 : * NB: It is not ok to commit without either finishing or aborting a
6106 : * speculative insertion. We could treat speculative tuples of committed
6107 : * transactions implicitly as completed, but then we would have to be prepared
6108 : * to deal with speculative tokens on committed tuples. That wouldn't be
6109 : * difficult - no-one looks at the ctid field of a tuple with invalid xmax -
6110 : * but clearing the token at completion isn't very expensive either.
6111 : * An explicit confirmation WAL record also makes logical decoding simpler.
6112 : */
6113 : void
6114 4158 : heap_finish_speculative(Relation relation, const ItemPointerData *tid)
6115 : {
6116 : Buffer buffer;
6117 : Page page;
6118 : OffsetNumber offnum;
6119 4158 : ItemId lp = NULL;
6120 : HeapTupleHeader htup;
6121 :
6122 4158 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
6123 4158 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6124 4158 : page = BufferGetPage(buffer);
6125 :
6126 4158 : offnum = ItemPointerGetOffsetNumber(tid);
6127 4158 : if (PageGetMaxOffsetNumber(page) >= offnum)
6128 4158 : lp = PageGetItemId(page, offnum);
6129 :
6130 4158 : if (PageGetMaxOffsetNumber(page) < offnum || !ItemIdIsNormal(lp))
6131 0 : elog(ERROR, "invalid lp");
6132 :
6133 4158 : htup = (HeapTupleHeader) PageGetItem(page, lp);
6134 :
6135 : /* NO EREPORT(ERROR) from here till changes are logged */
6136 4158 : START_CRIT_SECTION();
6137 :
6138 : Assert(HeapTupleHeaderIsSpeculative(htup));
6139 :
6140 4158 : MarkBufferDirty(buffer);
6141 :
6142 : /*
6143 : * Replace the speculative insertion token with a real t_ctid, pointing to
6144 : * itself like it does on regular tuples.
6145 : */
6146 4158 : htup->t_ctid = *tid;
6147 :
6148 : /* XLOG stuff */
6149 4158 : if (RelationNeedsWAL(relation))
6150 : {
6151 : xl_heap_confirm xlrec;
6152 : XLogRecPtr recptr;
6153 :
6154 4124 : xlrec.offnum = ItemPointerGetOffsetNumber(tid);
6155 :
6156 4124 : XLogBeginInsert();
6157 :
6158 : /* We want the same filtering on this as on a plain insert */
6159 4124 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
6160 :
6161 4124 : XLogRegisterData(&xlrec, SizeOfHeapConfirm);
6162 4124 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
6163 :
6164 4124 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_CONFIRM);
6165 :
6166 4124 : PageSetLSN(page, recptr);
6167 : }
6168 :
6169 4158 : END_CRIT_SECTION();
6170 :
6171 4158 : UnlockReleaseBuffer(buffer);
6172 4158 : }
6173 :
6174 : /*
6175 : * heap_abort_speculative - kill a speculatively inserted tuple
6176 : *
6177 : * Marks a tuple that was speculatively inserted in the same command as dead,
6178 : * by setting its xmin as invalid. That makes it immediately appear as dead
6179 : * to all transactions, including our own. In particular, it makes
6180 : * HeapTupleSatisfiesDirty() regard the tuple as dead, so that another backend
6181 : * inserting a duplicate key value won't unnecessarily wait for our whole
6182 : * transaction to finish (it'll just wait for our speculative insertion to
6183 : * finish).
6184 : *
6185 : * Killing the tuple prevents "unprincipled deadlocks", which are deadlocks
6186 : * that arise due to a mutual dependency that is not user visible. By
6187 : * definition, unprincipled deadlocks cannot be prevented by the user
6188 : * reordering lock acquisition in client code, because the implementation level
6189 : * lock acquisitions are not under the user's direct control. If speculative
6190 : * inserters did not take this precaution, then under high concurrency they
6191 : * could deadlock with each other, which would not be acceptable.
6192 : *
6193 : * This is somewhat redundant with heap_delete, but we prefer to have a
6194 : * dedicated routine with stripped down requirements. Note that this is also
6195 : * used to delete the TOAST tuples created during speculative insertion.
6196 : *
6197 : * This routine does not affect logical decoding as it only looks at
6198 : * confirmation records.
6199 : */
6200 : void
6201 28 : heap_abort_speculative(Relation relation, const ItemPointerData *tid)
6202 : {
6203 28 : TransactionId xid = GetCurrentTransactionId();
6204 : ItemId lp;
6205 : HeapTupleData tp;
6206 : Page page;
6207 : BlockNumber block;
6208 : Buffer buffer;
6209 :
6210 : Assert(ItemPointerIsValid(tid));
6211 :
6212 28 : block = ItemPointerGetBlockNumber(tid);
6213 28 : buffer = ReadBuffer(relation, block);
6214 28 : page = BufferGetPage(buffer);
6215 :
6216 28 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6217 :
6218 : /*
6219 : * Page can't be all visible, we just inserted into it, and are still
6220 : * running.
6221 : */
6222 : Assert(!PageIsAllVisible(page));
6223 :
6224 28 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
6225 : Assert(ItemIdIsNormal(lp));
6226 :
6227 28 : tp.t_tableOid = RelationGetRelid(relation);
6228 28 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
6229 28 : tp.t_len = ItemIdGetLength(lp);
6230 28 : tp.t_self = *tid;
6231 :
6232 : /*
6233 : * Sanity check that the tuple really is a speculatively inserted tuple,
6234 : * inserted by us.
6235 : */
6236 28 : if (tp.t_data->t_choice.t_heap.t_xmin != xid)
6237 0 : elog(ERROR, "attempted to kill a tuple inserted by another transaction");
6238 28 : if (!(IsToastRelation(relation) || HeapTupleHeaderIsSpeculative(tp.t_data)))
6239 0 : elog(ERROR, "attempted to kill a non-speculative tuple");
6240 : Assert(!HeapTupleHeaderIsHeapOnly(tp.t_data));
6241 :
6242 : /*
6243 : * No need to check for serializable conflicts here. There is never a
6244 : * need for a combo CID, either. No need to extract replica identity, or
6245 : * do anything special with infomask bits.
6246 : */
6247 :
6248 28 : START_CRIT_SECTION();
6249 :
6250 : /*
6251 : * The tuple will become DEAD immediately. Flag that this page is a
6252 : * candidate for pruning by setting xmin to TransactionXmin. While not
6253 : * immediately prunable, it is the oldest xid we can cheaply determine
6254 : * that's safe against wraparound / being older than the table's
6255 : * relfrozenxid. To defend against the unlikely case of a new relation
6256 : * having a newer relfrozenxid than our TransactionXmin, use relfrozenxid
6257 : * if so (vacuum can't subsequently move relfrozenxid to beyond
6258 : * TransactionXmin, so there's no race here).
6259 : */
6260 : Assert(TransactionIdIsValid(TransactionXmin));
6261 : {
6262 28 : TransactionId relfrozenxid = relation->rd_rel->relfrozenxid;
6263 : TransactionId prune_xid;
6264 :
6265 28 : if (TransactionIdPrecedes(TransactionXmin, relfrozenxid))
6266 0 : prune_xid = relfrozenxid;
6267 : else
6268 28 : prune_xid = TransactionXmin;
6269 28 : PageSetPrunable(page, prune_xid);
6270 : }
6271 :
6272 : /* store transaction information of xact deleting the tuple */
6273 28 : tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
6274 28 : tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
6275 :
6276 : /*
6277 : * Set the tuple header xmin to InvalidTransactionId. This makes the
6278 : * tuple immediately invisible everyone. (In particular, to any
6279 : * transactions waiting on the speculative token, woken up later.)
6280 : */
6281 28 : HeapTupleHeaderSetXmin(tp.t_data, InvalidTransactionId);
6282 :
6283 : /* Clear the speculative insertion token too */
6284 28 : tp.t_data->t_ctid = tp.t_self;
6285 :
6286 28 : MarkBufferDirty(buffer);
6287 :
6288 : /*
6289 : * XLOG stuff
6290 : *
6291 : * The WAL records generated here match heap_delete(). The same recovery
6292 : * routines are used.
6293 : */
6294 28 : if (RelationNeedsWAL(relation))
6295 : {
6296 : xl_heap_delete xlrec;
6297 : XLogRecPtr recptr;
6298 :
6299 20 : xlrec.flags = XLH_DELETE_IS_SUPER;
6300 40 : xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
6301 20 : tp.t_data->t_infomask2);
6302 20 : xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
6303 20 : xlrec.xmax = xid;
6304 :
6305 20 : XLogBeginInsert();
6306 20 : XLogRegisterData(&xlrec, SizeOfHeapDelete);
6307 20 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
6308 :
6309 : /* No replica identity & replication origin logged */
6310 :
6311 20 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
6312 :
6313 20 : PageSetLSN(page, recptr);
6314 : }
6315 :
6316 28 : END_CRIT_SECTION();
6317 :
6318 28 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6319 :
6320 28 : if (HeapTupleHasExternal(&tp))
6321 : {
6322 : Assert(!IsToastRelation(relation));
6323 2 : heap_toast_delete(relation, &tp, true);
6324 : }
6325 :
6326 : /*
6327 : * Never need to mark tuple for invalidation, since catalogs don't support
6328 : * speculative insertion
6329 : */
6330 :
6331 : /* Now we can release the buffer */
6332 28 : ReleaseBuffer(buffer);
6333 :
6334 : /* count deletion, as we counted the insertion too */
6335 28 : pgstat_count_heap_delete(relation);
6336 28 : }
6337 :
6338 : /*
6339 : * heap_inplace_lock - protect inplace update from concurrent heap_update()
6340 : *
6341 : * Evaluate whether the tuple's state is compatible with a no-key update.
6342 : * Current transaction rowmarks are fine, as is KEY SHARE from any
6343 : * transaction. If compatible, return true with the buffer exclusive-locked,
6344 : * and the caller must release that by calling
6345 : * heap_inplace_update_and_unlock(), calling heap_inplace_unlock(), or raising
6346 : * an error. Otherwise, call release_callback(arg), wait for blocking
6347 : * transactions to end, and return false.
6348 : *
6349 : * Since this is intended for system catalogs and SERIALIZABLE doesn't cover
6350 : * DDL, this doesn't guarantee any particular predicate locking.
6351 : *
6352 : * One could modify this to return true for tuples with delete in progress,
6353 : * All inplace updaters take a lock that conflicts with DROP. If explicit
6354 : * "DELETE FROM pg_class" is in progress, we'll wait for it like we would an
6355 : * update.
6356 : *
6357 : * Readers of inplace-updated fields expect changes to those fields are
6358 : * durable. For example, vac_truncate_clog() reads datfrozenxid from
6359 : * pg_database tuples via catalog snapshots. A future snapshot must not
6360 : * return a lower datfrozenxid for the same database OID (lower in the
6361 : * FullTransactionIdPrecedes() sense). We achieve that since no update of a
6362 : * tuple can start while we hold a lock on its buffer. In cases like
6363 : * BEGIN;GRANT;CREATE INDEX;COMMIT we're inplace-updating a tuple visible only
6364 : * to this transaction. ROLLBACK then is one case where it's okay to lose
6365 : * inplace updates. (Restoring relhasindex=false on ROLLBACK is fine, since
6366 : * any concurrent CREATE INDEX would have blocked, then inplace-updated the
6367 : * committed tuple.)
6368 : *
6369 : * In principle, we could avoid waiting by overwriting every tuple in the
6370 : * updated tuple chain. Reader expectations permit updating a tuple only if
6371 : * it's aborted, is the tail of the chain, or we already updated the tuple
6372 : * referenced in its t_ctid. Hence, we would need to overwrite the tuples in
6373 : * order from tail to head. That would imply either (a) mutating all tuples
6374 : * in one critical section or (b) accepting a chance of partial completion.
6375 : * Partial completion of a relfrozenxid update would have the weird
6376 : * consequence that the table's next VACUUM could see the table's relfrozenxid
6377 : * move forward between vacuum_get_cutoffs() and finishing.
6378 : */
6379 : bool
6380 357166 : heap_inplace_lock(Relation relation,
6381 : HeapTuple oldtup_ptr, Buffer buffer,
6382 : void (*release_callback) (void *), void *arg)
6383 : {
6384 357166 : HeapTupleData oldtup = *oldtup_ptr; /* minimize diff vs. heap_update() */
6385 : TM_Result result;
6386 : bool ret;
6387 :
6388 : #ifdef USE_ASSERT_CHECKING
6389 : if (RelationGetRelid(relation) == RelationRelationId)
6390 : check_inplace_rel_lock(oldtup_ptr);
6391 : #endif
6392 :
6393 : Assert(BufferIsValid(buffer));
6394 :
6395 : /*
6396 : * Construct shared cache inval if necessary. Because we pass a tuple
6397 : * version without our own inplace changes or inplace changes other
6398 : * sessions complete while we wait for locks, inplace update mustn't
6399 : * change catcache lookup keys. But we aren't bothering with index
6400 : * updates either, so that's true a fortiori. After LockBuffer(), it
6401 : * would be too late, because this might reach a
6402 : * CatalogCacheInitializeCache() that locks "buffer".
6403 : */
6404 357166 : CacheInvalidateHeapTupleInplace(relation, oldtup_ptr, NULL);
6405 :
6406 357166 : LockTuple(relation, &oldtup.t_self, InplaceUpdateTupleLock);
6407 357166 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6408 :
6409 : /*----------
6410 : * Interpret HeapTupleSatisfiesUpdate() like heap_update() does, except:
6411 : *
6412 : * - wait unconditionally
6413 : * - already locked tuple above, since inplace needs that unconditionally
6414 : * - don't recheck header after wait: simpler to defer to next iteration
6415 : * - don't try to continue even if the updater aborts: likewise
6416 : * - no crosscheck
6417 : */
6418 357166 : result = HeapTupleSatisfiesUpdate(&oldtup, GetCurrentCommandId(false),
6419 : buffer);
6420 :
6421 357166 : if (result == TM_Invisible)
6422 : {
6423 : /* no known way this can happen */
6424 0 : ereport(ERROR,
6425 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6426 : errmsg_internal("attempted to overwrite invisible tuple")));
6427 : }
6428 357166 : else if (result == TM_SelfModified)
6429 : {
6430 : /*
6431 : * CREATE INDEX might reach this if an expression is silly enough to
6432 : * call e.g. SELECT ... FROM pg_class FOR SHARE. C code of other SQL
6433 : * statements might get here after a heap_update() of the same row, in
6434 : * the absence of an intervening CommandCounterIncrement().
6435 : */
6436 0 : ereport(ERROR,
6437 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6438 : errmsg("tuple to be updated was already modified by an operation triggered by the current command")));
6439 : }
6440 357166 : else if (result == TM_BeingModified)
6441 : {
6442 : TransactionId xwait;
6443 : uint16 infomask;
6444 :
6445 54 : xwait = HeapTupleHeaderGetRawXmax(oldtup.t_data);
6446 54 : infomask = oldtup.t_data->t_infomask;
6447 :
6448 54 : if (infomask & HEAP_XMAX_IS_MULTI)
6449 : {
6450 10 : LockTupleMode lockmode = LockTupleNoKeyExclusive;
6451 10 : MultiXactStatus mxact_status = MultiXactStatusNoKeyUpdate;
6452 : int remain;
6453 :
6454 10 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
6455 : lockmode, NULL))
6456 : {
6457 4 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6458 4 : release_callback(arg);
6459 4 : ret = false;
6460 4 : MultiXactIdWait((MultiXactId) xwait, mxact_status, infomask,
6461 : relation, &oldtup.t_self, XLTW_Update,
6462 : &remain);
6463 : }
6464 : else
6465 6 : ret = true;
6466 : }
6467 44 : else if (TransactionIdIsCurrentTransactionId(xwait))
6468 2 : ret = true;
6469 42 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask))
6470 2 : ret = true;
6471 : else
6472 : {
6473 40 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6474 40 : release_callback(arg);
6475 40 : ret = false;
6476 40 : XactLockTableWait(xwait, relation, &oldtup.t_self,
6477 : XLTW_Update);
6478 : }
6479 : }
6480 : else
6481 : {
6482 357112 : ret = (result == TM_Ok);
6483 357112 : if (!ret)
6484 : {
6485 2 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6486 2 : release_callback(arg);
6487 : }
6488 : }
6489 :
6490 : /*
6491 : * GetCatalogSnapshot() relies on invalidation messages to know when to
6492 : * take a new snapshot. COMMIT of xwait is responsible for sending the
6493 : * invalidation. We're not acquiring heavyweight locks sufficient to
6494 : * block if not yet sent, so we must take a new snapshot to ensure a later
6495 : * attempt has a fair chance. While we don't need this if xwait aborted,
6496 : * don't bother optimizing that.
6497 : */
6498 357166 : if (!ret)
6499 : {
6500 46 : UnlockTuple(relation, &oldtup.t_self, InplaceUpdateTupleLock);
6501 46 : ForgetInplace_Inval();
6502 46 : InvalidateCatalogSnapshot();
6503 : }
6504 357166 : return ret;
6505 : }
6506 :
6507 : /*
6508 : * heap_inplace_update_and_unlock - core of systable_inplace_update_finish
6509 : *
6510 : * The tuple cannot change size, and therefore its header fields and null
6511 : * bitmap (if any) don't change either.
6512 : *
6513 : * Since we hold LOCKTAG_TUPLE, no updater has a local copy of this tuple.
