Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * heapam.c
4 : * heap access method code
5 : *
6 : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/access/heap/heapam.c
12 : *
13 : *
14 : * INTERFACE ROUTINES
15 : * heap_beginscan - begin relation scan
16 : * heap_rescan - restart a relation scan
17 : * heap_endscan - end relation scan
18 : * heap_getnext - retrieve next tuple in scan
19 : * heap_fetch - retrieve tuple with given tid
20 : * heap_insert - insert tuple into a relation
21 : * heap_multi_insert - insert multiple tuples into a relation
22 : * heap_delete - delete a tuple from a relation
23 : * heap_update - replace a tuple in a relation with another tuple
24 : *
25 : * NOTES
26 : * This file contains the heap_ routines which implement
27 : * the POSTGRES heap access method used for all POSTGRES
28 : * relations.
29 : *
30 : *-------------------------------------------------------------------------
31 : */
32 : #include "postgres.h"
33 :
34 : #include "access/heapam.h"
35 : #include "access/heaptoast.h"
36 : #include "access/hio.h"
37 : #include "access/multixact.h"
38 : #include "access/subtrans.h"
39 : #include "access/syncscan.h"
40 : #include "access/valid.h"
41 : #include "access/visibilitymap.h"
42 : #include "access/xloginsert.h"
43 : #include "catalog/pg_database.h"
44 : #include "catalog/pg_database_d.h"
45 : #include "commands/vacuum.h"
46 : #include "pgstat.h"
47 : #include "port/pg_bitutils.h"
48 : #include "storage/lmgr.h"
49 : #include "storage/predicate.h"
50 : #include "storage/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,
89 : uint16 prior_infomask,
90 : TransactionId prior_raw_xmax,
91 : const ItemPointerData *prior_ctid,
92 : TransactionId xid,
93 : LockTupleMode mode);
94 : static void GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
95 : uint16 *new_infomask2);
96 : static TransactionId MultiXactIdGetUpdateXid(TransactionId xmax,
97 : uint16 t_infomask);
98 : static bool DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
99 : LockTupleMode lockmode, bool *current_is_member);
100 : static void MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
101 : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
102 : int *remaining);
103 : static bool ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status,
104 : uint16 infomask, Relation rel, int *remaining,
105 : bool logLockFailure);
106 : static void index_delete_sort(TM_IndexDeleteOp *delstate);
107 : static int bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate);
108 : static XLogRecPtr log_heap_new_cid(Relation relation, HeapTuple tup);
109 : static HeapTuple ExtractReplicaIdentity(Relation relation, HeapTuple tp, bool key_required,
110 : bool *copy);
111 :
112 :
113 : /*
114 : * Each tuple lock mode has a corresponding heavyweight lock, and one or two
115 : * corresponding MultiXactStatuses (one to merely lock tuples, another one to
116 : * update them). This table (and the macros below) helps us determine the
117 : * heavyweight lock mode and MultiXactStatus values to use for any particular
118 : * tuple lock strength.
119 : *
120 : * These interact with InplaceUpdateTupleLock, an alias for ExclusiveLock.
121 : *
122 : * Don't look at lockstatus/updstatus directly! Use get_mxact_status_for_lock
123 : * instead.
124 : */
125 : static const struct
126 : {
127 : LOCKMODE hwlock;
128 : int lockstatus;
129 : int updstatus;
130 : }
131 :
132 : tupleLockExtraInfo[MaxLockTupleMode + 1] =
133 : {
134 : { /* LockTupleKeyShare */
135 : AccessShareLock,
136 : MultiXactStatusForKeyShare,
137 : -1 /* KeyShare does not allow updating tuples */
138 : },
139 : { /* LockTupleShare */
140 : RowShareLock,
141 : MultiXactStatusForShare,
142 : -1 /* Share does not allow updating tuples */
143 : },
144 : { /* LockTupleNoKeyExclusive */
145 : ExclusiveLock,
146 : MultiXactStatusForNoKeyUpdate,
147 : MultiXactStatusNoKeyUpdate
148 : },
149 : { /* LockTupleExclusive */
150 : AccessExclusiveLock,
151 : MultiXactStatusForUpdate,
152 : MultiXactStatusUpdate
153 : }
154 : };
155 :
156 : /* Get the LOCKMODE for a given MultiXactStatus */
157 : #define LOCKMODE_from_mxstatus(status) \
158 : (tupleLockExtraInfo[TUPLOCK_from_mxstatus((status))].hwlock)
159 :
160 : /*
161 : * Acquire heavyweight locks on tuples, using a LockTupleMode strength value.
162 : * This is more readable than having every caller translate it to lock.h's
163 : * LOCKMODE.
164 : */
165 : #define LockTupleTuplock(rel, tup, mode) \
166 : LockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
167 : #define UnlockTupleTuplock(rel, tup, mode) \
168 : UnlockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
169 : #define ConditionalLockTupleTuplock(rel, tup, mode, log) \
170 : ConditionalLockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock, (log))
171 :
172 : #ifdef USE_PREFETCH
173 : /*
174 : * heap_index_delete_tuples and index_delete_prefetch_buffer use this
175 : * structure to coordinate prefetching activity
176 : */
177 : typedef struct
178 : {
179 : BlockNumber cur_hblkno;
180 : int next_item;
181 : int ndeltids;
182 : TM_IndexDelete *deltids;
183 : } IndexDeletePrefetchState;
184 : #endif
185 :
186 : /* heap_index_delete_tuples bottom-up index deletion costing constants */
187 : #define BOTTOMUP_MAX_NBLOCKS 6
188 : #define BOTTOMUP_TOLERANCE_NBLOCKS 3
189 :
190 : /*
191 : * heap_index_delete_tuples uses this when determining which heap blocks it
192 : * must visit to help its bottom-up index deletion caller
193 : */
194 : typedef struct IndexDeleteCounts
195 : {
196 : int16 npromisingtids; /* Number of "promising" TIDs in group */
197 : int16 ntids; /* Number of TIDs in group */
198 : int16 ifirsttid; /* Offset to group's first deltid */
199 : } IndexDeleteCounts;
200 :
201 : /*
202 : * This table maps tuple lock strength values for each particular
203 : * MultiXactStatus value.
204 : */
205 : static const int MultiXactStatusLock[MaxMultiXactStatus + 1] =
206 : {
207 : LockTupleKeyShare, /* ForKeyShare */
208 : LockTupleShare, /* ForShare */
209 : LockTupleNoKeyExclusive, /* ForNoKeyUpdate */
210 : LockTupleExclusive, /* ForUpdate */
211 : LockTupleNoKeyExclusive, /* NoKeyUpdate */
212 : LockTupleExclusive /* Update */
213 : };
214 :
215 : /* Get the LockTupleMode for a given MultiXactStatus */
216 : #define TUPLOCK_from_mxstatus(status) \
217 : (MultiXactStatusLock[(status)])
218 :
219 : /*
220 : * Check that we have a valid snapshot if we might need TOAST access.
221 : */
222 : static inline void
223 21189384 : AssertHasSnapshotForToast(Relation rel)
224 : {
225 : #ifdef USE_ASSERT_CHECKING
226 :
227 : /* bootstrap mode in particular breaks this rule */
228 : if (!IsNormalProcessingMode())
229 : return;
230 :
231 : /* if the relation doesn't have a TOAST table, we are good */
232 : if (!OidIsValid(rel->rd_rel->reltoastrelid))
233 : return;
234 :
235 : Assert(HaveRegisteredOrActiveSnapshot());
236 :
237 : #endif /* USE_ASSERT_CHECKING */
238 21189384 : }
239 :
240 : /* ----------------------------------------------------------------
241 : * heap support routines
242 : * ----------------------------------------------------------------
243 : */
244 :
245 : /*
246 : * Streaming read API callback for parallel sequential scans. Returns the next
247 : * block the caller wants from the read stream or InvalidBlockNumber when done.
248 : */
249 : static BlockNumber
250 202564 : heap_scan_stream_read_next_parallel(ReadStream *stream,
251 : void *callback_private_data,
252 : void *per_buffer_data)
253 : {
254 202564 : HeapScanDesc scan = (HeapScanDesc) callback_private_data;
255 :
256 : Assert(ScanDirectionIsForward(scan->rs_dir));
257 : Assert(scan->rs_base.rs_parallel);
258 :
259 202564 : if (unlikely(!scan->rs_inited))
260 : {
261 : /* parallel scan */
262 3172 : table_block_parallelscan_startblock_init(scan->rs_base.rs_rd,
263 3172 : scan->rs_parallelworkerdata,
264 3172 : (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel,
265 : scan->rs_startblock,
266 : scan->rs_numblocks);
267 :
268 : /* may return InvalidBlockNumber if there are no more blocks */
269 6344 : scan->rs_prefetch_block = table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
270 3172 : scan->rs_parallelworkerdata,
271 3172 : (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel);
272 3172 : scan->rs_inited = true;
273 : }
274 : else
275 : {
276 199392 : scan->rs_prefetch_block = table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
277 199392 : scan->rs_parallelworkerdata, (ParallelBlockTableScanDesc)
278 199392 : scan->rs_base.rs_parallel);
279 : }
280 :
281 202564 : return scan->rs_prefetch_block;
282 : }
283 :
284 : /*
285 : * Streaming read API callback for serial sequential and TID range scans.
286 : * Returns the next block the caller wants from the read stream or
287 : * InvalidBlockNumber when done.
288 : */
289 : static BlockNumber
290 7657150 : heap_scan_stream_read_next_serial(ReadStream *stream,
291 : void *callback_private_data,
292 : void *per_buffer_data)
293 : {
294 7657150 : HeapScanDesc scan = (HeapScanDesc) callback_private_data;
295 :
296 7657150 : if (unlikely(!scan->rs_inited))
297 : {
298 1989682 : scan->rs_prefetch_block = heapgettup_initial_block(scan, scan->rs_dir);
299 1989682 : scan->rs_inited = true;
300 : }
301 : else
302 5667468 : scan->rs_prefetch_block = heapgettup_advance_block(scan,
303 : scan->rs_prefetch_block,
304 : scan->rs_dir);
305 :
306 7657150 : return scan->rs_prefetch_block;
307 : }
308 :
309 : /*
310 : * Read stream API callback for bitmap heap scans.
311 : * Returns the next block the caller wants from the read stream or
312 : * InvalidBlockNumber when done.
313 : */
314 : static BlockNumber
315 431478 : bitmapheap_stream_read_next(ReadStream *pgsr, void *private_data,
316 : void *per_buffer_data)
317 : {
318 431478 : TBMIterateResult *tbmres = per_buffer_data;
319 431478 : BitmapHeapScanDesc bscan = (BitmapHeapScanDesc) private_data;
320 431478 : HeapScanDesc hscan = (HeapScanDesc) bscan;
321 431478 : TableScanDesc sscan = &hscan->rs_base;
322 :
323 : for (;;)
324 : {
325 431478 : CHECK_FOR_INTERRUPTS();
326 :
327 : /* no more entries in the bitmap */
328 431478 : if (!tbm_iterate(&sscan->st.rs_tbmiterator, tbmres))
329 25570 : return InvalidBlockNumber;
330 :
331 : /*
332 : * Ignore any claimed entries past what we think is the end of the
333 : * relation. It may have been extended after the start of our scan (we
334 : * only hold an AccessShareLock, and it could be inserts from this
335 : * backend). We don't take this optimization in SERIALIZABLE
336 : * isolation though, as we need to examine all invisible tuples
337 : * reachable by the index.
338 : */
339 405908 : if (!IsolationIsSerializable() &&
340 405690 : tbmres->blockno >= hscan->rs_nblocks)
341 0 : continue;
342 :
343 405908 : return tbmres->blockno;
344 : }
345 :
346 : /* not reachable */
347 : Assert(false);
348 : }
349 :
350 : /* ----------------
351 : * initscan - scan code common to heap_beginscan and heap_rescan
352 : * ----------------
353 : */
354 : static void
355 2038820 : initscan(HeapScanDesc scan, ScanKey key, bool keep_startblock)
356 : {
357 2038820 : ParallelBlockTableScanDesc bpscan = NULL;
358 : bool allow_strat;
359 : bool allow_sync;
360 :
361 : /*
362 : * Determine the number of blocks we have to scan.
363 : *
364 : * It is sufficient to do this once at scan start, since any tuples added
365 : * while the scan is in progress will be invisible to my snapshot anyway.
366 : * (That is not true when using a non-MVCC snapshot. However, we couldn't
367 : * guarantee to return tuples added after scan start anyway, since they
368 : * might go into pages we already scanned. To guarantee consistent
369 : * results for a non-MVCC snapshot, the caller must hold some higher-level
370 : * lock that ensures the interesting tuple(s) won't change.)
371 : */
372 2038820 : if (scan->rs_base.rs_parallel != NULL)
373 : {
374 4338 : bpscan = (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel;
375 4338 : scan->rs_nblocks = bpscan->phs_nblocks;
376 : }
377 : else
378 2034482 : scan->rs_nblocks = RelationGetNumberOfBlocks(scan->rs_base.rs_rd);
379 :
380 : /*
381 : * If the table is large relative to NBuffers, use a bulk-read access
382 : * strategy and enable synchronized scanning (see syncscan.c). Although
383 : * the thresholds for these features could be different, we make them the
384 : * same so that there are only two behaviors to tune rather than four.
385 : * (However, some callers need to be able to disable one or both of these
386 : * behaviors, independently of the size of the table; also there is a GUC
387 : * variable that can disable synchronized scanning.)
388 : *
389 : * Note that table_block_parallelscan_initialize has a very similar test;
390 : * if you change this, consider changing that one, too.
391 : */
392 2038816 : if (!RelationUsesLocalBuffers(scan->rs_base.rs_rd) &&
393 2024174 : scan->rs_nblocks > NBuffers / 4)
394 : {
395 28602 : allow_strat = (scan->rs_base.rs_flags & SO_ALLOW_STRAT) != 0;
396 28602 : allow_sync = (scan->rs_base.rs_flags & SO_ALLOW_SYNC) != 0;
397 : }
398 : else
399 2010214 : allow_strat = allow_sync = false;
400 :
401 2038816 : if (allow_strat)
402 : {
403 : /* During a rescan, keep the previous strategy object. */
404 25928 : if (scan->rs_strategy == NULL)
405 25562 : scan->rs_strategy = GetAccessStrategy(BAS_BULKREAD);
406 : }
407 : else
408 : {
409 2012888 : if (scan->rs_strategy != NULL)
410 0 : FreeAccessStrategy(scan->rs_strategy);
411 2012888 : scan->rs_strategy = NULL;
412 : }
413 :
414 2038816 : if (scan->rs_base.rs_parallel != NULL)
415 : {
416 : /* For parallel scan, believe whatever ParallelTableScanDesc says. */
417 4338 : if (scan->rs_base.rs_parallel->phs_syncscan)
418 4 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
419 : else
420 4334 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
421 :
422 : /*
423 : * If not rescanning, initialize the startblock. Finding the actual
424 : * start location is done in table_block_parallelscan_startblock_init,
425 : * based on whether an alternative start location has been set with
426 : * heap_setscanlimits, or using the syncscan location, when syncscan
427 : * is enabled.
428 : */
429 4338 : if (!keep_startblock)
430 4110 : scan->rs_startblock = InvalidBlockNumber;
431 : }
432 : else
433 : {
434 2034478 : if (keep_startblock)
435 : {
436 : /*
437 : * When rescanning, we want to keep the previous startblock
438 : * setting, so that rewinding a cursor doesn't generate surprising
439 : * results. Reset the active syncscan setting, though.
440 : */
441 1245922 : if (allow_sync && synchronize_seqscans)
442 100 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
443 : else
444 1245822 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
445 : }
446 788556 : else if (allow_sync && synchronize_seqscans)
447 : {
448 140 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
449 140 : scan->rs_startblock = ss_get_location(scan->rs_base.rs_rd, scan->rs_nblocks);
450 : }
451 : else
452 : {
453 788416 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
454 788416 : scan->rs_startblock = 0;
455 : }
456 : }
457 :
458 2038816 : scan->rs_numblocks = InvalidBlockNumber;
459 2038816 : scan->rs_inited = false;
460 2038816 : scan->rs_ctup.t_data = NULL;
461 2038816 : ItemPointerSetInvalid(&scan->rs_ctup.t_self);
462 2038816 : scan->rs_cbuf = InvalidBuffer;
463 2038816 : scan->rs_cblock = InvalidBlockNumber;
464 2038816 : scan->rs_ntuples = 0;
465 2038816 : scan->rs_cindex = 0;
466 :
467 : /*
468 : * Initialize to ForwardScanDirection because it is most common and
469 : * because heap scans go forward before going backward (e.g. CURSORs).
470 : */
471 2038816 : scan->rs_dir = ForwardScanDirection;
472 2038816 : scan->rs_prefetch_block = InvalidBlockNumber;
473 :
474 : /* page-at-a-time fields are always invalid when not rs_inited */
475 :
476 : /*
477 : * copy the scan key, if appropriate
478 : */
479 2038816 : if (key != NULL && scan->rs_base.rs_nkeys > 0)
480 457644 : memcpy(scan->rs_base.rs_key, key, scan->rs_base.rs_nkeys * sizeof(ScanKeyData));
481 :
482 : /*
483 : * Currently, we only have a stats counter for sequential heap scans (but
484 : * e.g for bitmap scans the underlying bitmap index scans will be counted,
485 : * and for sample scans we update stats for tuple fetches).
486 : */
487 2038816 : if (scan->rs_base.rs_flags & SO_TYPE_SEQSCAN)
488 1992078 : pgstat_count_heap_scan(scan->rs_base.rs_rd);
489 2038816 : }
490 :
491 : /*
492 : * heap_setscanlimits - restrict range of a heapscan
493 : *
494 : * startBlk is the page to start at
495 : * numBlks is number of pages to scan (InvalidBlockNumber means "all")
496 : */
497 : void
498 5772 : heap_setscanlimits(TableScanDesc sscan, BlockNumber startBlk, BlockNumber numBlks)
499 : {
500 5772 : HeapScanDesc scan = (HeapScanDesc) sscan;
501 :
502 : Assert(!scan->rs_inited); /* else too late to change */
503 : /* else rs_startblock is significant */
504 : Assert(!(scan->rs_base.rs_flags & SO_ALLOW_SYNC));
505 :
506 : /* Check startBlk is valid (but allow case of zero blocks...) */
507 : Assert(startBlk == 0 || startBlk < scan->rs_nblocks);
508 :
509 5772 : scan->rs_startblock = startBlk;
510 5772 : scan->rs_numblocks = numBlks;
511 5772 : }
512 :
513 : /*
514 : * Per-tuple loop for heap_prepare_pagescan(). Pulled out so it can be called
515 : * multiple times, with constant arguments for all_visible,
516 : * check_serializable.
517 : */
518 : pg_attribute_always_inline
519 : static int
520 5644686 : page_collect_tuples(HeapScanDesc scan, Snapshot snapshot,
521 : Page page, Buffer buffer,
522 : BlockNumber block, int lines,
523 : bool all_visible, bool check_serializable)
524 : {
525 5644686 : int ntup = 0;
526 : OffsetNumber lineoff;
527 :
528 286697648 : for (lineoff = FirstOffsetNumber; lineoff <= lines; lineoff++)
529 : {
530 281052978 : ItemId lpp = PageGetItemId(page, lineoff);
531 : HeapTupleData loctup;
532 : bool valid;
533 :
534 281052978 : if (!ItemIdIsNormal(lpp))
535 58577408 : continue;
536 :
537 222475570 : loctup.t_data = (HeapTupleHeader) PageGetItem(page, lpp);
538 222475570 : loctup.t_len = ItemIdGetLength(lpp);
539 222475570 : loctup.t_tableOid = RelationGetRelid(scan->rs_base.rs_rd);
540 222475570 : ItemPointerSet(&(loctup.t_self), block, lineoff);
541 :
542 222475570 : if (all_visible)
543 89814160 : valid = true;
544 : else
545 132661410 : valid = HeapTupleSatisfiesVisibility(&loctup, snapshot, buffer);
546 :
547 222475570 : if (check_serializable)
548 2842 : HeapCheckForSerializableConflictOut(valid, scan->rs_base.rs_rd,
549 : &loctup, buffer, snapshot);
550 :
551 222475554 : if (valid)
552 : {
553 206209966 : scan->rs_vistuples[ntup] = lineoff;
554 206209966 : ntup++;
555 : }
556 : }
557 :
558 : Assert(ntup <= MaxHeapTuplesPerPage);
559 :
560 5644670 : return ntup;
561 : }
562 :
563 : /*
564 : * heap_prepare_pagescan - Prepare current scan page to be scanned in pagemode
565 : *
566 : * Preparation currently consists of 1. prune the scan's rs_cbuf page, and 2.
567 : * fill the rs_vistuples[] array with the OffsetNumbers of visible tuples.
568 : */
569 : void
570 5644686 : heap_prepare_pagescan(TableScanDesc sscan)
571 : {
572 5644686 : HeapScanDesc scan = (HeapScanDesc) sscan;
573 5644686 : Buffer buffer = scan->rs_cbuf;
574 5644686 : BlockNumber block = scan->rs_cblock;
575 : Snapshot snapshot;
576 : Page page;
577 : int lines;
578 : bool all_visible;
579 : bool check_serializable;
580 :
581 : Assert(BufferGetBlockNumber(buffer) == block);
582 :
583 : /* ensure we're not accidentally being used when not in pagemode */
584 : Assert(scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE);
585 5644686 : snapshot = scan->rs_base.rs_snapshot;
586 :
587 : /*
588 : * Prune and repair fragmentation for the whole page, if possible.
589 : */
590 5644686 : heap_page_prune_opt(scan->rs_base.rs_rd, buffer);
591 :
592 : /*
593 : * We must hold share lock on the buffer content while examining tuple
594 : * visibility. Afterwards, however, the tuples we have found to be
595 : * visible are guaranteed good as long as we hold the buffer pin.
596 : */
597 5644686 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
598 :
599 5644686 : page = BufferGetPage(buffer);
600 5644686 : lines = PageGetMaxOffsetNumber(page);
601 :
602 : /*
603 : * If the all-visible flag indicates that all tuples on the page are
604 : * visible to everyone, we can skip the per-tuple visibility tests.
605 : *
606 : * Note: In hot standby, a tuple that's already visible to all
607 : * transactions on the primary might still be invisible to a read-only
608 : * transaction in the standby. We partly handle this problem by tracking
609 : * the minimum xmin of visible tuples as the cut-off XID while marking a
610 : * page all-visible on the primary and WAL log that along with the
611 : * visibility map SET operation. In hot standby, we wait for (or abort)
612 : * all transactions that can potentially may not see one or more tuples on
613 : * the page. That's how index-only scans work fine in hot standby. A
614 : * crucial difference between index-only scans and heap scans is that the
615 : * index-only scan completely relies on the visibility map where as heap
616 : * scan looks at the page-level PD_ALL_VISIBLE flag. We are not sure if
617 : * the page-level flag can be trusted in the same way, because it might
618 : * get propagated somehow without being explicitly WAL-logged, e.g. via a
619 : * full page write. Until we can prove that beyond doubt, let's check each
620 : * tuple for visibility the hard way.
621 : */
622 5644686 : all_visible = PageIsAllVisible(page) && !snapshot->takenDuringRecovery;
623 : check_serializable =
624 5644686 : CheckForSerializableConflictOutNeeded(scan->rs_base.rs_rd, snapshot);
625 :
626 : /*
627 : * We call page_collect_tuples() with constant arguments, to get the
628 : * compiler to constant fold the constant arguments. Separate calls with
629 : * constant arguments, rather than variables, are needed on several
630 : * compilers to actually perform constant folding.
631 : */
632 5644686 : if (likely(all_visible))
633 : {
634 2124412 : if (likely(!check_serializable))
635 2124412 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
636 : block, lines, true, false);
637 : else
638 0 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
639 : block, lines, true, true);
640 : }
641 : else
642 : {
643 3520274 : if (likely(!check_serializable))
644 3519010 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
645 : block, lines, false, false);
646 : else
647 1264 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
648 : block, lines, false, true);
649 : }
650 :
651 5644670 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
652 5644670 : }
653 :
654 : /*
655 : * heap_fetch_next_buffer - read and pin the next block from MAIN_FORKNUM.
656 : *
657 : * Read the next block of the scan relation from the read stream and save it
658 : * in the scan descriptor. It is already pinned.
659 : */
660 : static inline void
661 7475240 : heap_fetch_next_buffer(HeapScanDesc scan, ScanDirection dir)
662 : {
663 : Assert(scan->rs_read_stream);
664 :
665 : /* release previous scan buffer, if any */
666 7475240 : if (BufferIsValid(scan->rs_cbuf))
667 : {
668 5482382 : ReleaseBuffer(scan->rs_cbuf);
669 5482382 : scan->rs_cbuf = InvalidBuffer;
670 : }
671 :
672 : /*
673 : * Be sure to check for interrupts at least once per page. Checks at
674 : * higher code levels won't be able to stop a seqscan that encounters many
675 : * pages' worth of consecutive dead tuples.
676 : */
677 7475240 : CHECK_FOR_INTERRUPTS();
678 :
679 : /*
680 : * If the scan direction is changing, reset the prefetch block to the
681 : * current block. Otherwise, we will incorrectly prefetch the blocks
682 : * between the prefetch block and the current block again before
683 : * prefetching blocks in the new, correct scan direction.
684 : */
685 7475236 : if (unlikely(scan->rs_dir != dir))
686 : {
687 154 : scan->rs_prefetch_block = scan->rs_cblock;
688 154 : read_stream_reset(scan->rs_read_stream);
689 : }
690 :
691 7475236 : scan->rs_dir = dir;
692 :
693 7475236 : scan->rs_cbuf = read_stream_next_buffer(scan->rs_read_stream, NULL);
694 7475186 : if (BufferIsValid(scan->rs_cbuf))
695 5821490 : scan->rs_cblock = BufferGetBlockNumber(scan->rs_cbuf);
696 7475186 : }
697 :
698 : /*
699 : * heapgettup_initial_block - return the first BlockNumber to scan
700 : *
701 : * Returns InvalidBlockNumber when there are no blocks to scan. This can
702 : * occur with empty tables and in parallel scans when parallel workers get all
703 : * of the pages before we can get a chance to get our first page.
704 : */
705 : static pg_noinline BlockNumber
706 1989682 : heapgettup_initial_block(HeapScanDesc scan, ScanDirection dir)
707 : {
708 : Assert(!scan->rs_inited);
709 : Assert(scan->rs_base.rs_parallel == NULL);
710 :
711 : /* When there are no pages to scan, return InvalidBlockNumber */
712 1989682 : if (scan->rs_nblocks == 0 || scan->rs_numblocks == 0)
713 1008876 : return InvalidBlockNumber;
714 :
715 980806 : if (ScanDirectionIsForward(dir))
716 : {
717 980742 : return scan->rs_startblock;
718 : }
719 : else
720 : {
721 : /*
722 : * Disable reporting to syncscan logic in a backwards scan; it's not
723 : * very likely anyone else is doing the same thing at the same time,
724 : * and much more likely that we'll just bollix things for forward
725 : * scanners.
726 : */
727 64 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
728 :
729 : /*
730 : * Start from last page of the scan. Ensure we take into account
731 : * rs_numblocks if it's been adjusted by heap_setscanlimits().
732 : */
733 64 : if (scan->rs_numblocks != InvalidBlockNumber)
734 6 : return (scan->rs_startblock + scan->rs_numblocks - 1) % scan->rs_nblocks;
735 :
736 58 : if (scan->rs_startblock > 0)
737 0 : return scan->rs_startblock - 1;
738 :
739 58 : return scan->rs_nblocks - 1;
740 : }
741 : }
742 :
743 :
744 : /*
745 : * heapgettup_start_page - helper function for heapgettup()
746 : *
747 : * Return the next page to scan based on the scan->rs_cbuf and set *linesleft
748 : * to the number of tuples on this page. Also set *lineoff to the first
749 : * offset to scan with forward scans getting the first offset and backward
750 : * getting the final offset on the page.
751 : */
752 : static Page
753 185358 : heapgettup_start_page(HeapScanDesc scan, ScanDirection dir, int *linesleft,
754 : OffsetNumber *lineoff)
755 : {
756 : Page page;
757 :
758 : Assert(scan->rs_inited);
759 : Assert(BufferIsValid(scan->rs_cbuf));
760 :
761 : /* Caller is responsible for ensuring buffer is locked if needed */
762 185358 : page = BufferGetPage(scan->rs_cbuf);
763 :
764 185358 : *linesleft = PageGetMaxOffsetNumber(page) - FirstOffsetNumber + 1;
765 :
766 185358 : if (ScanDirectionIsForward(dir))
767 185358 : *lineoff = FirstOffsetNumber;
768 : else
769 0 : *lineoff = (OffsetNumber) (*linesleft);
770 :
771 : /* lineoff now references the physically previous or next tid */
772 185358 : return page;
773 : }
774 :
775 :
776 : /*
777 : * heapgettup_continue_page - helper function for heapgettup()
778 : *
779 : * Return the next page to scan based on the scan->rs_cbuf and set *linesleft
780 : * to the number of tuples left to scan on this page. Also set *lineoff to
781 : * the next offset to scan according to the ScanDirection in 'dir'.
782 : */
783 : static inline Page
784 14982482 : heapgettup_continue_page(HeapScanDesc scan, ScanDirection dir, int *linesleft,
785 : OffsetNumber *lineoff)
786 : {
787 : Page page;
788 :
789 : Assert(scan->rs_inited);
790 : Assert(BufferIsValid(scan->rs_cbuf));
791 :
792 : /* Caller is responsible for ensuring buffer is locked if needed */
793 14982482 : page = BufferGetPage(scan->rs_cbuf);
794 :
795 14982482 : if (ScanDirectionIsForward(dir))
796 : {
797 14982482 : *lineoff = OffsetNumberNext(scan->rs_coffset);
798 14982482 : *linesleft = PageGetMaxOffsetNumber(page) - (*lineoff) + 1;
799 : }
800 : else
801 : {
802 : /*
803 : * The previous returned tuple may have been vacuumed since the
804 : * previous scan when we use a non-MVCC snapshot, so we must
805 : * re-establish the lineoff <= PageGetMaxOffsetNumber(page) invariant
806 : */
807 0 : *lineoff = Min(PageGetMaxOffsetNumber(page), OffsetNumberPrev(scan->rs_coffset));
808 0 : *linesleft = *lineoff;
809 : }
810 :
811 : /* lineoff now references the physically previous or next tid */
812 14982482 : return page;
813 : }
814 :
815 : /*
816 : * heapgettup_advance_block - helper for heap_fetch_next_buffer()
817 : *
818 : * Given the current block number, the scan direction, and various information
819 : * contained in the scan descriptor, calculate the BlockNumber to scan next
820 : * and return it. If there are no further blocks to scan, return
821 : * InvalidBlockNumber to indicate this fact to the caller.
822 : *
823 : * This should not be called to determine the initial block number -- only for
824 : * subsequent blocks.
825 : *
826 : * This also adjusts rs_numblocks when a limit has been imposed by
827 : * heap_setscanlimits().
828 : */
829 : static inline BlockNumber
830 5667468 : heapgettup_advance_block(HeapScanDesc scan, BlockNumber block, ScanDirection dir)
831 : {
832 : Assert(scan->rs_base.rs_parallel == NULL);
833 :
834 5667468 : if (likely(ScanDirectionIsForward(dir)))
835 : {
836 5667350 : block++;
837 :
838 : /* wrap back to the start of the heap */
839 5667350 : if (block >= scan->rs_nblocks)
840 769678 : block = 0;
841 :
842 : /*
843 : * Report our new scan position for synchronization purposes. We don't
844 : * do that when moving backwards, however. That would just mess up any
845 : * other forward-moving scanners.
846 : *
847 : * Note: we do this before checking for end of scan so that the final
848 : * state of the position hint is back at the start of the rel. That's
849 : * not strictly necessary, but otherwise when you run the same query
850 : * multiple times the starting position would shift a little bit
851 : * backwards on every invocation, which is confusing. We don't
852 : * guarantee any specific ordering in general, though.
853 : */
854 5667350 : if (scan->rs_base.rs_flags & SO_ALLOW_SYNC)
855 22184 : ss_report_location(scan->rs_base.rs_rd, block);
856 :
857 : /* we're done if we're back at where we started */
858 5667350 : if (block == scan->rs_startblock)
859 769596 : return InvalidBlockNumber;
860 :
861 : /* check if the limit imposed by heap_setscanlimits() is met */
862 4897754 : if (scan->rs_numblocks != InvalidBlockNumber)
863 : {
864 4972 : if (--scan->rs_numblocks == 0)
865 3096 : return InvalidBlockNumber;
866 : }
867 :
868 4894658 : return block;
869 : }
870 : else
871 : {
872 : /* we're done if the last block is the start position */
873 118 : if (block == scan->rs_startblock)
874 118 : return InvalidBlockNumber;
875 :
876 : /* check if the limit imposed by heap_setscanlimits() is met */
877 0 : if (scan->rs_numblocks != InvalidBlockNumber)
878 : {
879 0 : if (--scan->rs_numblocks == 0)
880 0 : return InvalidBlockNumber;
881 : }
882 :
883 : /* wrap to the end of the heap when the last page was page 0 */
884 0 : if (block == 0)
885 0 : block = scan->rs_nblocks;
886 :
887 0 : block--;
888 :
889 0 : return block;
890 : }
891 : }
892 :
893 : /* ----------------
894 : * heapgettup - fetch next heap tuple
895 : *
896 : * Initialize the scan if not already done; then advance to the next
897 : * tuple as indicated by "dir"; return the next tuple in scan->rs_ctup,
898 : * or set scan->rs_ctup.t_data = NULL if no more tuples.
899 : *
900 : * Note: the reason nkeys/key are passed separately, even though they are
901 : * kept in the scan descriptor, is that the caller may not want us to check
902 : * the scankeys.
903 : *
904 : * Note: when we fall off the end of the scan in either direction, we
905 : * reset rs_inited. This means that a further request with the same
906 : * scan direction will restart the scan, which is a bit odd, but a
907 : * request with the opposite scan direction will start a fresh scan
908 : * in the proper direction. The latter is required behavior for cursors,
909 : * while the former case is generally undefined behavior in Postgres
910 : * so we don't care too much.
911 : * ----------------
912 : */
913 : static void
914 15024438 : heapgettup(HeapScanDesc scan,
915 : ScanDirection dir,
916 : int nkeys,
917 : ScanKey key)
918 : {
919 15024438 : HeapTuple tuple = &(scan->rs_ctup);
920 : Page page;
921 : OffsetNumber lineoff;
922 : int linesleft;
923 :
924 15024438 : if (likely(scan->rs_inited))
925 : {
926 : /* continue from previously returned page/tuple */
927 14982482 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
928 14982482 : page = heapgettup_continue_page(scan, dir, &linesleft, &lineoff);
929 14982482 : goto continue_page;
930 : }
931 :
932 : /*
933 : * advance the scan until we find a qualifying tuple or run out of stuff
934 : * to scan
935 : */
936 : while (true)
937 : {
938 227018 : heap_fetch_next_buffer(scan, dir);
939 :
940 : /* did we run out of blocks to scan? */
941 227018 : if (!BufferIsValid(scan->rs_cbuf))
942 41660 : break;
943 :
944 : Assert(BufferGetBlockNumber(scan->rs_cbuf) == scan->rs_cblock);
945 :
946 185358 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
947 185358 : page = heapgettup_start_page(scan, dir, &linesleft, &lineoff);
948 15167840 : continue_page:
949 :
950 : /*
951 : * Only continue scanning the page while we have lines left.
952 : *
953 : * Note that this protects us from accessing line pointers past
954 : * PageGetMaxOffsetNumber(); both for forward scans when we resume the
955 : * table scan, and for when we start scanning a new page.
956 : */
957 15242196 : for (; linesleft > 0; linesleft--, lineoff += dir)
958 : {
959 : bool visible;
960 15057134 : ItemId lpp = PageGetItemId(page, lineoff);
961 :
962 15057134 : if (!ItemIdIsNormal(lpp))
963 63920 : continue;
964 :
965 14993214 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
966 14993214 : tuple->t_len = ItemIdGetLength(lpp);
967 14993214 : ItemPointerSet(&(tuple->t_self), scan->rs_cblock, lineoff);
968 :
969 14993214 : visible = HeapTupleSatisfiesVisibility(tuple,
970 : scan->rs_base.rs_snapshot,
971 : scan->rs_cbuf);
972 :
973 14993214 : HeapCheckForSerializableConflictOut(visible, scan->rs_base.rs_rd,
974 : tuple, scan->rs_cbuf,
975 : scan->rs_base.rs_snapshot);
976 :
977 : /* skip tuples not visible to this snapshot */
978 14993214 : if (!visible)
979 10436 : continue;
980 :
981 : /* skip any tuples that don't match the scan key */
982 14982778 : if (key != NULL &&
983 0 : !HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
984 : nkeys, key))
985 0 : continue;
986 :
987 14982778 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
988 14982778 : scan->rs_coffset = lineoff;
989 14982778 : return;
990 : }
991 :
992 : /*
993 : * if we get here, it means we've exhausted the items on this page and
994 : * it's time to move to the next.
995 : */
996 185062 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
997 : }
998 :
999 : /* end of scan */
1000 41660 : if (BufferIsValid(scan->rs_cbuf))
1001 0 : ReleaseBuffer(scan->rs_cbuf);
1002 :
1003 41660 : scan->rs_cbuf = InvalidBuffer;
1004 41660 : scan->rs_cblock = InvalidBlockNumber;
1005 41660 : scan->rs_prefetch_block = InvalidBlockNumber;
1006 41660 : tuple->t_data = NULL;
1007 41660 : scan->rs_inited = false;
1008 : }
1009 :
1010 : /* ----------------
1011 : * heapgettup_pagemode - fetch next heap tuple in page-at-a-time mode
1012 : *
1013 : * Same API as heapgettup, but used in page-at-a-time mode
1014 : *
1015 : * The internal logic is much the same as heapgettup's too, but there are some
1016 : * differences: we do not take the buffer content lock (that only needs to
1017 : * happen inside heap_prepare_pagescan), and we iterate through just the
1018 : * tuples listed in rs_vistuples[] rather than all tuples on the page. Notice
1019 : * that lineindex is 0-based, where the corresponding loop variable lineoff in
1020 : * heapgettup is 1-based.
1021 : * ----------------
1022 : */
1023 : static void
1024 99886736 : heapgettup_pagemode(HeapScanDesc scan,
1025 : ScanDirection dir,
1026 : int nkeys,
1027 : ScanKey key)
1028 : {
1029 99886736 : HeapTuple tuple = &(scan->rs_ctup);
1030 : Page page;
1031 : uint32 lineindex;
1032 : uint32 linesleft;
1033 :
1034 99886736 : if (likely(scan->rs_inited))
1035 : {
1036 : /* continue from previously returned page/tuple */
1037 97935834 : page = BufferGetPage(scan->rs_cbuf);
1038 :
1039 97935834 : lineindex = scan->rs_cindex + dir;
1040 97935834 : if (ScanDirectionIsForward(dir))
1041 97935176 : linesleft = scan->rs_ntuples - lineindex;
1042 : else
1043 658 : linesleft = scan->rs_cindex;
1044 : /* lineindex now references the next or previous visible tid */
1045 :
1046 97935834 : goto continue_page;
1047 : }
1048 :
1049 : /*
1050 : * advance the scan until we find a qualifying tuple or run out of stuff
1051 : * to scan
1052 : */
1053 : while (true)
1054 : {
1055 7248222 : heap_fetch_next_buffer(scan, dir);
1056 :
1057 : /* did we run out of blocks to scan? */
1058 7248168 : if (!BufferIsValid(scan->rs_cbuf))
1059 1612036 : break;
1060 :
1061 : Assert(BufferGetBlockNumber(scan->rs_cbuf) == scan->rs_cblock);
1062 :
1063 : /* prune the page and determine visible tuple offsets */
1064 5636132 : heap_prepare_pagescan((TableScanDesc) scan);
1065 5636116 : page = BufferGetPage(scan->rs_cbuf);
1066 5636116 : linesleft = scan->rs_ntuples;
1067 5636116 : lineindex = ScanDirectionIsForward(dir) ? 0 : linesleft - 1;
1068 :
1069 : /* block is the same for all tuples, set it once outside the loop */
1070 5636116 : ItemPointerSetBlockNumber(&tuple->t_self, scan->rs_cblock);
1071 :
1072 : /* lineindex now references the next or previous visible tid */
1073 103571950 : continue_page:
1074 :
1075 204894072 : for (; linesleft > 0; linesleft--, lineindex += dir)
1076 : {
1077 : ItemId lpp;
1078 : OffsetNumber lineoff;
1079 :
1080 : Assert(lineindex < scan->rs_ntuples);
1081 199596752 : lineoff = scan->rs_vistuples[lineindex];
1082 199596752 : lpp = PageGetItemId(page, lineoff);
1083 : Assert(ItemIdIsNormal(lpp));
1084 :
1085 199596752 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
1086 199596752 : tuple->t_len = ItemIdGetLength(lpp);
1087 199596752 : ItemPointerSetOffsetNumber(&tuple->t_self, lineoff);
1088 :
1089 : /* skip any tuples that don't match the scan key */
1090 199596752 : if (key != NULL &&
1091 102115554 : !HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
1092 : nkeys, key))
1093 101322122 : continue;
1094 :
1095 98274630 : scan->rs_cindex = lineindex;
1096 98274630 : return;
1097 : }
1098 : }
1099 :
1100 : /* end of scan */
1101 1612036 : if (BufferIsValid(scan->rs_cbuf))
1102 0 : ReleaseBuffer(scan->rs_cbuf);
1103 1612036 : scan->rs_cbuf = InvalidBuffer;
1104 1612036 : scan->rs_cblock = InvalidBlockNumber;
1105 1612036 : scan->rs_prefetch_block = InvalidBlockNumber;
1106 1612036 : tuple->t_data = NULL;
1107 1612036 : scan->rs_inited = false;
1108 : }
1109 :
1110 :
1111 : /* ----------------------------------------------------------------
1112 : * heap access method interface
1113 : * ----------------------------------------------------------------
1114 : */
1115 :
1116 :
1117 : TableScanDesc
1118 792670 : heap_beginscan(Relation relation, Snapshot snapshot,
1119 : int nkeys, ScanKey key,
1120 : ParallelTableScanDesc parallel_scan,
1121 : uint32 flags)
1122 : {
1123 : HeapScanDesc scan;
1124 :
1125 : /*
1126 : * increment relation ref count while scanning relation
1127 : *
1128 : * This is just to make really sure the relcache entry won't go away while
1129 : * the scan has a pointer to it. Caller should be holding the rel open
1130 : * anyway, so this is redundant in all normal scenarios...
1131 : */
1132 792670 : RelationIncrementReferenceCount(relation);
1133 :
1134 : /*
1135 : * allocate and initialize scan descriptor
1136 : */
1137 792670 : if (flags & SO_TYPE_BITMAPSCAN)
1138 : {
1139 21388 : BitmapHeapScanDesc bscan = palloc_object(BitmapHeapScanDescData);
1140 :
1141 : /*
1142 : * Bitmap Heap scans do not have any fields that a normal Heap Scan
1143 : * does not have, so no special initializations required here.
1144 : */
1145 21388 : scan = (HeapScanDesc) bscan;
1146 : }
1147 : else
1148 771282 : scan = (HeapScanDesc) palloc_object(HeapScanDescData);
1149 :
1150 792670 : scan->rs_base.rs_rd = relation;
1151 792670 : scan->rs_base.rs_snapshot = snapshot;
1152 792670 : scan->rs_base.rs_nkeys = nkeys;
1153 792670 : scan->rs_base.rs_flags = flags;
1154 792670 : scan->rs_base.rs_parallel = parallel_scan;
1155 792670 : scan->rs_strategy = NULL; /* set in initscan */
1156 792670 : scan->rs_cbuf = InvalidBuffer;
1157 :
1158 : /*
1159 : * Disable page-at-a-time mode if it's not a MVCC-safe snapshot.
1160 : */
1161 792670 : if (!(snapshot && IsMVCCSnapshot(snapshot)))
1162 59928 : scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
1163 :
1164 : /* Check that a historic snapshot is not used for non-catalog tables */
1165 792670 : if (snapshot &&
1166 774698 : IsHistoricMVCCSnapshot(snapshot) &&
1167 1354 : !RelationIsAccessibleInLogicalDecoding(relation))
1168 : {
1169 0 : ereport(ERROR,
1170 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
1171 : errmsg("cannot query non-catalog table \"%s\" during logical decoding",
1172 : RelationGetRelationName(relation))));
1173 : }
1174 :
1175 : /*
1176 : * For seqscan and sample scans in a serializable transaction, acquire a
1177 : * predicate lock on the entire relation. This is required not only to
1178 : * lock all the matching tuples, but also to conflict with new insertions
1179 : * into the table. In an indexscan, we take page locks on the index pages
1180 : * covering the range specified in the scan qual, but in a heap scan there
1181 : * is nothing more fine-grained to lock. A bitmap scan is a different
1182 : * story, there we have already scanned the index and locked the index
1183 : * pages covering the predicate. But in that case we still have to lock
1184 : * any matching heap tuples. For sample scan we could optimize the locking
1185 : * to be at least page-level granularity, but we'd need to add per-tuple
1186 : * locking for that.
1187 : */
1188 792670 : if (scan->rs_base.rs_flags & (SO_TYPE_SEQSCAN | SO_TYPE_SAMPLESCAN))
1189 : {
1190 : /*
1191 : * Ensure a missing snapshot is noticed reliably, even if the
1192 : * isolation mode means predicate locking isn't performed (and
1193 : * therefore the snapshot isn't used here).
1194 : */
1195 : Assert(snapshot);
1196 750544 : PredicateLockRelation(relation, snapshot);
1197 : }
1198 :
1199 : /* we only need to set this up once */
1200 792670 : scan->rs_ctup.t_tableOid = RelationGetRelid(relation);
1201 :
1202 : /*
1203 : * Allocate memory to keep track of page allocation for parallel workers
1204 : * when doing a parallel scan.
1205 : */
1206 792670 : if (parallel_scan != NULL)
1207 4110 : scan->rs_parallelworkerdata = palloc_object(ParallelBlockTableScanWorkerData);
1208 : else
1209 788560 : scan->rs_parallelworkerdata = NULL;
1210 :
1211 : /*
1212 : * we do this here instead of in initscan() because heap_rescan also calls
1213 : * initscan() and we don't want to allocate memory again
1214 : */
1215 792670 : if (nkeys > 0)
1216 457644 : scan->rs_base.rs_key = palloc_array(ScanKeyData, nkeys);
1217 : else
1218 335026 : scan->rs_base.rs_key = NULL;
1219 :
1220 792670 : initscan(scan, key, false);
1221 :
1222 792666 : scan->rs_read_stream = NULL;
1223 :
1224 : /*
1225 : * Set up a read stream for sequential scans and TID range scans. This
1226 : * should be done after initscan() because initscan() allocates the
1227 : * BufferAccessStrategy object passed to the read stream API.
1228 : */
1229 792666 : if (scan->rs_base.rs_flags & SO_TYPE_SEQSCAN ||
1230 42272 : scan->rs_base.rs_flags & SO_TYPE_TIDRANGESCAN)
1231 752374 : {
1232 : ReadStreamBlockNumberCB cb;
1233 :
1234 752374 : if (scan->rs_base.rs_parallel)
1235 4110 : cb = heap_scan_stream_read_next_parallel;
1236 : else
1237 748264 : cb = heap_scan_stream_read_next_serial;
1238 :
1239 : /* ---
1240 : * It is safe to use batchmode as the only locks taken by `cb`
1241 : * are never taken while waiting for IO:
1242 : * - SyncScanLock is used in the non-parallel case
1243 : * - in the parallel case, only spinlocks and atomics are used
1244 : * ---
1245 : */
1246 752374 : scan->rs_read_stream = read_stream_begin_relation(READ_STREAM_SEQUENTIAL |
1247 : READ_STREAM_USE_BATCHING,
1248 : scan->rs_strategy,
1249 : scan->rs_base.rs_rd,
1250 : MAIN_FORKNUM,
1251 : cb,
1252 : scan,
1253 : 0);
1254 : }
1255 40292 : else if (scan->rs_base.rs_flags & SO_TYPE_BITMAPSCAN)
1256 : {
1257 21388 : scan->rs_read_stream = read_stream_begin_relation(READ_STREAM_DEFAULT |
1258 : READ_STREAM_USE_BATCHING,
1259 : scan->rs_strategy,
1260 : scan->rs_base.rs_rd,
1261 : MAIN_FORKNUM,
1262 : bitmapheap_stream_read_next,
1263 : scan,
1264 : sizeof(TBMIterateResult));
1265 : }
1266 :
1267 :
1268 792666 : return (TableScanDesc) scan;
1269 : }
1270 :
1271 : void
1272 1246150 : heap_rescan(TableScanDesc sscan, ScanKey key, bool set_params,
1273 : bool allow_strat, bool allow_sync, bool allow_pagemode)
1274 : {
1275 1246150 : HeapScanDesc scan = (HeapScanDesc) sscan;
1276 :
1277 1246150 : if (set_params)
1278 : {
1279 30 : if (allow_strat)
1280 30 : scan->rs_base.rs_flags |= SO_ALLOW_STRAT;
1281 : else
1282 0 : scan->rs_base.rs_flags &= ~SO_ALLOW_STRAT;
1283 :
1284 30 : if (allow_sync)
1285 12 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
1286 : else
1287 18 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
1288 :
1289 30 : if (allow_pagemode && scan->rs_base.rs_snapshot &&
1290 30 : IsMVCCSnapshot(scan->rs_base.rs_snapshot))
1291 30 : scan->rs_base.rs_flags |= SO_ALLOW_PAGEMODE;
1292 : else
1293 0 : scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
1294 : }
1295 :
1296 : /*
1297 : * unpin scan buffers
1298 : */
1299 1246150 : if (BufferIsValid(scan->rs_cbuf))
1300 : {
1301 3452 : ReleaseBuffer(scan->rs_cbuf);
1302 3452 : scan->rs_cbuf = InvalidBuffer;
1303 : }
1304 :
1305 : /*
1306 : * SO_TYPE_BITMAPSCAN would be cleaned up here, but it does not hold any
1307 : * additional data vs a normal HeapScan
1308 : */
1309 :
1310 : /*
1311 : * The read stream is reset on rescan. This must be done before
1312 : * initscan(), as some state referred to by read_stream_reset() is reset
1313 : * in initscan().
1314 : */
1315 1246150 : if (scan->rs_read_stream)
1316 1246114 : read_stream_reset(scan->rs_read_stream);
1317 :
1318 : /*
1319 : * reinitialize scan descriptor
1320 : */
1321 1246150 : initscan(scan, key, true);
1322 1246150 : }
1323 :
1324 : void
1325 787930 : heap_endscan(TableScanDesc sscan)
1326 : {
1327 787930 : HeapScanDesc scan = (HeapScanDesc) sscan;
1328 :
1329 : /* Note: no locking manipulations needed */
1330 :
1331 : /*
1332 : * unpin scan buffers
1333 : */
1334 787930 : if (BufferIsValid(scan->rs_cbuf))
1335 332396 : ReleaseBuffer(scan->rs_cbuf);
1336 :
1337 : /*
1338 : * Must free the read stream before freeing the BufferAccessStrategy.
1339 : */
1340 787930 : if (scan->rs_read_stream)
1341 769132 : read_stream_end(scan->rs_read_stream);
1342 :
1343 : /*
1344 : * decrement relation reference count and free scan descriptor storage
1345 : */
1346 787930 : RelationDecrementReferenceCount(scan->rs_base.rs_rd);
1347 :
1348 787930 : if (scan->rs_base.rs_key)
1349 457588 : pfree(scan->rs_base.rs_key);
1350 :
1351 787930 : if (scan->rs_strategy != NULL)
1352 25544 : FreeAccessStrategy(scan->rs_strategy);
1353 :
1354 787930 : if (scan->rs_parallelworkerdata != NULL)
1355 4110 : pfree(scan->rs_parallelworkerdata);
1356 :
1357 787930 : if (scan->rs_base.rs_flags & SO_TEMP_SNAPSHOT)
1358 78798 : UnregisterSnapshot(scan->rs_base.rs_snapshot);
1359 :
1360 787930 : pfree(scan);
1361 787930 : }
1362 :
1363 : HeapTuple
1364 19553324 : heap_getnext(TableScanDesc sscan, ScanDirection direction)
1365 : {
1366 19553324 : HeapScanDesc scan = (HeapScanDesc) sscan;
1367 :
1368 : /*
1369 : * This is still widely used directly, without going through table AM, so
1370 : * add a safety check. It's possible we should, at a later point,
1371 : * downgrade this to an assert. The reason for checking the AM routine,
1372 : * rather than the AM oid, is that this allows to write regression tests
1373 : * that create another AM reusing the heap handler.
1374 : */
1375 19553324 : if (unlikely(sscan->rs_rd->rd_tableam != GetHeapamTableAmRoutine()))
1376 0 : ereport(ERROR,
1377 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1378 : errmsg_internal("only heap AM is supported")));
1379 :
1380 : /*
1381 : * We don't expect direct calls to heap_getnext with valid CheckXidAlive
1382 : * for catalog or regular tables. See detailed comments in xact.c where
1383 : * these variables are declared. Normally we have such a check at tableam
1384 : * level API but this is called from many places so we need to ensure it
1385 : * here.
1386 : */
1387 19553324 : if (unlikely(TransactionIdIsValid(CheckXidAlive) && !bsysscan))
1388 0 : elog(ERROR, "unexpected heap_getnext call during logical decoding");
1389 :
1390 : /* Note: no locking manipulations needed */
1391 :
1392 19553324 : if (scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE)
1393 5505128 : heapgettup_pagemode(scan, direction,
1394 5505128 : scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1395 : else
1396 14048196 : heapgettup(scan, direction,
1397 14048196 : scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1398 :
1399 19553322 : if (scan->rs_ctup.t_data == NULL)
1400 132492 : return NULL;
1401 :
1402 : /*
1403 : * if we get here it means we have a new current scan tuple, so point to
1404 : * the proper return buffer and return the tuple.
1405 : */
1406 :
1407 19420830 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1408 :
1409 19420830 : return &scan->rs_ctup;
1410 : }
1411 :
1412 : bool
1413 95346392 : heap_getnextslot(TableScanDesc sscan, ScanDirection direction, TupleTableSlot *slot)
1414 : {
1415 95346392 : HeapScanDesc scan = (HeapScanDesc) sscan;
1416 :
1417 : /* Note: no locking manipulations needed */
1418 :
1419 95346392 : if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1420 94370150 : heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1421 : else
1422 976242 : heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1423 :
1424 95346340 : if (scan->rs_ctup.t_data == NULL)
1425 : {
1426 1520996 : ExecClearTuple(slot);
1427 1520996 : return false;
1428 : }
1429 :
1430 : /*
1431 : * if we get here it means we have a new current scan tuple, so point to
1432 : * the proper return buffer and return the tuple.
1433 : */
1434 :
1435 93825344 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1436 :
1437 93825344 : ExecStoreBufferHeapTuple(&scan->rs_ctup, slot,
1438 : scan->rs_cbuf);
1439 93825344 : return true;
1440 : }
1441 :
1442 : void
1443 2070 : heap_set_tidrange(TableScanDesc sscan, ItemPointer mintid,
1444 : ItemPointer maxtid)
1445 : {
1446 2070 : HeapScanDesc scan = (HeapScanDesc) sscan;
1447 : BlockNumber startBlk;
1448 : BlockNumber numBlks;
1449 : ItemPointerData highestItem;
1450 : ItemPointerData lowestItem;
1451 :
1452 : /*
1453 : * For relations without any pages, we can simply leave the TID range
1454 : * unset. There will be no tuples to scan, therefore no tuples outside
1455 : * the given TID range.
1456 : */
1457 2070 : if (scan->rs_nblocks == 0)
1458 48 : return;
1459 :
1460 : /*
1461 : * Set up some ItemPointers which point to the first and last possible
1462 : * tuples in the heap.
1463 : */
1464 2058 : ItemPointerSet(&highestItem, scan->rs_nblocks - 1, MaxOffsetNumber);
1465 2058 : ItemPointerSet(&lowestItem, 0, FirstOffsetNumber);
1466 :
1467 : /*
1468 : * If the given maximum TID is below the highest possible TID in the
1469 : * relation, then restrict the range to that, otherwise we scan to the end
1470 : * of the relation.
1471 : */
1472 2058 : if (ItemPointerCompare(maxtid, &highestItem) < 0)
1473 260 : ItemPointerCopy(maxtid, &highestItem);
1474 :
1475 : /*
1476 : * If the given minimum TID is above the lowest possible TID in the
1477 : * relation, then restrict the range to only scan for TIDs above that.
1478 : */
1479 2058 : if (ItemPointerCompare(mintid, &lowestItem) > 0)
1480 1822 : ItemPointerCopy(mintid, &lowestItem);
1481 :
1482 : /*
1483 : * Check for an empty range and protect from would be negative results
1484 : * from the numBlks calculation below.
1485 : */
1486 2058 : if (ItemPointerCompare(&highestItem, &lowestItem) < 0)
1487 : {
1488 : /* Set an empty range of blocks to scan */
1489 36 : heap_setscanlimits(sscan, 0, 0);
1490 36 : return;
1491 : }
1492 :
1493 : /*
1494 : * Calculate the first block and the number of blocks we must scan. We
1495 : * could be more aggressive here and perform some more validation to try
1496 : * and further narrow the scope of blocks to scan by checking if the
1497 : * lowestItem has an offset above MaxOffsetNumber. In this case, we could
1498 : * advance startBlk by one. Likewise, if highestItem has an offset of 0
1499 : * we could scan one fewer blocks. However, such an optimization does not
1500 : * seem worth troubling over, currently.
1501 : */
1502 2022 : startBlk = ItemPointerGetBlockNumberNoCheck(&lowestItem);
1503 :
1504 2022 : numBlks = ItemPointerGetBlockNumberNoCheck(&highestItem) -
1505 2022 : ItemPointerGetBlockNumberNoCheck(&lowestItem) + 1;
1506 :
1507 : /* Set the start block and number of blocks to scan */
1508 2022 : heap_setscanlimits(sscan, startBlk, numBlks);
1509 :
1510 : /* Finally, set the TID range in sscan */
1511 2022 : ItemPointerCopy(&lowestItem, &sscan->st.tidrange.rs_mintid);
1512 2022 : ItemPointerCopy(&highestItem, &sscan->st.tidrange.rs_maxtid);
1513 : }
1514 :
1515 : bool
1516 11272 : heap_getnextslot_tidrange(TableScanDesc sscan, ScanDirection direction,
1517 : TupleTableSlot *slot)
1518 : {
1519 11272 : HeapScanDesc scan = (HeapScanDesc) sscan;
1520 11272 : ItemPointer mintid = &sscan->st.tidrange.rs_mintid;
1521 11272 : ItemPointer maxtid = &sscan->st.tidrange.rs_maxtid;
1522 :
1523 : /* Note: no locking manipulations needed */
1524 : for (;;)
1525 : {
1526 11458 : if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1527 11458 : heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1528 : else
1529 0 : heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1530 :
1531 11442 : if (scan->rs_ctup.t_data == NULL)
1532 : {
1533 208 : ExecClearTuple(slot);
1534 208 : return false;
1535 : }
1536 :
1537 : /*
1538 : * heap_set_tidrange will have used heap_setscanlimits to limit the
1539 : * range of pages we scan to only ones that can contain the TID range
1540 : * we're scanning for. Here we must filter out any tuples from these
1541 : * pages that are outside of that range.
1542 : */
1543 11234 : if (ItemPointerCompare(&scan->rs_ctup.t_self, mintid) < 0)
1544 : {
1545 186 : ExecClearTuple(slot);
1546 :
1547 : /*
1548 : * When scanning backwards, the TIDs will be in descending order.
1549 : * Future tuples in this direction will be lower still, so we can
1550 : * just return false to indicate there will be no more tuples.
1551 : */
1552 186 : if (ScanDirectionIsBackward(direction))
1553 0 : return false;
1554 :
1555 186 : continue;
1556 : }
1557 :
1558 : /*
1559 : * Likewise for the final page, we must filter out TIDs greater than
1560 : * maxtid.
1561 : */
1562 11048 : if (ItemPointerCompare(&scan->rs_ctup.t_self, maxtid) > 0)
1563 : {
1564 112 : ExecClearTuple(slot);
1565 :
1566 : /*
1567 : * When scanning forward, the TIDs will be in ascending order.
1568 : * Future tuples in this direction will be higher still, so we can
1569 : * just return false to indicate there will be no more tuples.
1570 : */
1571 112 : if (ScanDirectionIsForward(direction))
1572 112 : return false;
1573 0 : continue;
1574 : }
1575 :
1576 10936 : break;
1577 : }
1578 :
1579 : /*
1580 : * if we get here it means we have a new current scan tuple, so point to
1581 : * the proper return buffer and return the tuple.
1582 : */
1583 10936 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1584 :
1585 10936 : ExecStoreBufferHeapTuple(&scan->rs_ctup, slot, scan->rs_cbuf);
1586 10936 : return true;
1587 : }
1588 :
1589 : /*
1590 : * heap_fetch - retrieve tuple with given tid
1591 : *
1592 : * On entry, tuple->t_self is the TID to fetch. We pin the buffer holding
1593 : * the tuple, fill in the remaining fields of *tuple, and check the tuple
1594 : * against the specified snapshot.
1595 : *
1596 : * If successful (tuple found and passes snapshot time qual), then *userbuf
1597 : * is set to the buffer holding the tuple and true is returned. The caller
1598 : * must unpin the buffer when done with the tuple.
1599 : *
1600 : * If the tuple is not found (ie, item number references a deleted slot),
1601 : * then tuple->t_data is set to NULL, *userbuf is set to InvalidBuffer,
1602 : * and false is returned.
1603 : *
1604 : * If the tuple is found but fails the time qual check, then the behavior
1605 : * depends on the keep_buf parameter. If keep_buf is false, the results
1606 : * are the same as for the tuple-not-found case. If keep_buf is true,
1607 : * then tuple->t_data and *userbuf are returned as for the success case,
1608 : * and again the caller must unpin the buffer; but false is returned.
1609 : *
1610 : * heap_fetch does not follow HOT chains: only the exact TID requested will
1611 : * be fetched.
1612 : *
1613 : * It is somewhat inconsistent that we ereport() on invalid block number but
1614 : * return false on invalid item number. There are a couple of reasons though.
1615 : * One is that the caller can relatively easily check the block number for
1616 : * validity, but cannot check the item number without reading the page
1617 : * himself. Another is that when we are following a t_ctid link, we can be
1618 : * reasonably confident that the page number is valid (since VACUUM shouldn't
1619 : * truncate off the destination page without having killed the referencing
1620 : * tuple first), but the item number might well not be good.