6514 : */
6515 : void
6516 163324 : heap_inplace_update_and_unlock(Relation relation,
6517 : HeapTuple oldtup, HeapTuple tuple,
6518 : Buffer buffer)
6519 : {
6520 163324 : HeapTupleHeader htup = oldtup->t_data;
6521 : uint32 oldlen;
6522 : uint32 newlen;
6523 : char *dst;
6524 : char *src;
6525 163324 : int nmsgs = 0;
6526 163324 : SharedInvalidationMessage *invalMessages = NULL;
6527 163324 : bool RelcacheInitFileInval = false;
6528 :
6529 : Assert(ItemPointerEquals(&oldtup->t_self, &tuple->t_self));
6530 163324 : oldlen = oldtup->t_len - htup->t_hoff;
6531 163324 : newlen = tuple->t_len - tuple->t_data->t_hoff;
6532 163324 : if (oldlen != newlen || htup->t_hoff != tuple->t_data->t_hoff)
6533 0 : elog(ERROR, "wrong tuple length");
6534 :
6535 163324 : dst = (char *) htup + htup->t_hoff;
6536 163324 : src = (char *) tuple->t_data + tuple->t_data->t_hoff;
6537 :
6538 : /* Like RecordTransactionCommit(), log only if needed */
6539 163324 : if (XLogStandbyInfoActive())
6540 99776 : nmsgs = inplaceGetInvalidationMessages(&invalMessages,
6541 : &RelcacheInitFileInval);
6542 :
6543 : /*
6544 : * Unlink relcache init files as needed. If unlinking, acquire
6545 : * RelCacheInitLock until after associated invalidations. By doing this
6546 : * in advance, if we checkpoint and then crash between inplace
6547 : * XLogInsert() and inval, we don't rely on StartupXLOG() ->
6548 : * RelationCacheInitFileRemove(). That uses elevel==LOG, so replay would
6549 : * neglect to PANIC on EIO.
6550 : */
6551 163324 : PreInplace_Inval();
6552 :
6553 : /*----------
6554 : * NO EREPORT(ERROR) from here till changes are complete
6555 : *
6556 : * Our buffer lock won't stop a reader having already pinned and checked
6557 : * visibility for this tuple. Hence, we write WAL first, then mutate the
6558 : * buffer. Like in MarkBufferDirtyHint() or RecordTransactionCommit(),
6559 : * checkpoint delay makes that acceptable. With the usual order of
6560 : * changes, a crash after memcpy() and before XLogInsert() could allow
6561 : * datfrozenxid to overtake relfrozenxid:
6562 : *
6563 : * ["D" is a VACUUM (ONLY_DATABASE_STATS)]
6564 : * ["R" is a VACUUM tbl]
6565 : * D: vac_update_datfrozenxid() -> systable_beginscan(pg_class)
6566 : * D: systable_getnext() returns pg_class tuple of tbl
6567 : * R: memcpy() into pg_class tuple of tbl
6568 : * D: raise pg_database.datfrozenxid, XLogInsert(), finish
6569 : * [crash]
6570 : * [recovery restores datfrozenxid w/o relfrozenxid]
6571 : *
6572 : * Mimic MarkBufferDirtyHint() subroutine XLogSaveBufferForHint().
6573 : * Specifically, use DELAY_CHKPT_START, and copy the buffer to the stack.
6574 : * The stack copy facilitates a FPI of the post-mutation block before we
6575 : * accept other sessions seeing it. DELAY_CHKPT_START allows us to
6576 : * XLogInsert() before MarkBufferDirty(). Since XLogSaveBufferForHint()
6577 : * can operate under BUFFER_LOCK_SHARED, it can't avoid DELAY_CHKPT_START.
6578 : * This function, however, likely could avoid it with the following order
6579 : * of operations: MarkBufferDirty(), XLogInsert(), memcpy(). Opt to use
6580 : * DELAY_CHKPT_START here, too, as a way to have fewer distinct code
6581 : * patterns to analyze. Inplace update isn't so frequent that it should
6582 : * pursue the small optimization of skipping DELAY_CHKPT_START.
6583 : */
6584 : Assert((MyProc->delayChkptFlags & DELAY_CHKPT_START) == 0);
6585 163324 : START_CRIT_SECTION();
6586 163324 : MyProc->delayChkptFlags |= DELAY_CHKPT_START;
6587 :
6588 : /* XLOG stuff */
6589 163324 : if (RelationNeedsWAL(relation))
6590 : {
6591 : xl_heap_inplace xlrec;
6592 : PGAlignedBlock copied_buffer;
6593 163308 : char *origdata = (char *) BufferGetBlock(buffer);
6594 163308 : Page page = BufferGetPage(buffer);
6595 163308 : uint16 lower = ((PageHeader) page)->pd_lower;
6596 163308 : uint16 upper = ((PageHeader) page)->pd_upper;
6597 : uintptr_t dst_offset_in_block;
6598 : RelFileLocator rlocator;
6599 : ForkNumber forkno;
6600 : BlockNumber blkno;
6601 : XLogRecPtr recptr;
6602 :
6603 163308 : xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
6604 163308 : xlrec.dbId = MyDatabaseId;
6605 163308 : xlrec.tsId = MyDatabaseTableSpace;
6606 163308 : xlrec.relcacheInitFileInval = RelcacheInitFileInval;
6607 163308 : xlrec.nmsgs = nmsgs;
6608 :
6609 163308 : XLogBeginInsert();
6610 163308 : XLogRegisterData(&xlrec, MinSizeOfHeapInplace);
6611 163308 : if (nmsgs != 0)
6612 70476 : XLogRegisterData(invalMessages,
6613 : nmsgs * sizeof(SharedInvalidationMessage));
6614 :
6615 : /* register block matching what buffer will look like after changes */
6616 163308 : memcpy(copied_buffer.data, origdata, lower);
6617 163308 : memcpy(copied_buffer.data + upper, origdata + upper, BLCKSZ - upper);
6618 163308 : dst_offset_in_block = dst - origdata;
6619 163308 : memcpy(copied_buffer.data + dst_offset_in_block, src, newlen);
6620 163308 : BufferGetTag(buffer, &rlocator, &forkno, &blkno);
6621 : Assert(forkno == MAIN_FORKNUM);
6622 163308 : XLogRegisterBlock(0, &rlocator, forkno, blkno, copied_buffer.data,
6623 : REGBUF_STANDARD);
6624 163308 : XLogRegisterBufData(0, src, newlen);
6625 :
6626 : /* inplace updates aren't decoded atm, don't log the origin */
6627 :
6628 163308 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_INPLACE);
6629 :
6630 163308 : PageSetLSN(page, recptr);
6631 : }
6632 :
6633 163324 : memcpy(dst, src, newlen);
6634 :
6635 163324 : MarkBufferDirty(buffer);
6636 :
6637 163324 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6638 :
6639 : /*
6640 : * Send invalidations to shared queue. SearchSysCacheLocked1() assumes we
6641 : * do this before UnlockTuple().
6642 : *
6643 : * If we're mutating a tuple visible only to this transaction, there's an
6644 : * equivalent transactional inval from the action that created the tuple,
6645 : * and this inval is superfluous.
6646 : */
6647 163324 : AtInplace_Inval();
6648 :
6649 163324 : MyProc->delayChkptFlags &= ~DELAY_CHKPT_START;
6650 163324 : END_CRIT_SECTION();
6651 163324 : UnlockTuple(relation, &tuple->t_self, InplaceUpdateTupleLock);
6652 :
6653 163324 : AcceptInvalidationMessages(); /* local processing of just-sent inval */
6654 :
6655 : /*
6656 : * Queue a transactional inval. The immediate invalidation we just sent
6657 : * is the only one known to be necessary. To reduce risk from the
6658 : * transition to immediate invalidation, continue sending a transactional
6659 : * invalidation like we've long done. Third-party code might rely on it.
6660 : */
6661 163324 : if (!IsBootstrapProcessingMode())
6662 134024 : CacheInvalidateHeapTuple(relation, tuple, NULL);
6663 163324 : }
6664 :
6665 : /*
6666 : * heap_inplace_unlock - reverse of heap_inplace_lock
6667 : */
6668 : void
6669 193796 : heap_inplace_unlock(Relation relation,
6670 : HeapTuple oldtup, Buffer buffer)
6671 : {
6672 193796 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6673 193796 : UnlockTuple(relation, &oldtup->t_self, InplaceUpdateTupleLock);
6674 193796 : ForgetInplace_Inval();
6675 193796 : }
6676 :
6677 : #define FRM_NOOP 0x0001
6678 : #define FRM_INVALIDATE_XMAX 0x0002
6679 : #define FRM_RETURN_IS_XID 0x0004
6680 : #define FRM_RETURN_IS_MULTI 0x0008
6681 : #define FRM_MARK_COMMITTED 0x0010
6682 :
6683 : /*
6684 : * FreezeMultiXactId
6685 : * Determine what to do during freezing when a tuple is marked by a
6686 : * MultiXactId.
6687 : *
6688 : * "flags" is an output value; it's used to tell caller what to do on return.
6689 : * "pagefrz" is an input/output value, used to manage page level freezing.
6690 : *
6691 : * Possible values that we can set in "flags":
6692 : * FRM_NOOP
6693 : * don't do anything -- keep existing Xmax
6694 : * FRM_INVALIDATE_XMAX
6695 : * mark Xmax as InvalidTransactionId and set XMAX_INVALID flag.
6696 : * FRM_RETURN_IS_XID
6697 : * The Xid return value is a single update Xid to set as xmax.
6698 : * FRM_MARK_COMMITTED
6699 : * Xmax can be marked as HEAP_XMAX_COMMITTED
6700 : * FRM_RETURN_IS_MULTI
6701 : * The return value is a new MultiXactId to set as new Xmax.
6702 : * (caller must obtain proper infomask bits using GetMultiXactIdHintBits)
6703 : *
6704 : * Caller delegates control of page freezing to us. In practice we always
6705 : * force freezing of caller's page unless FRM_NOOP processing is indicated.
6706 : * We help caller ensure that XIDs < FreezeLimit and MXIDs < MultiXactCutoff
6707 : * can never be left behind. We freely choose when and how to process each
6708 : * Multi, without ever violating the cutoff postconditions for freezing.
6709 : *
6710 : * It's useful to remove Multis on a proactive timeline (relative to freezing
6711 : * XIDs) to keep MultiXact member SLRU buffer misses to a minimum. It can also
6712 : * be cheaper in the short run, for us, since we too can avoid SLRU buffer
6713 : * misses through eager processing.
6714 : *
6715 : * NB: Creates a _new_ MultiXactId when FRM_RETURN_IS_MULTI is set, though only
6716 : * when FreezeLimit and/or MultiXactCutoff cutoffs leave us with no choice.
6717 : * This can usually be put off, which is usually enough to avoid it altogether.
6718 : * Allocating new multis during VACUUM should be avoided on general principle;
6719 : * only VACUUM can advance relminmxid, so allocating new Multis here comes with
6720 : * its own special risks.
6721 : *
6722 : * NB: Caller must maintain "no freeze" NewRelfrozenXid/NewRelminMxid trackers
6723 : * using heap_tuple_should_freeze when we haven't forced page-level freezing.
6724 : *
6725 : * NB: Caller should avoid needlessly calling heap_tuple_should_freeze when we
6726 : * have already forced page-level freezing, since that might incur the same
6727 : * SLRU buffer misses that we specifically intended to avoid by freezing.
6728 : */
6729 : static TransactionId
6730 14 : FreezeMultiXactId(MultiXactId multi, uint16 t_infomask,
6731 : const struct VacuumCutoffs *cutoffs, uint16 *flags,
6732 : HeapPageFreeze *pagefrz)
6733 : {
6734 : TransactionId newxmax;
6735 : MultiXactMember *members;
6736 : int nmembers;
6737 : bool need_replace;
6738 : int nnewmembers;
6739 : MultiXactMember *newmembers;
6740 : bool has_lockers;
6741 : TransactionId update_xid;
6742 : bool update_committed;
6743 : TransactionId FreezePageRelfrozenXid;
6744 :
6745 14 : *flags = 0;
6746 :
6747 : /* We should only be called in Multis */
6748 : Assert(t_infomask & HEAP_XMAX_IS_MULTI);
6749 :
6750 28 : if (!MultiXactIdIsValid(multi) ||
6751 14 : HEAP_LOCKED_UPGRADED(t_infomask))
6752 : {
6753 0 : *flags |= FRM_INVALIDATE_XMAX;
6754 0 : pagefrz->freeze_required = true;
6755 0 : return InvalidTransactionId;
6756 : }
6757 14 : else if (MultiXactIdPrecedes(multi, cutoffs->relminmxid))
6758 0 : ereport(ERROR,
6759 : (errcode(ERRCODE_DATA_CORRUPTED),
6760 : errmsg_internal("found multixact %u from before relminmxid %u",
6761 : multi, cutoffs->relminmxid)));
6762 14 : else if (MultiXactIdPrecedes(multi, cutoffs->OldestMxact))
6763 : {
6764 : TransactionId update_xact;
6765 :
6766 : /*
6767 : * This old multi cannot possibly have members still running, but
6768 : * verify just in case. If it was a locker only, it can be removed
6769 : * without any further consideration; but if it contained an update,
6770 : * we might need to preserve it.
6771 : */
6772 10 : if (MultiXactIdIsRunning(multi,
6773 10 : HEAP_XMAX_IS_LOCKED_ONLY(t_infomask)))
6774 0 : ereport(ERROR,
6775 : (errcode(ERRCODE_DATA_CORRUPTED),
6776 : errmsg_internal("multixact %u from before multi freeze cutoff %u found to be still running",
6777 : multi, cutoffs->OldestMxact)));
6778 :
6779 10 : if (HEAP_XMAX_IS_LOCKED_ONLY(t_infomask))
6780 : {
6781 10 : *flags |= FRM_INVALIDATE_XMAX;
6782 10 : pagefrz->freeze_required = true;
6783 10 : return InvalidTransactionId;
6784 : }
6785 :
6786 : /* replace multi with single XID for its updater? */
6787 0 : update_xact = MultiXactIdGetUpdateXid(multi, t_infomask);
6788 0 : if (TransactionIdPrecedes(update_xact, cutoffs->relfrozenxid))
6789 0 : ereport(ERROR,
6790 : (errcode(ERRCODE_DATA_CORRUPTED),
6791 : errmsg_internal("multixact %u contains update XID %u from before relfrozenxid %u",
6792 : multi, update_xact,
6793 : cutoffs->relfrozenxid)));
6794 0 : else if (TransactionIdPrecedes(update_xact, cutoffs->OldestXmin))
6795 : {
6796 : /*
6797 : * Updater XID has to have aborted (otherwise the tuple would have
6798 : * been pruned away instead, since updater XID is < OldestXmin).
6799 : * Just remove xmax.
6800 : */
6801 0 : if (TransactionIdDidCommit(update_xact))
6802 0 : ereport(ERROR,
6803 : (errcode(ERRCODE_DATA_CORRUPTED),
6804 : errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
6805 : multi, update_xact,
6806 : cutoffs->OldestXmin)));
6807 0 : *flags |= FRM_INVALIDATE_XMAX;
6808 0 : pagefrz->freeze_required = true;
6809 0 : return InvalidTransactionId;
6810 : }
6811 :
6812 : /* Have to keep updater XID as new xmax */
6813 0 : *flags |= FRM_RETURN_IS_XID;
6814 0 : pagefrz->freeze_required = true;
6815 0 : return update_xact;
6816 : }
6817 :
6818 : /*
6819 : * Some member(s) of this Multi may be below FreezeLimit xid cutoff, so we
6820 : * need to walk the whole members array to figure out what to do, if
6821 : * anything.
6822 : */
6823 : nmembers =
6824 4 : GetMultiXactIdMembers(multi, &members, false,
6825 4 : HEAP_XMAX_IS_LOCKED_ONLY(t_infomask));
6826 4 : if (nmembers <= 0)
6827 : {
6828 : /* Nothing worth keeping */
6829 0 : *flags |= FRM_INVALIDATE_XMAX;
6830 0 : pagefrz->freeze_required = true;
6831 0 : return InvalidTransactionId;
6832 : }
6833 :
6834 : /*
6835 : * The FRM_NOOP case is the only case where we might need to ratchet back
6836 : * FreezePageRelfrozenXid or FreezePageRelminMxid. It is also the only
6837 : * case where our caller might ratchet back its NoFreezePageRelfrozenXid
6838 : * or NoFreezePageRelminMxid "no freeze" trackers to deal with a multi.
6839 : * FRM_NOOP handling should result in the NewRelfrozenXid/NewRelminMxid
6840 : * trackers managed by VACUUM being ratcheting back by xmax to the degree
6841 : * required to make it safe to leave xmax undisturbed, independent of
6842 : * whether or not page freezing is triggered somewhere else.
6843 : *
6844 : * Our policy is to force freezing in every case other than FRM_NOOP,
6845 : * which obviates the need to maintain either set of trackers, anywhere.
6846 : * Every other case will reliably execute a freeze plan for xmax that
6847 : * either replaces xmax with an XID/MXID >= OldestXmin/OldestMxact, or
6848 : * sets xmax to an InvalidTransactionId XID, rendering xmax fully frozen.
6849 : * (VACUUM's NewRelfrozenXid/NewRelminMxid trackers are initialized with
6850 : * OldestXmin/OldestMxact, so later values never need to be tracked here.)
6851 : */
6852 4 : need_replace = false;
6853 4 : FreezePageRelfrozenXid = pagefrz->FreezePageRelfrozenXid;
6854 8 : for (int i = 0; i < nmembers; i++)
6855 : {
6856 6 : TransactionId xid = members[i].xid;
6857 :
6858 : Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
6859 :
6860 6 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
6861 : {
6862 : /* Can't violate the FreezeLimit postcondition */
6863 2 : need_replace = true;
6864 2 : break;
6865 : }
6866 4 : if (TransactionIdPrecedes(xid, FreezePageRelfrozenXid))
6867 0 : FreezePageRelfrozenXid = xid;
6868 : }
6869 :
6870 : /* Can't violate the MultiXactCutoff postcondition, either */
6871 4 : if (!need_replace)
6872 2 : need_replace = MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff);
6873 :
6874 4 : if (!need_replace)
6875 : {
6876 : /*
6877 : * vacuumlazy.c might ratchet back NewRelminMxid, NewRelfrozenXid, or
6878 : * both together to make it safe to retain this particular multi after
6879 : * freezing its page
6880 : */
6881 2 : *flags |= FRM_NOOP;
6882 2 : pagefrz->FreezePageRelfrozenXid = FreezePageRelfrozenXid;
6883 2 : if (MultiXactIdPrecedes(multi, pagefrz->FreezePageRelminMxid))
6884 0 : pagefrz->FreezePageRelminMxid = multi;
6885 2 : pfree(members);
6886 2 : return multi;
6887 : }
6888 :
6889 : /*
6890 : * Do a more thorough second pass over the multi to figure out which
6891 : * member XIDs actually need to be kept. Checking the precise status of
6892 : * individual members might even show that we don't need to keep anything.