1621 : */
1622 : bool
1623 362450 : heap_fetch(Relation relation,
1624 : Snapshot snapshot,
1625 : HeapTuple tuple,
1626 : Buffer *userbuf,
1627 : bool keep_buf)
1628 : {
1629 362450 : ItemPointer tid = &(tuple->t_self);
1630 : ItemId lp;
1631 : Buffer buffer;
1632 : Page page;
1633 : OffsetNumber offnum;
1634 : bool valid;
1635 :
1636 : /*
1637 : * Fetch and pin the appropriate page of the relation.
1638 : */
1639 362450 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
1640 :
1641 : /*
1642 : * Need share lock on buffer to examine tuple commit status.
1643 : */
1644 362434 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
1645 362434 : page = BufferGetPage(buffer);
1646 :
1647 : /*
1648 : * We'd better check for out-of-range offnum in case of VACUUM since the
1649 : * TID was obtained.
1650 : */
1651 362434 : offnum = ItemPointerGetOffsetNumber(tid);
1652 362434 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1653 : {
1654 6 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1655 6 : ReleaseBuffer(buffer);
1656 6 : *userbuf = InvalidBuffer;
1657 6 : tuple->t_data = NULL;
1658 6 : return false;
1659 : }
1660 :
1661 : /*
1662 : * get the item line pointer corresponding to the requested tid
1663 : */
1664 362428 : lp = PageGetItemId(page, offnum);
1665 :
1666 : /*
1667 : * Must check for deleted tuple.
1668 : */
1669 362428 : if (!ItemIdIsNormal(lp))
1670 : {
1671 668 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1672 668 : ReleaseBuffer(buffer);
1673 668 : *userbuf = InvalidBuffer;
1674 668 : tuple->t_data = NULL;
1675 668 : return false;
1676 : }
1677 :
1678 : /*
1679 : * fill in *tuple fields
1680 : */
1681 361760 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
1682 361760 : tuple->t_len = ItemIdGetLength(lp);
1683 361760 : tuple->t_tableOid = RelationGetRelid(relation);
1684 :
1685 : /*
1686 : * check tuple visibility, then release lock
1687 : */
1688 361760 : valid = HeapTupleSatisfiesVisibility(tuple, snapshot, buffer);
1689 :
1690 361760 : if (valid)
1691 361654 : PredicateLockTID(relation, &(tuple->t_self), snapshot,
1692 361654 : HeapTupleHeaderGetXmin(tuple->t_data));
1693 :
1694 361760 : HeapCheckForSerializableConflictOut(valid, relation, tuple, buffer, snapshot);
1695 :
1696 361760 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1697 :
1698 361760 : if (valid)
1699 : {
1700 : /*
1701 : * All checks passed, so return the tuple as valid. Caller is now
1702 : * responsible for releasing the buffer.
1703 : */
1704 361654 : *userbuf = buffer;
1705 :
1706 361654 : return true;
1707 : }
1708 :
1709 : /* Tuple failed time qual, but maybe caller wants to see it anyway. */
1710 106 : if (keep_buf)
1711 68 : *userbuf = buffer;
1712 : else
1713 : {
1714 38 : ReleaseBuffer(buffer);
1715 38 : *userbuf = InvalidBuffer;
1716 38 : tuple->t_data = NULL;
1717 : }
1718 :
1719 106 : return false;
1720 : }
1721 :
1722 : /*
1723 : * heap_hot_search_buffer - search HOT chain for tuple satisfying snapshot
1724 : *
1725 : * On entry, *tid is the TID of a tuple (either a simple tuple, or the root
1726 : * of a HOT chain), and buffer is the buffer holding this tuple. We search
1727 : * for the first chain member satisfying the given snapshot. If one is
1728 : * found, we update *tid to reference that tuple's offset number, and
1729 : * return true. If no match, return false without modifying *tid.
1730 : *
1731 : * heapTuple is a caller-supplied buffer. When a match is found, we return
1732 : * the tuple here, in addition to updating *tid. If no match is found, the
1733 : * contents of this buffer on return are undefined.
1734 : *
1735 : * If all_dead is not NULL, we check non-visible tuples to see if they are
1736 : * globally dead; *all_dead is set true if all members of the HOT chain
1737 : * are vacuumable, false if not.
1738 : *
1739 : * Unlike heap_fetch, the caller must already have pin and (at least) share
1740 : * lock on the buffer; it is still pinned/locked at exit.
1741 : */
1742 : bool
1743 45449598 : heap_hot_search_buffer(ItemPointer tid, Relation relation, Buffer buffer,
1744 : Snapshot snapshot, HeapTuple heapTuple,
1745 : bool *all_dead, bool first_call)
1746 : {
1747 45449598 : Page page = BufferGetPage(buffer);
1748 45449598 : TransactionId prev_xmax = InvalidTransactionId;
1749 : BlockNumber blkno;
1750 : OffsetNumber offnum;
1751 : bool at_chain_start;
1752 : bool valid;
1753 : bool skip;
1754 45449598 : GlobalVisState *vistest = NULL;
1755 :
1756 : /* If this is not the first call, previous call returned a (live!) tuple */
1757 45449598 : if (all_dead)
1758 38813544 : *all_dead = first_call;
1759 :
1760 45449598 : blkno = ItemPointerGetBlockNumber(tid);
1761 45449598 : offnum = ItemPointerGetOffsetNumber(tid);
1762 45449598 : at_chain_start = first_call;
1763 45449598 : skip = !first_call;
1764 :
1765 : /* XXX: we should assert that a snapshot is pushed or registered */
1766 : Assert(TransactionIdIsValid(RecentXmin));
1767 : Assert(BufferGetBlockNumber(buffer) == blkno);
1768 :
1769 : /* Scan through possible multiple members of HOT-chain */
1770 : for (;;)
1771 3014926 : {
1772 : ItemId lp;
1773 :
1774 : /* check for bogus TID */
1775 48464524 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1776 : break;
1777 :
1778 48464524 : lp = PageGetItemId(page, offnum);
1779 :
1780 : /* check for unused, dead, or redirected items */
1781 48464524 : if (!ItemIdIsNormal(lp))
1782 : {
1783 : /* We should only see a redirect at start of chain */
1784 1698572 : if (ItemIdIsRedirected(lp) && at_chain_start)
1785 : {
1786 : /* Follow the redirect */
1787 982622 : offnum = ItemIdGetRedirect(lp);
1788 982622 : at_chain_start = false;
1789 982622 : continue;
1790 : }
1791 : /* else must be end of chain */
1792 715950 : break;
1793 : }
1794 :
1795 : /*
1796 : * Update heapTuple to point to the element of the HOT chain we're
1797 : * currently investigating. Having t_self set correctly is important
1798 : * because the SSI checks and the *Satisfies routine for historical
1799 : * MVCC snapshots need the correct tid to decide about the visibility.
1800 : */
1801 46765952 : heapTuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
1802 46765952 : heapTuple->t_len = ItemIdGetLength(lp);
1803 46765952 : heapTuple->t_tableOid = RelationGetRelid(relation);
1804 46765952 : ItemPointerSet(&heapTuple->t_self, blkno, offnum);
1805 :
1806 : /*
1807 : * Shouldn't see a HEAP_ONLY tuple at chain start.
1808 : */
1809 46765952 : if (at_chain_start && HeapTupleIsHeapOnly(heapTuple))
1810 0 : break;
1811 :
1812 : /*
1813 : * The xmin should match the previous xmax value, else chain is
1814 : * broken.
1815 : */
1816 48798256 : if (TransactionIdIsValid(prev_xmax) &&
1817 2032304 : !TransactionIdEquals(prev_xmax,
1818 : HeapTupleHeaderGetXmin(heapTuple->t_data)))
1819 0 : break;
1820 :
1821 : /*
1822 : * When first_call is true (and thus, skip is initially false) we'll
1823 : * return the first tuple we find. But on later passes, heapTuple
1824 : * will initially be pointing to the tuple we returned last time.
1825 : * Returning it again would be incorrect (and would loop forever), so
1826 : * we skip it and return the next match we find.
1827 : */
1828 46765952 : if (!skip)
1829 : {
1830 : /* If it's visible per the snapshot, we must return it */
1831 46591636 : valid = HeapTupleSatisfiesVisibility(heapTuple, snapshot, buffer);
1832 46591636 : HeapCheckForSerializableConflictOut(valid, relation, heapTuple,
1833 : buffer, snapshot);
1834 :
1835 46591626 : if (valid)
1836 : {
1837 32271722 : ItemPointerSetOffsetNumber(tid, offnum);
1838 32271722 : PredicateLockTID(relation, &heapTuple->t_self, snapshot,
1839 32271722 : HeapTupleHeaderGetXmin(heapTuple->t_data));
1840 32271722 : if (all_dead)
1841 26195770 : *all_dead = false;
1842 32271722 : return true;
1843 : }
1844 : }
1845 14494220 : skip = false;
1846 :
1847 : /*
1848 : * If we can't see it, maybe no one else can either. At caller
1849 : * request, check whether all chain members are dead to all
1850 : * transactions.
1851 : *
1852 : * Note: if you change the criterion here for what is "dead", fix the
1853 : * planner's get_actual_variable_range() function to match.
1854 : */
1855 14494220 : if (all_dead && *all_dead)
1856 : {
1857 12959906 : if (!vistest)
1858 12709834 : vistest = GlobalVisTestFor(relation);
1859 :
1860 12959906 : if (!HeapTupleIsSurelyDead(heapTuple, vistest))
1861 12240772 : *all_dead = false;
1862 : }
1863 :
1864 : /*
1865 : * Check to see if HOT chain continues past this tuple; if so fetch
1866 : * the next offnum and loop around.
1867 : */
1868 14494220 : if (HeapTupleIsHotUpdated(heapTuple))
1869 : {
1870 : Assert(ItemPointerGetBlockNumber(&heapTuple->t_data->t_ctid) ==
1871 : blkno);
1872 2032304 : offnum = ItemPointerGetOffsetNumber(&heapTuple->t_data->t_ctid);
1873 2032304 : at_chain_start = false;
1874 2032304 : prev_xmax = HeapTupleHeaderGetUpdateXid(heapTuple->t_data);
1875 : }
1876 : else
1877 12461916 : break; /* end of chain */
1878 : }
1879 :
1880 13177866 : return false;
1881 : }
1882 :
1883 : /*
1884 : * heap_get_latest_tid - get the latest tid of a specified tuple
1885 : *
1886 : * Actually, this gets the latest version that is visible according to the
1887 : * scan's snapshot. Create a scan using SnapshotDirty to get the very latest,
1888 : * possibly uncommitted version.
1889 : *
1890 : * *tid is both an input and an output parameter: it is updated to
1891 : * show the latest version of the row. Note that it will not be changed
1892 : * if no version of the row passes the snapshot test.
1893 : */
1894 : void
1895 300 : heap_get_latest_tid(TableScanDesc sscan,
1896 : ItemPointer tid)
1897 : {
1898 300 : Relation relation = sscan->rs_rd;
1899 300 : Snapshot snapshot = sscan->rs_snapshot;
1900 : ItemPointerData ctid;
1901 : TransactionId priorXmax;
1902 :
1903 : /*
1904 : * table_tuple_get_latest_tid() verified that the passed in tid is valid.
1905 : * Assume that t_ctid links are valid however - there shouldn't be invalid
1906 : * ones in the table.
1907 : */
1908 : Assert(ItemPointerIsValid(tid));
1909 :
1910 : /*
1911 : * Loop to chase down t_ctid links. At top of loop, ctid is the tuple we
1912 : * need to examine, and *tid is the TID we will return if ctid turns out
1913 : * to be bogus.
1914 : *
1915 : * Note that we will loop until we reach the end of the t_ctid chain.
1916 : * Depending on the snapshot passed, there might be at most one visible
1917 : * version of the row, but we don't try to optimize for that.
1918 : */
1919 300 : ctid = *tid;
1920 300 : priorXmax = InvalidTransactionId; /* cannot check first XMIN */
1921 : for (;;)
1922 90 : {
1923 : Buffer buffer;
1924 : Page page;
1925 : OffsetNumber offnum;
1926 : ItemId lp;
1927 : HeapTupleData tp;
1928 : bool valid;
1929 :
1930 : /*
1931 : * Read, pin, and lock the page.
1932 : */
1933 390 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(&ctid));
1934 390 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
1935 390 : page = BufferGetPage(buffer);
1936 :
1937 : /*
1938 : * Check for bogus item number. This is not treated as an error
1939 : * condition because it can happen while following a t_ctid link. We
1940 : * just assume that the prior tid is OK and return it unchanged.
1941 : */
1942 390 : offnum = ItemPointerGetOffsetNumber(&ctid);
1943 390 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1944 : {
1945 0 : UnlockReleaseBuffer(buffer);
1946 0 : break;
1947 : }
1948 390 : lp = PageGetItemId(page, offnum);
1949 390 : if (!ItemIdIsNormal(lp))
1950 : {
1951 0 : UnlockReleaseBuffer(buffer);
1952 0 : break;
1953 : }
1954 :
1955 : /* OK to access the tuple */
1956 390 : tp.t_self = ctid;
1957 390 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
1958 390 : tp.t_len = ItemIdGetLength(lp);
1959 390 : tp.t_tableOid = RelationGetRelid(relation);
1960 :
1961 : /*
1962 : * After following a t_ctid link, we might arrive at an unrelated
1963 : * tuple. Check for XMIN match.
1964 : */
1965 480 : if (TransactionIdIsValid(priorXmax) &&
1966 90 : !TransactionIdEquals(priorXmax, HeapTupleHeaderGetXmin(tp.t_data)))
1967 : {
1968 0 : UnlockReleaseBuffer(buffer);
1969 0 : break;
1970 : }
1971 :
1972 : /*
1973 : * Check tuple visibility; if visible, set it as the new result
1974 : * candidate.
1975 : */
1976 390 : valid = HeapTupleSatisfiesVisibility(&tp, snapshot, buffer);
1977 390 : HeapCheckForSerializableConflictOut(valid, relation, &tp, buffer, snapshot);
1978 390 : if (valid)
1979 276 : *tid = ctid;
1980 :
1981 : /*
1982 : * If there's a valid t_ctid link, follow it, else we're done.
1983 : */
1984 552 : if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
1985 276 : HeapTupleHeaderIsOnlyLocked(tp.t_data) ||
1986 228 : HeapTupleHeaderIndicatesMovedPartitions(tp.t_data) ||
1987 114 : ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
1988 : {
1989 300 : UnlockReleaseBuffer(buffer);
1990 300 : break;
1991 : }
1992 :
1993 90 : ctid = tp.t_data->t_ctid;
1994 90 : priorXmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
1995 90 : UnlockReleaseBuffer(buffer);
1996 : } /* end of loop */
1997 300 : }
1998 :
1999 :
2000 : /*
2001 : * UpdateXmaxHintBits - update tuple hint bits after xmax transaction ends
2002 : *
2003 : * This is called after we have waited for the XMAX transaction to terminate.
2004 : * If the transaction aborted, we guarantee the XMAX_INVALID hint bit will
2005 : * be set on exit. If the transaction committed, we set the XMAX_COMMITTED
2006 : * hint bit if possible --- but beware that that may not yet be possible,
2007 : * if the transaction committed asynchronously.
2008 : *
2009 : * Note that if the transaction was a locker only, we set HEAP_XMAX_INVALID
2010 : * even if it commits.
2011 : *
2012 : * Hence callers should look only at XMAX_INVALID.
2013 : *
2014 : * Note this is not allowed for tuples whose xmax is a multixact.
2015 : */
2016 : static void
2017 438 : UpdateXmaxHintBits(HeapTupleHeader tuple, Buffer buffer, TransactionId xid)
2018 : {
2019 : Assert(TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple), xid));
2020 : Assert(!(tuple->t_infomask & HEAP_XMAX_IS_MULTI));
2021 :
2022 438 : if (!(tuple->t_infomask & (HEAP_XMAX_COMMITTED | HEAP_XMAX_INVALID)))
2023 : {
2024 782 : if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask) &&
2025 346 : TransactionIdDidCommit(xid))
2026 292 : HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_COMMITTED,
2027 : xid);
2028 : else
2029 144 : HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_INVALID,
2030 : InvalidTransactionId);
2031 : }
2032 438 : }
2033 :
2034 :
2035 : /*
2036 : * GetBulkInsertState - prepare status object for a bulk insert
2037 : */
2038 : BulkInsertState
2039 5454 : GetBulkInsertState(void)
2040 : {
2041 : BulkInsertState bistate;
2042 :
2043 5454 : bistate = (BulkInsertState) palloc_object(BulkInsertStateData);
2044 5454 : bistate->strategy = GetAccessStrategy(BAS_BULKWRITE);
2045 5454 : bistate->current_buf = InvalidBuffer;
2046 5454 : bistate->next_free = InvalidBlockNumber;
2047 5454 : bistate->last_free = InvalidBlockNumber;
2048 5454 : bistate->already_extended_by = 0;
2049 5454 : return bistate;
2050 : }
2051 :
2052 : /*
2053 : * FreeBulkInsertState - clean up after finishing a bulk insert
2054 : */
2055 : void
2056 5164 : FreeBulkInsertState(BulkInsertState bistate)
2057 : {
2058 5164 : if (bistate->current_buf != InvalidBuffer)
2059 4006 : ReleaseBuffer(bistate->current_buf);
2060 5164 : FreeAccessStrategy(bistate->strategy);
2061 5164 : pfree(bistate);
2062 5164 : }
2063 :
2064 : /*
2065 : * ReleaseBulkInsertStatePin - release a buffer currently held in bistate
2066 : */
2067 : void
2068 161516 : ReleaseBulkInsertStatePin(BulkInsertState bistate)
2069 : {
2070 161516 : if (bistate->current_buf != InvalidBuffer)
2071 60042 : ReleaseBuffer(bistate->current_buf);
2072 161516 : bistate->current_buf = InvalidBuffer;
2073 :
2074 : /*
2075 : * Despite the name, we also reset bulk relation extension state.
2076 : * Otherwise we can end up erroring out due to looking for free space in
2077 : * ->next_free of one partition, even though ->next_free was set when
2078 : * extending another partition. It could obviously also be bad for
2079 : * efficiency to look at existing blocks at offsets from another
2080 : * partition, even if we don't error out.
2081 : */
2082 161516 : bistate->next_free = InvalidBlockNumber;
2083 161516 : bistate->last_free = InvalidBlockNumber;
2084 161516 : }
2085 :
2086 :
2087 : /*
2088 : * heap_insert - insert tuple into a heap
2089 : *
2090 : * The new tuple is stamped with current transaction ID and the specified
2091 : * command ID.
2092 : *
2093 : * See table_tuple_insert for comments about most of the input flags, except
2094 : * that this routine directly takes a tuple rather than a slot.
2095 : *
2096 : * There's corresponding HEAP_INSERT_ options to all the TABLE_INSERT_
2097 : * options, and there additionally is HEAP_INSERT_SPECULATIVE which is used to
2098 : * implement table_tuple_insert_speculative().
2099 : *
2100 : * On return the header fields of *tup are updated to match the stored tuple;
2101 : * in particular tup->t_self receives the actual TID where the tuple was
2102 : * stored. But note that any toasting of fields within the tuple data is NOT
2103 : * reflected into *tup.
2104 : */
2105 : void
2106 16767298 : heap_insert(Relation relation, HeapTuple tup, CommandId cid,
2107 : int options, BulkInsertState bistate)
2108 : {
2109 16767298 : TransactionId xid = GetCurrentTransactionId();
2110 : HeapTuple heaptup;
2111 : Buffer buffer;
2112 16767284 : Buffer vmbuffer = InvalidBuffer;
2113 16767284 : bool all_visible_cleared = false;
2114 :
2115 : /* Cheap, simplistic check that the tuple matches the rel's rowtype. */
2116 : Assert(HeapTupleHeaderGetNatts(tup->t_data) <=
2117 : RelationGetNumberOfAttributes(relation));
2118 :
2119 16767284 : AssertHasSnapshotForToast(relation);
2120 :
2121 : /*
2122 : * Fill in tuple header fields and toast the tuple if necessary.
2123 : *
2124 : * Note: below this point, heaptup is the data we actually intend to store
2125 : * into the relation; tup is the caller's original untoasted data.
2126 : */
2127 16767284 : heaptup = heap_prepare_insert(relation, tup, xid, cid, options);
2128 :
2129 : /*
2130 : * Find buffer to insert this tuple into. If the page is all visible,
2131 : * this will also pin the requisite visibility map page.
2132 : */
2133 16767284 : buffer = RelationGetBufferForTuple(relation, heaptup->t_len,
2134 : InvalidBuffer, options, bistate,
2135 : &vmbuffer, NULL,
2136 : 0);
2137 :
2138 : /*
2139 : * We're about to do the actual insert -- but check for conflict first, to
2140 : * avoid possibly having to roll back work we've just done.
2141 : *
2142 : * This is safe without a recheck as long as there is no possibility of
2143 : * another process scanning the page between this check and the insert
2144 : * being visible to the scan (i.e., an exclusive buffer content lock is
2145 : * continuously held from this point until the tuple insert is visible).
2146 : *
2147 : * For a heap insert, we only need to check for table-level SSI locks. Our
2148 : * new tuple can't possibly conflict with existing tuple locks, and heap
2149 : * page locks are only consolidated versions of tuple locks; they do not
2150 : * lock "gaps" as index page locks do. So we don't need to specify a
2151 : * buffer when making the call, which makes for a faster check.
2152 : */
2153 16767284 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2154 :
2155 : /* NO EREPORT(ERROR) from here till changes are logged */
2156 16767260 : START_CRIT_SECTION();
2157 :
2158 16767260 : RelationPutHeapTuple(relation, buffer, heaptup,
2159 16767260 : (options & HEAP_INSERT_SPECULATIVE) != 0);
2160 :
2161 16767260 : if (PageIsAllVisible(BufferGetPage(buffer)))
2162 : {
2163 15396 : all_visible_cleared = true;
2164 15396 : PageClearAllVisible(BufferGetPage(buffer));
2165 15396 : visibilitymap_clear(relation,
2166 15396 : ItemPointerGetBlockNumber(&(heaptup->t_self)),
2167 : vmbuffer, VISIBILITYMAP_VALID_BITS);
2168 : }
2169 :
2170 : /*
2171 : * XXX Should we set PageSetPrunable on this page ?
2172 : *
2173 : * The inserting transaction may eventually abort thus making this tuple
2174 : * DEAD and hence available for pruning. Though we don't want to optimize
2175 : * for aborts, if no other tuple in this page is UPDATEd/DELETEd, the
2176 : * aborted tuple will never be pruned until next vacuum is triggered.
2177 : *
2178 : * If you do add PageSetPrunable here, add it in heap_xlog_insert too.
2179 : */
2180 :
2181 16767260 : MarkBufferDirty(buffer);
2182 :
2183 : /* XLOG stuff */
2184 16767260 : if (RelationNeedsWAL(relation))
2185 : {
2186 : xl_heap_insert xlrec;
2187 : xl_heap_header xlhdr;
2188 : XLogRecPtr recptr;
2189 14988064 : Page page = BufferGetPage(buffer);
2190 14988064 : uint8 info = XLOG_HEAP_INSERT;
2191 14988064 : int bufflags = 0;
2192 :
2193 : /*
2194 : * If this is a catalog, we need to transmit combo CIDs to properly
2195 : * decode, so log that as well.
2196 : */
2197 14988064 : if (RelationIsAccessibleInLogicalDecoding(relation))
2198 6858 : log_heap_new_cid(relation, heaptup);
2199 :
2200 : /*
2201 : * If this is the single and first tuple on page, we can reinit the
2202 : * page instead of restoring the whole thing. Set flag, and hide
2203 : * buffer references from XLogInsert.
2204 : */
2205 15181334 : if (ItemPointerGetOffsetNumber(&(heaptup->t_self)) == FirstOffsetNumber &&
2206 193270 : PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
2207 : {
2208 191298 : info |= XLOG_HEAP_INIT_PAGE;
2209 191298 : bufflags |= REGBUF_WILL_INIT;
2210 : }
2211 :
2212 14988064 : xlrec.offnum = ItemPointerGetOffsetNumber(&heaptup->t_self);
2213 14988064 : xlrec.flags = 0;
2214 14988064 : if (all_visible_cleared)
2215 15390 : xlrec.flags |= XLH_INSERT_ALL_VISIBLE_CLEARED;
2216 14988064 : if (options & HEAP_INSERT_SPECULATIVE)
2217 4156 : xlrec.flags |= XLH_INSERT_IS_SPECULATIVE;
2218 : Assert(ItemPointerGetBlockNumber(&heaptup->t_self) == BufferGetBlockNumber(buffer));
2219 :
2220 : /*
2221 : * For logical decoding, we need the tuple even if we're doing a full
2222 : * page write, so make sure it's included even if we take a full-page
2223 : * image. (XXX We could alternatively store a pointer into the FPW).
2224 : */
2225 14988064 : if (RelationIsLogicallyLogged(relation) &&
2226 501056 : !(options & HEAP_INSERT_NO_LOGICAL))
2227 : {
2228 501002 : xlrec.flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
2229 501002 : bufflags |= REGBUF_KEEP_DATA;
2230 :
2231 501002 : if (IsToastRelation(relation))
2232 3572 : xlrec.flags |= XLH_INSERT_ON_TOAST_RELATION;
2233 : }
2234 :
2235 14988064 : XLogBeginInsert();
2236 14988064 : XLogRegisterData(&xlrec, SizeOfHeapInsert);
2237 :
2238 14988064 : xlhdr.t_infomask2 = heaptup->t_data->t_infomask2;
2239 14988064 : xlhdr.t_infomask = heaptup->t_data->t_infomask;
2240 14988064 : xlhdr.t_hoff = heaptup->t_data->t_hoff;
2241 :
2242 : /*
2243 : * note we mark xlhdr as belonging to buffer; if XLogInsert decides to
2244 : * write the whole page to the xlog, we don't need to store
2245 : * xl_heap_header in the xlog.
2246 : */
2247 14988064 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD | bufflags);
2248 14988064 : XLogRegisterBufData(0, &xlhdr, SizeOfHeapHeader);
2249 : /* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
2250 14988064 : XLogRegisterBufData(0,
2251 14988064 : (char *) heaptup->t_data + SizeofHeapTupleHeader,
2252 14988064 : heaptup->t_len - SizeofHeapTupleHeader);
2253 :
2254 : /* filtering by origin on a row level is much more efficient */
2255 14988064 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
2256 :
2257 14988064 : recptr = XLogInsert(RM_HEAP_ID, info);
2258 :
2259 14988064 : PageSetLSN(page, recptr);
2260 : }
2261 :
2262 16767260 : END_CRIT_SECTION();
2263 :
2264 16767260 : UnlockReleaseBuffer(buffer);
2265 16767260 : if (vmbuffer != InvalidBuffer)
2266 15952 : ReleaseBuffer(vmbuffer);
2267 :
2268 : /*
2269 : * If tuple is cacheable, mark it for invalidation from the caches in case
2270 : * we abort. Note it is OK to do this after releasing the buffer, because
2271 : * the heaptup data structure is all in local memory, not in the shared
2272 : * buffer.
2273 : */
2274 16767260 : CacheInvalidateHeapTuple(relation, heaptup, NULL);
2275 :
2276 : /* Note: speculative insertions are counted too, even if aborted later */
2277 16767260 : pgstat_count_heap_insert(relation, 1);
2278 :
2279 : /*
2280 : * If heaptup is a private copy, release it. Don't forget to copy t_self
2281 : * back to the caller's image, too.
2282 : */
2283 16767260 : if (heaptup != tup)
2284 : {
2285 36922 : tup->t_self = heaptup->t_self;
2286 36922 : heap_freetuple(heaptup);
2287 : }
2288 16767260 : }
2289 :
2290 : /*
2291 : * Subroutine for heap_insert(). Prepares a tuple for insertion. This sets the
2292 : * tuple header fields and toasts the tuple if necessary. Returns a toasted
2293 : * version of the tuple if it was toasted, or the original tuple if not. Note
2294 : * that in any case, the header fields are also set in the original tuple.
2295 : */
2296 : static HeapTuple
2297 19778522 : heap_prepare_insert(Relation relation, HeapTuple tup, TransactionId xid,
2298 : CommandId cid, int options)
2299 : {
2300 : /*
2301 : * To allow parallel inserts, we need to ensure that they are safe to be
2302 : * performed in workers. We have the infrastructure to allow parallel
2303 : * inserts in general except for the cases where inserts generate a new
2304 : * CommandId (eg. inserts into a table having a foreign key column).
2305 : */
2306 19778522 : if (IsParallelWorker())
2307 0 : ereport(ERROR,
2308 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
2309 : errmsg("cannot insert tuples in a parallel worker")));
2310 :
2311 19778522 : tup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
2312 19778522 : tup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
2313 19778522 : tup->t_data->t_infomask |= HEAP_XMAX_INVALID;
2314 19778522 : HeapTupleHeaderSetXmin(tup->t_data, xid);
2315 19778522 : if (options & HEAP_INSERT_FROZEN)
2316 204176 : HeapTupleHeaderSetXminFrozen(tup->t_data);
2317 :
2318 19778522 : HeapTupleHeaderSetCmin(tup->t_data, cid);
2319 19778522 : HeapTupleHeaderSetXmax(tup->t_data, 0); /* for cleanliness */
2320 19778522 : tup->t_tableOid = RelationGetRelid(relation);
2321 :
2322 : /*
2323 : * If the new tuple is too big for storage or contains already toasted
2324 : * out-of-line attributes from some other relation, invoke the toaster.
2325 : */
2326 19778522 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
2327 62902 : relation->rd_rel->relkind != RELKIND_MATVIEW)
2328 : {
2329 : /* toast table entries should never be recursively toasted */
2330 : Assert(!HeapTupleHasExternal(tup));
2331 62806 : return tup;
2332 : }
2333 19715716 : else if (HeapTupleHasExternal(tup) || tup->t_len > TOAST_TUPLE_THRESHOLD)
2334 37010 : return heap_toast_insert_or_update(relation, tup, NULL, options);
2335 : else
2336 19678706 : return tup;
2337 : }
2338 :
2339 : /*
2340 : * Helper for heap_multi_insert() that computes the number of entire pages
2341 : * that inserting the remaining heaptuples requires. Used to determine how
2342 : * much the relation needs to be extended by.
2343 : */
2344 : static int
2345 760542 : heap_multi_insert_pages(HeapTuple *heaptuples, int done, int ntuples, Size saveFreeSpace)
2346 : {
2347 760542 : size_t page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
2348 760542 : int npages = 1;
2349 :
2350 4958666 : for (int i = done; i < ntuples; i++)
2351 : {
2352 4198124 : size_t tup_sz = sizeof(ItemIdData) + MAXALIGN(heaptuples[i]->t_len);
2353 :
2354 4198124 : if (page_avail < tup_sz)
2355 : {
2356 31236 : npages++;
2357 31236 : page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
2358 : }
2359 4198124 : page_avail -= tup_sz;
2360 : }
2361 :
2362 760542 : return npages;
2363 : }
2364 :
2365 : /*
2366 : * heap_multi_insert - insert multiple tuples into a heap
2367 : *
2368 : * This is like heap_insert(), but inserts multiple tuples in one operation.
2369 : * That's faster than calling heap_insert() in a loop, because when multiple
2370 : * tuples can be inserted on a single page, we can write just a single WAL
2371 : * record covering all of them, and only need to lock/unlock the page once.
2372 : *
2373 : * Note: this leaks memory into the current memory context. You can create a
2374 : * temporary context before calling this, if that's a problem.
2375 : */
2376 : void
2377 747250 : heap_multi_insert(Relation relation, TupleTableSlot **slots, int ntuples,
2378 : CommandId cid, int options, BulkInsertState bistate)
2379 : {
2380 747250 : TransactionId xid = GetCurrentTransactionId();
2381 : HeapTuple *heaptuples;
2382 : int i;
2383 : int ndone;
2384 : PGAlignedBlock scratch;
2385 : Page page;
2386 747250 : Buffer vmbuffer = InvalidBuffer;
2387 : bool needwal;
2388 : Size saveFreeSpace;
2389 747250 : bool need_tuple_data = RelationIsLogicallyLogged(relation);
2390 747250 : bool need_cids = RelationIsAccessibleInLogicalDecoding(relation);
2391 747250 : bool starting_with_empty_page = false;
2392 747250 : int npages = 0;
2393 747250 : int npages_used = 0;
2394 :
2395 : /* currently not needed (thus unsupported) for heap_multi_insert() */
2396 : Assert(!(options & HEAP_INSERT_NO_LOGICAL));
2397 :
2398 747250 : AssertHasSnapshotForToast(relation);
2399 :
2400 747250 : needwal = RelationNeedsWAL(relation);
2401 747250 : saveFreeSpace = RelationGetTargetPageFreeSpace(relation,
2402 : HEAP_DEFAULT_FILLFACTOR);
2403 :
2404 : /* Toast and set header data in all the slots */
2405 747250 : heaptuples = palloc(ntuples * sizeof(HeapTuple));
2406 3758488 : for (i = 0; i < ntuples; i++)
2407 : {
2408 : HeapTuple tuple;
2409 :
2410 3011238 : tuple = ExecFetchSlotHeapTuple(slots[i], true, NULL);
2411 3011238 : slots[i]->tts_tableOid = RelationGetRelid(relation);
2412 3011238 : tuple->t_tableOid = slots[i]->tts_tableOid;
2413 3011238 : heaptuples[i] = heap_prepare_insert(relation, tuple, xid, cid,
2414 : options);
2415 : }
2416 :
2417 : /*
2418 : * We're about to do the actual inserts -- but check for conflict first,
2419 : * to minimize the possibility of having to roll back work we've just
2420 : * done.
2421 : *
2422 : * A check here does not definitively prevent a serialization anomaly;
2423 : * that check MUST be done at least past the point of acquiring an
2424 : * exclusive buffer content lock on every buffer that will be affected,
2425 : * and MAY be done after all inserts are reflected in the buffers and
2426 : * those locks are released; otherwise there is a race condition. Since
2427 : * multiple buffers can be locked and unlocked in the loop below, and it
2428 : * would not be feasible to identify and lock all of those buffers before
2429 : * the loop, we must do a final check at the end.
2430 : *
2431 : * The check here could be omitted with no loss of correctness; it is
2432 : * present strictly as an optimization.
2433 : *
2434 : * For heap inserts, we only need to check for table-level SSI locks. Our
2435 : * new tuples can't possibly conflict with existing tuple locks, and heap
2436 : * page locks are only consolidated versions of tuple locks; they do not
2437 : * lock "gaps" as index page locks do. So we don't need to specify a
2438 : * buffer when making the call, which makes for a faster check.
2439 : */
2440 747250 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2441 :
2442 747250 : ndone = 0;
2443 1524206 : while (ndone < ntuples)
2444 : {
2445 : Buffer buffer;
2446 776956 : bool all_visible_cleared = false;
2447 776956 : bool all_frozen_set = false;
2448 : int nthispage;
2449 :
2450 776956 : CHECK_FOR_INTERRUPTS();
2451 :
2452 : /*
2453 : * Compute number of pages needed to fit the to-be-inserted tuples in
2454 : * the worst case. This will be used to determine how much to extend
2455 : * the relation by in RelationGetBufferForTuple(), if needed. If we
2456 : * filled a prior page from scratch, we can just update our last
2457 : * computation, but if we started with a partially filled page,
2458 : * recompute from scratch, the number of potentially required pages
2459 : * can vary due to tuples needing to fit onto the page, page headers
2460 : * etc.
2461 : */
2462 776956 : if (ndone == 0 || !starting_with_empty_page)
2463 : {
2464 760542 : npages = heap_multi_insert_pages(heaptuples, ndone, ntuples,
2465 : saveFreeSpace);
2466 760542 : npages_used = 0;
2467 : }
2468 : else
2469 16414 : npages_used++;
2470 :
2471 : /*
2472 : * Find buffer where at least the next tuple will fit. If the page is
2473 : * all-visible, this will also pin the requisite visibility map page.
2474 : *
2475 : * Also pin visibility map page if COPY FREEZE inserts tuples into an
2476 : * empty page. See all_frozen_set below.
2477 : */
2478 776956 : buffer = RelationGetBufferForTuple(relation, heaptuples[ndone]->t_len,
2479 : InvalidBuffer, options, bistate,
2480 : &vmbuffer, NULL,
2481 : npages - npages_used);
2482 776956 : page = BufferGetPage(buffer);
2483 :
2484 776956 : starting_with_empty_page = PageGetMaxOffsetNumber(page) == 0;
2485 :
2486 776956 : if (starting_with_empty_page && (options & HEAP_INSERT_FROZEN))
2487 : {
2488 3322 : all_frozen_set = true;
2489 : /* Lock the vmbuffer before entering the critical section */
2490 3322 : LockBuffer(vmbuffer, BUFFER_LOCK_EXCLUSIVE);
2491 : }
2492 :
2493 : /* NO EREPORT(ERROR) from here till changes are logged */
2494 776956 : START_CRIT_SECTION();
2495 :
2496 : /*
2497 : * RelationGetBufferForTuple has ensured that the first tuple fits.
2498 : * Put that on the page, and then as many other tuples as fit.
2499 : */
2500 776956 : RelationPutHeapTuple(relation, buffer, heaptuples[ndone], false);
2501 :
2502 : /*
2503 : * For logical decoding we need combo CIDs to properly decode the
2504 : * catalog.
2505 : */
2506 776956 : if (needwal && need_cids)
2507 10100 : log_heap_new_cid(relation, heaptuples[ndone]);
2508 :
2509 3011238 : for (nthispage = 1; ndone + nthispage < ntuples; nthispage++)
2510 : {
2511 2263988 : HeapTuple heaptup = heaptuples[ndone + nthispage];
2512 :
2513 2263988 : if (PageGetHeapFreeSpace(page) < MAXALIGN(heaptup->t_len) + saveFreeSpace)
2514 29706 : break;
2515 :
2516 2234282 : RelationPutHeapTuple(relation, buffer, heaptup, false);
2517 :
2518 : /*
2519 : * For logical decoding we need combo CIDs to properly decode the
2520 : * catalog.
2521 : */
2522 2234282 : if (needwal && need_cids)
2523 9480 : log_heap_new_cid(relation, heaptup);
2524 : }
2525 :
2526 : /*
2527 : * If the page is all visible, need to clear that, unless we're only
2528 : * going to add further frozen rows to it.
2529 : *
2530 : * If we're only adding already frozen rows to a previously empty
2531 : * page, mark it as all-frozen and update the visibility map. We're
2532 : * already holding a pin on the vmbuffer.
2533 : */
2534 776956 : if (PageIsAllVisible(page) && !(options & HEAP_INSERT_FROZEN))
2535 : {
2536 6698 : all_visible_cleared = true;
2537 6698 : PageClearAllVisible(page);
2538 6698 : visibilitymap_clear(relation,
2539 : BufferGetBlockNumber(buffer),
2540 : vmbuffer, VISIBILITYMAP_VALID_BITS);
2541 : }
2542 770258 : else if (all_frozen_set)
2543 : {
2544 3322 : PageSetAllVisible(page);
2545 3322 : visibilitymap_set_vmbits(BufferGetBlockNumber(buffer),
2546 : vmbuffer,
2547 : VISIBILITYMAP_ALL_VISIBLE |
2548 : VISIBILITYMAP_ALL_FROZEN,
2549 : relation->rd_locator);
2550 : }
2551 :
2552 : /*
2553 : * XXX Should we set PageSetPrunable on this page ? See heap_insert()
2554 : */
2555 :
2556 776956 : MarkBufferDirty(buffer);
2557 :
2558 : /* XLOG stuff */
2559 776956 : if (needwal)
2560 : {
2561 : XLogRecPtr recptr;
2562 : xl_heap_multi_insert *xlrec;
2563 769316 : uint8 info = XLOG_HEAP2_MULTI_INSERT;
2564 : char *tupledata;
2565 : int totaldatalen;
2566 769316 : char *scratchptr = scratch.data;
2567 : bool init;
2568 769316 : int bufflags = 0;
2569 :
2570 : /*
2571 : * If the page was previously empty, we can reinit the page
2572 : * instead of restoring the whole thing.
2573 : */
2574 769316 : init = starting_with_empty_page;
2575 :
2576 : /* allocate xl_heap_multi_insert struct from the scratch area */
2577 769316 : xlrec = (xl_heap_multi_insert *) scratchptr;
2578 769316 : scratchptr += SizeOfHeapMultiInsert;
2579 :
2580 : /*
2581 : * Allocate offsets array. Unless we're reinitializing the page,
2582 : * in that case the tuples are stored in order starting at
2583 : * FirstOffsetNumber and we don't need to store the offsets
2584 : * explicitly.
2585 : */
2586 769316 : if (!init)
2587 742994 : scratchptr += nthispage * sizeof(OffsetNumber);
2588 :
2589 : /* the rest of the scratch space is used for tuple data */
2590 769316 : tupledata = scratchptr;
2591 :
2592 : /* check that the mutually exclusive flags are not both set */
2593 : Assert(!(all_visible_cleared && all_frozen_set));
2594 :
2595 769316 : xlrec->flags = 0;
2596 769316 : if (all_visible_cleared)
2597 6698 : xlrec->flags = XLH_INSERT_ALL_VISIBLE_CLEARED;
2598 :
2599 : /*
2600 : * We don't have to worry about including a conflict xid in the
2601 : * WAL record, as HEAP_INSERT_FROZEN intentionally violates
2602 : * visibility rules.
2603 : */
2604 769316 : if (all_frozen_set)
2605 34 : xlrec->flags = XLH_INSERT_ALL_FROZEN_SET;
2606 :
2607 769316 : xlrec->ntuples = nthispage;
2608 :
2609 : /*
2610 : * Write out an xl_multi_insert_tuple and the tuple data itself
2611 : * for each tuple.
2612 : */
2613 3369756 : for (i = 0; i < nthispage; i++)
2614 : {
2615 2600440 : HeapTuple heaptup = heaptuples[ndone + i];
2616 : xl_multi_insert_tuple *tuphdr;
2617 : int datalen;
2618 :
2619 2600440 : if (!init)
2620 1548104 : xlrec->offsets[i] = ItemPointerGetOffsetNumber(&heaptup->t_self);
2621 : /* xl_multi_insert_tuple needs two-byte alignment. */
2622 2600440 : tuphdr = (xl_multi_insert_tuple *) SHORTALIGN(scratchptr);
2623 2600440 : scratchptr = ((char *) tuphdr) + SizeOfMultiInsertTuple;
2624 :
2625 2600440 : tuphdr->t_infomask2 = heaptup->t_data->t_infomask2;
2626 2600440 : tuphdr->t_infomask = heaptup->t_data->t_infomask;
2627 2600440 : tuphdr->t_hoff = heaptup->t_data->t_hoff;
2628 :
2629 : /* write bitmap [+ padding] [+ oid] + data */
2630 2600440 : datalen = heaptup->t_len - SizeofHeapTupleHeader;
2631 2600440 : memcpy(scratchptr,
2632 2600440 : (char *) heaptup->t_data + SizeofHeapTupleHeader,
2633 : datalen);
2634 2600440 : tuphdr->datalen = datalen;
2635 2600440 : scratchptr += datalen;
2636 : }
2637 769316 : totaldatalen = scratchptr - tupledata;
2638 : Assert((scratchptr - scratch.data) < BLCKSZ);
2639 :
2640 769316 : if (need_tuple_data)
2641 144 : xlrec->flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
2642 :
2643 : /*
2644 : * Signal that this is the last xl_heap_multi_insert record
2645 : * emitted by this call to heap_multi_insert(). Needed for logical
2646 : * decoding so it knows when to cleanup temporary data.
2647 : */
2648 769316 : if (ndone + nthispage == ntuples)
2649 746428 : xlrec->flags |= XLH_INSERT_LAST_IN_MULTI;
2650 :
2651 769316 : if (init)
2652 : {
2653 26322 : info |= XLOG_HEAP_INIT_PAGE;
2654 26322 : bufflags |= REGBUF_WILL_INIT;
2655 : }
2656 :
2657 : /*
2658 : * If we're doing logical decoding, include the new tuple data
2659 : * even if we take a full-page image of the page.
2660 : */
2661 769316 : if (need_tuple_data)
2662 144 : bufflags |= REGBUF_KEEP_DATA;
2663 :
2664 769316 : XLogBeginInsert();
2665 769316 : XLogRegisterData(xlrec, tupledata - scratch.data);
2666 769316 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD | bufflags);
2667 769316 : if (all_frozen_set)
2668 34 : XLogRegisterBuffer(1, vmbuffer, 0);
2669 :
2670 769316 : XLogRegisterBufData(0, tupledata, totaldatalen);
2671 :
2672 : /* filtering by origin on a row level is much more efficient */
2673 769316 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
2674 :
2675 769316 : recptr = XLogInsert(RM_HEAP2_ID, info);
2676 :
2677 769316 : PageSetLSN(page, recptr);
2678 769316 : if (all_frozen_set)
2679 : {
2680 : Assert(BufferIsDirty(vmbuffer));
2681 34 : PageSetLSN(BufferGetPage(vmbuffer), recptr);
2682 : }
2683 : }
2684 :
2685 776956 : END_CRIT_SECTION();
2686 :
2687 776956 : if (all_frozen_set)
2688 3322 : LockBuffer(vmbuffer, BUFFER_LOCK_UNLOCK);
2689 :
2690 776956 : UnlockReleaseBuffer(buffer);
2691 776956 : ndone += nthispage;
2692 :
2693 : /*
2694 : * NB: Only release vmbuffer after inserting all tuples - it's fairly
2695 : * likely that we'll insert into subsequent heap pages that are likely
2696 : * to use the same vm page.
2697 : */
2698 : }
2699 :
2700 : /* We're done with inserting all tuples, so release the last vmbuffer. */
2701 747250 : if (vmbuffer != InvalidBuffer)
2702 6888 : ReleaseBuffer(vmbuffer);
2703 :
2704 : /*
2705 : * We're done with the actual inserts. Check for conflicts again, to
2706 : * ensure that all rw-conflicts in to these inserts are detected. Without
2707 : * this final check, a sequential scan of the heap may have locked the
2708 : * table after the "before" check, missing one opportunity to detect the
2709 : * conflict, and then scanned the table before the new tuples were there,
2710 : * missing the other chance to detect the conflict.
2711 : *
2712 : * For heap inserts, we only need to check for table-level SSI locks. Our
2713 : * new tuples can't possibly conflict with existing tuple locks, and heap
2714 : * page locks are only consolidated versions of tuple locks; they do not
2715 : * lock "gaps" as index page locks do. So we don't need to specify a
2716 : * buffer when making the call.
2717 : */
2718 747250 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2719 :
2720 : /*
2721 : * If tuples are cacheable, mark them for invalidation from the caches in
2722 : * case we abort. Note it is OK to do this after releasing the buffer,
2723 : * because the heaptuples data structure is all in local memory, not in
2724 : * the shared buffer.
2725 : */
2726 747250 : if (IsCatalogRelation(relation))
2727 : {
2728 2551398 : for (i = 0; i < ntuples; i++)
2729 1806584 : CacheInvalidateHeapTuple(relation, heaptuples[i], NULL);
2730 : }
2731 :
2732 : /* copy t_self fields back to the caller's slots */
2733 3758488 : for (i = 0; i < ntuples; i++)
2734 3011238 : slots[i]->tts_tid = heaptuples[i]->t_self;
2735 :
2736 747250 : pgstat_count_heap_insert(relation, ntuples);
2737 747250 : }
2738 :
2739 : /*
2740 : * simple_heap_insert - insert a tuple
2741 : *
2742 : * Currently, this routine differs from heap_insert only in supplying
2743 : * a default command ID and not allowing access to the speedup options.
2744 : *
2745 : * This should be used rather than using heap_insert directly in most places
2746 : * where we are modifying system catalogs.
2747 : */
2748 : void
2749 1854896 : simple_heap_insert(Relation relation, HeapTuple tup)
2750 : {
2751 1854896 : heap_insert(relation, tup, GetCurrentCommandId(true), 0, NULL);
2752 1854896 : }
2753 :
2754 : /*
2755 : * Given infomask/infomask2, compute the bits that must be saved in the
2756 : * "infobits" field of xl_heap_delete, xl_heap_update, xl_heap_lock,
2757 : * xl_heap_lock_updated WAL records.
2758 : *
2759 : * See fix_infomask_from_infobits.
2760 : */
2761 : static uint8
2762 4127386 : compute_infobits(uint16 infomask, uint16 infomask2)
2763 : {
2764 : return
2765 4127386 : ((infomask & HEAP_XMAX_IS_MULTI) != 0 ? XLHL_XMAX_IS_MULTI : 0) |
2766 4127386 : ((infomask & HEAP_XMAX_LOCK_ONLY) != 0 ? XLHL_XMAX_LOCK_ONLY : 0) |
2767 4127386 : ((infomask & HEAP_XMAX_EXCL_LOCK) != 0 ? XLHL_XMAX_EXCL_LOCK : 0) |
2768 : /* note we ignore HEAP_XMAX_SHR_LOCK here */
2769 8254772 : ((infomask & HEAP_XMAX_KEYSHR_LOCK) != 0 ? XLHL_XMAX_KEYSHR_LOCK : 0) |
2770 : ((infomask2 & HEAP_KEYS_UPDATED) != 0 ?
2771 4127386 : XLHL_KEYS_UPDATED : 0);
2772 : }
2773 :
2774 : /*
2775 : * Given two versions of the same t_infomask for a tuple, compare them and
2776 : * return whether the relevant status for a tuple Xmax has changed. This is
2777 : * used after a buffer lock has been released and reacquired: we want to ensure
2778 : * that the tuple state continues to be the same it was when we previously
2779 : * examined it.
2780 : *
2781 : * Note the Xmax field itself must be compared separately.
2782 : */
2783 : static inline bool
2784 10750 : xmax_infomask_changed(uint16 new_infomask, uint16 old_infomask)
2785 : {
2786 10750 : const uint16 interesting =
2787 : HEAP_XMAX_IS_MULTI | HEAP_XMAX_LOCK_ONLY | HEAP_LOCK_MASK;
2788 :
2789 10750 : if ((new_infomask & interesting) != (old_infomask & interesting))
2790 30 : return true;
2791 :
2792 10720 : return false;
2793 : }
2794 :
2795 : /*
2796 : * heap_delete - delete a tuple
2797 : *
2798 : * See table_tuple_delete() for an explanation of the parameters, except that
2799 : * this routine directly takes a tuple rather than a slot.
2800 : *
2801 : * In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
2802 : * t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
2803 : * only for TM_SelfModified, since we cannot obtain cmax from a combo CID
2804 : * generated by another transaction).
2805 : */
2806 : TM_Result
2807 3050542 : heap_delete(Relation relation, const ItemPointerData *tid,
2808 : CommandId cid, Snapshot crosscheck, bool wait,
2809 : TM_FailureData *tmfd, bool changingPart)
2810 : {
2811 : TM_Result result;
2812 3050542 : TransactionId xid = GetCurrentTransactionId();
2813 : ItemId lp;
2814 : HeapTupleData tp;
2815 : Page page;
2816 : BlockNumber block;
2817 : Buffer buffer;
2818 3050542 : Buffer vmbuffer = InvalidBuffer;
2819 : TransactionId new_xmax;
2820 : uint16 new_infomask,
2821 : new_infomask2;
2822 3050542 : bool have_tuple_lock = false;
2823 : bool iscombo;
2824 3050542 : bool all_visible_cleared = false;
2825 3050542 : HeapTuple old_key_tuple = NULL; /* replica identity of the tuple */
2826 3050542 : bool old_key_copied = false;
2827 :
2828 : Assert(ItemPointerIsValid(tid));
2829 :
2830 3050542 : AssertHasSnapshotForToast(relation);
2831 :
2832 : /*
2833 : * Forbid this during a parallel operation, lest it allocate a combo CID.
2834 : * Other workers might need that combo CID for visibility checks, and we
2835 : * have no provision for broadcasting it to them.
2836 : */
2837 3050542 : if (IsInParallelMode())
2838 0 : ereport(ERROR,
2839 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
2840 : errmsg("cannot delete tuples during a parallel operation")));
2841 :
2842 3050542 : block = ItemPointerGetBlockNumber(tid);
2843 3050542 : buffer = ReadBuffer(relation, block);
2844 3050542 : page = BufferGetPage(buffer);
2845 :
2846 : /*
2847 : * Before locking the buffer, pin the visibility map page if it appears to
2848 : * be necessary. Since we haven't got the lock yet, someone else might be
2849 : * in the middle of changing this, so we'll need to recheck after we have
2850 : * the lock.
2851 : */
2852 3050542 : if (PageIsAllVisible(page))
2853 456 : visibilitymap_pin(relation, block, &vmbuffer);
2854 :
2855 3050542 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2856 :
2857 3050542 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
2858 : Assert(ItemIdIsNormal(lp));
2859 :
2860 3050542 : tp.t_tableOid = RelationGetRelid(relation);
2861 3050542 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
2862 3050542 : tp.t_len = ItemIdGetLength(lp);
2863 3050542 : tp.t_self = *tid;
2864 :
2865 2 : l1:
2866 :
2867 : /*
2868 : * If we didn't pin the visibility map page and the page has become all
2869 : * visible while we were busy locking the buffer, we'll have to unlock and
2870 : * re-lock, to avoid holding the buffer lock across an I/O. That's a bit
2871 : * unfortunate, but hopefully shouldn't happen often.
2872 : */
2873 3050544 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
2874 : {
2875 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2876 0 : visibilitymap_pin(relation, block, &vmbuffer);
2877 0 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2878 : }
2879 :
2880 3050544 : result = HeapTupleSatisfiesUpdate(&tp, cid, buffer);
2881 :
2882 3050544 : if (result == TM_Invisible)
2883 : {
2884 0 : UnlockReleaseBuffer(buffer);
2885 0 : ereport(ERROR,
2886 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2887 : errmsg("attempted to delete invisible tuple")));
2888 : }
2889 3050544 : else if (result == TM_BeingModified && wait)
2890 : {
2891 : TransactionId xwait;
2892 : uint16 infomask;
2893 :
2894 : /* must copy state data before unlocking buffer */
2895 81118 : xwait = HeapTupleHeaderGetRawXmax(tp.t_data);
2896 81118 : infomask = tp.t_data->t_infomask;
2897 :
2898 : /*
2899 : * Sleep until concurrent transaction ends -- except when there's a
2900 : * single locker and it's our own transaction. Note we don't care
2901 : * which lock mode the locker has, because we need the strongest one.
2902 : *
2903 : * Before sleeping, we need to acquire tuple lock to establish our
2904 : * priority for the tuple (see heap_lock_tuple). LockTuple will
2905 : * release us when we are next-in-line for the tuple.
2906 : *
2907 : * If we are forced to "start over" below, we keep the tuple lock;
2908 : * this arranges that we stay at the head of the line while rechecking
2909 : * tuple state.
2910 : */
2911 81118 : if (infomask & HEAP_XMAX_IS_MULTI)
2912 : {
2913 16 : bool current_is_member = false;
2914 :
2915 16 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
2916 : LockTupleExclusive, ¤t_is_member))
2917 : {
2918 16 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2919 :
2920 : /*
2921 : * Acquire the lock, if necessary (but skip it when we're
2922 : * requesting a lock and already have one; avoids deadlock).
2923 : */
2924 16 : if (!current_is_member)
2925 12 : heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
2926 : LockWaitBlock, &have_tuple_lock);
2927 :
2928 : /* wait for multixact */
2929 16 : MultiXactIdWait((MultiXactId) xwait, MultiXactStatusUpdate, infomask,
2930 : relation, &(tp.t_self), XLTW_Delete,
2931 : NULL);
2932 16 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2933 :
2934 : /*
2935 : * If xwait had just locked the tuple then some other xact
2936 : * could update this tuple before we get to this point. Check
2937 : * for xmax change, and start over if so.
2938 : *
2939 : * We also must start over if we didn't pin the VM page, and
2940 : * the page has become all visible.
2941 : */
2942 32 : if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
2943 32 : xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
2944 16 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
2945 : xwait))
2946 0 : goto l1;
2947 : }
2948 :
2949 : /*
2950 : * You might think the multixact is necessarily done here, but not
2951 : * so: it could have surviving members, namely our own xact or
2952 : * other subxacts of this backend. It is legal for us to delete
2953 : * the tuple in either case, however (the latter case is
2954 : * essentially a situation of upgrading our former shared lock to
2955 : * exclusive). We don't bother changing the on-disk hint bits
2956 : * since we are about to overwrite the xmax altogether.
2957 : */
2958 : }
2959 81102 : else if (!TransactionIdIsCurrentTransactionId(xwait))
2960 : {
2961 : /*
2962 : * Wait for regular transaction to end; but first, acquire tuple
2963 : * lock.
2964 : */
2965 104 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2966 104 : heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
2967 : LockWaitBlock, &have_tuple_lock);
2968 104 : XactLockTableWait(xwait, relation, &(tp.t_self), XLTW_Delete);
2969 96 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2970 :
2971 : /*
2972 : * xwait is done, but if xwait had just locked the tuple then some
2973 : * other xact could update this tuple before we get to this point.
2974 : * Check for xmax change, and start over if so.
2975 : *
2976 : * We also must start over if we didn't pin the VM page, and the
2977 : * page has become all visible.
2978 : */
2979 192 : if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
2980 190 : xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
2981 94 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
2982 : xwait))
2983 2 : goto l1;
2984 :
2985 : /* Otherwise check if it committed or aborted */
2986 94 : UpdateXmaxHintBits(tp.t_data, buffer, xwait);
2987 : }
2988 :
2989 : /*
2990 : * We may overwrite if previous xmax aborted, or if it committed but
2991 : * only locked the tuple without updating it.
2992 : */
2993 162176 : if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
2994 81130 : HEAP_XMAX_IS_LOCKED_ONLY(tp.t_data->t_infomask) ||
2995 62 : HeapTupleHeaderIsOnlyLocked(tp.t_data))
2996 81054 : result = TM_Ok;
2997 54 : else if (!ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
2998 46 : result = TM_Updated;
2999 : else
3000 8 : result = TM_Deleted;
3001 : }
3002 :
3003 : /* sanity check the result HeapTupleSatisfiesUpdate() and the logic above */
3004 : if (result != TM_Ok)
3005 : {
3006 : Assert(result == TM_SelfModified ||
3007 : result == TM_Updated ||
3008 : result == TM_Deleted ||
3009 : result == TM_BeingModified);
3010 : Assert(!(tp.t_data->t_infomask & HEAP_XMAX_INVALID));
3011 : Assert(result != TM_Updated ||
3012 : !ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid));
3013 : }
3014 :
3015 3050534 : if (crosscheck != InvalidSnapshot && result == TM_Ok)
3016 : {
3017 : /* Perform additional check for transaction-snapshot mode RI updates */
3018 2 : if (!HeapTupleSatisfiesVisibility(&tp, crosscheck, buffer))
3019 2 : result = TM_Updated;
3020 : }
3021 :
3022 3050534 : if (result != TM_Ok)
3023 : {
3024 124 : tmfd->ctid = tp.t_data->t_ctid;
3025 124 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
3026 124 : if (result == TM_SelfModified)
3027 42 : tmfd->cmax = HeapTupleHeaderGetCmax(tp.t_data);
3028 : else
3029 82 : tmfd->cmax = InvalidCommandId;
3030 124 : UnlockReleaseBuffer(buffer);
3031 124 : if (have_tuple_lock)
3032 54 : UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
3033 124 : if (vmbuffer != InvalidBuffer)
3034 0 : ReleaseBuffer(vmbuffer);
3035 124 : return result;
3036 : }
3037 :
3038 : /*
3039 : * We're about to do the actual delete -- check for conflict first, to
3040 : * avoid possibly having to roll back work we've just done.
3041 : *
3042 : * This is safe without a recheck as long as there is no possibility of
3043 : * another process scanning the page between this check and the delete
3044 : * being visible to the scan (i.e., an exclusive buffer content lock is
3045 : * continuously held from this point until the tuple delete is visible).
3046 : */
3047 3050410 : CheckForSerializableConflictIn(relation, tid, BufferGetBlockNumber(buffer));
3048 :
3049 : /* replace cid with a combo CID if necessary */
3050 3050382 : HeapTupleHeaderAdjustCmax(tp.t_data, &cid, &iscombo);
3051 :
3052 : /*
3053 : * Compute replica identity tuple before entering the critical section so
3054 : * we don't PANIC upon a memory allocation failure.
3055 : */
3056 3050382 : old_key_tuple = ExtractReplicaIdentity(relation, &tp, true, &old_key_copied);
3057 :
3058 : /*
3059 : * If this is the first possibly-multixact-able operation in the current
3060 : * transaction, set my per-backend OldestMemberMXactId setting. We can be
3061 : * certain that the transaction will never become a member of any older
3062 : * MultiXactIds than that. (We have to do this even if we end up just
3063 : * using our own TransactionId below, since some other backend could
3064 : * incorporate our XID into a MultiXact immediately afterwards.)
3065 : */
3066 3050382 : MultiXactIdSetOldestMember();
3067 :
3068 3050382 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(tp.t_data),
3069 3050382 : tp.t_data->t_infomask, tp.t_data->t_infomask2,
3070 : xid, LockTupleExclusive, true,
3071 : &new_xmax, &new_infomask, &new_infomask2);
3072 :
3073 3050382 : START_CRIT_SECTION();
3074 :
3075 : /*
3076 : * If this transaction commits, the tuple will become DEAD sooner or
3077 : * later. Set flag that this page is a candidate for pruning once our xid
3078 : * falls below the OldestXmin horizon. If the transaction finally aborts,
3079 : * the subsequent page pruning will be a no-op and the hint will be
3080 : * cleared.
3081 : */
3082 3050382 : PageSetPrunable(page, xid);
3083 :
3084 3050382 : if (PageIsAllVisible(page))
3085 : {
3086 456 : all_visible_cleared = true;
3087 456 : PageClearAllVisible(page);
3088 456 : visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
3089 : vmbuffer, VISIBILITYMAP_VALID_BITS);
3090 : }
3091 :
3092 : /* store transaction information of xact deleting the tuple */
3093 3050382 : tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
3094 3050382 : tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
3095 3050382 : tp.t_data->t_infomask |= new_infomask;
3096 3050382 : tp.t_data->t_infomask2 |= new_infomask2;
3097 3050382 : HeapTupleHeaderClearHotUpdated(tp.t_data);
3098 3050382 : HeapTupleHeaderSetXmax(tp.t_data, new_xmax);
3099 3050382 : HeapTupleHeaderSetCmax(tp.t_data, cid, iscombo);
3100 : /* Make sure there is no forward chain link in t_ctid */
3101 3050382 : tp.t_data->t_ctid = tp.t_self;
3102 :
3103 : /* Signal that this is actually a move into another partition */
3104 3050382 : if (changingPart)
3105 984 : HeapTupleHeaderSetMovedPartitions(tp.t_data);
3106 :
3107 3050382 : MarkBufferDirty(buffer);
3108 :
3109 : /*
3110 : * XLOG stuff
3111 : *
3112 : * NB: heap_abort_speculative() uses the same xlog record and replay
3113 : * routines.