6893 : * That is quite possible even though the Multi must be >= OldestMxact,
6894 : * since our second pass only keeps member XIDs when it's truly necessary;
6895 : * even member XIDs >= OldestXmin often won't be kept by second pass.
6896 : */
6897 2 : nnewmembers = 0;
6898 2 : newmembers = palloc(sizeof(MultiXactMember) * nmembers);
6899 2 : has_lockers = false;
6900 2 : update_xid = InvalidTransactionId;
6901 2 : update_committed = false;
6902 :
6903 : /*
6904 : * Determine whether to keep each member xid, or to ignore it instead
6905 : */
6906 6 : for (int i = 0; i < nmembers; i++)
6907 : {
6908 4 : TransactionId xid = members[i].xid;
6909 4 : MultiXactStatus mstatus = members[i].status;
6910 :
6911 : Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
6912 :
6913 4 : if (!ISUPDATE_from_mxstatus(mstatus))
6914 : {
6915 : /*
6916 : * Locker XID (not updater XID). We only keep lockers that are
6917 : * still running.
6918 : */
6919 8 : if (TransactionIdIsCurrentTransactionId(xid) ||
6920 4 : TransactionIdIsInProgress(xid))
6921 : {
6922 2 : if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
6923 0 : ereport(ERROR,
6924 : (errcode(ERRCODE_DATA_CORRUPTED),
6925 : errmsg_internal("multixact %u contains running locker XID %u from before removable cutoff %u",
6926 : multi, xid,
6927 : cutoffs->OldestXmin)));
6928 2 : newmembers[nnewmembers++] = members[i];
6929 2 : has_lockers = true;
6930 : }
6931 :
6932 4 : continue;
6933 : }
6934 :
6935 : /*
6936 : * Updater XID (not locker XID). Should we keep it?
6937 : *
6938 : * Since the tuple wasn't totally removed when vacuum pruned, the
6939 : * update Xid cannot possibly be older than OldestXmin cutoff unless
6940 : * the updater XID aborted. If the updater transaction is known
6941 : * aborted or crashed then it's okay to ignore it, otherwise not.
6942 : *
6943 : * In any case the Multi should never contain two updaters, whatever
6944 : * their individual commit status. Check for that first, in passing.
6945 : */
6946 0 : if (TransactionIdIsValid(update_xid))
6947 0 : ereport(ERROR,
6948 : (errcode(ERRCODE_DATA_CORRUPTED),
6949 : errmsg_internal("multixact %u has two or more updating members",
6950 : multi),
6951 : errdetail_internal("First updater XID=%u second updater XID=%u.",
6952 : update_xid, xid)));
6953 :
6954 : /*
6955 : * As with all tuple visibility routines, it's critical to test
6956 : * TransactionIdIsInProgress before TransactionIdDidCommit, because of
6957 : * race conditions explained in detail in heapam_visibility.c.
6958 : */
6959 0 : if (TransactionIdIsCurrentTransactionId(xid) ||
6960 0 : TransactionIdIsInProgress(xid))
6961 0 : update_xid = xid;
6962 0 : else if (TransactionIdDidCommit(xid))
6963 : {
6964 : /*
6965 : * The transaction committed, so we can tell caller to set
6966 : * HEAP_XMAX_COMMITTED. (We can only do this because we know the
6967 : * transaction is not running.)
6968 : */
6969 0 : update_committed = true;
6970 0 : update_xid = xid;
6971 : }
6972 : else
6973 : {
6974 : /*
6975 : * Not in progress, not committed -- must be aborted or crashed;
6976 : * we can ignore it.
6977 : */
6978 0 : continue;
6979 : }
6980 :
6981 : /*
6982 : * We determined that updater must be kept -- add it to pending new
6983 : * members list
6984 : */
6985 0 : if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
6986 0 : ereport(ERROR,
6987 : (errcode(ERRCODE_DATA_CORRUPTED),
6988 : errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
6989 : multi, xid, cutoffs->OldestXmin)));
6990 0 : newmembers[nnewmembers++] = members[i];
6991 : }
6992 :
6993 2 : pfree(members);
6994 :
6995 : /*
6996 : * Determine what to do with caller's multi based on information gathered
6997 : * during our second pass
6998 : */
6999 2 : if (nnewmembers == 0)
7000 : {
7001 : /* Nothing worth keeping */
7002 0 : *flags |= FRM_INVALIDATE_XMAX;
7003 0 : newxmax = InvalidTransactionId;
7004 : }
7005 2 : else if (TransactionIdIsValid(update_xid) && !has_lockers)
7006 : {
7007 : /*
7008 : * If there's a single member and it's an update, pass it back alone
7009 : * without creating a new Multi. (XXX we could do this when there's a
7010 : * single remaining locker, too, but that would complicate the API too
7011 : * much; moreover, the case with the single updater is more
7012 : * interesting, because those are longer-lived.)
7013 : */
7014 : Assert(nnewmembers == 1);
7015 0 : *flags |= FRM_RETURN_IS_XID;
7016 0 : if (update_committed)
7017 0 : *flags |= FRM_MARK_COMMITTED;
7018 0 : newxmax = update_xid;
7019 : }
7020 : else
7021 : {
7022 : /*
7023 : * Create a new multixact with the surviving members of the previous
7024 : * one, to set as new Xmax in the tuple
7025 : */
7026 2 : newxmax = MultiXactIdCreateFromMembers(nnewmembers, newmembers);
7027 2 : *flags |= FRM_RETURN_IS_MULTI;
7028 : }
7029 :
7030 2 : pfree(newmembers);
7031 :
7032 2 : pagefrz->freeze_required = true;
7033 2 : return newxmax;
7034 : }
7035 :
7036 : /*
7037 : * heap_prepare_freeze_tuple
7038 : *
7039 : * Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
7040 : * are older than the OldestXmin and/or OldestMxact freeze cutoffs. If so,
7041 : * setup enough state (in the *frz output argument) to enable caller to
7042 : * process this tuple as part of freezing its page, and return true. Return
7043 : * false if nothing can be changed about the tuple right now.
7044 : *
7045 : * Also sets *totally_frozen to true if the tuple will be totally frozen once
7046 : * caller executes returned freeze plan (or if the tuple was already totally
7047 : * frozen by an earlier VACUUM). This indicates that there are no remaining
7048 : * XIDs or MultiXactIds that will need to be processed by a future VACUUM.
7049 : *
7050 : * VACUUM caller must assemble HeapTupleFreeze freeze plan entries for every
7051 : * tuple that we returned true for, and then execute freezing. Caller must
7052 : * initialize pagefrz fields for page as a whole before first call here for
7053 : * each heap page.
7054 : *
7055 : * VACUUM caller decides on whether or not to freeze the page as a whole.
7056 : * We'll often prepare freeze plans for a page that caller just discards.
7057 : * However, VACUUM doesn't always get to make a choice; it must freeze when
7058 : * pagefrz.freeze_required is set, to ensure that any XIDs < FreezeLimit (and
7059 : * MXIDs < MultiXactCutoff) can never be left behind. We help to make sure
7060 : * that VACUUM always follows that rule.
7061 : *
7062 : * We sometimes force freezing of xmax MultiXactId values long before it is
7063 : * strictly necessary to do so just to ensure the FreezeLimit postcondition.
7064 : * It's worth processing MultiXactIds proactively when it is cheap to do so,
7065 : * and it's convenient to make that happen by piggy-backing it on the "force
7066 : * freezing" mechanism. Conversely, we sometimes delay freezing MultiXactIds
7067 : * because it is expensive right now (though only when it's still possible to
7068 : * do so without violating the FreezeLimit/MultiXactCutoff postcondition).
7069 : *
7070 : * It is assumed that the caller has checked the tuple with
7071 : * HeapTupleSatisfiesVacuum() and determined that it is not HEAPTUPLE_DEAD
7072 : * (else we should be removing the tuple, not freezing it).
7073 : *
7074 : * NB: This function has side effects: it might allocate a new MultiXactId.
7075 : * It will be set as tuple's new xmax when our *frz output is processed within
7076 : * heap_execute_freeze_tuple later on. If the tuple is in a shared buffer
7077 : * then caller had better have an exclusive lock on it already.
7078 : */
7079 : bool
7080 39062084 : heap_prepare_freeze_tuple(HeapTupleHeader tuple,
7081 : const struct VacuumCutoffs *cutoffs,
7082 : HeapPageFreeze *pagefrz,
7083 : HeapTupleFreeze *frz, bool *totally_frozen)
7084 : {
7085 39062084 : bool xmin_already_frozen = false,
7086 39062084 : xmax_already_frozen = false;
7087 39062084 : bool freeze_xmin = false,
7088 39062084 : replace_xvac = false,
7089 39062084 : replace_xmax = false,
7090 39062084 : freeze_xmax = false;
7091 : TransactionId xid;
7092 :
7093 39062084 : frz->xmax = HeapTupleHeaderGetRawXmax(tuple);
7094 39062084 : frz->t_infomask2 = tuple->t_infomask2;
7095 39062084 : frz->t_infomask = tuple->t_infomask;
7096 39062084 : frz->frzflags = 0;
7097 39062084 : frz->checkflags = 0;
7098 :
7099 : /*
7100 : * Process xmin, while keeping track of whether it's already frozen, or
7101 : * will become frozen iff our freeze plan is executed by caller (could be
7102 : * neither).
7103 : */
7104 39062084 : xid = HeapTupleHeaderGetXmin(tuple);
7105 39062084 : if (!TransactionIdIsNormal(xid))
7106 33307882 : xmin_already_frozen = true;
7107 : else
7108 : {
7109 5754202 : if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
7110 0 : ereport(ERROR,
7111 : (errcode(ERRCODE_DATA_CORRUPTED),
7112 : errmsg_internal("found xmin %u from before relfrozenxid %u",
7113 : xid, cutoffs->relfrozenxid)));
7114 :
7115 : /* Will set freeze_xmin flags in freeze plan below */
7116 5754202 : freeze_xmin = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
7117 :
7118 : /* Verify that xmin committed if and when freeze plan is executed */
7119 5754202 : if (freeze_xmin)
7120 4534658 : frz->checkflags |= HEAP_FREEZE_CHECK_XMIN_COMMITTED;
7121 : }
7122 :
7123 : /*
7124 : * Old-style VACUUM FULL is gone, but we have to process xvac for as long
7125 : * as we support having MOVED_OFF/MOVED_IN tuples in the database
7126 : */
7127 39062084 : xid = HeapTupleHeaderGetXvac(tuple);
7128 39062084 : if (TransactionIdIsNormal(xid))
7129 : {
7130 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7131 : Assert(TransactionIdPrecedes(xid, cutoffs->OldestXmin));
7132 :
7133 : /*
7134 : * For Xvac, we always freeze proactively. This allows totally_frozen
7135 : * tracking to ignore xvac.
7136 : */
7137 0 : replace_xvac = pagefrz->freeze_required = true;
7138 :
7139 : /* Will set replace_xvac flags in freeze plan below */
7140 : }
7141 :
7142 : /* Now process xmax */
7143 39062084 : xid = frz->xmax;
7144 39062084 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7145 : {
7146 : /* Raw xmax is a MultiXactId */
7147 : TransactionId newxmax;
7148 : uint16 flags;
7149 :
7150 : /*
7151 : * We will either remove xmax completely (in the "freeze_xmax" path),
7152 : * process xmax by replacing it (in the "replace_xmax" path), or
7153 : * perform no-op xmax processing. The only constraint is that the
7154 : * FreezeLimit/MultiXactCutoff postcondition must never be violated.
7155 : */
7156 14 : newxmax = FreezeMultiXactId(xid, tuple->t_infomask, cutoffs,
7157 : &flags, pagefrz);
7158 :
7159 14 : if (flags & FRM_NOOP)
7160 : {
7161 : /*
7162 : * xmax is a MultiXactId, and nothing about it changes for now.
7163 : * This is the only case where 'freeze_required' won't have been
7164 : * set for us by FreezeMultiXactId, as well as the only case where
7165 : * neither freeze_xmax nor replace_xmax are set (given a multi).
7166 : *
7167 : * This is a no-op, but the call to FreezeMultiXactId might have
7168 : * ratcheted back NewRelfrozenXid and/or NewRelminMxid trackers
7169 : * for us (the "freeze page" variants, specifically). That'll
7170 : * make it safe for our caller to freeze the page later on, while
7171 : * leaving this particular xmax undisturbed.
7172 : *
7173 : * FreezeMultiXactId is _not_ responsible for the "no freeze"
7174 : * NewRelfrozenXid/NewRelminMxid trackers, though -- that's our
7175 : * job. A call to heap_tuple_should_freeze for this same tuple
7176 : * will take place below if 'freeze_required' isn't set already.
7177 : * (This repeats work from FreezeMultiXactId, but allows "no
7178 : * freeze" tracker maintenance to happen in only one place.)
7179 : */
7180 : Assert(!MultiXactIdPrecedes(newxmax, cutoffs->MultiXactCutoff));
7181 : Assert(MultiXactIdIsValid(newxmax) && xid == newxmax);
7182 : }
7183 12 : else if (flags & FRM_RETURN_IS_XID)
7184 : {
7185 : /*
7186 : * xmax will become an updater Xid (original MultiXact's updater
7187 : * member Xid will be carried forward as a simple Xid in Xmax).
7188 : */
7189 : Assert(!TransactionIdPrecedes(newxmax, cutoffs->OldestXmin));
7190 :
7191 : /*
7192 : * NB -- some of these transformations are only valid because we
7193 : * know the return Xid is a tuple updater (i.e. not merely a
7194 : * locker.) Also note that the only reason we don't explicitly
7195 : * worry about HEAP_KEYS_UPDATED is because it lives in
7196 : * t_infomask2 rather than t_infomask.
7197 : */
7198 0 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7199 0 : frz->xmax = newxmax;
7200 0 : if (flags & FRM_MARK_COMMITTED)
7201 0 : frz->t_infomask |= HEAP_XMAX_COMMITTED;
7202 0 : replace_xmax = true;
7203 : }
7204 12 : else if (flags & FRM_RETURN_IS_MULTI)
7205 : {
7206 : uint16 newbits;
7207 : uint16 newbits2;
7208 :
7209 : /*
7210 : * xmax is an old MultiXactId that we have to replace with a new
7211 : * MultiXactId, to carry forward two or more original member XIDs.
7212 : */
7213 : Assert(!MultiXactIdPrecedes(newxmax, cutoffs->OldestMxact));
7214 :
7215 : /*
7216 : * We can't use GetMultiXactIdHintBits directly on the new multi
7217 : * here; that routine initializes the masks to all zeroes, which
7218 : * would lose other bits we need. Doing it this way ensures all
7219 : * unrelated bits remain untouched.
7220 : */
7221 2 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7222 2 : frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
7223 2 : GetMultiXactIdHintBits(newxmax, &newbits, &newbits2);
7224 2 : frz->t_infomask |= newbits;
7225 2 : frz->t_infomask2 |= newbits2;
7226 2 : frz->xmax = newxmax;
7227 2 : replace_xmax = true;
7228 : }
7229 : else
7230 : {
7231 : /*
7232 : * Freeze plan for tuple "freezes xmax" in the strictest sense:
7233 : * it'll leave nothing in xmax (neither an Xid nor a MultiXactId).
7234 : */
7235 : Assert(flags & FRM_INVALIDATE_XMAX);
7236 : Assert(!TransactionIdIsValid(newxmax));
7237 :
7238 : /* Will set freeze_xmax flags in freeze plan below */
7239 10 : freeze_xmax = true;
7240 : }
7241 :
7242 : /* MultiXactId processing forces freezing (barring FRM_NOOP case) */
7243 : Assert(pagefrz->freeze_required || (!freeze_xmax && !replace_xmax));
7244 : }
7245 39062070 : else if (TransactionIdIsNormal(xid))
7246 : {
7247 : /* Raw xmax is normal XID */
7248 11737412 : if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
7249 0 : ereport(ERROR,
7250 : (errcode(ERRCODE_DATA_CORRUPTED),
7251 : errmsg_internal("found xmax %u from before relfrozenxid %u",
7252 : xid, cutoffs->relfrozenxid)));
7253 :
7254 : /* Will set freeze_xmax flags in freeze plan below */
7255 11737412 : freeze_xmax = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
7256 :
7257 : /*
7258 : * Verify that xmax aborted if and when freeze plan is executed,
7259 : * provided it's from an update. (A lock-only xmax can be removed
7260 : * independent of this, since the lock is released at xact end.)
7261 : */
7262 11737412 : if (freeze_xmax && !HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask))
7263 1322 : frz->checkflags |= HEAP_FREEZE_CHECK_XMAX_ABORTED;
7264 : }
7265 27324658 : else if (!TransactionIdIsValid(xid))
7266 : {
7267 : /* Raw xmax is InvalidTransactionId XID */
7268 : Assert((tuple->t_infomask & HEAP_XMAX_IS_MULTI) == 0);
7269 27324658 : xmax_already_frozen = true;
7270 : }
7271 : else
7272 0 : ereport(ERROR,
7273 : (errcode(ERRCODE_DATA_CORRUPTED),
7274 : errmsg_internal("found raw xmax %u (infomask 0x%04x) not invalid and not multi",
7275 : xid, tuple->t_infomask)));
7276 :
7277 39062084 : if (freeze_xmin)
7278 : {
7279 : Assert(!xmin_already_frozen);
7280 :
7281 4534658 : frz->t_infomask |= HEAP_XMIN_FROZEN;
7282 : }
7283 39062084 : if (replace_xvac)
7284 : {
7285 : /*
7286 : * If a MOVED_OFF tuple is not dead, the xvac transaction must have
7287 : * failed; whereas a non-dead MOVED_IN tuple must mean the xvac
7288 : * transaction succeeded.