3114 : */
3115 3050382 : if (RelationNeedsWAL(relation))
3116 : {
3117 : xl_heap_delete xlrec;
3118 : xl_heap_header xlhdr;
3119 : XLogRecPtr recptr;
3120 :
3121 : /*
3122 : * For logical decode we need combo CIDs to properly decode the
3123 : * catalog
3124 : */
3125 2925194 : if (RelationIsAccessibleInLogicalDecoding(relation))
3126 12546 : log_heap_new_cid(relation, &tp);
3127 :
3128 2925194 : xlrec.flags = 0;
3129 2925194 : if (all_visible_cleared)
3130 456 : xlrec.flags |= XLH_DELETE_ALL_VISIBLE_CLEARED;
3131 2925194 : if (changingPart)
3132 984 : xlrec.flags |= XLH_DELETE_IS_PARTITION_MOVE;
3133 5850388 : xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
3134 2925194 : tp.t_data->t_infomask2);
3135 2925194 : xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
3136 2925194 : xlrec.xmax = new_xmax;
3137 :
3138 2925194 : if (old_key_tuple != NULL)
3139 : {
3140 94038 : if (relation->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
3141 264 : xlrec.flags |= XLH_DELETE_CONTAINS_OLD_TUPLE;
3142 : else
3143 93774 : xlrec.flags |= XLH_DELETE_CONTAINS_OLD_KEY;
3144 : }
3145 :
3146 2925194 : XLogBeginInsert();
3147 2925194 : XLogRegisterData(&xlrec, SizeOfHeapDelete);
3148 :
3149 2925194 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
3150 :
3151 : /*
3152 : * Log replica identity of the deleted tuple if there is one
3153 : */
3154 2925194 : if (old_key_tuple != NULL)
3155 : {
3156 94038 : xlhdr.t_infomask2 = old_key_tuple->t_data->t_infomask2;
3157 94038 : xlhdr.t_infomask = old_key_tuple->t_data->t_infomask;
3158 94038 : xlhdr.t_hoff = old_key_tuple->t_data->t_hoff;
3159 :
3160 94038 : XLogRegisterData(&xlhdr, SizeOfHeapHeader);
3161 94038 : XLogRegisterData((char *) old_key_tuple->t_data
3162 : + SizeofHeapTupleHeader,
3163 94038 : old_key_tuple->t_len
3164 : - SizeofHeapTupleHeader);
3165 : }
3166 :
3167 : /* filtering by origin on a row level is much more efficient */
3168 2925194 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
3169 :
3170 2925194 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
3171 :
3172 2925194 : PageSetLSN(page, recptr);
3173 : }
3174 :
3175 3050382 : END_CRIT_SECTION();
3176 :
3177 3050382 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3178 :
3179 3050382 : if (vmbuffer != InvalidBuffer)
3180 456 : ReleaseBuffer(vmbuffer);
3181 :
3182 : /*
3183 : * If the tuple has toasted out-of-line attributes, we need to delete
3184 : * those items too. We have to do this before releasing the buffer
3185 : * because we need to look at the contents of the tuple, but it's OK to
3186 : * release the content lock on the buffer first.
3187 : */
3188 3050382 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
3189 5206 : relation->rd_rel->relkind != RELKIND_MATVIEW)
3190 : {
3191 : /* toast table entries should never be recursively toasted */
3192 : Assert(!HeapTupleHasExternal(&tp));
3193 : }
3194 3045196 : else if (HeapTupleHasExternal(&tp))
3195 598 : heap_toast_delete(relation, &tp, false);
3196 :
3197 : /*
3198 : * Mark tuple for invalidation from system caches at next command
3199 : * boundary. We have to do this before releasing the buffer because we
3200 : * need to look at the contents of the tuple.
3201 : */
3202 3050382 : CacheInvalidateHeapTuple(relation, &tp, NULL);
3203 :
3204 : /* Now we can release the buffer */
3205 3050382 : ReleaseBuffer(buffer);
3206 :
3207 : /*
3208 : * Release the lmgr tuple lock, if we had it.
3209 : */
3210 3050382 : if (have_tuple_lock)
3211 52 : UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
3212 :
3213 3050382 : pgstat_count_heap_delete(relation);
3214 :
3215 3050382 : if (old_key_tuple != NULL && old_key_copied)
3216 93776 : heap_freetuple(old_key_tuple);
3217 :
3218 3050382 : return TM_Ok;
3219 : }
3220 :
3221 : /*
3222 : * simple_heap_delete - delete a tuple
3223 : *
3224 : * This routine may be used to delete a tuple when concurrent updates of
3225 : * the target tuple are not expected (for example, because we have a lock
3226 : * on the relation associated with the tuple). Any failure is reported
3227 : * via ereport().
3228 : */
3229 : void
3230 1318594 : simple_heap_delete(Relation relation, const ItemPointerData *tid)
3231 : {
3232 : TM_Result result;
3233 : TM_FailureData tmfd;
3234 :
3235 1318594 : result = heap_delete(relation, tid,
3236 : GetCurrentCommandId(true), InvalidSnapshot,
3237 : true /* wait for commit */ ,
3238 : &tmfd, false /* changingPart */ );
3239 1318594 : switch (result)
3240 : {
3241 0 : case TM_SelfModified:
3242 : /* Tuple was already updated in current command? */
3243 0 : elog(ERROR, "tuple already updated by self");
3244 : break;
3245 :
3246 1318594 : case TM_Ok:
3247 : /* done successfully */
3248 1318594 : break;
3249 :
3250 0 : case TM_Updated:
3251 0 : elog(ERROR, "tuple concurrently updated");
3252 : break;
3253 :
3254 0 : case TM_Deleted:
3255 0 : elog(ERROR, "tuple concurrently deleted");
3256 : break;
3257 :
3258 0 : default:
3259 0 : elog(ERROR, "unrecognized heap_delete status: %u", result);
3260 : break;
3261 : }
3262 1318594 : }
3263 :
3264 : /*
3265 : * heap_update - replace a tuple
3266 : *
3267 : * See table_tuple_update() for an explanation of the parameters, except that
3268 : * this routine directly takes a tuple rather than a slot.
3269 : *
3270 : * In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
3271 : * t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
3272 : * only for TM_SelfModified, since we cannot obtain cmax from a combo CID
3273 : * generated by another transaction).
3274 : */
3275 : TM_Result
3276 624308 : heap_update(Relation relation, const ItemPointerData *otid, HeapTuple newtup,
3277 : CommandId cid, Snapshot crosscheck, bool wait,
3278 : TM_FailureData *tmfd, LockTupleMode *lockmode,
3279 : TU_UpdateIndexes *update_indexes)
3280 : {
3281 : TM_Result result;
3282 624308 : TransactionId xid = GetCurrentTransactionId();
3283 : Bitmapset *hot_attrs;
3284 : Bitmapset *sum_attrs;
3285 : Bitmapset *key_attrs;
3286 : Bitmapset *id_attrs;
3287 : Bitmapset *interesting_attrs;
3288 : Bitmapset *modified_attrs;
3289 : ItemId lp;
3290 : HeapTupleData oldtup;
3291 : HeapTuple heaptup;
3292 624308 : HeapTuple old_key_tuple = NULL;
3293 624308 : bool old_key_copied = false;
3294 : Page page;
3295 : BlockNumber block;
3296 : MultiXactStatus mxact_status;
3297 : Buffer buffer,
3298 : newbuf,
3299 624308 : vmbuffer = InvalidBuffer,
3300 624308 : vmbuffer_new = InvalidBuffer;
3301 : bool need_toast;
3302 : Size newtupsize,
3303 : pagefree;
3304 624308 : bool have_tuple_lock = false;
3305 : bool iscombo;
3306 624308 : bool use_hot_update = false;
3307 624308 : bool summarized_update = false;
3308 : bool key_intact;
3309 624308 : bool all_visible_cleared = false;
3310 624308 : bool all_visible_cleared_new = false;
3311 : bool checked_lockers;
3312 : bool locker_remains;
3313 624308 : bool id_has_external = false;
3314 : TransactionId xmax_new_tuple,
3315 : xmax_old_tuple;
3316 : uint16 infomask_old_tuple,
3317 : infomask2_old_tuple,
3318 : infomask_new_tuple,
3319 : infomask2_new_tuple;
3320 :
3321 : Assert(ItemPointerIsValid(otid));
3322 :
3323 : /* Cheap, simplistic check that the tuple matches the rel's rowtype. */
3324 : Assert(HeapTupleHeaderGetNatts(newtup->t_data) <=
3325 : RelationGetNumberOfAttributes(relation));
3326 :
3327 624308 : AssertHasSnapshotForToast(relation);
3328 :
3329 : /*
3330 : * Forbid this during a parallel operation, lest it allocate a combo CID.
3331 : * Other workers might need that combo CID for visibility checks, and we
3332 : * have no provision for broadcasting it to them.
3333 : */
3334 624308 : if (IsInParallelMode())
3335 0 : ereport(ERROR,
3336 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
3337 : errmsg("cannot update tuples during a parallel operation")));
3338 :
3339 : #ifdef USE_ASSERT_CHECKING
3340 : check_lock_if_inplace_updateable_rel(relation, otid, newtup);
3341 : #endif
3342 :
3343 : /*
3344 : * Fetch the list of attributes to be checked for various operations.
3345 : *
3346 : * For HOT considerations, this is wasted effort if we fail to update or
3347 : * have to put the new tuple on a different page. But we must compute the
3348 : * list before obtaining buffer lock --- in the worst case, if we are
3349 : * doing an update on one of the relevant system catalogs, we could
3350 : * deadlock if we try to fetch the list later. In any case, the relcache
3351 : * caches the data so this is usually pretty cheap.
3352 : *
3353 : * We also need columns used by the replica identity and columns that are
3354 : * considered the "key" of rows in the table.
3355 : *
3356 : * Note that we get copies of each bitmap, so we need not worry about
3357 : * relcache flush happening midway through.
3358 : */
3359 624308 : hot_attrs = RelationGetIndexAttrBitmap(relation,
3360 : INDEX_ATTR_BITMAP_HOT_BLOCKING);
3361 624308 : sum_attrs = RelationGetIndexAttrBitmap(relation,
3362 : INDEX_ATTR_BITMAP_SUMMARIZED);
3363 624308 : key_attrs = RelationGetIndexAttrBitmap(relation, INDEX_ATTR_BITMAP_KEY);
3364 624308 : id_attrs = RelationGetIndexAttrBitmap(relation,
3365 : INDEX_ATTR_BITMAP_IDENTITY_KEY);
3366 624308 : interesting_attrs = NULL;
3367 624308 : interesting_attrs = bms_add_members(interesting_attrs, hot_attrs);
3368 624308 : interesting_attrs = bms_add_members(interesting_attrs, sum_attrs);
3369 624308 : interesting_attrs = bms_add_members(interesting_attrs, key_attrs);
3370 624308 : interesting_attrs = bms_add_members(interesting_attrs, id_attrs);
3371 :
3372 624308 : block = ItemPointerGetBlockNumber(otid);
3373 624308 : INJECTION_POINT("heap_update-before-pin", NULL);
3374 624308 : buffer = ReadBuffer(relation, block);
3375 624308 : page = BufferGetPage(buffer);
3376 :
3377 : /*
3378 : * Before locking the buffer, pin the visibility map page if it appears to
3379 : * be necessary. Since we haven't got the lock yet, someone else might be
3380 : * in the middle of changing this, so we'll need to recheck after we have
3381 : * the lock.
3382 : */
3383 624308 : if (PageIsAllVisible(page))
3384 3120 : visibilitymap_pin(relation, block, &vmbuffer);
3385 :
3386 624308 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3387 :
3388 624308 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(otid));
3389 :
3390 : /*
3391 : * Usually, a buffer pin and/or snapshot blocks pruning of otid, ensuring
3392 : * we see LP_NORMAL here. When the otid origin is a syscache, we may have
3393 : * neither a pin nor a snapshot. Hence, we may see other LP_ states, each
3394 : * of which indicates concurrent pruning.
3395 : *
3396 : * Failing with TM_Updated would be most accurate. However, unlike other
3397 : * TM_Updated scenarios, we don't know the successor ctid in LP_UNUSED and
3398 : * LP_DEAD cases. While the distinction between TM_Updated and TM_Deleted
3399 : * does matter to SQL statements UPDATE and MERGE, those SQL statements
3400 : * hold a snapshot that ensures LP_NORMAL. Hence, the choice between
3401 : * TM_Updated and TM_Deleted affects only the wording of error messages.
3402 : * Settle on TM_Deleted, for two reasons. First, it avoids complicating
3403 : * the specification of when tmfd->ctid is valid. Second, it creates
3404 : * error log evidence that we took this branch.
3405 : *
3406 : * Since it's possible to see LP_UNUSED at otid, it's also possible to see
3407 : * LP_NORMAL for a tuple that replaced LP_UNUSED. If it's a tuple for an
3408 : * unrelated row, we'll fail with "duplicate key value violates unique".
3409 : * XXX if otid is the live, newer version of the newtup row, we'll discard
3410 : * changes originating in versions of this catalog row after the version
3411 : * the caller got from syscache. See syscache-update-pruned.spec.
3412 : */
3413 624308 : if (!ItemIdIsNormal(lp))
3414 : {
3415 : Assert(RelationSupportsSysCache(RelationGetRelid(relation)));
3416 :
3417 2 : UnlockReleaseBuffer(buffer);
3418 : Assert(!have_tuple_lock);
3419 2 : if (vmbuffer != InvalidBuffer)
3420 2 : ReleaseBuffer(vmbuffer);
3421 2 : tmfd->ctid = *otid;
3422 2 : tmfd->xmax = InvalidTransactionId;
3423 2 : tmfd->cmax = InvalidCommandId;
3424 2 : *update_indexes = TU_None;
3425 :
3426 2 : bms_free(hot_attrs);
3427 2 : bms_free(sum_attrs);
3428 2 : bms_free(key_attrs);
3429 2 : bms_free(id_attrs);
3430 : /* modified_attrs not yet initialized */
3431 2 : bms_free(interesting_attrs);
3432 2 : return TM_Deleted;
3433 : }
3434 :
3435 : /*
3436 : * Fill in enough data in oldtup for HeapDetermineColumnsInfo to work
3437 : * properly.
3438 : */
3439 624306 : oldtup.t_tableOid = RelationGetRelid(relation);
3440 624306 : oldtup.t_data = (HeapTupleHeader) PageGetItem(page, lp);
3441 624306 : oldtup.t_len = ItemIdGetLength(lp);
3442 624306 : oldtup.t_self = *otid;
3443 :
3444 : /* the new tuple is ready, except for this: */
3445 624306 : newtup->t_tableOid = RelationGetRelid(relation);
3446 :
3447 : /*
3448 : * Determine columns modified by the update. Additionally, identify
3449 : * whether any of the unmodified replica identity key attributes in the
3450 : * old tuple is externally stored or not. This is required because for
3451 : * such attributes the flattened value won't be WAL logged as part of the
3452 : * new tuple so we must include it as part of the old_key_tuple. See
3453 : * ExtractReplicaIdentity.
3454 : */
3455 624306 : modified_attrs = HeapDetermineColumnsInfo(relation, interesting_attrs,
3456 : id_attrs, &oldtup,
3457 : newtup, &id_has_external);
3458 :
3459 : /*
3460 : * If we're not updating any "key" column, we can grab a weaker lock type.
3461 : * This allows for more concurrency when we are running simultaneously
3462 : * with foreign key checks.
3463 : *
3464 : * Note that if a column gets detoasted while executing the update, but
3465 : * the value ends up being the same, this test will fail and we will use
3466 : * the stronger lock. This is acceptable; the important case to optimize
3467 : * is updates that don't manipulate key columns, not those that
3468 : * serendipitously arrive at the same key values.
3469 : */
3470 624306 : if (!bms_overlap(modified_attrs, key_attrs))
3471 : {
3472 615684 : *lockmode = LockTupleNoKeyExclusive;
3473 615684 : mxact_status = MultiXactStatusNoKeyUpdate;
3474 615684 : key_intact = true;
3475 :
3476 : /*
3477 : * If this is the first possibly-multixact-able operation in the
3478 : * current transaction, set my per-backend OldestMemberMXactId
3479 : * setting. We can be certain that the transaction will never become a
3480 : * member of any older MultiXactIds than that. (We have to do this
3481 : * even if we end up just using our own TransactionId below, since
3482 : * some other backend could incorporate our XID into a MultiXact
3483 : * immediately afterwards.)
3484 : */
3485 615684 : MultiXactIdSetOldestMember();
3486 : }
3487 : else
3488 : {
3489 8622 : *lockmode = LockTupleExclusive;
3490 8622 : mxact_status = MultiXactStatusUpdate;
3491 8622 : key_intact = false;
3492 : }
3493 :
3494 : /*
3495 : * Note: beyond this point, use oldtup not otid to refer to old tuple.
3496 : * otid may very well point at newtup->t_self, which we will overwrite
3497 : * with the new tuple's location, so there's great risk of confusion if we
3498 : * use otid anymore.
3499 : */
3500 :
3501 624306 : l2:
3502 624308 : checked_lockers = false;
3503 624308 : locker_remains = false;
3504 624308 : result = HeapTupleSatisfiesUpdate(&oldtup, cid, buffer);
3505 :
3506 : /* see below about the "no wait" case */
3507 : Assert(result != TM_BeingModified || wait);
3508 :
3509 624308 : if (result == TM_Invisible)
3510 : {
3511 0 : UnlockReleaseBuffer(buffer);
3512 0 : ereport(ERROR,
3513 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3514 : errmsg("attempted to update invisible tuple")));
3515 : }
3516 624308 : else if (result == TM_BeingModified && wait)
3517 : {
3518 : TransactionId xwait;
3519 : uint16 infomask;
3520 72204 : bool can_continue = false;
3521 :
3522 : /*
3523 : * XXX note that we don't consider the "no wait" case here. This
3524 : * isn't a problem currently because no caller uses that case, but it
3525 : * should be fixed if such a caller is introduced. It wasn't a
3526 : * problem previously because this code would always wait, but now
3527 : * that some tuple locks do not conflict with one of the lock modes we
3528 : * use, it is possible that this case is interesting to handle
3529 : * specially.
3530 : *
3531 : * This may cause failures with third-party code that calls
3532 : * heap_update directly.
3533 : */
3534 :
3535 : /* must copy state data before unlocking buffer */
3536 72204 : xwait = HeapTupleHeaderGetRawXmax(oldtup.t_data);
3537 72204 : infomask = oldtup.t_data->t_infomask;
3538 :
3539 : /*
3540 : * Now we have to do something about the existing locker. If it's a
3541 : * multi, sleep on it; we might be awakened before it is completely
3542 : * gone (or even not sleep at all in some cases); we need to preserve
3543 : * it as locker, unless it is gone completely.
3544 : *
3545 : * If it's not a multi, we need to check for sleeping conditions
3546 : * before actually going to sleep. If the update doesn't conflict
3547 : * with the locks, we just continue without sleeping (but making sure
3548 : * it is preserved).
3549 : *
3550 : * Before sleeping, we need to acquire tuple lock to establish our
3551 : * priority for the tuple (see heap_lock_tuple). LockTuple will
3552 : * release us when we are next-in-line for the tuple. Note we must
3553 : * not acquire the tuple lock until we're sure we're going to sleep;
3554 : * otherwise we're open for race conditions with other transactions
3555 : * holding the tuple lock which sleep on us.
3556 : *
3557 : * If we are forced to "start over" below, we keep the tuple lock;
3558 : * this arranges that we stay at the head of the line while rechecking
3559 : * tuple state.
3560 : */
3561 72204 : if (infomask & HEAP_XMAX_IS_MULTI)
3562 : {
3563 : TransactionId update_xact;
3564 : int remain;
3565 358 : bool current_is_member = false;
3566 :
3567 358 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
3568 : *lockmode, ¤t_is_member))
3569 : {
3570 16 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3571 :
3572 : /*
3573 : * Acquire the lock, if necessary (but skip it when we're
3574 : * requesting a lock and already have one; avoids deadlock).
3575 : */
3576 16 : if (!current_is_member)
3577 0 : heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3578 : LockWaitBlock, &have_tuple_lock);
3579 :
3580 : /* wait for multixact */
3581 16 : MultiXactIdWait((MultiXactId) xwait, mxact_status, infomask,
3582 : relation, &oldtup.t_self, XLTW_Update,
3583 : &remain);
3584 16 : checked_lockers = true;
3585 16 : locker_remains = remain != 0;
3586 16 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3587 :
3588 : /*
3589 : * If xwait had just locked the tuple then some other xact
3590 : * could update this tuple before we get to this point. Check
3591 : * for xmax change, and start over if so.
3592 : */
3593 16 : if (xmax_infomask_changed(oldtup.t_data->t_infomask,
3594 16 : infomask) ||
3595 16 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3596 : xwait))
3597 0 : goto l2;
3598 : }
3599 :
3600 : /*
3601 : * Note that the multixact may not be done by now. It could have
3602 : * surviving members; our own xact or other subxacts of this
3603 : * backend, and also any other concurrent transaction that locked
3604 : * the tuple with LockTupleKeyShare if we only got
3605 : * LockTupleNoKeyExclusive. If this is the case, we have to be
3606 : * careful to mark the updated tuple with the surviving members in
3607 : * Xmax.
3608 : *
3609 : * Note that there could have been another update in the
3610 : * MultiXact. In that case, we need to check whether it committed
3611 : * or aborted. If it aborted we are safe to update it again;
3612 : * otherwise there is an update conflict, and we have to return
3613 : * TableTuple{Deleted, Updated} below.
3614 : *
3615 : * In the LockTupleExclusive case, we still need to preserve the
3616 : * surviving members: those would include the tuple locks we had
3617 : * before this one, which are important to keep in case this
3618 : * subxact aborts.
3619 : */
3620 358 : if (!HEAP_XMAX_IS_LOCKED_ONLY(oldtup.t_data->t_infomask))
3621 16 : update_xact = HeapTupleGetUpdateXid(oldtup.t_data);
3622 : else
3623 342 : update_xact = InvalidTransactionId;
3624 :
3625 : /*
3626 : * There was no UPDATE in the MultiXact; or it aborted. No
3627 : * TransactionIdIsInProgress() call needed here, since we called
3628 : * MultiXactIdWait() above.
3629 : */
3630 374 : if (!TransactionIdIsValid(update_xact) ||
3631 16 : TransactionIdDidAbort(update_xact))
3632 344 : can_continue = true;
3633 : }
3634 71846 : else if (TransactionIdIsCurrentTransactionId(xwait))
3635 : {
3636 : /*
3637 : * The only locker is ourselves; we can avoid grabbing the tuple
3638 : * lock here, but must preserve our locking information.
3639 : */
3640 71634 : checked_lockers = true;
3641 71634 : locker_remains = true;
3642 71634 : can_continue = true;
3643 : }
3644 212 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) && key_intact)
3645 : {
3646 : /*
3647 : * If it's just a key-share locker, and we're not changing the key
3648 : * columns, we don't need to wait for it to end; but we need to
3649 : * preserve it as locker.
3650 : */
3651 58 : checked_lockers = true;
3652 58 : locker_remains = true;
3653 58 : can_continue = true;
3654 : }
3655 : else
3656 : {
3657 : /*
3658 : * Wait for regular transaction to end; but first, acquire tuple
3659 : * lock.
3660 : */
3661 154 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3662 154 : heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3663 : LockWaitBlock, &have_tuple_lock);
3664 154 : XactLockTableWait(xwait, relation, &oldtup.t_self,
3665 : XLTW_Update);
3666 154 : checked_lockers = true;
3667 154 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3668 :
3669 : /*
3670 : * xwait is done, but if xwait had just locked the tuple then some
3671 : * other xact could update this tuple before we get to this point.
3672 : * Check for xmax change, and start over if so.
3673 : */
3674 306 : if (xmax_infomask_changed(oldtup.t_data->t_infomask, infomask) ||
3675 152 : !TransactionIdEquals(xwait,
3676 : HeapTupleHeaderGetRawXmax(oldtup.t_data)))
3677 2 : goto l2;
3678 :
3679 : /* Otherwise check if it committed or aborted */
3680 152 : UpdateXmaxHintBits(oldtup.t_data, buffer, xwait);
3681 152 : if (oldtup.t_data->t_infomask & HEAP_XMAX_INVALID)
3682 44 : can_continue = true;
3683 : }
3684 :
3685 72202 : if (can_continue)
3686 72080 : result = TM_Ok;
3687 122 : else if (!ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid))
3688 112 : result = TM_Updated;
3689 : else
3690 10 : result = TM_Deleted;
3691 : }
3692 :
3693 : /* Sanity check the result HeapTupleSatisfiesUpdate() and the logic above */
3694 : if (result != TM_Ok)
3695 : {
3696 : Assert(result == TM_SelfModified ||
3697 : result == TM_Updated ||
3698 : result == TM_Deleted ||
3699 : result == TM_BeingModified);
3700 : Assert(!(oldtup.t_data->t_infomask & HEAP_XMAX_INVALID));
3701 : Assert(result != TM_Updated ||
3702 : !ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid));
3703 : }
3704 :
3705 624306 : if (crosscheck != InvalidSnapshot && result == TM_Ok)
3706 : {
3707 : /* Perform additional check for transaction-snapshot mode RI updates */
3708 2 : if (!HeapTupleSatisfiesVisibility(&oldtup, crosscheck, buffer))
3709 2 : result = TM_Updated;
3710 : }
3711 :
3712 624306 : if (result != TM_Ok)
3713 : {
3714 318 : tmfd->ctid = oldtup.t_data->t_ctid;
3715 318 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(oldtup.t_data);
3716 318 : if (result == TM_SelfModified)
3717 104 : tmfd->cmax = HeapTupleHeaderGetCmax(oldtup.t_data);
3718 : else
3719 214 : tmfd->cmax = InvalidCommandId;
3720 318 : UnlockReleaseBuffer(buffer);
3721 318 : if (have_tuple_lock)
3722 108 : UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
3723 318 : if (vmbuffer != InvalidBuffer)
3724 0 : ReleaseBuffer(vmbuffer);
3725 318 : *update_indexes = TU_None;
3726 :
3727 318 : bms_free(hot_attrs);
3728 318 : bms_free(sum_attrs);
3729 318 : bms_free(key_attrs);
3730 318 : bms_free(id_attrs);
3731 318 : bms_free(modified_attrs);
3732 318 : bms_free(interesting_attrs);
3733 318 : return result;
3734 : }
3735 :
3736 : /*
3737 : * If we didn't pin the visibility map page and the page has become all
3738 : * visible while we were busy locking the buffer, or during some
3739 : * subsequent window during which we had it unlocked, we'll have to unlock
3740 : * and re-lock, to avoid holding the buffer lock across an I/O. That's a
3741 : * bit unfortunate, especially since we'll now have to recheck whether the
3742 : * tuple has been locked or updated under us, but hopefully it won't
3743 : * happen very often.
3744 : */
3745 623988 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
3746 : {
3747 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3748 0 : visibilitymap_pin(relation, block, &vmbuffer);
3749 0 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3750 0 : goto l2;
3751 : }
3752 :
3753 : /* Fill in transaction status data */
3754 :
3755 : /*
3756 : * If the tuple we're updating is locked, we need to preserve the locking
3757 : * info in the old tuple's Xmax. Prepare a new Xmax value for this.
3758 : */
3759 623988 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3760 623988 : oldtup.t_data->t_infomask,
3761 623988 : oldtup.t_data->t_infomask2,
3762 : xid, *lockmode, true,
3763 : &xmax_old_tuple, &infomask_old_tuple,
3764 : &infomask2_old_tuple);
3765 :
3766 : /*
3767 : * And also prepare an Xmax value for the new copy of the tuple. If there
3768 : * was no xmax previously, or there was one but all lockers are now gone,
3769 : * then use InvalidTransactionId; otherwise, get the xmax from the old
3770 : * tuple. (In rare cases that might also be InvalidTransactionId and yet
3771 : * not have the HEAP_XMAX_INVALID bit set; that's fine.)
3772 : */
3773 696024 : if ((oldtup.t_data->t_infomask & HEAP_XMAX_INVALID) ||
3774 144072 : HEAP_LOCKED_UPGRADED(oldtup.t_data->t_infomask) ||
3775 71694 : (checked_lockers && !locker_remains))
3776 551952 : xmax_new_tuple = InvalidTransactionId;
3777 : else
3778 72036 : xmax_new_tuple = HeapTupleHeaderGetRawXmax(oldtup.t_data);
3779 :
3780 623988 : if (!TransactionIdIsValid(xmax_new_tuple))
3781 : {
3782 551952 : infomask_new_tuple = HEAP_XMAX_INVALID;
3783 551952 : infomask2_new_tuple = 0;
3784 : }
3785 : else
3786 : {
3787 : /*
3788 : * If we found a valid Xmax for the new tuple, then the infomask bits
3789 : * to use on the new tuple depend on what was there on the old one.
3790 : * Note that since we're doing an update, the only possibility is that
3791 : * the lockers had FOR KEY SHARE lock.
3792 : */
3793 72036 : if (oldtup.t_data->t_infomask & HEAP_XMAX_IS_MULTI)
3794 : {
3795 344 : GetMultiXactIdHintBits(xmax_new_tuple, &infomask_new_tuple,
3796 : &infomask2_new_tuple);
3797 : }
3798 : else
3799 : {
3800 71692 : infomask_new_tuple = HEAP_XMAX_KEYSHR_LOCK | HEAP_XMAX_LOCK_ONLY;
3801 71692 : infomask2_new_tuple = 0;
3802 : }
3803 : }
3804 :
3805 : /*
3806 : * Prepare the new tuple with the appropriate initial values of Xmin and
3807 : * Xmax, as well as initial infomask bits as computed above.
3808 : */
3809 623988 : newtup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
3810 623988 : newtup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
3811 623988 : HeapTupleHeaderSetXmin(newtup->t_data, xid);
3812 623988 : HeapTupleHeaderSetCmin(newtup->t_data, cid);
3813 623988 : newtup->t_data->t_infomask |= HEAP_UPDATED | infomask_new_tuple;
3814 623988 : newtup->t_data->t_infomask2 |= infomask2_new_tuple;
3815 623988 : HeapTupleHeaderSetXmax(newtup->t_data, xmax_new_tuple);
3816 :
3817 : /*
3818 : * Replace cid with a combo CID if necessary. Note that we already put
3819 : * the plain cid into the new tuple.
3820 : */
3821 623988 : HeapTupleHeaderAdjustCmax(oldtup.t_data, &cid, &iscombo);
3822 :
3823 : /*
3824 : * If the toaster needs to be activated, OR if the new tuple will not fit
3825 : * on the same page as the old, then we need to release the content lock
3826 : * (but not the pin!) on the old tuple's buffer while we are off doing
3827 : * TOAST and/or table-file-extension work. We must mark the old tuple to
3828 : * show that it's locked, else other processes may try to update it
3829 : * themselves.
3830 : *
3831 : * We need to invoke the toaster if there are already any out-of-line
3832 : * toasted values present, or if the new tuple is over-threshold.
3833 : */
3834 623988 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
3835 0 : relation->rd_rel->relkind != RELKIND_MATVIEW)
3836 : {
3837 : /* toast table entries should never be recursively toasted */
3838 : Assert(!HeapTupleHasExternal(&oldtup));
3839 : Assert(!HeapTupleHasExternal(newtup));
3840 0 : need_toast = false;
3841 : }
3842 : else
3843 1871188 : need_toast = (HeapTupleHasExternal(&oldtup) ||
3844 1247200 : HeapTupleHasExternal(newtup) ||
3845 623164 : newtup->t_len > TOAST_TUPLE_THRESHOLD);
3846 :
3847 623988 : pagefree = PageGetHeapFreeSpace(page);
3848 :
3849 623988 : newtupsize = MAXALIGN(newtup->t_len);
3850 :
3851 623988 : if (need_toast || newtupsize > pagefree)
3852 304300 : {
3853 : TransactionId xmax_lock_old_tuple;
3854 : uint16 infomask_lock_old_tuple,
3855 : infomask2_lock_old_tuple;
3856 304300 : bool cleared_all_frozen = false;
3857 :
3858 : /*
3859 : * To prevent concurrent sessions from updating the tuple, we have to
3860 : * temporarily mark it locked, while we release the page-level lock.