7289 : */
7290 : Assert(pagefrz->freeze_required);
7291 0 : if (tuple->t_infomask & HEAP_MOVED_OFF)
7292 0 : frz->frzflags |= XLH_INVALID_XVAC;
7293 : else
7294 0 : frz->frzflags |= XLH_FREEZE_XVAC;
7295 : }
7296 : if (replace_xmax)
7297 : {
7298 : Assert(!xmax_already_frozen && !freeze_xmax);
7299 : Assert(pagefrz->freeze_required);
7300 :
7301 : /* Already set replace_xmax flags in freeze plan earlier */
7302 : }
7303 39062084 : if (freeze_xmax)
7304 : {
7305 : Assert(!xmax_already_frozen && !replace_xmax);
7306 :
7307 3074 : frz->xmax = InvalidTransactionId;
7308 :
7309 : /*
7310 : * The tuple might be marked either XMAX_INVALID or XMAX_COMMITTED +
7311 : * LOCKED. Normalize to INVALID just to be sure no one gets confused.
7312 : * Also get rid of the HEAP_KEYS_UPDATED bit.
7313 : */
7314 3074 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7315 3074 : frz->t_infomask |= HEAP_XMAX_INVALID;
7316 3074 : frz->t_infomask2 &= ~HEAP_HOT_UPDATED;
7317 3074 : frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
7318 : }
7319 :
7320 : /*
7321 : * Determine if this tuple is already totally frozen, or will become
7322 : * totally frozen (provided caller executes freeze plans for the page)
7323 : */
7324 76901550 : *totally_frozen = ((freeze_xmin || xmin_already_frozen) &&
7325 37839466 : (freeze_xmax || xmax_already_frozen));
7326 :
7327 39062084 : if (!pagefrz->freeze_required && !(xmin_already_frozen &&
7328 : xmax_already_frozen))
7329 : {
7330 : /*
7331 : * So far no previous tuple from the page made freezing mandatory.
7332 : * Does this tuple force caller to freeze the entire page?
7333 : */
7334 14474848 : pagefrz->freeze_required =
7335 14474848 : heap_tuple_should_freeze(tuple, cutoffs,
7336 : &pagefrz->NoFreezePageRelfrozenXid,
7337 : &pagefrz->NoFreezePageRelminMxid);
7338 : }
7339 :
7340 : /* Tell caller if this tuple has a usable freeze plan set in *frz */
7341 39062084 : return freeze_xmin || replace_xvac || replace_xmax || freeze_xmax;
7342 : }
7343 :
7344 : /*
7345 : * Perform xmin/xmax XID status sanity checks before actually executing freeze
7346 : * plans.
7347 : *
7348 : * heap_prepare_freeze_tuple doesn't perform these checks directly because
7349 : * pg_xact lookups are relatively expensive. They shouldn't be repeated by
7350 : * successive VACUUMs that each decide against freezing the same page.
7351 : */
7352 : void
7353 44808 : heap_pre_freeze_checks(Buffer buffer,
7354 : HeapTupleFreeze *tuples, int ntuples)
7355 : {
7356 44808 : Page page = BufferGetPage(buffer);
7357 :
7358 1942564 : for (int i = 0; i < ntuples; i++)
7359 : {
7360 1897756 : HeapTupleFreeze *frz = tuples + i;
7361 1897756 : ItemId itemid = PageGetItemId(page, frz->offset);
7362 : HeapTupleHeader htup;
7363 :
7364 1897756 : htup = (HeapTupleHeader) PageGetItem(page, itemid);
7365 :
7366 : /* Deliberately avoid relying on tuple hint bits here */
7367 1897756 : if (frz->checkflags & HEAP_FREEZE_CHECK_XMIN_COMMITTED)
7368 : {
7369 1897754 : TransactionId xmin = HeapTupleHeaderGetRawXmin(htup);
7370 :
7371 : Assert(!HeapTupleHeaderXminFrozen(htup));
7372 1897754 : if (unlikely(!TransactionIdDidCommit(xmin)))
7373 0 : ereport(ERROR,
7374 : (errcode(ERRCODE_DATA_CORRUPTED),
7375 : errmsg_internal("uncommitted xmin %u needs to be frozen",
7376 : xmin)));
7377 : }
7378 :
7379 : /*
7380 : * TransactionIdDidAbort won't work reliably in the presence of XIDs
7381 : * left behind by transactions that were in progress during a crash,
7382 : * so we can only check that xmax didn't commit
7383 : */
7384 1897756 : if (frz->checkflags & HEAP_FREEZE_CHECK_XMAX_ABORTED)
7385 : {
7386 330 : TransactionId xmax = HeapTupleHeaderGetRawXmax(htup);
7387 :
7388 : Assert(TransactionIdIsNormal(xmax));
7389 330 : if (unlikely(TransactionIdDidCommit(xmax)))
7390 0 : ereport(ERROR,
7391 : (errcode(ERRCODE_DATA_CORRUPTED),
7392 : errmsg_internal("cannot freeze committed xmax %u",
7393 : xmax)));
7394 : }
7395 : }
7396 44808 : }
7397 :
7398 : /*
7399 : * Helper which executes freezing of one or more heap tuples on a page on
7400 : * behalf of caller. Caller passes an array of tuple plans from
7401 : * heap_prepare_freeze_tuple. Caller must set 'offset' in each plan for us.
7402 : * Must be called in a critical section that also marks the buffer dirty and,
7403 : * if needed, emits WAL.
7404 : */
7405 : void
7406 44808 : heap_freeze_prepared_tuples(Buffer buffer, HeapTupleFreeze *tuples, int ntuples)
7407 : {
7408 44808 : Page page = BufferGetPage(buffer);
7409 :
7410 1942564 : for (int i = 0; i < ntuples; i++)
7411 : {
7412 1897756 : HeapTupleFreeze *frz = tuples + i;
7413 1897756 : ItemId itemid = PageGetItemId(page, frz->offset);
7414 : HeapTupleHeader htup;
7415 :
7416 1897756 : htup = (HeapTupleHeader) PageGetItem(page, itemid);
7417 1897756 : heap_execute_freeze_tuple(htup, frz);
7418 : }
7419 44808 : }
7420 :
7421 : /*
7422 : * heap_freeze_tuple
7423 : * Freeze tuple in place, without WAL logging.
7424 : *
7425 : * Useful for callers like CLUSTER that perform their own WAL logging.
7426 : */
7427 : bool
7428 732682 : heap_freeze_tuple(HeapTupleHeader tuple,
7429 : TransactionId relfrozenxid, TransactionId relminmxid,
7430 : TransactionId FreezeLimit, TransactionId MultiXactCutoff)
7431 : {
7432 : HeapTupleFreeze frz;
7433 : bool do_freeze;
7434 : bool totally_frozen;
7435 : struct VacuumCutoffs cutoffs;
7436 : HeapPageFreeze pagefrz;
7437 :
7438 732682 : cutoffs.relfrozenxid = relfrozenxid;
7439 732682 : cutoffs.relminmxid = relminmxid;
7440 732682 : cutoffs.OldestXmin = FreezeLimit;
7441 732682 : cutoffs.OldestMxact = MultiXactCutoff;
7442 732682 : cutoffs.FreezeLimit = FreezeLimit;
7443 732682 : cutoffs.MultiXactCutoff = MultiXactCutoff;
7444 :
7445 732682 : pagefrz.freeze_required = true;
7446 732682 : pagefrz.FreezePageRelfrozenXid = FreezeLimit;
7447 732682 : pagefrz.FreezePageRelminMxid = MultiXactCutoff;
7448 732682 : pagefrz.NoFreezePageRelfrozenXid = FreezeLimit;
7449 732682 : pagefrz.NoFreezePageRelminMxid = MultiXactCutoff;
7450 :
7451 732682 : do_freeze = heap_prepare_freeze_tuple(tuple, &cutoffs,
7452 : &pagefrz, &frz, &totally_frozen);
7453 :
7454 : /*
7455 : * Note that because this is not a WAL-logged operation, we don't need to
7456 : * fill in the offset in the freeze record.
7457 : */
7458 :
7459 732682 : if (do_freeze)
7460 547996 : heap_execute_freeze_tuple(tuple, &frz);
7461 732682 : return do_freeze;
7462 : }
7463 :
7464 : /*
7465 : * For a given MultiXactId, return the hint bits that should be set in the
7466 : * tuple's infomask.
7467 : *
7468 : * Normally this should be called for a multixact that was just created, and
7469 : * so is on our local cache, so the GetMembers call is fast.
7470 : */
7471 : static void
7472 2374 : GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
7473 : uint16 *new_infomask2)
7474 : {
7475 : int nmembers;
7476 : MultiXactMember *members;
7477 : int i;
7478 2374 : uint16 bits = HEAP_XMAX_IS_MULTI;
7479 2374 : uint16 bits2 = 0;
7480 2374 : bool has_update = false;
7481 2374 : LockTupleMode strongest = LockTupleKeyShare;
7482 :
7483 : /*
7484 : * We only use this in multis we just created, so they cannot be values
7485 : * pre-pg_upgrade.
7486 : */
7487 2374 : nmembers = GetMultiXactIdMembers(multi, &members, false, false);
7488 :
7489 7246 : for (i = 0; i < nmembers; i++)
7490 : {
7491 : LockTupleMode mode;
7492 :
7493 : /*
7494 : * Remember the strongest lock mode held by any member of the
7495 : * multixact.
7496 : */
7497 4872 : mode = TUPLOCK_from_mxstatus(members[i].status);
7498 4872 : if (mode > strongest)
7499 1322 : strongest = mode;
7500 :
7501 : /* See what other bits we need */
7502 4872 : switch (members[i].status)
7503 : {
7504 4490 : case MultiXactStatusForKeyShare:
7505 : case MultiXactStatusForShare:
7506 : case MultiXactStatusForNoKeyUpdate:
7507 4490 : break;
7508 :
7509 104 : case MultiXactStatusForUpdate:
7510 104 : bits2 |= HEAP_KEYS_UPDATED;
7511 104 : break;
7512 :
7513 258 : case MultiXactStatusNoKeyUpdate:
7514 258 : has_update = true;
7515 258 : break;
7516 :
7517 20 : case MultiXactStatusUpdate:
7518 20 : bits2 |= HEAP_KEYS_UPDATED;
7519 20 : has_update = true;
7520 20 : break;
7521 : }
7522 : }
7523 :
7524 2374 : if (strongest == LockTupleExclusive ||
7525 : strongest == LockTupleNoKeyExclusive)
7526 438 : bits |= HEAP_XMAX_EXCL_LOCK;
7527 1936 : else if (strongest == LockTupleShare)
7528 878 : bits |= HEAP_XMAX_SHR_LOCK;
7529 1058 : else if (strongest == LockTupleKeyShare)
7530 1058 : bits |= HEAP_XMAX_KEYSHR_LOCK;
7531 :
7532 2374 : if (!has_update)
7533 2096 : bits |= HEAP_XMAX_LOCK_ONLY;
7534 :
7535 2374 : if (nmembers > 0)
7536 2374 : pfree(members);
7537 :
7538 2374 : *new_infomask = bits;
7539 2374 : *new_infomask2 = bits2;
7540 2374 : }
7541 :
7542 : /*
7543 : * MultiXactIdGetUpdateXid
7544 : *
7545 : * Given a multixact Xmax and corresponding infomask, which does not have the
7546 : * HEAP_XMAX_LOCK_ONLY bit set, obtain and return the Xid of the updating
7547 : * transaction.
7548 : *
7549 : * Caller is expected to check the status of the updating transaction, if
7550 : * necessary.
7551 : */
7552 : static TransactionId
7553 980 : MultiXactIdGetUpdateXid(TransactionId xmax, uint16 t_infomask)
7554 : {
7555 980 : TransactionId update_xact = InvalidTransactionId;
7556 : MultiXactMember *members;
7557 : int nmembers;
7558 :
7559 : Assert(!(t_infomask & HEAP_XMAX_LOCK_ONLY));
7560 : Assert(t_infomask & HEAP_XMAX_IS_MULTI);
7561 :
7562 : /*
7563 : * Since we know the LOCK_ONLY bit is not set, this cannot be a multi from
7564 : * pre-pg_upgrade.
7565 : */
7566 980 : nmembers = GetMultiXactIdMembers(xmax, &members, false, false);
7567 :
7568 980 : if (nmembers > 0)
7569 : {
7570 : int i;
7571 :
7572 2548 : for (i = 0; i < nmembers; i++)
7573 : {
7574 : /* Ignore lockers */
7575 2548 : if (!ISUPDATE_from_mxstatus(members[i].status))
7576 1568 : continue;
7577 :
7578 : /* there can be at most one updater */
7579 : Assert(update_xact == InvalidTransactionId);
7580 980 : update_xact = members[i].xid;
7581 : #ifndef USE_ASSERT_CHECKING
7582 :
7583 : /*
7584 : * in an assert-enabled build, walk the whole array to ensure
7585 : * there's no other updater.
7586 : */
7587 980 : break;
7588 : #endif
7589 : }
7590 :
7591 980 : pfree(members);
7592 : }
7593 :
7594 980 : return update_xact;
7595 : }
7596 :
7597 : /*
7598 : * HeapTupleGetUpdateXid
7599 : * As above, but use a HeapTupleHeader
7600 : *
7601 : * See also HeapTupleHeaderGetUpdateXid, which can be used without previously
7602 : * checking the hint bits.
7603 : */
7604 : TransactionId
7605 964 : HeapTupleGetUpdateXid(const HeapTupleHeaderData *tup)
7606 : {
7607 964 : return MultiXactIdGetUpdateXid(HeapTupleHeaderGetRawXmax(tup),
7608 964 : tup->t_infomask);
7609 : }
7610 :
7611 : /*
7612 : * Does the given multixact conflict with the current transaction grabbing a
7613 : * tuple lock of the given strength?
7614 : *
7615 : * The passed infomask pairs up with the given multixact in the tuple header.
7616 : *
7617 : * If current_is_member is not NULL, it is set to 'true' if the current
7618 : * transaction is a member of the given multixact.
7619 : */
7620 : static bool
7621 198 : DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
7622 : LockTupleMode lockmode, bool *current_is_member)
7623 : {
7624 : int nmembers;
7625 : MultiXactMember *members;
7626 198 : bool result = false;
7627 198 : LOCKMODE wanted = tupleLockExtraInfo[lockmode].hwlock;
7628 :
7629 198 : if (HEAP_LOCKED_UPGRADED(infomask))
7630 0 : return false;
7631 :
7632 198 : nmembers = GetMultiXactIdMembers(multi, &members, false,
7633 198 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
7634 198 : if (nmembers >= 0)
7635 : {
7636 : int i;
7637 :
7638 618 : for (i = 0; i < nmembers; i++)
7639 : {
7640 : TransactionId memxid;
7641 : LOCKMODE memlockmode;
7642 :
7643 434 : if (result && (current_is_member == NULL || *current_is_member))
7644 : break;
7645 :
7646 420 : memlockmode = LOCKMODE_from_mxstatus(members[i].status);
7647 :
7648 : /* ignore members from current xact (but track their presence) */
7649 420 : memxid = members[i].xid;
7650 420 : if (TransactionIdIsCurrentTransactionId(memxid))
7651 : {
7652 184 : if (current_is_member != NULL)
7653 156 : *current_is_member = true;
7654 184 : continue;
7655 : }
7656 236 : else if (result)
7657 16 : continue;
7658 :
7659 : /* ignore members that don't conflict with the lock we want */
7660 220 : if (!DoLockModesConflict(memlockmode, wanted))
7661 142 : continue;
7662 :
7663 78 : if (ISUPDATE_from_mxstatus(members[i].status))
7664 : {
7665 : /* ignore aborted updaters */
7666 34 : if (TransactionIdDidAbort(memxid))
7667 2 : continue;
7668 : }
7669 : else
7670 : {
7671 : /* ignore lockers-only that are no longer in progress */
7672 44 : if (!TransactionIdIsInProgress(memxid))
7673 14 : continue;
7674 : }
7675 :
7676 : /*
7677 : * Whatever remains are either live lockers that conflict with our
7678 : * wanted lock, and updaters that are not aborted. Those conflict
7679 : * with what we want. Set up to return true, but keep going to
7680 : * look for the current transaction among the multixact members,
7681 : * if needed.
7682 : */
7683 62 : result = true;
7684 : }
7685 198 : pfree(members);
7686 : }
7687 :
7688 198 : return result;
7689 : }
7690 :
7691 : /*
7692 : * Do_MultiXactIdWait
7693 : * Actual implementation for the two functions below.
7694 : *
7695 : * 'multi', 'status' and 'infomask' indicate what to sleep on (the status is
7696 : * needed to ensure we only sleep on conflicting members, and the infomask is
7697 : * used to optimize multixact access in case it's a lock-only multi); 'nowait'
7698 : * indicates whether to use conditional lock acquisition, to allow callers to
7699 : * fail if lock is unavailable. 'rel', 'ctid' and 'oper' are used to set up
7700 : * context information for error messages. 'remaining', if not NULL, receives
7701 : * the number of members that are still running, including any (non-aborted)
7702 : * subtransactions of our own transaction. 'logLockFailure' indicates whether
7703 : * to log details when a lock acquisition fails with 'nowait' enabled.
7704 : *
7705 : * We do this by sleeping on each member using XactLockTableWait. Any
7706 : * members that belong to the current backend are *not* waited for, however;
7707 : * this would not merely be useless but would lead to Assert failure inside
7708 : * XactLockTableWait. By the time this returns, it is certain that all
7709 : * transactions *of other backends* that were members of the MultiXactId
7710 : * that conflict with the requested status are dead (and no new ones can have
7711 : * been added, since it is not legal to add members to an existing
7712 : * MultiXactId).
7713 : *
7714 : * But by the time we finish sleeping, someone else may have changed the Xmax
7715 : * of the containing tuple, so the caller needs to iterate on us somehow.
7716 : *
7717 : * Note that in case we return false, the number of remaining members is
7718 : * not to be trusted.