3861 : *
3862 : * To satisfy the rule that any xid potentially appearing in a buffer
3863 : * written out to disk, we unfortunately have to WAL log this
3864 : * temporary modification. We can reuse xl_heap_lock for this
3865 : * purpose. If we crash/error before following through with the
3866 : * actual update, xmax will be of an aborted transaction, allowing
3867 : * other sessions to proceed.
3868 : */
3869 :
3870 : /*
3871 : * Compute xmax / infomask appropriate for locking the tuple. This has
3872 : * to be done separately from the combo that's going to be used for
3873 : * updating, because the potentially created multixact would otherwise
3874 : * be wrong.
3875 : */
3876 304300 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3877 304300 : oldtup.t_data->t_infomask,
3878 304300 : oldtup.t_data->t_infomask2,
3879 : xid, *lockmode, false,
3880 : &xmax_lock_old_tuple, &infomask_lock_old_tuple,
3881 : &infomask2_lock_old_tuple);
3882 :
3883 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(infomask_lock_old_tuple));
3884 :
3885 304300 : START_CRIT_SECTION();
3886 :
3887 : /* Clear obsolete visibility flags ... */
3888 304300 : oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
3889 304300 : oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
3890 304300 : HeapTupleClearHotUpdated(&oldtup);
3891 : /* ... and store info about transaction updating this tuple */
3892 : Assert(TransactionIdIsValid(xmax_lock_old_tuple));
3893 304300 : HeapTupleHeaderSetXmax(oldtup.t_data, xmax_lock_old_tuple);
3894 304300 : oldtup.t_data->t_infomask |= infomask_lock_old_tuple;
3895 304300 : oldtup.t_data->t_infomask2 |= infomask2_lock_old_tuple;
3896 304300 : HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
3897 :
3898 : /* temporarily make it look not-updated, but locked */
3899 304300 : oldtup.t_data->t_ctid = oldtup.t_self;
3900 :
3901 : /*
3902 : * Clear all-frozen bit on visibility map if needed. We could
3903 : * immediately reset ALL_VISIBLE, but given that the WAL logging
3904 : * overhead would be unchanged, that doesn't seem necessarily
3905 : * worthwhile.
3906 : */
3907 306212 : if (PageIsAllVisible(page) &&
3908 1912 : visibilitymap_clear(relation, block, vmbuffer,
3909 : VISIBILITYMAP_ALL_FROZEN))
3910 1560 : cleared_all_frozen = true;
3911 :
3912 304300 : MarkBufferDirty(buffer);
3913 :
3914 304300 : if (RelationNeedsWAL(relation))
3915 : {
3916 : xl_heap_lock xlrec;
3917 : XLogRecPtr recptr;
3918 :
3919 284050 : XLogBeginInsert();
3920 284050 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
3921 :
3922 284050 : xlrec.offnum = ItemPointerGetOffsetNumber(&oldtup.t_self);
3923 284050 : xlrec.xmax = xmax_lock_old_tuple;
3924 568100 : xlrec.infobits_set = compute_infobits(oldtup.t_data->t_infomask,
3925 284050 : oldtup.t_data->t_infomask2);
3926 284050 : xlrec.flags =
3927 284050 : cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
3928 284050 : XLogRegisterData(&xlrec, SizeOfHeapLock);
3929 284050 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
3930 284050 : PageSetLSN(page, recptr);
3931 : }
3932 :
3933 304300 : END_CRIT_SECTION();
3934 :
3935 304300 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3936 :
3937 : /*
3938 : * Let the toaster do its thing, if needed.
3939 : *
3940 : * Note: below this point, heaptup is the data we actually intend to
3941 : * store into the relation; newtup is the caller's original untoasted
3942 : * data.
3943 : */
3944 304300 : if (need_toast)
3945 : {
3946 : /* Note we always use WAL and FSM during updates */
3947 3426 : heaptup = heap_toast_insert_or_update(relation, newtup, &oldtup, 0);
3948 3426 : newtupsize = MAXALIGN(heaptup->t_len);
3949 : }
3950 : else
3951 300874 : heaptup = newtup;
3952 :
3953 : /*
3954 : * Now, do we need a new page for the tuple, or not? This is a bit
3955 : * tricky since someone else could have added tuples to the page while
3956 : * we weren't looking. We have to recheck the available space after
3957 : * reacquiring the buffer lock. But don't bother to do that if the
3958 : * former amount of free space is still not enough; it's unlikely
3959 : * there's more free now than before.
3960 : *
3961 : * What's more, if we need to get a new page, we will need to acquire
3962 : * buffer locks on both old and new pages. To avoid deadlock against
3963 : * some other backend trying to get the same two locks in the other
3964 : * order, we must be consistent about the order we get the locks in.
3965 : * We use the rule "lock the lower-numbered page of the relation
3966 : * first". To implement this, we must do RelationGetBufferForTuple
3967 : * while not holding the lock on the old page, and we must rely on it
3968 : * to get the locks on both pages in the correct order.
3969 : *
3970 : * Another consideration is that we need visibility map page pin(s) if
3971 : * we will have to clear the all-visible flag on either page. If we
3972 : * call RelationGetBufferForTuple, we rely on it to acquire any such
3973 : * pins; but if we don't, we have to handle that here. Hence we need
3974 : * a loop.
3975 : */
3976 : for (;;)
3977 : {
3978 304300 : if (newtupsize > pagefree)
3979 : {
3980 : /* It doesn't fit, must use RelationGetBufferForTuple. */
3981 303158 : newbuf = RelationGetBufferForTuple(relation, heaptup->t_len,
3982 : buffer, 0, NULL,
3983 : &vmbuffer_new, &vmbuffer,
3984 : 0);
3985 : /* We're all done. */
3986 303158 : break;
3987 : }
3988 : /* Acquire VM page pin if needed and we don't have it. */
3989 1142 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
3990 0 : visibilitymap_pin(relation, block, &vmbuffer);
3991 : /* Re-acquire the lock on the old tuple's page. */
3992 1142 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3993 : /* Re-check using the up-to-date free space */
3994 1142 : pagefree = PageGetHeapFreeSpace(page);
3995 1142 : if (newtupsize > pagefree ||
3996 1142 : (vmbuffer == InvalidBuffer && PageIsAllVisible(page)))
3997 : {
3998 : /*
3999 : * Rats, it doesn't fit anymore, or somebody just now set the
4000 : * all-visible flag. We must now unlock and loop to avoid
4001 : * deadlock. Fortunately, this path should seldom be taken.
4002 : */
4003 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
4004 : }
4005 : else
4006 : {
4007 : /* We're all done. */
4008 1142 : newbuf = buffer;
4009 1142 : break;
4010 : }
4011 : }
4012 : }
4013 : else
4014 : {
4015 : /* No TOAST work needed, and it'll fit on same page */
4016 319688 : newbuf = buffer;
4017 319688 : heaptup = newtup;
4018 : }
4019 :
4020 : /*
4021 : * We're about to do the actual update -- check for conflict first, to
4022 : * avoid possibly having to roll back work we've just done.
4023 : *
4024 : * This is safe without a recheck as long as there is no possibility of
4025 : * another process scanning the pages between this check and the update
4026 : * being visible to the scan (i.e., exclusive buffer content lock(s) are
4027 : * continuously held from this point until the tuple update is visible).
4028 : *
4029 : * For the new tuple the only check needed is at the relation level, but
4030 : * since both tuples are in the same relation and the check for oldtup
4031 : * will include checking the relation level, there is no benefit to a
4032 : * separate check for the new tuple.
4033 : */
4034 623988 : CheckForSerializableConflictIn(relation, &oldtup.t_self,
4035 : BufferGetBlockNumber(buffer));
4036 :
4037 : /*
4038 : * At this point newbuf and buffer are both pinned and locked, and newbuf
4039 : * has enough space for the new tuple. If they are the same buffer, only
4040 : * one pin is held.
4041 : */
4042 :
4043 623964 : if (newbuf == buffer)
4044 : {
4045 : /*
4046 : * Since the new tuple is going into the same page, we might be able
4047 : * to do a HOT update. Check if any of the index columns have been
4048 : * changed.
4049 : */
4050 320806 : if (!bms_overlap(modified_attrs, hot_attrs))
4051 : {
4052 296110 : use_hot_update = true;
4053 :
4054 : /*
4055 : * If none of the columns that are used in hot-blocking indexes
4056 : * were updated, we can apply HOT, but we do still need to check
4057 : * if we need to update the summarizing indexes, and update those
4058 : * indexes if the columns were updated, or we may fail to detect
4059 : * e.g. value bound changes in BRIN minmax indexes.
4060 : */
4061 296110 : if (bms_overlap(modified_attrs, sum_attrs))
4062 3282 : summarized_update = true;
4063 : }
4064 : }
4065 : else
4066 : {
4067 : /* Set a hint that the old page could use prune/defrag */
4068 303158 : PageSetFull(page);
4069 : }
4070 :
4071 : /*
4072 : * Compute replica identity tuple before entering the critical section so
4073 : * we don't PANIC upon a memory allocation failure.
4074 : * ExtractReplicaIdentity() will return NULL if nothing needs to be
4075 : * logged. Pass old key required as true only if the replica identity key
4076 : * columns are modified or it has external data.
4077 : */
4078 623964 : old_key_tuple = ExtractReplicaIdentity(relation, &oldtup,
4079 623964 : bms_overlap(modified_attrs, id_attrs) ||
4080 : id_has_external,
4081 : &old_key_copied);
4082 :
4083 : /* NO EREPORT(ERROR) from here till changes are logged */
4084 623964 : START_CRIT_SECTION();
4085 :
4086 : /*
4087 : * If this transaction commits, the old tuple will become DEAD sooner or
4088 : * later. Set flag that this page is a candidate for pruning once our xid
4089 : * falls below the OldestXmin horizon. If the transaction finally aborts,
4090 : * the subsequent page pruning will be a no-op and the hint will be
4091 : * cleared.
4092 : *
4093 : * XXX Should we set hint on newbuf as well? If the transaction aborts,
4094 : * there would be a prunable tuple in the newbuf; but for now we choose
4095 : * not to optimize for aborts. Note that heap_xlog_update must be kept in
4096 : * sync if this decision changes.
4097 : */
4098 623964 : PageSetPrunable(page, xid);
4099 :
4100 623964 : if (use_hot_update)
4101 : {
4102 : /* Mark the old tuple as HOT-updated */
4103 296110 : HeapTupleSetHotUpdated(&oldtup);
4104 : /* And mark the new tuple as heap-only */
4105 296110 : HeapTupleSetHeapOnly(heaptup);
4106 : /* Mark the caller's copy too, in case different from heaptup */
4107 296110 : HeapTupleSetHeapOnly(newtup);
4108 : }
4109 : else
4110 : {
4111 : /* Make sure tuples are correctly marked as not-HOT */
4112 327854 : HeapTupleClearHotUpdated(&oldtup);
4113 327854 : HeapTupleClearHeapOnly(heaptup);
4114 327854 : HeapTupleClearHeapOnly(newtup);
4115 : }
4116 :
4117 623964 : RelationPutHeapTuple(relation, newbuf, heaptup, false); /* insert new tuple */
4118 :
4119 :
4120 : /* Clear obsolete visibility flags, possibly set by ourselves above... */
4121 623964 : oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
4122 623964 : oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
4123 : /* ... and store info about transaction updating this tuple */
4124 : Assert(TransactionIdIsValid(xmax_old_tuple));
4125 623964 : HeapTupleHeaderSetXmax(oldtup.t_data, xmax_old_tuple);
4126 623964 : oldtup.t_data->t_infomask |= infomask_old_tuple;
4127 623964 : oldtup.t_data->t_infomask2 |= infomask2_old_tuple;
4128 623964 : HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
4129 :
4130 : /* record address of new tuple in t_ctid of old one */
4131 623964 : oldtup.t_data->t_ctid = heaptup->t_self;
4132 :
4133 : /* clear PD_ALL_VISIBLE flags, reset all visibilitymap bits */
4134 623964 : if (PageIsAllVisible(BufferGetPage(buffer)))
4135 : {
4136 3118 : all_visible_cleared = true;
4137 3118 : PageClearAllVisible(BufferGetPage(buffer));
4138 3118 : visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
4139 : vmbuffer, VISIBILITYMAP_VALID_BITS);
4140 : }
4141 623964 : if (newbuf != buffer && PageIsAllVisible(BufferGetPage(newbuf)))
4142 : {
4143 1704 : all_visible_cleared_new = true;
4144 1704 : PageClearAllVisible(BufferGetPage(newbuf));
4145 1704 : visibilitymap_clear(relation, BufferGetBlockNumber(newbuf),
4146 : vmbuffer_new, VISIBILITYMAP_VALID_BITS);
4147 : }
4148 :
4149 623964 : if (newbuf != buffer)
4150 303158 : MarkBufferDirty(newbuf);
4151 623964 : MarkBufferDirty(buffer);
4152 :
4153 : /* XLOG stuff */
4154 623964 : if (RelationNeedsWAL(relation))
4155 : {
4156 : XLogRecPtr recptr;
4157 :
4158 : /*
4159 : * For logical decoding we need combo CIDs to properly decode the
4160 : * catalog.
4161 : */
4162 601236 : if (RelationIsAccessibleInLogicalDecoding(relation))
4163 : {
4164 5112 : log_heap_new_cid(relation, &oldtup);
4165 5112 : log_heap_new_cid(relation, heaptup);
4166 : }
4167 :
4168 601236 : recptr = log_heap_update(relation, buffer,
4169 : newbuf, &oldtup, heaptup,
4170 : old_key_tuple,
4171 : all_visible_cleared,
4172 : all_visible_cleared_new);
4173 601236 : if (newbuf != buffer)
4174 : {
4175 282920 : PageSetLSN(BufferGetPage(newbuf), recptr);
4176 : }
4177 601236 : PageSetLSN(BufferGetPage(buffer), recptr);
4178 : }
4179 :
4180 623964 : END_CRIT_SECTION();
4181 :
4182 623964 : if (newbuf != buffer)
4183 303158 : LockBuffer(newbuf, BUFFER_LOCK_UNLOCK);
4184 623964 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
4185 :
4186 : /*
4187 : * Mark old tuple for invalidation from system caches at next command
4188 : * boundary, and mark the new tuple for invalidation in case we abort. We
4189 : * have to do this before releasing the buffer because oldtup is in the
4190 : * buffer. (heaptup is all in local memory, but it's necessary to process
4191 : * both tuple versions in one call to inval.c so we can avoid redundant
4192 : * sinval messages.)
4193 : */
4194 623964 : CacheInvalidateHeapTuple(relation, &oldtup, heaptup);
4195 :
4196 : /* Now we can release the buffer(s) */
4197 623964 : if (newbuf != buffer)
4198 303158 : ReleaseBuffer(newbuf);
4199 623964 : ReleaseBuffer(buffer);
4200 623964 : if (BufferIsValid(vmbuffer_new))
4201 1704 : ReleaseBuffer(vmbuffer_new);
4202 623964 : if (BufferIsValid(vmbuffer))
4203 3118 : ReleaseBuffer(vmbuffer);
4204 :
4205 : /*
4206 : * Release the lmgr tuple lock, if we had it.
4207 : */
4208 623964 : if (have_tuple_lock)
4209 44 : UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
4210 :
4211 623964 : pgstat_count_heap_update(relation, use_hot_update, newbuf != buffer);
4212 :
4213 : /*
4214 : * If heaptup is a private copy, release it. Don't forget to copy t_self
4215 : * back to the caller's image, too.
4216 : */
4217 623964 : if (heaptup != newtup)
4218 : {
4219 3330 : newtup->t_self = heaptup->t_self;
4220 3330 : heap_freetuple(heaptup);
4221 : }
4222 :
4223 : /*
4224 : * If it is a HOT update, the update may still need to update summarized
4225 : * indexes, lest we fail to update those summaries and get incorrect
4226 : * results (for example, minmax bounds of the block may change with this
4227 : * update).
4228 : */
4229 623964 : if (use_hot_update)
4230 : {
4231 296110 : if (summarized_update)
4232 3282 : *update_indexes = TU_Summarizing;
4233 : else
4234 292828 : *update_indexes = TU_None;
4235 : }
4236 : else
4237 327854 : *update_indexes = TU_All;
4238 :
4239 623964 : if (old_key_tuple != NULL && old_key_copied)
4240 168 : heap_freetuple(old_key_tuple);
4241 :
4242 623964 : bms_free(hot_attrs);
4243 623964 : bms_free(sum_attrs);
4244 623964 : bms_free(key_attrs);
4245 623964 : bms_free(id_attrs);
4246 623964 : bms_free(modified_attrs);
4247 623964 : bms_free(interesting_attrs);
4248 :
4249 623964 : return TM_Ok;
4250 : }
4251 :
4252 : #ifdef USE_ASSERT_CHECKING
4253 : /*
4254 : * Confirm adequate lock held during heap_update(), per rules from
4255 : * README.tuplock section "Locking to write inplace-updated tables".
4256 : */
4257 : static void
4258 : check_lock_if_inplace_updateable_rel(Relation relation,
4259 : const ItemPointerData *otid,
4260 : HeapTuple newtup)
4261 : {
4262 : /* LOCKTAG_TUPLE acceptable for any catalog */
4263 : switch (RelationGetRelid(relation))
4264 : {
4265 : case RelationRelationId:
4266 : case DatabaseRelationId:
4267 : {
4268 : LOCKTAG tuptag;
4269 :
4270 : SET_LOCKTAG_TUPLE(tuptag,
4271 : relation->rd_lockInfo.lockRelId.dbId,
4272 : relation->rd_lockInfo.lockRelId.relId,
4273 : ItemPointerGetBlockNumber(otid),
4274 : ItemPointerGetOffsetNumber(otid));
4275 : if (LockHeldByMe(&tuptag, InplaceUpdateTupleLock, false))
4276 : return;
4277 : }
4278 : break;
4279 : default:
4280 : Assert(!IsInplaceUpdateRelation(relation));
4281 : return;
4282 : }
4283 :
4284 : switch (RelationGetRelid(relation))
4285 : {
4286 : case RelationRelationId:
4287 : {
4288 : /* LOCKTAG_TUPLE or LOCKTAG_RELATION ok */
4289 : Form_pg_class classForm = (Form_pg_class) GETSTRUCT(newtup);
4290 : Oid relid = classForm->oid;
4291 : Oid dbid;
4292 : LOCKTAG tag;
4293 :
4294 : if (IsSharedRelation(relid))
4295 : dbid = InvalidOid;
4296 : else
4297 : dbid = MyDatabaseId;
4298 :
4299 : if (classForm->relkind == RELKIND_INDEX)
4300 : {
4301 : Relation irel = index_open(relid, AccessShareLock);
4302 :
4303 : SET_LOCKTAG_RELATION(tag, dbid, irel->rd_index->indrelid);
4304 : index_close(irel, AccessShareLock);
4305 : }
4306 : else
4307 : SET_LOCKTAG_RELATION(tag, dbid, relid);
4308 :
4309 : if (!LockHeldByMe(&tag, ShareUpdateExclusiveLock, false) &&
4310 : !LockHeldByMe(&tag, ShareRowExclusiveLock, true))
4311 : elog(WARNING,
4312 : "missing lock for relation \"%s\" (OID %u, relkind %c) @ TID (%u,%u)",
4313 : NameStr(classForm->relname),
4314 : relid,
4315 : classForm->relkind,
4316 : ItemPointerGetBlockNumber(otid),
4317 : ItemPointerGetOffsetNumber(otid));
4318 : }
4319 : break;
4320 : case DatabaseRelationId:
4321 : {
4322 : /* LOCKTAG_TUPLE required */
4323 : Form_pg_database dbForm = (Form_pg_database) GETSTRUCT(newtup);
4324 :
4325 : elog(WARNING,
4326 : "missing lock on database \"%s\" (OID %u) @ TID (%u,%u)",
4327 : NameStr(dbForm->datname),
4328 : dbForm->oid,
4329 : ItemPointerGetBlockNumber(otid),
4330 : ItemPointerGetOffsetNumber(otid));
4331 : }
4332 : break;
4333 : }
4334 : }
4335 :
4336 : /*
4337 : * Confirm adequate relation lock held, per rules from README.tuplock section
4338 : * "Locking to write inplace-updated tables".
4339 : */
4340 : static void
4341 : check_inplace_rel_lock(HeapTuple oldtup)
4342 : {
4343 : Form_pg_class classForm = (Form_pg_class) GETSTRUCT(oldtup);
4344 : Oid relid = classForm->oid;
4345 : Oid dbid;
4346 : LOCKTAG tag;
4347 :
4348 : if (IsSharedRelation(relid))
4349 : dbid = InvalidOid;
4350 : else
4351 : dbid = MyDatabaseId;
4352 :
4353 : if (classForm->relkind == RELKIND_INDEX)
4354 : {
4355 : Relation irel = index_open(relid, AccessShareLock);
4356 :
4357 : SET_LOCKTAG_RELATION(tag, dbid, irel->rd_index->indrelid);
4358 : index_close(irel, AccessShareLock);
4359 : }
4360 : else
4361 : SET_LOCKTAG_RELATION(tag, dbid, relid);
4362 :
4363 : if (!LockHeldByMe(&tag, ShareUpdateExclusiveLock, true))
4364 : elog(WARNING,
4365 : "missing lock for relation \"%s\" (OID %u, relkind %c) @ TID (%u,%u)",
4366 : NameStr(classForm->relname),
4367 : relid,
4368 : classForm->relkind,
4369 : ItemPointerGetBlockNumber(&oldtup->t_self),
4370 : ItemPointerGetOffsetNumber(&oldtup->t_self));
4371 : }
4372 : #endif
4373 :
4374 : /*
4375 : * Check if the specified attribute's values are the same. Subroutine for
4376 : * HeapDetermineColumnsInfo.
4377 : */
4378 : static bool
4379 1531810 : heap_attr_equals(TupleDesc tupdesc, int attrnum, Datum value1, Datum value2,
4380 : bool isnull1, bool isnull2)
4381 : {
4382 : /*
4383 : * If one value is NULL and other is not, then they are certainly not
4384 : * equal
4385 : */
4386 1531810 : if (isnull1 != isnull2)
4387 90 : return false;
4388 :
4389 : /*
4390 : * If both are NULL, they can be considered equal.
4391 : */
4392 1531720 : if (isnull1)
4393 9982 : return true;
4394 :
4395 : /*
4396 : * We do simple binary comparison of the two datums. This may be overly
4397 : * strict because there can be multiple binary representations for the
4398 : * same logical value. But we should be OK as long as there are no false
4399 : * positives. Using a type-specific equality operator is messy because
4400 : * there could be multiple notions of equality in different operator
4401 : * classes; furthermore, we cannot safely invoke user-defined functions
4402 : * while holding exclusive buffer lock.
4403 : */
4404 1521738 : if (attrnum <= 0)
4405 : {
4406 : /* The only allowed system columns are OIDs, so do this */
4407 0 : return (DatumGetObjectId(value1) == DatumGetObjectId(value2));
4408 : }
4409 : else
4410 : {
4411 : CompactAttribute *att;
4412 :
4413 : Assert(attrnum <= tupdesc->natts);
4414 1521738 : att = TupleDescCompactAttr(tupdesc, attrnum - 1);
4415 1521738 : return datumIsEqual(value1, value2, att->attbyval, att->attlen);
4416 : }
4417 : }
4418 :
4419 : /*
4420 : * Check which columns are being updated.
4421 : *
4422 : * Given an updated tuple, determine (and return into the output bitmapset),
4423 : * from those listed as interesting, the set of columns that changed.
4424 : *
4425 : * has_external indicates if any of the unmodified attributes (from those
4426 : * listed as interesting) of the old tuple is a member of external_cols and is
4427 : * stored externally.
4428 : */
4429 : static Bitmapset *
4430 624306 : HeapDetermineColumnsInfo(Relation relation,
4431 : Bitmapset *interesting_cols,
4432 : Bitmapset *external_cols,
4433 : HeapTuple oldtup, HeapTuple newtup,
4434 : bool *has_external)
4435 : {
4436 : int attidx;
4437 624306 : Bitmapset *modified = NULL;
4438 624306 : TupleDesc tupdesc = RelationGetDescr(relation);
4439 :
4440 624306 : attidx = -1;
4441 2156116 : while ((attidx = bms_next_member(interesting_cols, attidx)) >= 0)
4442 : {
4443 : /* attidx is zero-based, attrnum is the normal attribute number */
4444 1531810 : AttrNumber attrnum = attidx + FirstLowInvalidHeapAttributeNumber;
4445 : Datum value1,
4446 : value2;
4447 : bool isnull1,
4448 : isnull2;
4449 :
4450 : /*
4451 : * If it's a whole-tuple reference, say "not equal". It's not really
4452 : * worth supporting this case, since it could only succeed after a
4453 : * no-op update, which is hardly a case worth optimizing for.
4454 : */
4455 1531810 : if (attrnum == 0)
4456 : {
4457 0 : modified = bms_add_member(modified, attidx);
4458 1468196 : continue;
4459 : }
4460 :
4461 : /*
4462 : * Likewise, automatically say "not equal" for any system attribute
4463 : * other than tableOID; we cannot expect these to be consistent in a
4464 : * HOT chain, or even to be set correctly yet in the new tuple.
4465 : */
4466 1531810 : if (attrnum < 0)
4467 : {
4468 0 : if (attrnum != TableOidAttributeNumber)
4469 : {
4470 0 : modified = bms_add_member(modified, attidx);
4471 0 : continue;
4472 : }
4473 : }
4474 :
4475 : /*
4476 : * Extract the corresponding values. XXX this is pretty inefficient
4477 : * if there are many indexed columns. Should we do a single
4478 : * heap_deform_tuple call on each tuple, instead? But that doesn't
4479 : * work for system columns ...
4480 : */
4481 1531810 : value1 = heap_getattr(oldtup, attrnum, tupdesc, &isnull1);
4482 1531810 : value2 = heap_getattr(newtup, attrnum, tupdesc, &isnull2);
4483 :
4484 1531810 : if (!heap_attr_equals(tupdesc, attrnum, value1,
4485 : value2, isnull1, isnull2))
4486 : {
4487 54426 : modified = bms_add_member(modified, attidx);
4488 54426 : continue;
4489 : }
4490 :
4491 : /*
4492 : * No need to check attributes that can't be stored externally. Note
4493 : * that system attributes can't be stored externally.
4494 : */
4495 1477384 : if (attrnum < 0 || isnull1 ||
4496 1467402 : TupleDescCompactAttr(tupdesc, attrnum - 1)->attlen != -1)
4497 1413770 : continue;
4498 :
4499 : /*
4500 : * Check if the old tuple's attribute is stored externally and is a
4501 : * member of external_cols.
4502 : */
4503 63624 : if (VARATT_IS_EXTERNAL((struct varlena *) DatumGetPointer(value1)) &&
4504 10 : bms_is_member(attidx, external_cols))
4505 4 : *has_external = true;
4506 : }
4507 :
4508 624306 : return modified;
4509 : }
4510 :
4511 : /*
4512 : * simple_heap_update - replace a tuple
4513 : *
4514 : * This routine may be used to update a tuple when concurrent updates of
4515 : * the target tuple are not expected (for example, because we have a lock
4516 : * on the relation associated with the tuple). Any failure is reported
4517 : * via ereport().
4518 : */
4519 : void
4520 235038 : simple_heap_update(Relation relation, const ItemPointerData *otid, HeapTuple tup,
4521 : TU_UpdateIndexes *update_indexes)
4522 : {
4523 : TM_Result result;
4524 : TM_FailureData tmfd;
4525 : LockTupleMode lockmode;
4526 :
4527 235038 : result = heap_update(relation, otid, tup,
4528 : GetCurrentCommandId(true), InvalidSnapshot,
4529 : true /* wait for commit */ ,
4530 : &tmfd, &lockmode, update_indexes);
4531 235038 : switch (result)
4532 : {
4533 0 : case TM_SelfModified:
4534 : /* Tuple was already updated in current command? */
4535 0 : elog(ERROR, "tuple already updated by self");
4536 : break;
4537 :
4538 235036 : case TM_Ok:
4539 : /* done successfully */
4540 235036 : break;
4541 :
4542 0 : case TM_Updated:
4543 0 : elog(ERROR, "tuple concurrently updated");
4544 : break;
4545 :
4546 2 : case TM_Deleted:
4547 2 : elog(ERROR, "tuple concurrently deleted");
4548 : break;
4549 :
4550 0 : default:
4551 0 : elog(ERROR, "unrecognized heap_update status: %u", result);
4552 : break;
4553 : }
4554 235036 : }
4555 :
4556 :
4557 : /*
4558 : * Return the MultiXactStatus corresponding to the given tuple lock mode.
4559 : */
4560 : static MultiXactStatus
4561 230870 : get_mxact_status_for_lock(LockTupleMode mode, bool is_update)
4562 : {
4563 : int retval;
4564 :
4565 230870 : if (is_update)
4566 430 : retval = tupleLockExtraInfo[mode].updstatus;
4567 : else
4568 230440 : retval = tupleLockExtraInfo[mode].lockstatus;
4569 :
4570 230870 : if (retval == -1)
4571 0 : elog(ERROR, "invalid lock tuple mode %d/%s", mode,
4572 : is_update ? "true" : "false");
4573 :
4574 230870 : return (MultiXactStatus) retval;
4575 : }
4576 :
4577 : /*
4578 : * heap_lock_tuple - lock a tuple in shared or exclusive mode
4579 : *
4580 : * Note that this acquires a buffer pin, which the caller must release.
4581 : *
4582 : * Input parameters:
4583 : * relation: relation containing tuple (caller must hold suitable lock)
4584 : * cid: current command ID (used for visibility test, and stored into
4585 : * tuple's cmax if lock is successful)
4586 : * mode: indicates if shared or exclusive tuple lock is desired
4587 : * wait_policy: what to do if tuple lock is not available
4588 : * follow_updates: if true, follow the update chain to also lock descendant
4589 : * tuples.
4590 : *
4591 : * Output parameters:
4592 : * *tuple: all fields filled in
4593 : * *buffer: set to buffer holding tuple (pinned but not locked at exit)
4594 : * *tmfd: filled in failure cases (see below)
4595 : *
4596 : * Function results are the same as the ones for table_tuple_lock().
4597 : *
4598 : * In the failure cases other than TM_Invisible, the routine fills
4599 : * *tmfd with the tuple's t_ctid, t_xmax (resolving a possible MultiXact,
4600 : * if necessary), and t_cmax (the last only for TM_SelfModified,
4601 : * since we cannot obtain cmax from a combo CID generated by another
4602 : * transaction).
4603 : * See comments for struct TM_FailureData for additional info.
4604 : *
4605 : * See README.tuplock for a thorough explanation of this mechanism.
4606 : */
4607 : TM_Result
4608 316848 : heap_lock_tuple(Relation relation, HeapTuple tuple,
4609 : CommandId cid, LockTupleMode mode, LockWaitPolicy wait_policy,
4610 : bool follow_updates,
4611 : Buffer *buffer, TM_FailureData *tmfd)
4612 : {
4613 : TM_Result result;
4614 316848 : ItemPointer tid = &(tuple->t_self);
4615 : ItemId lp;
4616 : Page page;
4617 316848 : Buffer vmbuffer = InvalidBuffer;
4618 : BlockNumber block;
4619 : TransactionId xid,
4620 : xmax;
4621 : uint16 old_infomask,
4622 : new_infomask,
4623 : new_infomask2;
4624 316848 : bool first_time = true;
4625 316848 : bool skip_tuple_lock = false;
4626 316848 : bool have_tuple_lock = false;
4627 316848 : bool cleared_all_frozen = false;
4628 :
4629 316848 : *buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
4630 316848 : block = ItemPointerGetBlockNumber(tid);
4631 :
4632 : /*
4633 : * Before locking the buffer, pin the visibility map page if it appears to
4634 : * be necessary. Since we haven't got the lock yet, someone else might be
4635 : * in the middle of changing this, so we'll need to recheck after we have
4636 : * the lock.