7719 : */
7720 : static bool
7721 116 : Do_MultiXactIdWait(MultiXactId multi, MultiXactStatus status,
7722 : uint16 infomask, bool nowait,
7723 : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
7724 : int *remaining, bool logLockFailure)
7725 : {
7726 116 : bool result = true;
7727 : MultiXactMember *members;
7728 : int nmembers;
7729 116 : int remain = 0;
7730 :
7731 : /* for pre-pg_upgrade tuples, no need to sleep at all */
7732 116 : nmembers = HEAP_LOCKED_UPGRADED(infomask) ? -1 :
7733 116 : GetMultiXactIdMembers(multi, &members, false,
7734 116 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
7735 :
7736 116 : if (nmembers >= 0)
7737 : {
7738 : int i;
7739 :
7740 374 : for (i = 0; i < nmembers; i++)
7741 : {
7742 266 : TransactionId memxid = members[i].xid;
7743 266 : MultiXactStatus memstatus = members[i].status;
7744 :
7745 266 : if (TransactionIdIsCurrentTransactionId(memxid))
7746 : {
7747 48 : remain++;
7748 48 : continue;
7749 : }
7750 :
7751 218 : if (!DoLockModesConflict(LOCKMODE_from_mxstatus(memstatus),
7752 218 : LOCKMODE_from_mxstatus(status)))
7753 : {
7754 44 : if (remaining && TransactionIdIsInProgress(memxid))
7755 16 : remain++;
7756 44 : continue;
7757 : }
7758 :
7759 : /*
7760 : * This member conflicts with our multi, so we have to sleep (or
7761 : * return failure, if asked to avoid waiting.)
7762 : *
7763 : * Note that we don't set up an error context callback ourselves,
7764 : * but instead we pass the info down to XactLockTableWait. This
7765 : * might seem a bit wasteful because the context is set up and
7766 : * tore down for each member of the multixact, but in reality it
7767 : * should be barely noticeable, and it avoids duplicate code.
7768 : */
7769 174 : if (nowait)
7770 : {
7771 8 : result = ConditionalXactLockTableWait(memxid, logLockFailure);
7772 8 : if (!result)
7773 8 : break;
7774 : }
7775 : else
7776 166 : XactLockTableWait(memxid, rel, ctid, oper);
7777 : }
7778 :
7779 116 : pfree(members);
7780 : }
7781 :
7782 116 : if (remaining)
7783 20 : *remaining = remain;
7784 :
7785 116 : return result;
7786 : }
7787 :
7788 : /*
7789 : * MultiXactIdWait
7790 : * Sleep on a MultiXactId.
7791 : *
7792 : * By the time we finish sleeping, someone else may have changed the Xmax
7793 : * of the containing tuple, so the caller needs to iterate on us somehow.
7794 : *
7795 : * We return (in *remaining, if not NULL) the number of members that are still
7796 : * running, including any (non-aborted) subtransactions of our own transaction.
7797 : */
7798 : static void
7799 108 : MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
7800 : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
7801 : int *remaining)
7802 : {
7803 108 : (void) Do_MultiXactIdWait(multi, status, infomask, false,
7804 : rel, ctid, oper, remaining, false);
7805 108 : }
7806 :
7807 : /*
7808 : * ConditionalMultiXactIdWait
7809 : * As above, but only lock if we can get the lock without blocking.
7810 : *
7811 : * By the time we finish sleeping, someone else may have changed the Xmax
7812 : * of the containing tuple, so the caller needs to iterate on us somehow.
7813 : *
7814 : * If the multixact is now all gone, return true. Returns false if some
7815 : * transactions might still be running.
7816 : *
7817 : * We return (in *remaining, if not NULL) the number of members that are still
7818 : * running, including any (non-aborted) subtransactions of our own transaction.
7819 : */
7820 : static bool
7821 8 : ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status,
7822 : uint16 infomask, Relation rel, int *remaining,
7823 : bool logLockFailure)
7824 : {
7825 8 : return Do_MultiXactIdWait(multi, status, infomask, true,
7826 : rel, NULL, XLTW_None, remaining, logLockFailure);
7827 : }
7828 :
7829 : /*
7830 : * heap_tuple_needs_eventual_freeze
7831 : *
7832 : * Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
7833 : * will eventually require freezing (if tuple isn't removed by pruning first).
7834 : */
7835 : bool
7836 253704 : heap_tuple_needs_eventual_freeze(HeapTupleHeader tuple)
7837 : {
7838 : TransactionId xid;
7839 :
7840 : /*
7841 : * If xmin is a normal transaction ID, this tuple is definitely not
7842 : * frozen.
7843 : */
7844 253704 : xid = HeapTupleHeaderGetXmin(tuple);
7845 253704 : if (TransactionIdIsNormal(xid))
7846 4844 : return true;
7847 :
7848 : /*
7849 : * If xmax is a valid xact or multixact, this tuple is also not frozen.
7850 : */
7851 248860 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7852 : {
7853 : MultiXactId multi;
7854 :
7855 0 : multi = HeapTupleHeaderGetRawXmax(tuple);
7856 0 : if (MultiXactIdIsValid(multi))
7857 0 : return true;
7858 : }
7859 : else
7860 : {
7861 248860 : xid = HeapTupleHeaderGetRawXmax(tuple);
7862 248860 : if (TransactionIdIsNormal(xid))
7863 14 : return true;
7864 : }
7865 :
7866 248846 : if (tuple->t_infomask & HEAP_MOVED)
7867 : {
7868 0 : xid = HeapTupleHeaderGetXvac(tuple);
7869 0 : if (TransactionIdIsNormal(xid))
7870 0 : return true;
7871 : }
7872 :
7873 248846 : return false;
7874 : }
7875 :
7876 : /*
7877 : * heap_tuple_should_freeze
7878 : *
7879 : * Return value indicates if heap_prepare_freeze_tuple sibling function would
7880 : * (or should) force freezing of the heap page that contains caller's tuple.
7881 : * Tuple header XIDs/MXIDs < FreezeLimit/MultiXactCutoff trigger freezing.
7882 : * This includes (xmin, xmax, xvac) fields, as well as MultiXact member XIDs.
7883 : *
7884 : * The *NoFreezePageRelfrozenXid and *NoFreezePageRelminMxid input/output
7885 : * arguments help VACUUM track the oldest extant XID/MXID remaining in rel.
7886 : * Our working assumption is that caller won't decide to freeze this tuple.
7887 : * It's up to caller to only ratchet back its own top-level trackers after the
7888 : * point that it fully commits to not freezing the tuple/page in question.
7889 : */
7890 : bool
7891 14478970 : heap_tuple_should_freeze(HeapTupleHeader tuple,
7892 : const struct VacuumCutoffs *cutoffs,
7893 : TransactionId *NoFreezePageRelfrozenXid,
7894 : MultiXactId *NoFreezePageRelminMxid)
7895 : {
7896 : TransactionId xid;
7897 : MultiXactId multi;
7898 14478970 : bool freeze = false;
7899 :
7900 : /* First deal with xmin */
7901 14478970 : xid = HeapTupleHeaderGetXmin(tuple);
7902 14478970 : if (TransactionIdIsNormal(xid))
7903 : {
7904 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7905 3401082 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7906 45660 : *NoFreezePageRelfrozenXid = xid;
7907 3401082 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7908 41908 : freeze = true;
7909 : }
7910 :
7911 : /* Now deal with xmax */
7912 14478970 : xid = InvalidTransactionId;
7913 14478970 : multi = InvalidMultiXactId;
7914 14478970 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7915 4 : multi = HeapTupleHeaderGetRawXmax(tuple);
7916 : else
7917 14478966 : xid = HeapTupleHeaderGetRawXmax(tuple);
7918 :
7919 14478970 : if (TransactionIdIsNormal(xid))
7920 : {
7921 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7922 : /* xmax is a non-permanent XID */
7923 11588786 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7924 4 : *NoFreezePageRelfrozenXid = xid;
7925 11588786 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7926 52 : freeze = true;
7927 : }
7928 2890184 : else if (!MultiXactIdIsValid(multi))
7929 : {
7930 : /* xmax is a permanent XID or invalid MultiXactId/XID */
7931 : }
7932 4 : else if (HEAP_LOCKED_UPGRADED(tuple->t_infomask))
7933 : {
7934 : /* xmax is a pg_upgrade'd MultiXact, which can't have updater XID */
7935 0 : if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
7936 0 : *NoFreezePageRelminMxid = multi;
7937 : /* heap_prepare_freeze_tuple always freezes pg_upgrade'd xmax */
7938 0 : freeze = true;
7939 : }
7940 : else
7941 : {
7942 : /* xmax is a MultiXactId that may have an updater XID */
7943 : MultiXactMember *members;
7944 : int nmembers;
7945 :
7946 : Assert(MultiXactIdPrecedesOrEquals(cutoffs->relminmxid, multi));
7947 4 : if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
7948 4 : *NoFreezePageRelminMxid = multi;
7949 4 : if (MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff))
7950 4 : freeze = true;
7951 :
7952 : /* need to check whether any member of the mxact is old */
7953 4 : nmembers = GetMultiXactIdMembers(multi, &members, false,
7954 4 : HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask));
7955 :
7956 10 : for (int i = 0; i < nmembers; i++)
7957 : {
7958 6 : xid = members[i].xid;
7959 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7960 6 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7961 0 : *NoFreezePageRelfrozenXid = xid;
7962 6 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7963 0 : freeze = true;
7964 : }
7965 4 : if (nmembers > 0)
7966 2 : pfree(members);
7967 : }
7968 :
7969 14478970 : if (tuple->t_infomask & HEAP_MOVED)
7970 : {
7971 0 : xid = HeapTupleHeaderGetXvac(tuple);
7972 0 : if (TransactionIdIsNormal(xid))
7973 : {
7974 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7975 0 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7976 0 : *NoFreezePageRelfrozenXid = xid;
7977 : /* heap_prepare_freeze_tuple forces xvac freezing */
7978 0 : freeze = true;
7979 : }
7980 : }
7981 :
7982 14478970 : return freeze;
7983 : }
7984 :
7985 : /*
7986 : * Maintain snapshotConflictHorizon for caller by ratcheting forward its value
7987 : * using any committed XIDs contained in 'tuple', an obsolescent heap tuple
7988 : * that caller is in the process of physically removing, e.g. via HOT pruning
7989 : * or index deletion.
7990 : *
7991 : * Caller must initialize its value to InvalidTransactionId, which is
7992 : * generally interpreted as "definitely no need for a recovery conflict".
7993 : * Final value must reflect all heap tuples that caller will physically remove
7994 : * (or remove TID references to) via its ongoing pruning/deletion operation.
7995 : * ResolveRecoveryConflictWithSnapshot() is passed the final value (taken from
7996 : * caller's WAL record) by REDO routine when it replays caller's operation.
7997 : */
7998 : void
7999 3060040 : HeapTupleHeaderAdvanceConflictHorizon(HeapTupleHeader tuple,
8000 : TransactionId *snapshotConflictHorizon)
8001 : {
8002 3060040 : TransactionId xmin = HeapTupleHeaderGetXmin(tuple);
8003 3060040 : TransactionId xmax = HeapTupleHeaderGetUpdateXid(tuple);
8004 3060040 : TransactionId xvac = HeapTupleHeaderGetXvac(tuple);
8005 :
8006 3060040 : if (tuple->t_infomask & HEAP_MOVED)
8007 : {
8008 0 : if (TransactionIdPrecedes(*snapshotConflictHorizon, xvac))
8009 0 : *snapshotConflictHorizon = xvac;
8010 : }
8011 :
8012 : /*
8013 : * Ignore tuples inserted by an aborted transaction or if the tuple was
8014 : * updated/deleted by the inserting transaction.
8015 : *
8016 : * Look for a committed hint bit, or if no xmin bit is set, check clog.
8017 : */
8018 3060040 : if (HeapTupleHeaderXminCommitted(tuple) ||
8019 205524 : (!HeapTupleHeaderXminInvalid(tuple) && TransactionIdDidCommit(xmin)))
8020 : {
8021 5501074 : if (xmax != xmin &&
8022 2595754 : TransactionIdFollows(xmax, *snapshotConflictHorizon))
8023 193456 : *snapshotConflictHorizon = xmax;
8024 : }
8025 3060040 : }
8026 :
8027 : #ifdef USE_PREFETCH
8028 : /*
8029 : * Helper function for heap_index_delete_tuples. Issues prefetch requests for
8030 : * prefetch_count buffers. The prefetch_state keeps track of all the buffers
8031 : * we can prefetch, and which have already been prefetched; each call to this
8032 : * function picks up where the previous call left off.
8033 : *
8034 : * Note: we expect the deltids array to be sorted in an order that groups TIDs
8035 : * by heap block, with all TIDs for each block appearing together in exactly
8036 : * one group.
8037 : */
8038 : static void
8039 38438 : index_delete_prefetch_buffer(Relation rel,
8040 : IndexDeletePrefetchState *prefetch_state,
8041 : int prefetch_count)
8042 : {
8043 38438 : BlockNumber cur_hblkno = prefetch_state->cur_hblkno;
8044 38438 : int count = 0;
8045 : int i;
8046 38438 : int ndeltids = prefetch_state->ndeltids;
8047 38438 : TM_IndexDelete *deltids = prefetch_state->deltids;
8048 :
8049 38438 : for (i = prefetch_state->next_item;
8050 1334180 : i < ndeltids && count < prefetch_count;
8051 1295742 : i++)
8052 : {
8053 1295742 : ItemPointer htid = &deltids[i].tid;
8054 :
8055 2580046 : if (cur_hblkno == InvalidBlockNumber ||
8056 1284304 : ItemPointerGetBlockNumber(htid) != cur_hblkno)
8057 : {
8058 34972 : cur_hblkno = ItemPointerGetBlockNumber(htid);
8059 34972 : PrefetchBuffer(rel, MAIN_FORKNUM, cur_hblkno);
8060 34972 : count++;
8061 : }
8062 : }
8063 :
8064 : /*
8065 : * Save the prefetch position so that next time we can continue from that
8066 : * position.
8067 : */
8068 38438 : prefetch_state->next_item = i;
8069 38438 : prefetch_state->cur_hblkno = cur_hblkno;
8070 38438 : }
8071 : #endif
8072 :
8073 : /*
8074 : * Helper function for heap_index_delete_tuples. Checks for index corruption
8075 : * involving an invalid TID in index AM caller's index page.
8076 : *
8077 : * This is an ideal place for these checks. The index AM must hold a buffer
8078 : * lock on the index page containing the TIDs we examine here, so we don't
8079 : * have to worry about concurrent VACUUMs at all. We can be sure that the
8080 : * index is corrupt when htid points directly to an LP_UNUSED item or
8081 : * heap-only tuple, which is not the case during standard index scans.
8082 : */
8083 : static inline void
8084 1073430 : index_delete_check_htid(TM_IndexDeleteOp *delstate,
8085 : Page page, OffsetNumber maxoff,
8086 : const ItemPointerData *htid, TM_IndexStatus *istatus)
8087 : {
8088 1073430 : OffsetNumber indexpagehoffnum = ItemPointerGetOffsetNumber(htid);
8089 : ItemId iid;
8090 :
8091 : Assert(OffsetNumberIsValid(istatus->idxoffnum));
8092 :
8093 1073430 : if (unlikely(indexpagehoffnum > maxoff))
8094 0 : ereport(ERROR,
8095 : (errcode(ERRCODE_INDEX_CORRUPTED),
8096 : 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\"",
8097 : ItemPointerGetBlockNumber(htid),
8098 : indexpagehoffnum,
8099 : istatus->idxoffnum, delstate->iblknum,
8100 : RelationGetRelationName(delstate->irel))));
8101 :
8102 1073430 : iid = PageGetItemId(page, indexpagehoffnum);
8103 1073430 : if (unlikely(!ItemIdIsUsed(iid)))
8104 0 : ereport(ERROR,
8105 : (errcode(ERRCODE_INDEX_CORRUPTED),
8106 : errmsg_internal("heap tid from index tuple (%u,%u) points to unused heap page item at offset %u of block %u in index \"%s\"",
8107 : ItemPointerGetBlockNumber(htid),
8108 : indexpagehoffnum,
8109 : istatus->idxoffnum, delstate->iblknum,
8110 : RelationGetRelationName(delstate->irel))));
8111 :
8112 1073430 : if (ItemIdHasStorage(iid))
8113 : {
8114 : HeapTupleHeader htup;
8115 :
8116 : Assert(ItemIdIsNormal(iid));
8117 636048 : htup = (HeapTupleHeader) PageGetItem(page, iid);
8118 :
8119 636048 : if (unlikely(HeapTupleHeaderIsHeapOnly(htup)))
8120 0 : ereport(ERROR,
8121 : (errcode(ERRCODE_INDEX_CORRUPTED),
8122 : errmsg_internal("heap tid from index tuple (%u,%u) points to heap-only tuple at offset %u of block %u in index \"%s\"",
8123 : ItemPointerGetBlockNumber(htid),
8124 : indexpagehoffnum,
8125 : istatus->idxoffnum, delstate->iblknum,
8126 : RelationGetRelationName(delstate->irel))));
8127 : }
8128 1073430 : }
8129 :
8130 : /*
8131 : * heapam implementation of tableam's index_delete_tuples interface.
8132 : *
8133 : * This helper function is called by index AMs during index tuple deletion.
8134 : * See tableam header comments for an explanation of the interface implemented
8135 : * here and a general theory of operation. Note that each call here is either
8136 : * a simple index deletion call, or a bottom-up index deletion call.
8137 : *
8138 : * It's possible for this to generate a fair amount of I/O, since we may be
8139 : * deleting hundreds of tuples from a single index block. To amortize that
8140 : * cost to some degree, this uses prefetching and combines repeat accesses to
8141 : * the same heap block.
8142 : */
8143 : TransactionId
8144 11438 : heap_index_delete_tuples(Relation rel, TM_IndexDeleteOp *delstate)
8145 : {
8146 : /* Initial assumption is that earlier pruning took care of conflict */
8147 11438 : TransactionId snapshotConflictHorizon = InvalidTransactionId;
8148 11438 : BlockNumber blkno = InvalidBlockNumber;
8149 11438 : Buffer buf = InvalidBuffer;
8150 11438 : Page page = NULL;
8151 11438 : OffsetNumber maxoff = InvalidOffsetNumber;
8152 : TransactionId priorXmax;
8153 : #ifdef USE_PREFETCH
8154 : IndexDeletePrefetchState prefetch_state;
8155 : int prefetch_distance;
8156 : #endif
8157 : SnapshotData SnapshotNonVacuumable;
8158 11438 : int finalndeltids = 0,
8159 11438 : nblocksaccessed = 0;
8160 :
8161 : /* State that's only used in bottom-up index deletion case */
8162 11438 : int nblocksfavorable = 0;
8163 11438 : int curtargetfreespace = delstate->bottomupfreespace,
8164 11438 : lastfreespace = 0,
8165 11438 : actualfreespace = 0;
8166 11438 : bool bottomup_final_block = false;
8167 :
8168 11438 : InitNonVacuumableSnapshot(SnapshotNonVacuumable, GlobalVisTestFor(rel));
8169 :
8170 : /* Sort caller's deltids array by TID for further processing */
8171 11438 : index_delete_sort(delstate);
8172 :
8173 : /*
8174 : * Bottom-up case: resort deltids array in an order attuned to where the
8175 : * greatest number of promising TIDs are to be found, and determine how
8176 : * many blocks from the start of sorted array should be considered
8177 : * favorable. This will also shrink the deltids array in order to
8178 : * eliminate completely unfavorable blocks up front.