4637 : */
4638 316848 : if (PageIsAllVisible(BufferGetPage(*buffer)))
4639 3332 : visibilitymap_pin(relation, block, &vmbuffer);
4640 :
4641 316848 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4642 :
4643 316848 : page = BufferGetPage(*buffer);
4644 316848 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
4645 : Assert(ItemIdIsNormal(lp));
4646 :
4647 316848 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
4648 316848 : tuple->t_len = ItemIdGetLength(lp);
4649 316848 : tuple->t_tableOid = RelationGetRelid(relation);
4650 :
4651 30 : l3:
4652 316878 : result = HeapTupleSatisfiesUpdate(tuple, cid, *buffer);
4653 :
4654 316878 : if (result == TM_Invisible)
4655 : {
4656 : /*
4657 : * This is possible, but only when locking a tuple for ON CONFLICT
4658 : * UPDATE. We return this value here rather than throwing an error in
4659 : * order to give that case the opportunity to throw a more specific
4660 : * error.
4661 : */
4662 24 : result = TM_Invisible;
4663 24 : goto out_locked;
4664 : }
4665 316854 : else if (result == TM_BeingModified ||
4666 154460 : result == TM_Updated ||
4667 : result == TM_Deleted)
4668 : {
4669 : TransactionId xwait;
4670 : uint16 infomask;
4671 : uint16 infomask2;
4672 : bool require_sleep;
4673 : ItemPointerData t_ctid;
4674 :
4675 : /* must copy state data before unlocking buffer */
4676 162396 : xwait = HeapTupleHeaderGetRawXmax(tuple->t_data);
4677 162396 : infomask = tuple->t_data->t_infomask;
4678 162396 : infomask2 = tuple->t_data->t_infomask2;
4679 162396 : ItemPointerCopy(&tuple->t_data->t_ctid, &t_ctid);
4680 :
4681 162396 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
4682 :
4683 : /*
4684 : * If any subtransaction of the current top transaction already holds
4685 : * a lock as strong as or stronger than what we're requesting, we
4686 : * effectively hold the desired lock already. We *must* succeed
4687 : * without trying to take the tuple lock, else we will deadlock
4688 : * against anyone wanting to acquire a stronger lock.
4689 : *
4690 : * Note we only do this the first time we loop on the HTSU result;
4691 : * there is no point in testing in subsequent passes, because
4692 : * evidently our own transaction cannot have acquired a new lock after
4693 : * the first time we checked.
4694 : */
4695 162396 : if (first_time)
4696 : {
4697 162378 : first_time = false;
4698 :
4699 162378 : if (infomask & HEAP_XMAX_IS_MULTI)
4700 : {
4701 : int i;
4702 : int nmembers;
4703 : MultiXactMember *members;
4704 :
4705 : /*
4706 : * We don't need to allow old multixacts here; if that had
4707 : * been the case, HeapTupleSatisfiesUpdate would have returned
4708 : * MayBeUpdated and we wouldn't be here.
4709 : */
4710 : nmembers =
4711 146588 : GetMultiXactIdMembers(xwait, &members, false,
4712 146588 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
4713 :
4714 2845308 : for (i = 0; i < nmembers; i++)
4715 : {
4716 : /* only consider members of our own transaction */
4717 2698748 : if (!TransactionIdIsCurrentTransactionId(members[i].xid))
4718 2698650 : continue;
4719 :
4720 98 : if (TUPLOCK_from_mxstatus(members[i].status) >= mode)
4721 : {
4722 28 : pfree(members);
4723 28 : result = TM_Ok;
4724 28 : goto out_unlocked;
4725 : }
4726 : else
4727 : {
4728 : /*
4729 : * Disable acquisition of the heavyweight tuple lock.
4730 : * Otherwise, when promoting a weaker lock, we might
4731 : * deadlock with another locker that has acquired the
4732 : * heavyweight tuple lock and is waiting for our
4733 : * transaction to finish.
4734 : *
4735 : * Note that in this case we still need to wait for
4736 : * the multixact if required, to avoid acquiring
4737 : * conflicting locks.
4738 : */
4739 70 : skip_tuple_lock = true;
4740 : }
4741 : }
4742 :
4743 146560 : if (members)
4744 146560 : pfree(members);
4745 : }
4746 15790 : else if (TransactionIdIsCurrentTransactionId(xwait))
4747 : {
4748 13166 : switch (mode)
4749 : {
4750 348 : case LockTupleKeyShare:
4751 : Assert(HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) ||
4752 : HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4753 : HEAP_XMAX_IS_EXCL_LOCKED(infomask));
4754 348 : result = TM_Ok;
4755 348 : goto out_unlocked;
4756 64 : case LockTupleShare:
4757 76 : if (HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4758 12 : HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4759 : {
4760 52 : result = TM_Ok;
4761 52 : goto out_unlocked;
4762 : }
4763 12 : break;
4764 144 : case LockTupleNoKeyExclusive:
4765 144 : if (HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4766 : {
4767 120 : result = TM_Ok;
4768 120 : goto out_unlocked;
4769 : }
4770 24 : break;
4771 12610 : case LockTupleExclusive:
4772 12610 : if (HEAP_XMAX_IS_EXCL_LOCKED(infomask) &&
4773 2530 : infomask2 & HEAP_KEYS_UPDATED)
4774 : {
4775 2488 : result = TM_Ok;
4776 2488 : goto out_unlocked;
4777 : }
4778 10122 : break;
4779 : }
4780 : }
4781 : }
4782 :
4783 : /*
4784 : * Initially assume that we will have to wait for the locking
4785 : * transaction(s) to finish. We check various cases below in which
4786 : * this can be turned off.
4787 : */
4788 159360 : require_sleep = true;
4789 159360 : if (mode == LockTupleKeyShare)
4790 : {
4791 : /*
4792 : * If we're requesting KeyShare, and there's no update present, we
4793 : * don't need to wait. Even if there is an update, we can still
4794 : * continue if the key hasn't been modified.
4795 : *
4796 : * However, if there are updates, we need to walk the update chain
4797 : * to mark future versions of the row as locked, too. That way,
4798 : * if somebody deletes that future version, we're protected
4799 : * against the key going away. This locking of future versions
4800 : * could block momentarily, if a concurrent transaction is
4801 : * deleting a key; or it could return a value to the effect that
4802 : * the transaction deleting the key has already committed. So we
4803 : * do this before re-locking the buffer; otherwise this would be
4804 : * prone to deadlocks.
4805 : *
4806 : * Note that the TID we're locking was grabbed before we unlocked
4807 : * the buffer. For it to change while we're not looking, the
4808 : * other properties we're testing for below after re-locking the
4809 : * buffer would also change, in which case we would restart this
4810 : * loop above.
4811 : */
4812 147712 : if (!(infomask2 & HEAP_KEYS_UPDATED))
4813 : {
4814 : bool updated;
4815 :
4816 147626 : updated = !HEAP_XMAX_IS_LOCKED_ONLY(infomask);
4817 :
4818 : /*
4819 : * If there are updates, follow the update chain; bail out if
4820 : * that cannot be done.
4821 : */
4822 147626 : if (follow_updates && updated &&
4823 4338 : !ItemPointerEquals(&tuple->t_self, &t_ctid))
4824 : {
4825 : TM_Result res;
4826 :
4827 4338 : res = heap_lock_updated_tuple(relation,
4828 : infomask, xwait, &t_ctid,
4829 : GetCurrentTransactionId(),
4830 : mode);
4831 4338 : if (res != TM_Ok)
4832 : {
4833 12 : result = res;
4834 : /* recovery code expects to have buffer lock held */
4835 12 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4836 386 : goto failed;
4837 : }
4838 : }
4839 :
4840 147614 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4841 :
4842 : /*
4843 : * Make sure it's still an appropriate lock, else start over.
4844 : * Also, if it wasn't updated before we released the lock, but
4845 : * is updated now, we start over too; the reason is that we
4846 : * now need to follow the update chain to lock the new
4847 : * versions.
4848 : */
4849 147614 : if (!HeapTupleHeaderIsOnlyLocked(tuple->t_data) &&
4850 4302 : ((tuple->t_data->t_infomask2 & HEAP_KEYS_UPDATED) ||
4851 4302 : !updated))
4852 30 : goto l3;
4853 :
4854 : /* Things look okay, so we can skip sleeping */
4855 147614 : require_sleep = false;
4856 :
4857 : /*
4858 : * Note we allow Xmax to change here; other updaters/lockers
4859 : * could have modified it before we grabbed the buffer lock.
4860 : * However, this is not a problem, because with the recheck we
4861 : * just did we ensure that they still don't conflict with the
4862 : * lock we want.
4863 : */
4864 : }
4865 : }
4866 11648 : else if (mode == LockTupleShare)
4867 : {
4868 : /*
4869 : * If we're requesting Share, we can similarly avoid sleeping if
4870 : * there's no update and no exclusive lock present.
4871 : */
4872 886 : if (HEAP_XMAX_IS_LOCKED_ONLY(infomask) &&
4873 886 : !HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4874 : {
4875 874 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4876 :
4877 : /*
4878 : * Make sure it's still an appropriate lock, else start over.
4879 : * See above about allowing xmax to change.
4880 : */
4881 1748 : if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
4882 874 : HEAP_XMAX_IS_EXCL_LOCKED(tuple->t_data->t_infomask))
4883 0 : goto l3;
4884 874 : require_sleep = false;
4885 : }
4886 : }
4887 10762 : else if (mode == LockTupleNoKeyExclusive)
4888 : {
4889 : /*
4890 : * If we're requesting NoKeyExclusive, we might also be able to
4891 : * avoid sleeping; just ensure that there no conflicting lock
4892 : * already acquired.
4893 : */
4894 326 : if (infomask & HEAP_XMAX_IS_MULTI)
4895 : {
4896 52 : if (!DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
4897 : mode, NULL))
4898 : {
4899 : /*
4900 : * No conflict, but if the xmax changed under us in the
4901 : * meantime, start over.
4902 : */
4903 26 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4904 52 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4905 26 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4906 : xwait))
4907 0 : goto l3;
4908 :
4909 : /* otherwise, we're good */
4910 26 : require_sleep = false;
4911 : }
4912 : }
4913 274 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask))
4914 : {
4915 36 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4916 :
4917 : /* if the xmax changed in the meantime, start over */
4918 72 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4919 36 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4920 : xwait))
4921 0 : goto l3;
4922 : /* otherwise, we're good */
4923 36 : require_sleep = false;
4924 : }
4925 : }
4926 :
4927 : /*
4928 : * As a check independent from those above, we can also avoid sleeping
4929 : * if the current transaction is the sole locker of the tuple. Note
4930 : * that the strength of the lock already held is irrelevant; this is
4931 : * not about recording the lock in Xmax (which will be done regardless
4932 : * of this optimization, below). Also, note that the cases where we
4933 : * hold a lock stronger than we are requesting are already handled
4934 : * above by not doing anything.
4935 : *
4936 : * Note we only deal with the non-multixact case here; MultiXactIdWait
4937 : * is well equipped to deal with this situation on its own.
4938 : */
4939 170064 : if (require_sleep && !(infomask & HEAP_XMAX_IS_MULTI) &&
4940 10716 : TransactionIdIsCurrentTransactionId(xwait))
4941 : {
4942 : /* ... but if the xmax changed in the meantime, start over */
4943 10122 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4944 20244 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4945 10122 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4946 : xwait))
4947 0 : goto l3;
4948 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask));
4949 10122 : require_sleep = false;
4950 : }
4951 :
4952 : /*
4953 : * Time to sleep on the other transaction/multixact, if necessary.
4954 : *
4955 : * If the other transaction is an update/delete that's already
4956 : * committed, then sleeping cannot possibly do any good: if we're
4957 : * required to sleep, get out to raise an error instead.
4958 : *
4959 : * By here, we either have already acquired the buffer exclusive lock,
4960 : * or we must wait for the locking transaction or multixact; so below
4961 : * we ensure that we grab buffer lock after the sleep.
4962 : */
4963 159348 : if (require_sleep && (result == TM_Updated || result == TM_Deleted))
4964 : {
4965 298 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4966 298 : goto failed;
4967 : }
4968 159050 : else if (require_sleep)
4969 : {
4970 : /*
4971 : * Acquire tuple lock to establish our priority for the tuple, or
4972 : * die trying. LockTuple will release us when we are next-in-line
4973 : * for the tuple. We must do this even if we are share-locking,
4974 : * but not if we already have a weaker lock on the tuple.
4975 : *
4976 : * If we are forced to "start over" below, we keep the tuple lock;
4977 : * this arranges that we stay at the head of the line while
4978 : * rechecking tuple state.
4979 : */
4980 378 : if (!skip_tuple_lock &&
4981 346 : !heap_acquire_tuplock(relation, tid, mode, wait_policy,
4982 : &have_tuple_lock))
4983 : {
4984 : /*
4985 : * This can only happen if wait_policy is Skip and the lock
4986 : * couldn't be obtained.
4987 : */
4988 2 : result = TM_WouldBlock;
4989 : /* recovery code expects to have buffer lock held */
4990 2 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4991 2 : goto failed;
4992 : }
4993 :
4994 374 : if (infomask & HEAP_XMAX_IS_MULTI)
4995 : {
4996 80 : MultiXactStatus status = get_mxact_status_for_lock(mode, false);
4997 :
4998 : /* We only ever lock tuples, never update them */
4999 80 : if (status >= MultiXactStatusNoKeyUpdate)
5000 0 : elog(ERROR, "invalid lock mode in heap_lock_tuple");
5001 :
5002 : /* wait for multixact to end, or die trying */
5003 80 : switch (wait_policy)
5004 : {
5005 72 : case LockWaitBlock:
5006 72 : MultiXactIdWait((MultiXactId) xwait, status, infomask,
5007 72 : relation, &tuple->t_self, XLTW_Lock, NULL);
5008 72 : break;
5009 4 : case LockWaitSkip:
5010 4 : if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
5011 : status, infomask, relation,
5012 : NULL, false))
5013 : {
5014 4 : result = TM_WouldBlock;
5015 : /* recovery code expects to have buffer lock held */
5016 4 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5017 4 : goto failed;
5018 : }
5019 0 : break;
5020 4 : case LockWaitError:
5021 4 : if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
5022 : status, infomask, relation,
5023 : NULL, log_lock_failures))
5024 4 : ereport(ERROR,
5025 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5026 : errmsg("could not obtain lock on row in relation \"%s\"",
5027 : RelationGetRelationName(relation))));
5028 :
5029 0 : break;
5030 : }
5031 :
5032 : /*
5033 : * Of course, the multixact might not be done here: if we're
5034 : * requesting a light lock mode, other transactions with light
5035 : * locks could still be alive, as well as locks owned by our
5036 : * own xact or other subxacts of this backend. We need to
5037 : * preserve the surviving MultiXact members. Note that it
5038 : * isn't absolutely necessary in the latter case, but doing so
5039 : * is simpler.
5040 : */
5041 : }
5042 : else
5043 : {
5044 : /* wait for regular transaction to end, or die trying */
5045 294 : switch (wait_policy)
5046 : {
5047 216 : case LockWaitBlock:
5048 216 : XactLockTableWait(xwait, relation, &tuple->t_self,
5049 : XLTW_Lock);
5050 216 : break;
5051 66 : case LockWaitSkip:
5052 66 : if (!ConditionalXactLockTableWait(xwait, false))
5053 : {
5054 66 : result = TM_WouldBlock;
5055 : /* recovery code expects to have buffer lock held */
5056 66 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5057 66 : goto failed;
5058 : }
5059 0 : break;
5060 12 : case LockWaitError:
5061 12 : if (!ConditionalXactLockTableWait(xwait, log_lock_failures))
5062 12 : ereport(ERROR,
5063 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5064 : errmsg("could not obtain lock on row in relation \"%s\"",
5065 : RelationGetRelationName(relation))));
5066 0 : break;
5067 : }
5068 : }
5069 :
5070 : /* if there are updates, follow the update chain */
5071 288 : if (follow_updates && !HEAP_XMAX_IS_LOCKED_ONLY(infomask) &&
5072 116 : !ItemPointerEquals(&tuple->t_self, &t_ctid))
5073 : {
5074 : TM_Result res;
5075 :
5076 90 : res = heap_lock_updated_tuple(relation,
5077 : infomask, xwait, &t_ctid,
5078 : GetCurrentTransactionId(),
5079 : mode);
5080 90 : if (res != TM_Ok)
5081 : {
5082 4 : result = res;
5083 : /* recovery code expects to have buffer lock held */
5084 4 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5085 4 : goto failed;
5086 : }
5087 : }
5088 :
5089 284 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5090 :
5091 : /*
5092 : * xwait is done, but if xwait had just locked the tuple then some
5093 : * other xact could update this tuple before we get to this point.
5094 : * Check for xmax change, and start over if so.
5095 : */
5096 542 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
5097 258 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
5098 : xwait))
5099 30 : goto l3;
5100 :
5101 254 : if (!(infomask & HEAP_XMAX_IS_MULTI))
5102 : {
5103 : /*
5104 : * Otherwise check if it committed or aborted. Note we cannot
5105 : * be here if the tuple was only locked by somebody who didn't
5106 : * conflict with us; that would have been handled above. So
5107 : * that transaction must necessarily be gone by now. But
5108 : * don't check for this in the multixact case, because some
5109 : * locker transactions might still be running.
5110 : */
5111 192 : UpdateXmaxHintBits(tuple->t_data, *buffer, xwait);
5112 : }
5113 : }
5114 :
5115 : /* By here, we're certain that we hold buffer exclusive lock again */
5116 :
5117 : /*
5118 : * We may lock if previous xmax aborted, or if it committed but only
5119 : * locked the tuple without updating it; or if we didn't have to wait
5120 : * at all for whatever reason.
5121 : */
5122 158926 : if (!require_sleep ||
5123 446 : (tuple->t_data->t_infomask & HEAP_XMAX_INVALID) ||
5124 354 : HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
5125 162 : HeapTupleHeaderIsOnlyLocked(tuple->t_data))
5126 158776 : result = TM_Ok;
5127 150 : else if (!ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid))
5128 114 : result = TM_Updated;
5129 : else
5130 36 : result = TM_Deleted;
5131 : }
5132 :
5133 154458 : failed:
5134 313770 : if (result != TM_Ok)
5135 : {
5136 : Assert(result == TM_SelfModified || result == TM_Updated ||
5137 : result == TM_Deleted || result == TM_WouldBlock);
5138 :
5139 : /*
5140 : * When locking a tuple under LockWaitSkip semantics and we fail with
5141 : * TM_WouldBlock above, it's possible for concurrent transactions to
5142 : * release the lock and set HEAP_XMAX_INVALID in the meantime. So
5143 : * this assert is slightly different from the equivalent one in
5144 : * heap_delete and heap_update.
5145 : */
5146 : Assert((result == TM_WouldBlock) ||
5147 : !(tuple->t_data->t_infomask & HEAP_XMAX_INVALID));
5148 : Assert(result != TM_Updated ||
5149 : !ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid));
5150 548 : tmfd->ctid = tuple->t_data->t_ctid;
5151 548 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(tuple->t_data);
5152 548 : if (result == TM_SelfModified)
5153 12 : tmfd->cmax = HeapTupleHeaderGetCmax(tuple->t_data);
5154 : else
5155 536 : tmfd->cmax = InvalidCommandId;
5156 548 : goto out_locked;
5157 : }
5158 :
5159 : /*
5160 : * If we didn't pin the visibility map page and the page has become all
5161 : * visible while we were busy locking the buffer, or during some
5162 : * subsequent window during which we had it unlocked, we'll have to unlock
5163 : * and re-lock, to avoid holding the buffer lock across I/O. That's a bit
5164 : * unfortunate, especially since we'll now have to recheck whether the
5165 : * tuple has been locked or updated under us, but hopefully it won't
5166 : * happen very often.
5167 : */
5168 313222 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
5169 : {
5170 0 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
5171 0 : visibilitymap_pin(relation, block, &vmbuffer);
5172 0 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5173 0 : goto l3;
5174 : }
5175 :
5176 313222 : xmax = HeapTupleHeaderGetRawXmax(tuple->t_data);
5177 313222 : old_infomask = tuple->t_data->t_infomask;
5178 :
5179 : /*
5180 : * If this is the first possibly-multixact-able operation in the current
5181 : * transaction, set my per-backend OldestMemberMXactId setting. We can be
5182 : * certain that the transaction will never become a member of any older
5183 : * MultiXactIds than that. (We have to do this even if we end up just
5184 : * using our own TransactionId below, since some other backend could
5185 : * incorporate our XID into a MultiXact immediately afterwards.)
5186 : */
5187 313222 : MultiXactIdSetOldestMember();
5188 :
5189 : /*
5190 : * Compute the new xmax and infomask to store into the tuple. Note we do
5191 : * not modify the tuple just yet, because that would leave it in the wrong
5192 : * state if multixact.c elogs.
5193 : */
5194 313222 : compute_new_xmax_infomask(xmax, old_infomask, tuple->t_data->t_infomask2,
5195 : GetCurrentTransactionId(), mode, false,
5196 : &xid, &new_infomask, &new_infomask2);
5197 :
5198 313222 : START_CRIT_SECTION();
5199 :
5200 : /*
5201 : * Store transaction information of xact locking the tuple.
5202 : *
5203 : * Note: Cmax is meaningless in this context, so don't set it; this avoids
5204 : * possibly generating a useless combo CID. Moreover, if we're locking a
5205 : * previously updated tuple, it's important to preserve the Cmax.
5206 : *
5207 : * Also reset the HOT UPDATE bit, but only if there's no update; otherwise
5208 : * we would break the HOT chain.
5209 : */
5210 313222 : tuple->t_data->t_infomask &= ~HEAP_XMAX_BITS;
5211 313222 : tuple->t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
5212 313222 : tuple->t_data->t_infomask |= new_infomask;
5213 313222 : tuple->t_data->t_infomask2 |= new_infomask2;
5214 313222 : if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
5215 308928 : HeapTupleHeaderClearHotUpdated(tuple->t_data);
5216 313222 : HeapTupleHeaderSetXmax(tuple->t_data, xid);
5217 :
5218 : /*
5219 : * Make sure there is no forward chain link in t_ctid. Note that in the
5220 : * cases where the tuple has been updated, we must not overwrite t_ctid,
5221 : * because it was set by the updater. Moreover, if the tuple has been
5222 : * updated, we need to follow the update chain to lock the new versions of
5223 : * the tuple as well.
5224 : */
5225 313222 : if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
5226 308928 : tuple->t_data->t_ctid = *tid;
5227 :
5228 : /* Clear only the all-frozen bit on visibility map if needed */
5229 316554 : if (PageIsAllVisible(page) &&
5230 3332 : visibilitymap_clear(relation, block, vmbuffer,
5231 : VISIBILITYMAP_ALL_FROZEN))
5232 28 : cleared_all_frozen = true;
5233 :
5234 :
5235 313222 : MarkBufferDirty(*buffer);
5236 :
5237 : /*
5238 : * XLOG stuff. You might think that we don't need an XLOG record because
5239 : * there is no state change worth restoring after a crash. You would be
5240 : * wrong however: we have just written either a TransactionId or a
5241 : * MultiXactId that may never have been seen on disk before, and we need
5242 : * to make sure that there are XLOG entries covering those ID numbers.
5243 : * Else the same IDs might be re-used after a crash, which would be
5244 : * disastrous if this page made it to disk before the crash. Essentially
5245 : * we have to enforce the WAL log-before-data rule even in this case.
5246 : * (Also, in a PITR log-shipping or 2PC environment, we have to have XLOG
5247 : * entries for everything anyway.)
5248 : */
5249 313222 : if (RelationNeedsWAL(relation))
5250 : {
5251 : xl_heap_lock xlrec;
5252 : XLogRecPtr recptr;
5253 :
5254 312520 : XLogBeginInsert();
5255 312520 : XLogRegisterBuffer(0, *buffer, REGBUF_STANDARD);
5256 :
5257 312520 : xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
5258 312520 : xlrec.xmax = xid;
5259 625040 : xlrec.infobits_set = compute_infobits(new_infomask,
5260 312520 : tuple->t_data->t_infomask2);
5261 312520 : xlrec.flags = cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
5262 312520 : XLogRegisterData(&xlrec, SizeOfHeapLock);
5263 :
5264 : /* we don't decode row locks atm, so no need to log the origin */
5265 :
5266 312520 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
5267 :
5268 312520 : PageSetLSN(page, recptr);
5269 : }
5270 :
5271 313222 : END_CRIT_SECTION();
5272 :
5273 313222 : result = TM_Ok;
5274 :
5275 313794 : out_locked:
5276 313794 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
5277 :
5278 316830 : out_unlocked:
5279 316830 : if (BufferIsValid(vmbuffer))
5280 3332 : ReleaseBuffer(vmbuffer);
5281 :
5282 : /*
5283 : * Don't update the visibility map here. Locking a tuple doesn't change
5284 : * visibility info.
5285 : */
5286 :
5287 : /*
5288 : * Now that we have successfully marked the tuple as locked, we can
5289 : * release the lmgr tuple lock, if we had it.
5290 : */
5291 316830 : if (have_tuple_lock)
5292 316 : UnlockTupleTuplock(relation, tid, mode);
5293 :
5294 316830 : return result;
5295 : }
5296 :
5297 : /*
5298 : * Acquire heavyweight lock on the given tuple, in preparation for acquiring
5299 : * its normal, Xmax-based tuple lock.
5300 : *
5301 : * have_tuple_lock is an input and output parameter: on input, it indicates
5302 : * whether the lock has previously been acquired (and this function does
5303 : * nothing in that case). If this function returns success, have_tuple_lock
5304 : * has been flipped to true.
5305 : *
5306 : * Returns false if it was unable to obtain the lock; this can only happen if
5307 : * wait_policy is Skip.
5308 : */
5309 : static bool
5310 616 : heap_acquire_tuplock(Relation relation, const ItemPointerData *tid, LockTupleMode mode,
5311 : LockWaitPolicy wait_policy, bool *have_tuple_lock)
5312 : {
5313 616 : if (*have_tuple_lock)
5314 18 : return true;
5315 :
5316 598 : switch (wait_policy)
5317 : {
5318 516 : case LockWaitBlock:
5319 516 : LockTupleTuplock(relation, tid, mode);
5320 516 : break;
5321 :
5322 68 : case LockWaitSkip:
5323 68 : if (!ConditionalLockTupleTuplock(relation, tid, mode, false))
5324 2 : return false;
5325 66 : break;
5326 :
5327 14 : case LockWaitError:
5328 14 : if (!ConditionalLockTupleTuplock(relation, tid, mode, log_lock_failures))
5329 2 : ereport(ERROR,
5330 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5331 : errmsg("could not obtain lock on row in relation \"%s\"",
5332 : RelationGetRelationName(relation))));
5333 12 : break;
5334 : }
5335 594 : *have_tuple_lock = true;
5336 :
5337 594 : return true;
5338 : }
5339 :
5340 : /*
5341 : * Given an original set of Xmax and infomask, and a transaction (identified by
5342 : * add_to_xmax) acquiring a new lock of some mode, compute the new Xmax and
5343 : * corresponding infomasks to use on the tuple.
5344 : *
5345 : * Note that this might have side effects such as creating a new MultiXactId.
5346 : *
5347 : * Most callers will have called HeapTupleSatisfiesUpdate before this function;
5348 : * that will have set the HEAP_XMAX_INVALID bit if the xmax was a MultiXactId
5349 : * but it was not running anymore. There is a race condition, which is that the
5350 : * MultiXactId may have finished since then, but that uncommon case is handled
5351 : * either here, or within MultiXactIdExpand.
5352 : *
5353 : * There is a similar race condition possible when the old xmax was a regular
5354 : * TransactionId. We test TransactionIdIsInProgress again just to narrow the
5355 : * window, but it's still possible to end up creating an unnecessary
5356 : * MultiXactId. Fortunately this is harmless.
5357 : */
5358 : static void
5359 4296254 : compute_new_xmax_infomask(TransactionId xmax, uint16 old_infomask,
5360 : uint16 old_infomask2, TransactionId add_to_xmax,
5361 : LockTupleMode mode, bool is_update,
5362 : TransactionId *result_xmax, uint16 *result_infomask,
5363 : uint16 *result_infomask2)
5364 : {
5365 : TransactionId new_xmax;
5366 : uint16 new_infomask,
5367 : new_infomask2;
5368 :
5369 : Assert(TransactionIdIsCurrentTransactionId(add_to_xmax));
5370 :
5371 208038 : l5:
5372 4504292 : new_infomask = 0;
5373 4504292 : new_infomask2 = 0;
5374 4504292 : if (old_infomask & HEAP_XMAX_INVALID)
5375 : {
5376 : /*
5377 : * No previous locker; we just insert our own TransactionId.
5378 : *
5379 : * Note that it's critical that this case be the first one checked,
5380 : * because there are several blocks below that come back to this one
5381 : * to implement certain optimizations; old_infomask might contain
5382 : * other dirty bits in those cases, but we don't really care.
5383 : */
5384 4143012 : if (is_update)
5385 : {
5386 3673940 : new_xmax = add_to_xmax;
5387 3673940 : if (mode == LockTupleExclusive)
5388 3123076 : new_infomask2 |= HEAP_KEYS_UPDATED;
5389 : }
5390 : else
5391 : {
5392 469072 : new_infomask |= HEAP_XMAX_LOCK_ONLY;
5393 469072 : switch (mode)
5394 : {
5395 5294 : case LockTupleKeyShare:
5396 5294 : new_xmax = add_to_xmax;
5397 5294 : new_infomask |= HEAP_XMAX_KEYSHR_LOCK;
5398 5294 : break;
5399 1510 : case LockTupleShare:
5400 1510 : new_xmax = add_to_xmax;
5401 1510 : new_infomask |= HEAP_XMAX_SHR_LOCK;
5402 1510 : break;
5403 270828 : case LockTupleNoKeyExclusive:
5404 270828 : new_xmax = add_to_xmax;
5405 270828 : new_infomask |= HEAP_XMAX_EXCL_LOCK;
5406 270828 : break;
5407 191440 : case LockTupleExclusive:
5408 191440 : new_xmax = add_to_xmax;
5409 191440 : new_infomask |= HEAP_XMAX_EXCL_LOCK;
5410 191440 : new_infomask2 |= HEAP_KEYS_UPDATED;
5411 191440 : break;
5412 0 : default:
5413 0 : new_xmax = InvalidTransactionId; /* silence compiler */
5414 0 : elog(ERROR, "invalid lock mode");
5415 : }
5416 : }
5417 : }
5418 361280 : else if (old_infomask & HEAP_XMAX_IS_MULTI)
5419 : {
5420 : MultiXactStatus new_status;
5421 :
5422 : /*
5423 : * Currently we don't allow XMAX_COMMITTED to be set for multis, so
5424 : * cross-check.
5425 : */
5426 : Assert(!(old_infomask & HEAP_XMAX_COMMITTED));
5427 :
5428 : /*
5429 : * A multixact together with LOCK_ONLY set but neither lock bit set
5430 : * (i.e. a pg_upgraded share locked tuple) cannot possibly be running
5431 : * anymore. This check is critical for databases upgraded by
5432 : * pg_upgrade; both MultiXactIdIsRunning and MultiXactIdExpand assume
5433 : * that such multis are never passed.