8179 : */
8180 11438 : if (delstate->bottomup)
8181 3866 : nblocksfavorable = bottomup_sort_and_shrink(delstate);
8182 :
8183 : #ifdef USE_PREFETCH
8184 : /* Initialize prefetch state. */
8185 11438 : prefetch_state.cur_hblkno = InvalidBlockNumber;
8186 11438 : prefetch_state.next_item = 0;
8187 11438 : prefetch_state.ndeltids = delstate->ndeltids;
8188 11438 : prefetch_state.deltids = delstate->deltids;
8189 :
8190 : /*
8191 : * Determine the prefetch distance that we will attempt to maintain.
8192 : *
8193 : * Since the caller holds a buffer lock somewhere in rel, we'd better make
8194 : * sure that isn't a catalog relation before we call code that does
8195 : * syscache lookups, to avoid risk of deadlock.
8196 : */
8197 11438 : if (IsCatalogRelation(rel))
8198 8248 : prefetch_distance = maintenance_io_concurrency;
8199 : else
8200 : prefetch_distance =
8201 3190 : get_tablespace_maintenance_io_concurrency(rel->rd_rel->reltablespace);
8202 :
8203 : /* Cap initial prefetch distance for bottom-up deletion caller */
8204 11438 : if (delstate->bottomup)
8205 : {
8206 : Assert(nblocksfavorable >= 1);
8207 : Assert(nblocksfavorable <= BOTTOMUP_MAX_NBLOCKS);
8208 3866 : prefetch_distance = Min(prefetch_distance, nblocksfavorable);
8209 : }
8210 :
8211 : /* Start prefetching. */
8212 11438 : index_delete_prefetch_buffer(rel, &prefetch_state, prefetch_distance);
8213 : #endif
8214 :
8215 : /* Iterate over deltids, determine which to delete, check their horizon */
8216 : Assert(delstate->ndeltids > 0);
8217 1084868 : for (int i = 0; i < delstate->ndeltids; i++)
8218 : {
8219 1077296 : TM_IndexDelete *ideltid = &delstate->deltids[i];
8220 1077296 : TM_IndexStatus *istatus = delstate->status + ideltid->id;
8221 1077296 : ItemPointer htid = &ideltid->tid;
8222 : OffsetNumber offnum;
8223 :
8224 : /*
8225 : * Read buffer, and perform required extra steps each time a new block
8226 : * is encountered. Avoid refetching if it's the same block as the one
8227 : * from the last htid.
8228 : */
8229 2143154 : if (blkno == InvalidBlockNumber ||
8230 1065858 : ItemPointerGetBlockNumber(htid) != blkno)
8231 : {
8232 : /*
8233 : * Consider giving up early for bottom-up index deletion caller
8234 : * first. (Only prefetch next-next block afterwards, when it
8235 : * becomes clear that we're at least going to access the next
8236 : * block in line.)
8237 : *
8238 : * Sometimes the first block frees so much space for bottom-up
8239 : * caller that the deletion process can end without accessing any
8240 : * more blocks. It is usually necessary to access 2 or 3 blocks
8241 : * per bottom-up deletion operation, though.
8242 : */
8243 30866 : if (delstate->bottomup)
8244 : {
8245 : /*
8246 : * We often allow caller to delete a few additional items
8247 : * whose entries we reached after the point that space target
8248 : * from caller was satisfied. The cost of accessing the page
8249 : * was already paid at that point, so it made sense to finish
8250 : * it off. When that happened, we finalize everything here
8251 : * (by finishing off the whole bottom-up deletion operation
8252 : * without needlessly paying the cost of accessing any more
8253 : * blocks).
8254 : */
8255 8526 : if (bottomup_final_block)
8256 282 : break;
8257 :
8258 : /*
8259 : * Give up when we didn't enable our caller to free any
8260 : * additional space as a result of processing the page that we
8261 : * just finished up with. This rule is the main way in which
8262 : * we keep the cost of bottom-up deletion under control.
8263 : */
8264 8244 : if (nblocksaccessed >= 1 && actualfreespace == lastfreespace)
8265 3584 : break;
8266 4660 : lastfreespace = actualfreespace; /* for next time */
8267 :
8268 : /*
8269 : * Deletion operation (which is bottom-up) will definitely
8270 : * access the next block in line. Prepare for that now.
8271 : *
8272 : * Decay target free space so that we don't hang on for too
8273 : * long with a marginal case. (Space target is only truly
8274 : * helpful when it allows us to recognize that we don't need
8275 : * to access more than 1 or 2 blocks to satisfy caller due to
8276 : * agreeable workload characteristics.)
8277 : *
8278 : * We are a bit more patient when we encounter contiguous
8279 : * blocks, though: these are treated as favorable blocks. The
8280 : * decay process is only applied when the next block in line
8281 : * is not a favorable/contiguous block. This is not an
8282 : * exception to the general rule; we still insist on finding
8283 : * at least one deletable item per block accessed. See
8284 : * bottomup_nblocksfavorable() for full details of the theory
8285 : * behind favorable blocks and heap block locality in general.
8286 : *
8287 : * Note: The first block in line is always treated as a
8288 : * favorable block, so the earliest possible point that the
8289 : * decay can be applied is just before we access the second
8290 : * block in line. The Assert() verifies this for us.
8291 : */
8292 : Assert(nblocksaccessed > 0 || nblocksfavorable > 0);
8293 4660 : if (nblocksfavorable > 0)
8294 4188 : nblocksfavorable--;
8295 : else
8296 472 : curtargetfreespace /= 2;
8297 : }
8298 :
8299 : /* release old buffer */
8300 27000 : if (BufferIsValid(buf))
8301 15562 : UnlockReleaseBuffer(buf);
8302 :
8303 27000 : blkno = ItemPointerGetBlockNumber(htid);
8304 27000 : buf = ReadBuffer(rel, blkno);
8305 27000 : nblocksaccessed++;
8306 : Assert(!delstate->bottomup ||
8307 : nblocksaccessed <= BOTTOMUP_MAX_NBLOCKS);
8308 :
8309 : #ifdef USE_PREFETCH
8310 :
8311 : /*
8312 : * To maintain the prefetch distance, prefetch one more page for
8313 : * each page we read.
8314 : */
8315 27000 : index_delete_prefetch_buffer(rel, &prefetch_state, 1);
8316 : #endif
8317 :
8318 27000 : LockBuffer(buf, BUFFER_LOCK_SHARE);
8319 :
8320 27000 : page = BufferGetPage(buf);
8321 27000 : maxoff = PageGetMaxOffsetNumber(page);
8322 : }
8323 :
8324 : /*
8325 : * In passing, detect index corruption involving an index page with a
8326 : * TID that points to a location in the heap that couldn't possibly be
8327 : * correct. We only do this with actual TIDs from caller's index page
8328 : * (not items reached by traversing through a HOT chain).
8329 : */
8330 1073430 : index_delete_check_htid(delstate, page, maxoff, htid, istatus);
8331 :
8332 1073430 : if (istatus->knowndeletable)
8333 : Assert(!delstate->bottomup && !istatus->promising);
8334 : else
8335 : {
8336 807330 : ItemPointerData tmp = *htid;
8337 : HeapTupleData heapTuple;
8338 :
8339 : /* Are any tuples from this HOT chain non-vacuumable? */
8340 807330 : if (heap_hot_search_buffer(&tmp, rel, buf, &SnapshotNonVacuumable,
8341 : &heapTuple, NULL, true))
8342 485638 : continue; /* can't delete entry */
8343 :
8344 : /* Caller will delete, since whole HOT chain is vacuumable */
8345 321692 : istatus->knowndeletable = true;
8346 :
8347 : /* Maintain index free space info for bottom-up deletion case */
8348 321692 : if (delstate->bottomup)
8349 : {
8350 : Assert(istatus->freespace > 0);
8351 16350 : actualfreespace += istatus->freespace;
8352 16350 : if (actualfreespace >= curtargetfreespace)
8353 4350 : bottomup_final_block = true;
8354 : }
8355 : }
8356 :
8357 : /*
8358 : * Maintain snapshotConflictHorizon value for deletion operation as a
8359 : * whole by advancing current value using heap tuple headers. This is
8360 : * loosely based on the logic for pruning a HOT chain.
8361 : */
8362 587792 : offnum = ItemPointerGetOffsetNumber(htid);
8363 587792 : priorXmax = InvalidTransactionId; /* cannot check first XMIN */
8364 : for (;;)
8365 41622 : {
8366 : ItemId lp;
8367 : HeapTupleHeader htup;
8368 :
8369 : /* Sanity check (pure paranoia) */
8370 629414 : if (offnum < FirstOffsetNumber)
8371 0 : break;
8372 :
8373 : /*
8374 : * An offset past the end of page's line pointer array is possible
8375 : * when the array was truncated
8376 : */
8377 629414 : if (offnum > maxoff)
8378 0 : break;
8379 :
8380 629414 : lp = PageGetItemId(page, offnum);
8381 629414 : if (ItemIdIsRedirected(lp))
8382 : {
8383 18730 : offnum = ItemIdGetRedirect(lp);
8384 18730 : continue;
8385 : }
8386 :
8387 : /*
8388 : * We'll often encounter LP_DEAD line pointers (especially with an
8389 : * entry marked knowndeletable by our caller up front). No heap
8390 : * tuple headers get examined for an htid that leads us to an
8391 : * LP_DEAD item. This is okay because the earlier pruning
8392 : * operation that made the line pointer LP_DEAD in the first place
8393 : * must have considered the original tuple header as part of
8394 : * generating its own snapshotConflictHorizon value.
8395 : *
8396 : * Relying on XLOG_HEAP2_PRUNE_VACUUM_SCAN records like this is
8397 : * the same strategy that index vacuuming uses in all cases. Index
8398 : * VACUUM WAL records don't even have a snapshotConflictHorizon
8399 : * field of their own for this reason.
8400 : */
8401 610684 : if (!ItemIdIsNormal(lp))
8402 387750 : break;
8403 :
8404 222934 : htup = (HeapTupleHeader) PageGetItem(page, lp);
8405 :
8406 : /*
8407 : * Check the tuple XMIN against prior XMAX, if any
8408 : */
8409 245826 : if (TransactionIdIsValid(priorXmax) &&
8410 22892 : !TransactionIdEquals(HeapTupleHeaderGetXmin(htup), priorXmax))
8411 0 : break;
8412 :
8413 222934 : HeapTupleHeaderAdvanceConflictHorizon(htup,
8414 : &snapshotConflictHorizon);
8415 :
8416 : /*
8417 : * If the tuple is not HOT-updated, then we are at the end of this
8418 : * HOT-chain. No need to visit later tuples from the same update
8419 : * chain (they get their own index entries) -- just move on to
8420 : * next htid from index AM caller.
8421 : */
8422 222934 : if (!HeapTupleHeaderIsHotUpdated(htup))
8423 200042 : break;
8424 :
8425 : /* Advance to next HOT chain member */
8426 : Assert(ItemPointerGetBlockNumber(&htup->t_ctid) == blkno);
8427 22892 : offnum = ItemPointerGetOffsetNumber(&htup->t_ctid);
8428 22892 : priorXmax = HeapTupleHeaderGetUpdateXid(htup);
8429 : }
8430 :
8431 : /* Enable further/final shrinking of deltids for caller */
8432 587792 : finalndeltids = i + 1;
8433 : }
8434 :
8435 11438 : UnlockReleaseBuffer(buf);
8436 :
8437 : /*
8438 : * Shrink deltids array to exclude non-deletable entries at the end. This
8439 : * is not just a minor optimization. Final deltids array size might be
8440 : * zero for a bottom-up caller. Index AM is explicitly allowed to rely on
8441 : * ndeltids being zero in all cases with zero total deletable entries.
8442 : */
8443 : Assert(finalndeltids > 0 || delstate->bottomup);
8444 11438 : delstate->ndeltids = finalndeltids;
8445 :
8446 11438 : return snapshotConflictHorizon;
8447 : }
8448 :
8449 : /*
8450 : * Specialized inlineable comparison function for index_delete_sort()
8451 : */
8452 : static inline int
8453 25145272 : index_delete_sort_cmp(TM_IndexDelete *deltid1, TM_IndexDelete *deltid2)
8454 : {
8455 25145272 : ItemPointer tid1 = &deltid1->tid;
8456 25145272 : ItemPointer tid2 = &deltid2->tid;
8457 :
8458 : {
8459 25145272 : BlockNumber blk1 = ItemPointerGetBlockNumber(tid1);
8460 25145272 : BlockNumber blk2 = ItemPointerGetBlockNumber(tid2);
8461 :
8462 25145272 : if (blk1 != blk2)
8463 10222084 : return (blk1 < blk2) ? -1 : 1;
8464 : }
8465 : {
8466 14923188 : OffsetNumber pos1 = ItemPointerGetOffsetNumber(tid1);
8467 14923188 : OffsetNumber pos2 = ItemPointerGetOffsetNumber(tid2);
8468 :
8469 14923188 : if (pos1 != pos2)
8470 14923188 : return (pos1 < pos2) ? -1 : 1;
8471 : }
8472 :
8473 : Assert(false);
8474 :
8475 0 : return 0;
8476 : }
8477 :
8478 : /*
8479 : * Sort deltids array from delstate by TID. This prepares it for further
8480 : * processing by heap_index_delete_tuples().
8481 : *
8482 : * This operation becomes a noticeable consumer of CPU cycles with some
8483 : * workloads, so we go to the trouble of specialization/micro optimization.
8484 : * We use shellsort for this because it's easy to specialize, compiles to
8485 : * relatively few instructions, and is adaptive to presorted inputs/subsets
8486 : * (which are typical here).
8487 : */
8488 : static void
8489 11438 : index_delete_sort(TM_IndexDeleteOp *delstate)
8490 : {
8491 11438 : TM_IndexDelete *deltids = delstate->deltids;
8492 11438 : int ndeltids = delstate->ndeltids;
8493 :
8494 : /*
8495 : * Shellsort gap sequence (taken from Sedgewick-Incerpi paper).
8496 : *
8497 : * This implementation is fast with array sizes up to ~4500. This covers
8498 : * all supported BLCKSZ values.
8499 : */
8500 11438 : const int gaps[9] = {1968, 861, 336, 112, 48, 21, 7, 3, 1};
8501 :
8502 : /* Think carefully before changing anything here -- keep swaps cheap */
8503 : StaticAssertDecl(sizeof(TM_IndexDelete) <= 8,
8504 : "element size exceeds 8 bytes");
8505 :
8506 114380 : for (int g = 0; g < lengthof(gaps); g++)
8507 : {
8508 14981242 : for (int hi = gaps[g], i = hi; i < ndeltids; i++)
8509 : {
8510 14878300 : TM_IndexDelete d = deltids[i];
8511 14878300 : int j = i;
8512 :
8513 25874916 : while (j >= hi && index_delete_sort_cmp(&deltids[j - hi], &d) >= 0)
8514 : {
8515 10996616 : deltids[j] = deltids[j - hi];
8516 10996616 : j -= hi;
8517 : }
8518 14878300 : deltids[j] = d;
8519 : }
8520 : }
8521 11438 : }
8522 :
8523 : /*
8524 : * Returns how many blocks should be considered favorable/contiguous for a
8525 : * bottom-up index deletion pass. This is a number of heap blocks that starts
8526 : * from and includes the first block in line.
8527 : *
8528 : * There is always at least one favorable block during bottom-up index
8529 : * deletion. In the worst case (i.e. with totally random heap blocks) the
8530 : * first block in line (the only favorable block) can be thought of as a
8531 : * degenerate array of contiguous blocks that consists of a single block.
8532 : * heap_index_delete_tuples() will expect this.
8533 : *
8534 : * Caller passes blockgroups, a description of the final order that deltids
8535 : * will be sorted in for heap_index_delete_tuples() bottom-up index deletion
8536 : * processing. Note that deltids need not actually be sorted just yet (caller
8537 : * only passes deltids to us so that we can interpret blockgroups).
8538 : *
8539 : * You might guess that the existence of contiguous blocks cannot matter much,
8540 : * since in general the main factor that determines which blocks we visit is
8541 : * the number of promising TIDs, which is a fixed hint from the index AM.
8542 : * We're not really targeting the general case, though -- the actual goal is
8543 : * to adapt our behavior to a wide variety of naturally occurring conditions.
8544 : * The effects of most of the heuristics we apply are only noticeable in the
8545 : * aggregate, over time and across many _related_ bottom-up index deletion
8546 : * passes.
8547 : *
8548 : * Deeming certain blocks favorable allows heapam to recognize and adapt to
8549 : * workloads where heap blocks visited during bottom-up index deletion can be
8550 : * accessed contiguously, in the sense that each newly visited block is the
8551 : * neighbor of the block that bottom-up deletion just finished processing (or
8552 : * close enough to it). It will likely be cheaper to access more favorable
8553 : * blocks sooner rather than later (e.g. in this pass, not across a series of
8554 : * related bottom-up passes). Either way it is probably only a matter of time
8555 : * (or a matter of further correlated version churn) before all blocks that
8556 : * appear together as a single large batch of favorable blocks get accessed by
8557 : * _some_ bottom-up pass. Large batches of favorable blocks tend to either
8558 : * appear almost constantly or not even once (it all depends on per-index
8559 : * workload characteristics).
8560 : *
8561 : * Note that the blockgroups sort order applies a power-of-two bucketing
8562 : * scheme that creates opportunities for contiguous groups of blocks to get
8563 : * batched together, at least with workloads that are naturally amenable to
8564 : * being driven by heap block locality. This doesn't just enhance the spatial
8565 : * locality of bottom-up heap block processing in the obvious way. It also
8566 : * enables temporal locality of access, since sorting by heap block number
8567 : * naturally tends to make the bottom-up processing order deterministic.
8568 : *
8569 : * Consider the following example to get a sense of how temporal locality
8570 : * might matter: There is a heap relation with several indexes, each of which
8571 : * is low to medium cardinality. It is subject to constant non-HOT updates.
8572 : * The updates are skewed (in one part of the primary key, perhaps). None of
8573 : * the indexes are logically modified by the UPDATE statements (if they were
8574 : * then bottom-up index deletion would not be triggered in the first place).
8575 : * Naturally, each new round of index tuples (for each heap tuple that gets a
8576 : * heap_update() call) will have the same heap TID in each and every index.
8577 : * Since these indexes are low cardinality and never get logically modified,
8578 : * heapam processing during bottom-up deletion passes will access heap blocks
8579 : * in approximately sequential order. Temporal locality of access occurs due
8580 : * to bottom-up deletion passes behaving very similarly across each of the
8581 : * indexes at any given moment. This keeps the number of buffer misses needed
8582 : * to visit heap blocks to a minimum.
8583 : */
8584 : static int
8585 3866 : bottomup_nblocksfavorable(IndexDeleteCounts *blockgroups, int nblockgroups,
8586 : TM_IndexDelete *deltids)
8587 : {
8588 3866 : int64 lastblock = -1;
8589 3866 : int nblocksfavorable = 0;
8590 :
8591 : Assert(nblockgroups >= 1);
8592 : Assert(nblockgroups <= BOTTOMUP_MAX_NBLOCKS);
8593 :
8594 : /*
8595 : * We tolerate heap blocks that will be accessed only slightly out of
8596 : * physical order. Small blips occur when a pair of almost-contiguous
8597 : * blocks happen to fall into different buckets (perhaps due only to a
8598 : * small difference in npromisingtids that the bucketing scheme didn't
8599 : * quite manage to ignore). We effectively ignore these blips by applying
8600 : * a small tolerance. The precise tolerance we use is a little arbitrary,
8601 : * but it works well enough in practice.
8602 : */
8603 12416 : for (int b = 0; b < nblockgroups; b++)
8604 : {
8605 11868 : IndexDeleteCounts *group = blockgroups + b;
8606 11868 : TM_IndexDelete *firstdtid = deltids + group->ifirsttid;
8607 11868 : BlockNumber block = ItemPointerGetBlockNumber(&firstdtid->tid);
8608 :
8609 11868 : if (lastblock != -1 &&
8610 8002 : ((int64) block < lastblock - BOTTOMUP_TOLERANCE_NBLOCKS ||
8611 7000 : (int64) block > lastblock + BOTTOMUP_TOLERANCE_NBLOCKS))
8612 : break;
8613 :
8614 8550 : nblocksfavorable++;
8615 8550 : lastblock = block;
8616 : }
8617 :
8618 : /* Always indicate that there is at least 1 favorable block */
8619 : Assert(nblocksfavorable >= 1);
8620 :
8621 3866 : return nblocksfavorable;
8622 : }
8623 :
8624 : /*
8625 : * qsort comparison function for bottomup_sort_and_shrink()
8626 : */
8627 : static int
8628 386732 : bottomup_sort_and_shrink_cmp(const void *arg1, const void *arg2)
8629 : {
8630 386732 : const IndexDeleteCounts *group1 = (const IndexDeleteCounts *) arg1;
8631 386732 : const IndexDeleteCounts *group2 = (const IndexDeleteCounts *) arg2;
8632 :
8633 : /*
8634 : * Most significant field is npromisingtids (which we invert the order of
8635 : * so as to sort in desc order).
8636 : *
8637 : * Caller should have already normalized npromisingtids fields into
8638 : * power-of-two values (buckets).
8639 : */
8640 386732 : if (group1->npromisingtids > group2->npromisingtids)
8641 18008 : return -1;
8642 368724 : if (group1->npromisingtids < group2->npromisingtids)
8643 21250 : return 1;
8644 :
8645 : /*
8646 : * Tiebreak: desc ntids sort order.
8647 : *
8648 : * We cannot expect power-of-two values for ntids fields. We should
8649 : * behave as if they were already rounded up for us instead.
8650 : */
8651 347474 : if (group1->ntids != group2->ntids)
8652 : {
8653 240704 : uint32 ntids1 = pg_nextpower2_32((uint32) group1->ntids);
8654 240704 : uint32 ntids2 = pg_nextpower2_32((uint32) group2->ntids);
8655 :
8656 240704 : if (ntids1 > ntids2)
8657 36556 : return -1;
8658 204148 : if (ntids1 < ntids2)
8659 45258 : return 1;
8660 : }
8661 :
8662 : /*
8663 : * Tiebreak: asc offset-into-deltids-for-block (offset to first TID for
8664 : * block in deltids array) order.
8665 : *
8666 : * This is equivalent to sorting in ascending heap block number order
8667 : * (among otherwise equal subsets of the array). This approach allows us
8668 : * to avoid accessing the out-of-line TID. (We rely on the assumption
8669 : * that the deltids array was sorted in ascending heap TID order when
8670 : * these offsets to the first TID from each heap block group were formed.)
8671 : */
8672 265660 : if (group1->ifirsttid > group2->ifirsttid)
8673 131516 : return 1;
8674 134144 : if (group1->ifirsttid < group2->ifirsttid)
8675 134144 : return -1;
8676 :
8677 0 : pg_unreachable();
8678 :
8679 : return 0;
8680 : }
8681 :
8682 : /*
8683 : * heap_index_delete_tuples() helper function for bottom-up deletion callers.
8684 : *
8685 : * Sorts deltids array in the order needed for useful processing by bottom-up
8686 : * deletion. The array should already be sorted in TID order when we're
8687 : * called. The sort process groups heap TIDs from deltids into heap block
8688 : * groupings. Earlier/more-promising groups/blocks are usually those that are
8689 : * known to have the most "promising" TIDs.
8690 : *
8691 : * Sets new size of deltids array (ndeltids) in state. deltids will only have
8692 : * TIDs from the BOTTOMUP_MAX_NBLOCKS most promising heap blocks when we
8693 : * return. This often means that deltids will be shrunk to a small fraction
8694 : * of its original size (we eliminate many heap blocks from consideration for
8695 : * caller up front).
8696 : *
8697 : * Returns the number of "favorable" blocks. See bottomup_nblocksfavorable()
8698 : * for a definition and full details.
8699 : */
8700 : static int
8701 3866 : bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate)
8702 : {
8703 : IndexDeleteCounts *blockgroups;
8704 : TM_IndexDelete *reordereddeltids;
8705 3866 : BlockNumber curblock = InvalidBlockNumber;
8706 3866 : int nblockgroups = 0;
8707 3866 : int ncopied = 0;
8708 3866 : int nblocksfavorable = 0;
8709 :
8710 : Assert(delstate->bottomup);
8711 : Assert(delstate->ndeltids > 0);
8712 :
8713 : /* Calculate per-heap-block count of TIDs */
8714 3866 : blockgroups = palloc(sizeof(IndexDeleteCounts) * delstate->ndeltids);
8715 1853906 : for (int i = 0; i < delstate->ndeltids; i++)
8716 : {
8717 1850040 : TM_IndexDelete *ideltid = &delstate->deltids[i];
8718 1850040 : TM_IndexStatus *istatus = delstate->status + ideltid->id;
8719 1850040 : ItemPointer htid = &ideltid->tid;
8720 1850040 : bool promising = istatus->promising;
8721 :
8722 1850040 : if (curblock != ItemPointerGetBlockNumber(htid))
8723 : {
8724 : /* New block group */
8725 75558 : nblockgroups++;
8726 :
8727 : Assert(curblock < ItemPointerGetBlockNumber(htid) ||
8728 : !BlockNumberIsValid(curblock));
8729 :
8730 75558 : curblock = ItemPointerGetBlockNumber(htid);
8731 75558 : blockgroups[nblockgroups - 1].ifirsttid = i;
8732 75558 : blockgroups[nblockgroups - 1].ntids = 1;
8733 75558 : blockgroups[nblockgroups - 1].npromisingtids = 0;
8734 : }
8735 : else
8736 : {
8737 1774482 : blockgroups[nblockgroups - 1].ntids++;
8738 : }
8739 :
8740 1850040 : if (promising)
8741 238290 : blockgroups[nblockgroups - 1].npromisingtids++;
8742 : }
8743 :
8744 : /*
8745 : * We're about ready to sort block groups to determine the optimal order
8746 : * for visiting heap blocks. But before we do, round the number of
8747 : * promising tuples for each block group up to the next power-of-two,
8748 : * unless it is very low (less than 4), in which case we round up to 4.
8749 : * npromisingtids is far too noisy to trust when choosing between a pair
8750 : * of block groups that both have very low values.
8751 : *
8752 : * This scheme divides heap blocks/block groups into buckets. Each bucket
8753 : * contains blocks that have _approximately_ the same number of promising
8754 : * TIDs as each other. The goal is to ignore relatively small differences
8755 : * in the total number of promising entries, so that the whole process can
8756 : * give a little weight to heapam factors (like heap block locality)
8757 : * instead. This isn't a trade-off, really -- we have nothing to lose. It
8758 : * would be foolish to interpret small differences in npromisingtids
8759 : * values as anything more than noise.
8760 : *
8761 : * We tiebreak on nhtids when sorting block group subsets that have the
8762 : * same npromisingtids, but this has the same issues as npromisingtids,
8763 : * and so nhtids is subject to the same power-of-two bucketing scheme. The
8764 : * only reason that we don't fix nhtids in the same way here too is that
8765 : * we'll need accurate nhtids values after the sort. We handle nhtids
8766 : * bucketization dynamically instead (in the sort comparator).
8767 : *
8768 : * See bottomup_nblocksfavorable() for a full explanation of when and how
8769 : * heap locality/favorable blocks can significantly influence when and how
8770 : * heap blocks are accessed.
8771 : */
8772 79424 : for (int b = 0; b < nblockgroups; b++)
8773 : {
8774 75558 : IndexDeleteCounts *group = blockgroups + b;
8775 :
8776 : /* Better off falling back on nhtids with low npromisingtids */
8777 75558 : if (group->npromisingtids <= 4)
8778 64748 : group->npromisingtids = 4;
8779 : else
8780 10810 : group->npromisingtids =
8781 10810 : pg_nextpower2_32((uint32) group->npromisingtids);
8782 : }
8783 :
8784 : /* Sort groups and rearrange caller's deltids array */
8785 3866 : qsort(blockgroups, nblockgroups, sizeof(IndexDeleteCounts),
8786 : bottomup_sort_and_shrink_cmp);
8787 3866 : reordereddeltids = palloc(delstate->ndeltids * sizeof(TM_IndexDelete));
8788 :
8789 3866 : nblockgroups = Min(BOTTOMUP_MAX_NBLOCKS, nblockgroups);
8790 : /* Determine number of favorable blocks at the start of final deltids */
8791 3866 : nblocksfavorable = bottomup_nblocksfavorable(blockgroups, nblockgroups,
8792 : delstate->deltids);
8793 :
8794 25840 : for (int b = 0; b < nblockgroups; b++)
8795 : {
8796 21974 : IndexDeleteCounts *group = blockgroups + b;
8797 21974 : TM_IndexDelete *firstdtid = delstate->deltids + group->ifirsttid;
8798 :
8799 21974 : memcpy(reordereddeltids + ncopied, firstdtid,
8800 21974 : sizeof(TM_IndexDelete) * group->ntids);
8801 21974 : ncopied += group->ntids;
8802 : }
8803 :
8804 : /* Copy final grouped and sorted TIDs back into start of caller's array */
8805 3866 : memcpy(delstate->deltids, reordereddeltids,
8806 : sizeof(TM_IndexDelete) * ncopied);
8807 3866 : delstate->ndeltids = ncopied;
8808 :
8809 3866 : pfree(reordereddeltids);
8810 3866 : pfree(blockgroups);
8811 :
8812 3866 : return nblocksfavorable;
8813 : }
8814 :
8815 : /*
8816 : * Perform XLogInsert for a heap-visible operation. 'block' is the block
8817 : * being marked all-visible, and vm_buffer is the buffer containing the
8818 : * corresponding visibility map block. Both should have already been modified
8819 : * and dirtied.
8820 : *
8821 : * snapshotConflictHorizon comes from the largest xmin on the page being
8822 : * marked all-visible. REDO routine uses it to generate recovery conflicts.
8823 : *
8824 : * If checksums or wal_log_hints are enabled, we may also generate a full-page
8825 : * image of heap_buffer. Otherwise, we optimize away the FPI (by specifying
8826 : * REGBUF_NO_IMAGE for the heap buffer), in which case the caller should *not*
8827 : * update the heap page's LSN.
8828 : */
8829 : XLogRecPtr
8830 66842 : log_heap_visible(Relation rel, Buffer heap_buffer, Buffer vm_buffer,
8831 : TransactionId snapshotConflictHorizon, uint8 vmflags)
8832 : {
8833 : xl_heap_visible xlrec;
8834 : XLogRecPtr recptr;
8835 : uint8 flags;
8836 :
8837 : Assert(BufferIsValid(heap_buffer));
8838 : Assert(BufferIsValid(vm_buffer));
8839 :
8840 66842 : xlrec.snapshotConflictHorizon = snapshotConflictHorizon;
8841 66842 : xlrec.flags = vmflags;
8842 66842 : if (RelationIsAccessibleInLogicalDecoding(rel))
8843 108 : xlrec.flags |= VISIBILITYMAP_XLOG_CATALOG_REL;
8844 66842 : XLogBeginInsert();
8845 66842 : XLogRegisterData(&xlrec, SizeOfHeapVisible);
8846 :
8847 66842 : XLogRegisterBuffer(0, vm_buffer, 0);
8848 :
8849 66842 : flags = REGBUF_STANDARD;
8850 66842 : if (!XLogHintBitIsNeeded())
8851 6132 : flags |= REGBUF_NO_IMAGE;
8852 66842 : XLogRegisterBuffer(1, heap_buffer, flags);
8853 :
8854 66842 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_VISIBLE);
8855 :
8856 66842 : return recptr;
8857 : }
8858 :
8859 : /*
8860 : * Perform XLogInsert for a heap-update operation. Caller must already
8861 : * have modified the buffer(s) and marked them dirty.
8862 : */
8863 : static XLogRecPtr
8864 584584 : log_heap_update(Relation reln, Buffer oldbuf,
8865 : Buffer newbuf, HeapTuple oldtup, HeapTuple newtup,
8866 : HeapTuple old_key_tuple,
8867 : bool all_visible_cleared, bool new_all_visible_cleared)
8868 : {
8869 : xl_heap_update xlrec;
8870 : xl_heap_header xlhdr;
8871 : xl_heap_header xlhdr_idx;
8872 : uint8 info;
8873 : uint16 prefix_suffix[2];
8874 584584 : uint16 prefixlen = 0,
8875 584584 : suffixlen = 0;
8876 : XLogRecPtr recptr;
8877 584584 : Page page = BufferGetPage(newbuf);
8878 584584 : bool need_tuple_data = RelationIsLogicallyLogged(reln);
8879 : bool init;
8880 : int bufflags;
8881 :
8882 : /* Caller should not call me on a non-WAL-logged relation */
8883 : Assert(RelationNeedsWAL(reln));
8884 :
8885 584584 : XLogBeginInsert();
8886 :
8887 584584 : if (HeapTupleIsHeapOnly(newtup))
8888 282772 : info = XLOG_HEAP_HOT_UPDATE;
8889 : else
8890 301812 : info = XLOG_HEAP_UPDATE;
8891 :
8892 : /*
8893 : * If the old and new tuple are on the same page, we only need to log the
8894 : * parts of the new tuple that were changed. That saves on the amount of
8895 : * WAL we need to write. Currently, we just count any unchanged bytes in
8896 : * the beginning and end of the tuple. That's quick to check, and
8897 : * perfectly covers the common case that only one field is updated.
8898 : *
8899 : * We could do this even if the old and new tuple are on different pages,
8900 : * but only if we don't make a full-page image of the old page, which is
8901 : * difficult to know in advance. Also, if the old tuple is corrupt for
8902 : * some reason, it would allow the corruption to propagate the new page,
8903 : * so it seems best to avoid. Under the general assumption that most
8904 : * updates tend to create the new tuple version on the same page, there
8905 : * isn't much to be gained by doing this across pages anyway.
8906 : *
8907 : * Skip this if we're taking a full-page image of the new page, as we
8908 : * don't include the new tuple in the WAL record in that case. Also
8909 : * disable if wal_level='logical', as logical decoding needs to be able to
8910 : * read the new tuple in whole from the WAL record alone.
8911 : */
8912 584584 : if (oldbuf == newbuf && !need_tuple_data &&
8913 283148 : !XLogCheckBufferNeedsBackup(newbuf))
8914 : {
8915 282014 : char *oldp = (char *) oldtup->t_data + oldtup->t_data->t_hoff;
8916 282014 : char *newp = (char *) newtup->t_data + newtup->t_data->t_hoff;
8917 282014 : int oldlen = oldtup->t_len - oldtup->t_data->t_hoff;
8918 282014 : int newlen = newtup->t_len - newtup->t_data->t_hoff;
8919 :
8920 : /* Check for common prefix between old and new tuple */
8921 23263836 : for (prefixlen = 0; prefixlen < Min(oldlen, newlen); prefixlen++)
8922 : {
8923 23212738 : if (newp[prefixlen] != oldp[prefixlen])
8924 230916 : break;
8925 : }
8926 :
8927 : /*
8928 : * Storing the length of the prefix takes 2 bytes, so we need to save
8929 : * at least 3 bytes or there's no point.
8930 : */
8931 282014 : if (prefixlen < 3)
8932 44226 : prefixlen = 0;
8933 :
8934 : /* Same for suffix */
8935 9114258 : for (suffixlen = 0; suffixlen < Min(oldlen, newlen) - prefixlen; suffixlen++)
8936 : {
8937 9062648 : if (newp[newlen - suffixlen - 1] != oldp[oldlen - suffixlen - 1])
8938 230404 : break;
8939 : }
8940 282014 : if (suffixlen < 3)
8941 70580 : suffixlen = 0;
8942 : }
8943 :
8944 : /* Prepare main WAL data chain */
8945 584584 : xlrec.flags = 0;
8946 584584 : if (all_visible_cleared)
8947 2840 : xlrec.flags |= XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED;
8948 584584 : if (new_all_visible_cleared)
8949 1660 : xlrec.flags |= XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED;
8950 584584 : if (prefixlen > 0)
8951 237788 : xlrec.flags |= XLH_UPDATE_PREFIX_FROM_OLD;
8952 584584 : if (suffixlen > 0)
8953 211434 : xlrec.flags |= XLH_UPDATE_SUFFIX_FROM_OLD;
8954 584584 : if (need_tuple_data)
8955 : {
8956 94052 : xlrec.flags |= XLH_UPDATE_CONTAINS_NEW_TUPLE;
8957 94052 : if (old_key_tuple)
8958 : {
8959 292 : if (reln->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
8960 130 : xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_TUPLE;
8961 : else
8962 162 : xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_KEY;
8963 : }
8964 : }
8965 :
8966 : /* If new tuple is the single and first tuple on page... */
8967 591394 : if (ItemPointerGetOffsetNumber(&(newtup->t_self)) == FirstOffsetNumber &&
8968 6810 : PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
8969 : {
8970 6364 : info |= XLOG_HEAP_INIT_PAGE;
8971 6364 : init = true;
8972 : }
8973 : else
8974 578220 : init = false;
8975 :
8976 : /* Prepare WAL data for the old page */
8977 584584 : xlrec.old_offnum = ItemPointerGetOffsetNumber(&oldtup->t_self);
8978 584584 : xlrec.old_xmax = HeapTupleHeaderGetRawXmax(oldtup->t_data);
8979 1169168 : xlrec.old_infobits_set = compute_infobits(oldtup->t_data->t_infomask,
8980 584584 : oldtup->t_data->t_infomask2);
8981 :
8982 : /* Prepare WAL data for the new page */
8983 584584 : xlrec.new_offnum = ItemPointerGetOffsetNumber(&newtup->t_self);
8984 584584 : xlrec.new_xmax = HeapTupleHeaderGetRawXmax(newtup->t_data);
8985 :
8986 584584 : bufflags = REGBUF_STANDARD;
8987 584584 : if (init)
8988 6364 : bufflags |= REGBUF_WILL_INIT;
8989 584584 : if (need_tuple_data)
8990 94052 : bufflags |= REGBUF_KEEP_DATA;
8991 :
8992 584584 : XLogRegisterBuffer(0, newbuf, bufflags);
8993 584584 : if (oldbuf != newbuf)
8994 277544 : XLogRegisterBuffer(1, oldbuf, REGBUF_STANDARD);
8995 :
8996 584584 : XLogRegisterData(&xlrec, SizeOfHeapUpdate);
8997 :
8998 : /*
8999 : * Prepare WAL data for the new tuple.
9000 : */
9001 584584 : if (prefixlen > 0 || suffixlen > 0)
9002 : {
9003 281084 : if (prefixlen > 0 && suffixlen > 0)
9004 : {
9005 168138 : prefix_suffix[0] = prefixlen;
9006 168138 : prefix_suffix[1] = suffixlen;
9007 168138 : XLogRegisterBufData(0, &prefix_suffix, sizeof(uint16) * 2);
9008 : }
9009 112946 : else if (prefixlen > 0)
9010 : {
9011 69650 : XLogRegisterBufData(0, &prefixlen, sizeof(uint16));
9012 : }
9013 : else
9014 : {
9015 43296 : XLogRegisterBufData(0, &suffixlen, sizeof(uint16));
9016 : }
9017 : }
9018 :
9019 584584 : xlhdr.t_infomask2 = newtup->t_data->t_infomask2;
9020 584584 : xlhdr.t_infomask = newtup->t_data->t_infomask;
9021 584584 : xlhdr.t_hoff = newtup->t_data->t_hoff;
9022 : Assert(SizeofHeapTupleHeader + prefixlen + suffixlen <= newtup->t_len);
9023 :
9024 : /*
9025 : * PG73FORMAT: write bitmap [+ padding] [+ oid] + data
9026 : *
9027 : * The 'data' doesn't include the common prefix or suffix.
9028 : */
9029 584584 : XLogRegisterBufData(0, &xlhdr, SizeOfHeapHeader);
9030 584584 : if (prefixlen == 0)
9031 : {
9032 346796 : XLogRegisterBufData(0,
9033 346796 : (char *) newtup->t_data + SizeofHeapTupleHeader,
9034 346796 : newtup->t_len - SizeofHeapTupleHeader - suffixlen);
9035 : }
9036 : else
9037 : {
9038 : /*
9039 : * Have to write the null bitmap and data after the common prefix as
9040 : * two separate rdata entries.
9041 : */
9042 : /* bitmap [+ padding] [+ oid] */
9043 237788 : if (newtup->t_data->t_hoff - SizeofHeapTupleHeader > 0)
9044 : {
9045 237788 : XLogRegisterBufData(0,
9046 237788 : (char *) newtup->t_data + SizeofHeapTupleHeader,
9047 237788 : newtup->t_data->t_hoff - SizeofHeapTupleHeader);
9048 : }
9049 :
9050 : /* data after common prefix */
9051 237788 : XLogRegisterBufData(0,
9052 237788 : (char *) newtup->t_data + newtup->t_data->t_hoff + prefixlen,
9053 237788 : newtup->t_len - newtup->t_data->t_hoff - prefixlen - suffixlen);
9054 : }
9055 :
9056 : /* We need to log a tuple identity */
9057 584584 : if (need_tuple_data && old_key_tuple)
9058 : {
9059 : /* don't really need this, but its more comfy to decode */
9060 292 : xlhdr_idx.t_infomask2 = old_key_tuple->t_data->t_infomask2;
9061 292 : xlhdr_idx.t_infomask = old_key_tuple->t_data->t_infomask;
9062 292 : xlhdr_idx.t_hoff = old_key_tuple->t_data->t_hoff;
9063 :
9064 292 : XLogRegisterData(&xlhdr_idx, SizeOfHeapHeader);
9065 :
9066 : /* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
9067 292 : XLogRegisterData((char *) old_key_tuple->t_data + SizeofHeapTupleHeader,
9068 292 : old_key_tuple->t_len - SizeofHeapTupleHeader);
9069 : }
9070 :
9071 : /* filtering by origin on a row level is much more efficient */
9072 584584 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
9073 :
9074 584584 : recptr = XLogInsert(RM_HEAP_ID, info);
9075 :
9076 584584 : return recptr;
9077 : }
9078 :
9079 : /*
9080 : * Perform XLogInsert of an XLOG_HEAP2_NEW_CID record
9081 : *
9082 : * This is only used in wal_level >= WAL_LEVEL_LOGICAL, and only for catalog
9083 : * tuples.
9084 : */
9085 : static XLogRecPtr
9086 49176 : log_heap_new_cid(Relation relation, HeapTuple tup)
9087 : {
9088 : xl_heap_new_cid xlrec;
9089 :
9090 : XLogRecPtr recptr;
9091 49176 : HeapTupleHeader hdr = tup->t_data;
9092 :
9093 : Assert(ItemPointerIsValid(&tup->t_self));
9094 : Assert(tup->t_tableOid != InvalidOid);
9095 :
9096 49176 : xlrec.top_xid = GetTopTransactionId();
9097 49176 : xlrec.target_locator = relation->rd_locator;
9098 49176 : xlrec.target_tid = tup->t_self;
9099 :
9100 : /*
9101 : * If the tuple got inserted & deleted in the same TX we definitely have a
9102 : * combo CID, set cmin and cmax.
9103 : */
9104 49176 : if (hdr->t_infomask & HEAP_COMBOCID)
9105 : {
9106 : Assert(!(hdr->t_infomask & HEAP_XMAX_INVALID));
9107 : Assert(!HeapTupleHeaderXminInvalid(hdr));
9108 4048 : xlrec.cmin = HeapTupleHeaderGetCmin(hdr);
9109 4048 : xlrec.cmax = HeapTupleHeaderGetCmax(hdr);
9110 4048 : xlrec.combocid = HeapTupleHeaderGetRawCommandId(hdr);
9111 : }
9112 : /* No combo CID, so only cmin or cmax can be set by this TX */
9113 : else
9114 : {
9115 : /*
9116 : * Tuple inserted.
9117 : *
9118 : * We need to check for LOCK ONLY because multixacts might be
9119 : * transferred to the new tuple in case of FOR KEY SHARE updates in
9120 : * which case there will be an xmax, although the tuple just got
9121 : * inserted.
9122 : */
9123 58740 : if (hdr->t_infomask & HEAP_XMAX_INVALID ||
9124 13612 : HEAP_XMAX_IS_LOCKED_ONLY(hdr->t_infomask))
9125 : {
9126 31518 : xlrec.cmin = HeapTupleHeaderGetRawCommandId(hdr);
9127 31518 : xlrec.cmax = InvalidCommandId;
9128 : }
9129 : /* Tuple from a different tx updated or deleted. */
9130 : else
9131 : {
9132 13610 : xlrec.cmin = InvalidCommandId;
9133 13610 : xlrec.cmax = HeapTupleHeaderGetRawCommandId(hdr);
9134 : }
9135 45128 : xlrec.combocid = InvalidCommandId;
9136 : }
9137 :
9138 : /*
9139 : * Note that we don't need to register the buffer here, because this
9140 : * operation does not modify the page. The insert/update/delete that
9141 : * called us certainly did, but that's WAL-logged separately.
9142 : */
9143 49176 : XLogBeginInsert();
9144 49176 : XLogRegisterData(&xlrec, SizeOfHeapNewCid);
9145 :
9146 : /* will be looked at irrespective of origin */
9147 :
9148 49176 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_NEW_CID);
9149 :
9150 49176 : return recptr;
9151 : }
9152 :
9153 : /*
9154 : * Build a heap tuple representing the configured REPLICA IDENTITY to represent
9155 : * the old tuple in an UPDATE or DELETE.
9156 : *
9157 : * Returns NULL if there's no need to log an identity or if there's no suitable
9158 : * key defined.
9159 : *
9160 : * Pass key_required true if any replica identity columns changed value, or if
9161 : * any of them have any external data. Delete must always pass true.
9162 : *
9163 : * *copy is set to true if the returned tuple is a modified copy rather than
9164 : * the same tuple that was passed in.
9165 : */
9166 : static HeapTuple
9167 3607192 : ExtractReplicaIdentity(Relation relation, HeapTuple tp, bool key_required,
9168 : bool *copy)
9169 : {
9170 3607192 : TupleDesc desc = RelationGetDescr(relation);
9171 3607192 : char replident = relation->rd_rel->relreplident;
9172 : Bitmapset *idattrs;
9173 : HeapTuple key_tuple;
9174 : bool nulls[MaxHeapAttributeNumber];
9175 : Datum values[MaxHeapAttributeNumber];
9176 :
9177 3607192 : *copy = false;
9178 :
9179 3607192 : if (!RelationIsLogicallyLogged(relation))
9180 3406596 : return NULL;
9181 :
9182 200596 : if (replident == REPLICA_IDENTITY_NOTHING)
9183 462 : return NULL;
9184 :
9185 200134 : if (replident == REPLICA_IDENTITY_FULL)
9186 : {
9187 : /*
9188 : * When logging the entire old tuple, it very well could contain
9189 : * toasted columns. If so, force them to be inlined.
9190 : */
9191 394 : if (HeapTupleHasExternal(tp))
9192 : {
9193 8 : *copy = true;
9194 8 : tp = toast_flatten_tuple(tp, desc);
9195 : }
9196 394 : return tp;
9197 : }
9198 :
9199 : /* if the key isn't required and we're only logging the key, we're done */
9200 199740 : if (!key_required)
9201 93760 : return NULL;
9202 :
9203 : /* find out the replica identity columns */
9204 105980 : idattrs = RelationGetIndexAttrBitmap(relation,
9205 : INDEX_ATTR_BITMAP_IDENTITY_KEY);
9206 :
9207 : /*
9208 : * If there's no defined replica identity columns, treat as !key_required.
9209 : * (This case should not be reachable from heap_update, since that should
9210 : * calculate key_required accurately. But heap_delete just passes
9211 : * constant true for key_required, so we can hit this case in deletes.)
9212 : */
9213 105980 : if (bms_is_empty(idattrs))
9214 12042 : return NULL;
9215 :
9216 : /*
9217 : * Construct a new tuple containing only the replica identity columns,
9218 : * with nulls elsewhere. While we're at it, assert that the replica
9219 : * identity columns aren't null.
9220 : */
9221 93938 : heap_deform_tuple(tp, desc, values, nulls);
9222 :
9223 301802 : for (int i = 0; i < desc->natts; i++)
9224 : {
9225 207864 : if (bms_is_member(i + 1 - FirstLowInvalidHeapAttributeNumber,
9226 : idattrs))
9227 : Assert(!nulls[i]);
9228 : else
9229 113902 : nulls[i] = true;
9230 : }
9231 :
9232 93938 : key_tuple = heap_form_tuple(desc, values, nulls);
9233 93938 : *copy = true;
9234 :
9235 93938 : bms_free(idattrs);
9236 :
9237 : /*
9238 : * If the tuple, which by here only contains indexed columns, still has
9239 : * toasted columns, force them to be inlined. This is somewhat unlikely
9240 : * since there's limits on the size of indexed columns, so we don't
9241 : * duplicate toast_flatten_tuple()s functionality in the above loop over
9242 : * the indexed columns, even if it would be more efficient.
9243 : */
9244 93938 : if (HeapTupleHasExternal(key_tuple))
9245 : {
9246 8 : HeapTuple oldtup = key_tuple;
9247 :
9248 8 : key_tuple = toast_flatten_tuple(oldtup, desc);
9249 8 : heap_freetuple(oldtup);
9250 : }
9251 :
9252 93938 : return key_tuple;
9253 : }
9254 :
9255 : /*
9256 : * HeapCheckForSerializableConflictOut
9257 : * We are reading a tuple. If it's not visible, there may be a
9258 : * rw-conflict out with the inserter. Otherwise, if it is visible to us
9259 : * but has been deleted, there may be a rw-conflict out with the deleter.
9260 : *
9261 : * We will determine the top level xid of the writing transaction with which
9262 : * we may be in conflict, and ask CheckForSerializableConflictOut() to check
9263 : * for overlap with our own transaction.
9264 : *
9265 : * This function should be called just about anywhere in heapam.c where a
9266 : * tuple has been read. The caller must hold at least a shared lock on the
9267 : * buffer, because this function might set hint bits on the tuple. There is
9268 : * currently no known reason to call this function from an index AM.
9269 : */
9270 : void
9271 62771000 : HeapCheckForSerializableConflictOut(bool visible, Relation relation,
9272 : HeapTuple tuple, Buffer buffer,
9273 : Snapshot snapshot)
9274 : {
9275 : TransactionId xid;
9276 : HTSV_Result htsvResult;
9277 :
9278 62771000 : if (!CheckForSerializableConflictOutNeeded(relation, snapshot))
9279 62720276 : return;
9280 :
9281 : /*
9282 : * Check to see whether the tuple has been written to by a concurrent
9283 : * transaction, either to create it not visible to us, or to delete it
9284 : * while it is visible to us. The "visible" bool indicates whether the
9285 : * tuple is visible to us, while HeapTupleSatisfiesVacuum checks what else
9286 : * is going on with it.
9287 : *
9288 : * In the event of a concurrently inserted tuple that also happens to have
9289 : * been concurrently updated (by a separate transaction), the xmin of the
9290 : * tuple will be used -- not the updater's xid.
9291 : */
9292 50724 : htsvResult = HeapTupleSatisfiesVacuum(tuple, TransactionXmin, buffer);
9293 50724 : switch (htsvResult)
9294 : {
9295 49098 : case HEAPTUPLE_LIVE:
9296 49098 : if (visible)
9297 49072 : return;
9298 26 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9299 26 : break;
9300 722 : case HEAPTUPLE_RECENTLY_DEAD:
9301 : case HEAPTUPLE_DELETE_IN_PROGRESS:
9302 722 : if (visible)
9303 570 : xid = HeapTupleHeaderGetUpdateXid(tuple->t_data);
9304 : else
9305 152 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9306 :
9307 722 : if (TransactionIdPrecedes(xid, TransactionXmin))
9308 : {
9309 : /* This is like the HEAPTUPLE_DEAD case */
9310 : Assert(!visible);
9311 138 : return;
9312 : }
9313 584 : break;
9314 656 : case HEAPTUPLE_INSERT_IN_PROGRESS:
9315 656 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9316 656 : break;
9317 248 : case HEAPTUPLE_DEAD:
9318 : Assert(!visible);
9319 248 : return;
9320 0 : default:
9321 :
9322 : /*
9323 : * The only way to get to this default clause is if a new value is
9324 : * added to the enum type without adding it to this switch
9325 : * statement. That's a bug, so elog.
9326 : */
9327 0 : elog(ERROR, "unrecognized return value from HeapTupleSatisfiesVacuum: %u", htsvResult);
9328 :
9329 : /*
9330 : * In spite of having all enum values covered and calling elog on
9331 : * this default, some compilers think this is a code path which
9332 : * allows xid to be used below without initialization. Silence
9333 : * that warning.
9334 : */
9335 : xid = InvalidTransactionId;
9336 : }
9337 :
9338 : Assert(TransactionIdIsValid(xid));
9339 : Assert(TransactionIdFollowsOrEquals(xid, TransactionXmin));
9340 :
9341 : /*
9342 : * Find top level xid. Bail out if xid is too early to be a conflict, or
9343 : * if it's our own xid.
9344 : */
9345 1266 : if (TransactionIdEquals(xid, GetTopTransactionIdIfAny()))
9346 128 : return;
9347 1138 : xid = SubTransGetTopmostTransaction(xid);
9348 1138 : if (TransactionIdPrecedes(xid, TransactionXmin))
9349 0 : return;
9350 :
9351 1138 : CheckForSerializableConflictOut(relation, xid, snapshot);
9352 : }
|