5434 : */
5435 151114 : if (HEAP_LOCKED_UPGRADED(old_infomask))
5436 : {
5437 0 : old_infomask &= ~HEAP_XMAX_IS_MULTI;
5438 0 : old_infomask |= HEAP_XMAX_INVALID;
5439 0 : goto l5;
5440 : }
5441 :
5442 : /*
5443 : * If the XMAX is already a MultiXactId, then we need to expand it to
5444 : * include add_to_xmax; but if all the members were lockers and are
5445 : * all gone, we can do away with the IS_MULTI bit and just set
5446 : * add_to_xmax as the only locker/updater. If all lockers are gone
5447 : * and we have an updater that aborted, we can also do without a
5448 : * multi.
5449 : *
5450 : * The cost of doing GetMultiXactIdMembers would be paid by
5451 : * MultiXactIdExpand if we weren't to do this, so this check is not
5452 : * incurring extra work anyhow.
5453 : */
5454 151114 : if (!MultiXactIdIsRunning(xmax, HEAP_XMAX_IS_LOCKED_ONLY(old_infomask)))
5455 : {
5456 46 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) ||
5457 16 : !TransactionIdDidCommit(MultiXactIdGetUpdateXid(xmax,
5458 : old_infomask)))
5459 : {
5460 : /*
5461 : * Reset these bits and restart; otherwise fall through to
5462 : * create a new multi below.
5463 : */
5464 46 : old_infomask &= ~HEAP_XMAX_IS_MULTI;
5465 46 : old_infomask |= HEAP_XMAX_INVALID;
5466 46 : goto l5;
5467 : }
5468 : }
5469 :
5470 151068 : new_status = get_mxact_status_for_lock(mode, is_update);
5471 :
5472 151068 : new_xmax = MultiXactIdExpand((MultiXactId) xmax, add_to_xmax,
5473 : new_status);
5474 151068 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5475 : }
5476 210166 : else if (old_infomask & HEAP_XMAX_COMMITTED)
5477 : {
5478 : /*
5479 : * It's a committed update, so we need to preserve him as updater of
5480 : * the tuple.
5481 : */
5482 : MultiXactStatus status;
5483 : MultiXactStatus new_status;
5484 :
5485 26 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5486 0 : status = MultiXactStatusUpdate;
5487 : else
5488 26 : status = MultiXactStatusNoKeyUpdate;
5489 :
5490 26 : new_status = get_mxact_status_for_lock(mode, is_update);
5491 :
5492 : /*
5493 : * since it's not running, it's obviously impossible for the old
5494 : * updater to be identical to the current one, so we need not check
5495 : * for that case as we do in the block above.
5496 : */
5497 26 : new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5498 26 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5499 : }
5500 210140 : else if (TransactionIdIsInProgress(xmax))
5501 : {
5502 : /*
5503 : * If the XMAX is a valid, in-progress TransactionId, then we need to
5504 : * create a new MultiXactId that includes both the old locker or
5505 : * updater and our own TransactionId.
5506 : */
5507 : MultiXactStatus new_status;
5508 : MultiXactStatus old_status;
5509 : LockTupleMode old_mode;
5510 :
5511 210122 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
5512 : {
5513 210070 : if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
5514 11348 : old_status = MultiXactStatusForKeyShare;
5515 198722 : else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
5516 866 : old_status = MultiXactStatusForShare;
5517 197856 : else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
5518 : {
5519 197856 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5520 185522 : old_status = MultiXactStatusForUpdate;
5521 : else
5522 12334 : old_status = MultiXactStatusForNoKeyUpdate;
5523 : }
5524 : else
5525 : {
5526 : /*
5527 : * LOCK_ONLY can be present alone only when a page has been
5528 : * upgraded by pg_upgrade. But in that case,
5529 : * TransactionIdIsInProgress() should have returned false. We
5530 : * assume it's no longer locked in this case.
5531 : */
5532 0 : elog(WARNING, "LOCK_ONLY found for Xid in progress %u", xmax);
5533 0 : old_infomask |= HEAP_XMAX_INVALID;
5534 0 : old_infomask &= ~HEAP_XMAX_LOCK_ONLY;
5535 0 : goto l5;
5536 : }
5537 : }
5538 : else
5539 : {
5540 : /* it's an update, but which kind? */
5541 52 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5542 0 : old_status = MultiXactStatusUpdate;
5543 : else
5544 52 : old_status = MultiXactStatusNoKeyUpdate;
5545 : }
5546 :
5547 210122 : old_mode = TUPLOCK_from_mxstatus(old_status);
5548 :
5549 : /*
5550 : * If the lock to be acquired is for the same TransactionId as the
5551 : * existing lock, there's an optimization possible: consider only the
5552 : * strongest of both locks as the only one present, and restart.
5553 : */
5554 210122 : if (xmax == add_to_xmax)
5555 : {
5556 : /*
5557 : * Note that it's not possible for the original tuple to be
5558 : * updated: we wouldn't be here because the tuple would have been
5559 : * invisible and we wouldn't try to update it. As a subtlety,
5560 : * this code can also run when traversing an update chain to lock
5561 : * future versions of a tuple. But we wouldn't be here either,
5562 : * because the add_to_xmax would be different from the original
5563 : * updater.
5564 : */
5565 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
5566 :
5567 : /* acquire the strongest of both */
5568 207976 : if (mode < old_mode)
5569 104394 : mode = old_mode;
5570 : /* mustn't touch is_update */
5571 :
5572 207976 : old_infomask |= HEAP_XMAX_INVALID;
5573 207976 : goto l5;
5574 : }
5575 :
5576 : /* otherwise, just fall back to creating a new multixact */
5577 2146 : new_status = get_mxact_status_for_lock(mode, is_update);
5578 2146 : new_xmax = MultiXactIdCreate(xmax, old_status,
5579 : add_to_xmax, new_status);
5580 2146 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5581 : }
5582 28 : else if (!HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) &&
5583 10 : TransactionIdDidCommit(xmax))
5584 2 : {
5585 : /*
5586 : * It's a committed update, so we gotta preserve him as updater of the
5587 : * tuple.
5588 : */
5589 : MultiXactStatus status;
5590 : MultiXactStatus new_status;
5591 :
5592 2 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5593 0 : status = MultiXactStatusUpdate;
5594 : else
5595 2 : status = MultiXactStatusNoKeyUpdate;
5596 :
5597 2 : new_status = get_mxact_status_for_lock(mode, is_update);
5598 :
5599 : /*
5600 : * since it's not running, it's obviously impossible for the old
5601 : * updater to be identical to the current one, so we need not check
5602 : * for that case as we do in the block above.
5603 : */
5604 2 : new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5605 2 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5606 : }
5607 : else
5608 : {
5609 : /*
5610 : * Can get here iff the locking/updating transaction was running when
5611 : * the infomask was extracted from the tuple, but finished before
5612 : * TransactionIdIsInProgress got to run. Deal with it as if there was
5613 : * no locker at all in the first place.
5614 : */
5615 16 : old_infomask |= HEAP_XMAX_INVALID;
5616 16 : goto l5;
5617 : }
5618 :
5619 4296254 : *result_infomask = new_infomask;
5620 4296254 : *result_infomask2 = new_infomask2;
5621 4296254 : *result_xmax = new_xmax;
5622 4296254 : }
5623 :
5624 : /*
5625 : * Subroutine for heap_lock_updated_tuple_rec.
5626 : *
5627 : * Given a hypothetical multixact status held by the transaction identified
5628 : * with the given xid, does the current transaction need to wait, fail, or can
5629 : * it continue if it wanted to acquire a lock of the given mode? "needwait"
5630 : * is set to true if waiting is necessary; if it can continue, then TM_Ok is
5631 : * returned. If the lock is already held by the current transaction, return
5632 : * TM_SelfModified. In case of a conflict with another transaction, a
5633 : * different HeapTupleSatisfiesUpdate return code is returned.
5634 : *
5635 : * The held status is said to be hypothetical because it might correspond to a
5636 : * lock held by a single Xid, i.e. not a real MultiXactId; we express it this
5637 : * way for simplicity of API.
5638 : */
5639 : static TM_Result
5640 77548 : test_lockmode_for_conflict(MultiXactStatus status, TransactionId xid,
5641 : LockTupleMode mode, HeapTuple tup,
5642 : bool *needwait)
5643 : {
5644 : MultiXactStatus wantedstatus;
5645 :
5646 77548 : *needwait = false;
5647 77548 : wantedstatus = get_mxact_status_for_lock(mode, false);
5648 :
5649 : /*
5650 : * Note: we *must* check TransactionIdIsInProgress before
5651 : * TransactionIdDidAbort/Commit; see comment at top of heapam_visibility.c
5652 : * for an explanation.
5653 : */
5654 77548 : if (TransactionIdIsCurrentTransactionId(xid))
5655 : {
5656 : /*
5657 : * The tuple has already been locked by our own transaction. This is
5658 : * very rare but can happen if multiple transactions are trying to
5659 : * lock an ancient version of the same tuple.
5660 : */
5661 0 : return TM_SelfModified;
5662 : }
5663 77548 : else if (TransactionIdIsInProgress(xid))
5664 : {
5665 : /*
5666 : * If the locking transaction is running, what we do depends on
5667 : * whether the lock modes conflict: if they do, then we must wait for
5668 : * it to finish; otherwise we can fall through to lock this tuple
5669 : * version without waiting.
5670 : */
5671 73078 : if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
5672 73078 : LOCKMODE_from_mxstatus(wantedstatus)))
5673 : {
5674 16 : *needwait = true;
5675 : }
5676 :
5677 : /*
5678 : * If we set needwait above, then this value doesn't matter;
5679 : * otherwise, this value signals to caller that it's okay to proceed.
5680 : */
5681 73078 : return TM_Ok;
5682 : }
5683 4470 : else if (TransactionIdDidAbort(xid))
5684 412 : return TM_Ok;
5685 4058 : else if (TransactionIdDidCommit(xid))
5686 : {
5687 : /*
5688 : * The other transaction committed. If it was only a locker, then the
5689 : * lock is completely gone now and we can return success; but if it
5690 : * was an update, then what we do depends on whether the two lock
5691 : * modes conflict. If they conflict, then we must report error to
5692 : * caller. But if they don't, we can fall through to allow the current
5693 : * transaction to lock the tuple.
5694 : *
5695 : * Note: the reason we worry about ISUPDATE here is because as soon as
5696 : * a transaction ends, all its locks are gone and meaningless, and
5697 : * thus we can ignore them; whereas its updates persist. In the
5698 : * TransactionIdIsInProgress case, above, we don't need to check
5699 : * because we know the lock is still "alive" and thus a conflict needs
5700 : * always be checked.
5701 : */
5702 4058 : if (!ISUPDATE_from_mxstatus(status))
5703 4040 : return TM_Ok;
5704 :
5705 18 : if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
5706 18 : LOCKMODE_from_mxstatus(wantedstatus)))
5707 : {
5708 : /* bummer */
5709 16 : if (!ItemPointerEquals(&tup->t_self, &tup->t_data->t_ctid))
5710 12 : return TM_Updated;
5711 : else
5712 4 : return TM_Deleted;
5713 : }
5714 :
5715 2 : return TM_Ok;
5716 : }
5717 :
5718 : /* Not in progress, not aborted, not committed -- must have crashed */
5719 0 : return TM_Ok;
5720 : }
5721 :
5722 :
5723 : /*
5724 : * Recursive part of heap_lock_updated_tuple
5725 : *
5726 : * Fetch the tuple pointed to by tid in rel, and mark it as locked by the given
5727 : * xid with the given mode; if this tuple is updated, recurse to lock the new
5728 : * version as well.
5729 : */
5730 : static TM_Result
5731 4424 : heap_lock_updated_tuple_rec(Relation rel, TransactionId priorXmax,
5732 : const ItemPointerData *tid, TransactionId xid,
5733 : LockTupleMode mode)
5734 : {
5735 : TM_Result result;
5736 : ItemPointerData tupid;
5737 : HeapTupleData mytup;
5738 : Buffer buf;
5739 : uint16 new_infomask,
5740 : new_infomask2,
5741 : old_infomask,
5742 : old_infomask2;
5743 : TransactionId xmax,
5744 : new_xmax;
5745 4424 : bool cleared_all_frozen = false;
5746 : bool pinned_desired_page;
5747 4424 : Buffer vmbuffer = InvalidBuffer;
5748 : BlockNumber block;
5749 :
5750 4424 : ItemPointerCopy(tid, &tupid);
5751 :
5752 : for (;;)
5753 : {
5754 4430 : new_infomask = 0;
5755 4430 : new_xmax = InvalidTransactionId;
5756 4430 : block = ItemPointerGetBlockNumber(&tupid);
5757 4430 : ItemPointerCopy(&tupid, &(mytup.t_self));
5758 :
5759 4430 : if (!heap_fetch(rel, SnapshotAny, &mytup, &buf, false))
5760 : {
5761 : /*
5762 : * if we fail to find the updated version of the tuple, it's
5763 : * because it was vacuumed/pruned away after its creator
5764 : * transaction aborted. So behave as if we got to the end of the
5765 : * chain, and there's no further tuple to lock: return success to
5766 : * caller.
5767 : */
5768 0 : result = TM_Ok;
5769 0 : goto out_unlocked;
5770 : }
5771 :
5772 4430 : l4:
5773 4446 : CHECK_FOR_INTERRUPTS();
5774 :
5775 : /*
5776 : * Before locking the buffer, pin the visibility map page if it
5777 : * appears to be necessary. Since we haven't got the lock yet,
5778 : * someone else might be in the middle of changing this, so we'll need
5779 : * to recheck after we have the lock.
5780 : */
5781 4446 : if (PageIsAllVisible(BufferGetPage(buf)))
5782 : {
5783 0 : visibilitymap_pin(rel, block, &vmbuffer);
5784 0 : pinned_desired_page = true;
5785 : }
5786 : else
5787 4446 : pinned_desired_page = false;
5788 :
5789 4446 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
5790 :
5791 : /*
5792 : * If we didn't pin the visibility map page and the page has become
5793 : * all visible while we were busy locking the buffer, we'll have to
5794 : * unlock and re-lock, to avoid holding the buffer lock across I/O.
5795 : * That's a bit unfortunate, but hopefully shouldn't happen often.
5796 : *
5797 : * Note: in some paths through this function, we will reach here
5798 : * holding a pin on a vm page that may or may not be the one matching
5799 : * this page. If this page isn't all-visible, we won't use the vm
5800 : * page, but we hold onto such a pin till the end of the function.
5801 : */
5802 4446 : if (!pinned_desired_page && PageIsAllVisible(BufferGetPage(buf)))
5803 : {
5804 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5805 0 : visibilitymap_pin(rel, block, &vmbuffer);
5806 0 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
5807 : }
5808 :
5809 : /*
5810 : * Check the tuple XMIN against prior XMAX, if any. If we reached the
5811 : * end of the chain, we're done, so return success.
5812 : */
5813 8892 : if (TransactionIdIsValid(priorXmax) &&
5814 4446 : !TransactionIdEquals(HeapTupleHeaderGetXmin(mytup.t_data),
5815 : priorXmax))
5816 : {
5817 4 : result = TM_Ok;
5818 4 : goto out_locked;
5819 : }
5820 :
5821 : /*
5822 : * Also check Xmin: if this tuple was created by an aborted
5823 : * (sub)transaction, then we already locked the last live one in the
5824 : * chain, thus we're done, so return success.
5825 : */
5826 4442 : if (TransactionIdDidAbort(HeapTupleHeaderGetXmin(mytup.t_data)))
5827 : {
5828 48 : result = TM_Ok;
5829 48 : goto out_locked;
5830 : }
5831 :
5832 4394 : old_infomask = mytup.t_data->t_infomask;
5833 4394 : old_infomask2 = mytup.t_data->t_infomask2;
5834 4394 : xmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
5835 :
5836 : /*
5837 : * If this tuple version has been updated or locked by some concurrent
5838 : * transaction(s), what we do depends on whether our lock mode
5839 : * conflicts with what those other transactions hold, and also on the
5840 : * status of them.
5841 : */
5842 4394 : if (!(old_infomask & HEAP_XMAX_INVALID))
5843 : {
5844 : TransactionId rawxmax;
5845 : bool needwait;
5846 :
5847 4276 : rawxmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
5848 4276 : if (old_infomask & HEAP_XMAX_IS_MULTI)
5849 : {
5850 : int nmembers;
5851 : int i;
5852 : MultiXactMember *members;
5853 :
5854 : /*
5855 : * We don't need a test for pg_upgrade'd tuples: this is only
5856 : * applied to tuples after the first in an update chain. Said
5857 : * first tuple in the chain may well be locked-in-9.2-and-
5858 : * pg_upgraded, but that one was already locked by our caller,
5859 : * not us; and any subsequent ones cannot be because our
5860 : * caller must necessarily have obtained a snapshot later than
5861 : * the pg_upgrade itself.
5862 : */
5863 : Assert(!HEAP_LOCKED_UPGRADED(mytup.t_data->t_infomask));
5864 :
5865 4218 : nmembers = GetMultiXactIdMembers(rawxmax, &members, false,
5866 4218 : HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
5867 81708 : for (i = 0; i < nmembers; i++)
5868 : {
5869 77490 : result = test_lockmode_for_conflict(members[i].status,
5870 77490 : members[i].xid,
5871 : mode,
5872 : &mytup,
5873 : &needwait);
5874 :
5875 : /*
5876 : * If the tuple was already locked by ourselves in a
5877 : * previous iteration of this (say heap_lock_tuple was
5878 : * forced to restart the locking loop because of a change
5879 : * in xmax), then we hold the lock already on this tuple
5880 : * version and we don't need to do anything; and this is
5881 : * not an error condition either. We just need to skip
5882 : * this tuple and continue locking the next version in the
5883 : * update chain.
5884 : */
5885 77490 : if (result == TM_SelfModified)
5886 : {
5887 0 : pfree(members);
5888 0 : goto next;
5889 : }
5890 :
5891 77490 : if (needwait)
5892 : {
5893 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5894 0 : XactLockTableWait(members[i].xid, rel,
5895 : &mytup.t_self,
5896 : XLTW_LockUpdated);
5897 0 : pfree(members);
5898 0 : goto l4;
5899 : }
5900 77490 : if (result != TM_Ok)
5901 : {
5902 0 : pfree(members);
5903 0 : goto out_locked;
5904 : }
5905 : }
5906 4218 : if (members)
5907 4218 : pfree(members);
5908 : }
5909 : else
5910 : {
5911 : MultiXactStatus status;
5912 :
5913 : /*
5914 : * For a non-multi Xmax, we first need to compute the
5915 : * corresponding MultiXactStatus by using the infomask bits.
5916 : */
5917 58 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
5918 : {
5919 20 : if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
5920 20 : status = MultiXactStatusForKeyShare;
5921 0 : else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
5922 0 : status = MultiXactStatusForShare;
5923 0 : else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
5924 : {
5925 0 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5926 0 : status = MultiXactStatusForUpdate;
5927 : else
5928 0 : status = MultiXactStatusForNoKeyUpdate;
5929 : }
5930 : else
5931 : {
5932 : /*
5933 : * LOCK_ONLY present alone (a pg_upgraded tuple marked
5934 : * as share-locked in the old cluster) shouldn't be
5935 : * seen in the middle of an update chain.
5936 : */
5937 0 : elog(ERROR, "invalid lock status in tuple");
5938 : }
5939 : }
5940 : else
5941 : {
5942 : /* it's an update, but which kind? */
5943 38 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5944 28 : status = MultiXactStatusUpdate;
5945 : else
5946 10 : status = MultiXactStatusNoKeyUpdate;
5947 : }
5948 :
5949 58 : result = test_lockmode_for_conflict(status, rawxmax, mode,
5950 : &mytup, &needwait);
5951 :
5952 : /*
5953 : * If the tuple was already locked by ourselves in a previous
5954 : * iteration of this (say heap_lock_tuple was forced to
5955 : * restart the locking loop because of a change in xmax), then
5956 : * we hold the lock already on this tuple version and we don't
5957 : * need to do anything; and this is not an error condition
5958 : * either. We just need to skip this tuple and continue
5959 : * locking the next version in the update chain.
5960 : */
5961 58 : if (result == TM_SelfModified)
5962 0 : goto next;
5963 :
5964 58 : if (needwait)
5965 : {
5966 16 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5967 16 : XactLockTableWait(rawxmax, rel, &mytup.t_self,
5968 : XLTW_LockUpdated);
5969 16 : goto l4;
5970 : }
5971 42 : if (result != TM_Ok)
5972 : {
5973 16 : goto out_locked;
5974 : }
5975 : }
5976 : }
5977 :
5978 : /* compute the new Xmax and infomask values for the tuple ... */
5979 4362 : compute_new_xmax_infomask(xmax, old_infomask, mytup.t_data->t_infomask2,
5980 : xid, mode, false,
5981 : &new_xmax, &new_infomask, &new_infomask2);
5982 :
5983 4362 : if (PageIsAllVisible(BufferGetPage(buf)) &&
5984 0 : visibilitymap_clear(rel, block, vmbuffer,
5985 : VISIBILITYMAP_ALL_FROZEN))
5986 0 : cleared_all_frozen = true;
5987 :
5988 4362 : START_CRIT_SECTION();
5989 :
5990 : /* ... and set them */
5991 4362 : HeapTupleHeaderSetXmax(mytup.t_data, new_xmax);
5992 4362 : mytup.t_data->t_infomask &= ~HEAP_XMAX_BITS;
5993 4362 : mytup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
5994 4362 : mytup.t_data->t_infomask |= new_infomask;
5995 4362 : mytup.t_data->t_infomask2 |= new_infomask2;
5996 :
5997 4362 : MarkBufferDirty(buf);
5998 :
5999 : /* XLOG stuff */
6000 4362 : if (RelationNeedsWAL(rel))
6001 : {
6002 : xl_heap_lock_updated xlrec;
6003 : XLogRecPtr recptr;
6004 4362 : Page page = BufferGetPage(buf);
6005 :
6006 4362 : XLogBeginInsert();
6007 4362 : XLogRegisterBuffer(0, buf, REGBUF_STANDARD);
6008 :
6009 4362 : xlrec.offnum = ItemPointerGetOffsetNumber(&mytup.t_self);
6010 4362 : xlrec.xmax = new_xmax;
6011 4362 : xlrec.infobits_set = compute_infobits(new_infomask, new_infomask2);
6012 4362 : xlrec.flags =
6013 4362 : cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
6014 :
6015 4362 : XLogRegisterData(&xlrec, SizeOfHeapLockUpdated);
6016 :
6017 4362 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_LOCK_UPDATED);
6018 :
6019 4362 : PageSetLSN(page, recptr);
6020 : }
6021 :
6022 4362 : END_CRIT_SECTION();
6023 :
6024 4362 : next:
6025 : /* if we find the end of update chain, we're done. */
6026 8724 : if (mytup.t_data->t_infomask & HEAP_XMAX_INVALID ||
6027 8724 : HeapTupleHeaderIndicatesMovedPartitions(mytup.t_data) ||
6028 4370 : ItemPointerEquals(&mytup.t_self, &mytup.t_data->t_ctid) ||
6029 8 : HeapTupleHeaderIsOnlyLocked(mytup.t_data))
6030 : {
6031 4356 : result = TM_Ok;
6032 4356 : goto out_locked;
6033 : }
6034 :
6035 : /* tail recursion */
6036 6 : priorXmax = HeapTupleHeaderGetUpdateXid(mytup.t_data);
6037 6 : ItemPointerCopy(&(mytup.t_data->t_ctid), &tupid);
6038 6 : UnlockReleaseBuffer(buf);
6039 : }
6040 :
6041 : result = TM_Ok;
6042 :
6043 4424 : out_locked:
6044 4424 : UnlockReleaseBuffer(buf);
6045 :
6046 4424 : out_unlocked:
6047 4424 : if (vmbuffer != InvalidBuffer)
6048 0 : ReleaseBuffer(vmbuffer);
6049 :
6050 4424 : return result;
6051 : }
6052 :
6053 : /*
6054 : * heap_lock_updated_tuple
6055 : * Follow update chain when locking an updated tuple, acquiring locks (row
6056 : * marks) on the updated versions.
6057 : *
6058 : * 'prior_infomask', 'prior_raw_xmax' and 'prior_ctid' are the corresponding
6059 : * fields from the initial tuple. We will lock the tuples starting from the
6060 : * one that 'prior_ctid' points to. Note: This function does not lock the
6061 : * initial tuple itself.
6062 : *
6063 : * This function doesn't check visibility, it just unconditionally marks the
6064 : * tuple(s) as locked. If any tuple in the updated chain is being deleted
6065 : * concurrently (or updated with the key being modified), sleep until the
6066 : * transaction doing it is finished.
6067 : *
6068 : * Note that we don't acquire heavyweight tuple locks on the tuples we walk
6069 : * when we have to wait for other transactions to release them, as opposed to
6070 : * what heap_lock_tuple does. The reason is that having more than one
6071 : * transaction walking the chain is probably uncommon enough that risk of
6072 : * starvation is not likely: one of the preconditions for being here is that
6073 : * the snapshot in use predates the update that created this tuple (because we
6074 : * started at an earlier version of the tuple), but at the same time such a
6075 : * transaction cannot be using repeatable read or serializable isolation
6076 : * levels, because that would lead to a serializability failure.
6077 : */
6078 : static TM_Result
6079 4428 : heap_lock_updated_tuple(Relation rel,
6080 : uint16 prior_infomask,
6081 : TransactionId prior_raw_xmax,
6082 : const ItemPointerData *prior_ctid,
6083 : TransactionId xid, LockTupleMode mode)
6084 : {
6085 4428 : INJECTION_POINT("heap_lock_updated_tuple", NULL);
6086 :
6087 : /*
6088 : * If the tuple has moved into another partition (effectively a delete)
6089 : * stop here.
6090 : */
6091 4428 : if (!ItemPointerIndicatesMovedPartitions(prior_ctid))
6092 : {
6093 : TransactionId prior_xmax;
6094 :
6095 : /*
6096 : * If this is the first possibly-multixact-able operation in the
6097 : * current transaction, set my per-backend OldestMemberMXactId
6098 : * setting. We can be certain that the transaction will never become a
6099 : * member of any older MultiXactIds than that. (We have to do this
6100 : * even if we end up just using our own TransactionId below, since
6101 : * some other backend could incorporate our XID into a MultiXact
6102 : * immediately afterwards.)
6103 : */
6104 4424 : MultiXactIdSetOldestMember();
6105 :
6106 8848 : prior_xmax = (prior_infomask & HEAP_XMAX_IS_MULTI) ?
6107 4424 : MultiXactIdGetUpdateXid(prior_raw_xmax, prior_infomask) : prior_raw_xmax;
6108 4424 : return heap_lock_updated_tuple_rec(rel, prior_xmax, prior_ctid, xid, mode);
6109 : }
6110 :
6111 : /* nothing to lock */
6112 4 : return TM_Ok;
6113 : }
6114 :
6115 : /*
6116 : * heap_finish_speculative - mark speculative insertion as successful
6117 : *
6118 : * To successfully finish a speculative insertion we have to clear speculative
6119 : * token from tuple. To do so the t_ctid field, which will contain a
6120 : * speculative token value, is modified in place to point to the tuple itself,
6121 : * which is characteristic of a newly inserted ordinary tuple.
6122 : *
6123 : * NB: It is not ok to commit without either finishing or aborting a
6124 : * speculative insertion. We could treat speculative tuples of committed
6125 : * transactions implicitly as completed, but then we would have to be prepared
6126 : * to deal with speculative tokens on committed tuples. That wouldn't be
6127 : * difficult - no-one looks at the ctid field of a tuple with invalid xmax -
6128 : * but clearing the token at completion isn't very expensive either.
6129 : * An explicit confirmation WAL record also makes logical decoding simpler.
6130 : */
6131 : void
6132 4168 : heap_finish_speculative(Relation relation, const ItemPointerData *tid)
6133 : {
6134 : Buffer buffer;
6135 : Page page;
6136 : OffsetNumber offnum;
6137 : ItemId lp;
6138 : HeapTupleHeader htup;
6139 :
6140 4168 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
6141 4168 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6142 4168 : page = BufferGetPage(buffer);
6143 :
6144 4168 : offnum = ItemPointerGetOffsetNumber(tid);
6145 4168 : if (offnum < 1 || offnum > PageGetMaxOffsetNumber(page))
6146 0 : elog(ERROR, "offnum out of range");
6147 4168 : lp = PageGetItemId(page, offnum);
6148 4168 : if (!ItemIdIsNormal(lp))
6149 0 : elog(ERROR, "invalid lp");
6150 :
6151 4168 : htup = (HeapTupleHeader) PageGetItem(page, lp);
6152 :
6153 : /* NO EREPORT(ERROR) from here till changes are logged */
6154 4168 : START_CRIT_SECTION();
6155 :
6156 : Assert(HeapTupleHeaderIsSpeculative(htup));
6157 :
6158 4168 : MarkBufferDirty(buffer);
6159 :
6160 : /*
6161 : * Replace the speculative insertion token with a real t_ctid, pointing to
6162 : * itself like it does on regular tuples.
6163 : */
6164 4168 : htup->t_ctid = *tid;
6165 :
6166 : /* XLOG stuff */
6167 4168 : if (RelationNeedsWAL(relation))
6168 : {
6169 : xl_heap_confirm xlrec;
6170 : XLogRecPtr recptr;
6171 :
6172 4136 : xlrec.offnum = ItemPointerGetOffsetNumber(tid);
6173 :
6174 4136 : XLogBeginInsert();
6175 :
6176 : /* We want the same filtering on this as on a plain insert */
6177 4136 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
6178 :
6179 4136 : XLogRegisterData(&xlrec, SizeOfHeapConfirm);
6180 4136 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
6181 :
6182 4136 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_CONFIRM);
6183 :
6184 4136 : PageSetLSN(page, recptr);
6185 : }
6186 :
6187 4168 : END_CRIT_SECTION();
6188 :
6189 4168 : UnlockReleaseBuffer(buffer);
6190 4168 : }
6191 :
6192 : /*
6193 : * heap_abort_speculative - kill a speculatively inserted tuple
6194 : *
6195 : * Marks a tuple that was speculatively inserted in the same command as dead,
6196 : * by setting its xmin as invalid. That makes it immediately appear as dead
6197 : * to all transactions, including our own. In particular, it makes
6198 : * HeapTupleSatisfiesDirty() regard the tuple as dead, so that another backend
6199 : * inserting a duplicate key value won't unnecessarily wait for our whole
6200 : * transaction to finish (it'll just wait for our speculative insertion to
6201 : * finish).
6202 : *
6203 : * Killing the tuple prevents "unprincipled deadlocks", which are deadlocks
6204 : * that arise due to a mutual dependency that is not user visible. By
6205 : * definition, unprincipled deadlocks cannot be prevented by the user
6206 : * reordering lock acquisition in client code, because the implementation level
6207 : * lock acquisitions are not under the user's direct control. If speculative
6208 : * inserters did not take this precaution, then under high concurrency they
6209 : * could deadlock with each other, which would not be acceptable.
6210 : *
6211 : * This is somewhat redundant with heap_delete, but we prefer to have a
6212 : * dedicated routine with stripped down requirements. Note that this is also
6213 : * used to delete the TOAST tuples created during speculative insertion.
6214 : *
6215 : * This routine does not affect logical decoding as it only looks at
6216 : * confirmation records.
6217 : */
6218 : void
6219 32 : heap_abort_speculative(Relation relation, const ItemPointerData *tid)
6220 : {
6221 32 : TransactionId xid = GetCurrentTransactionId();
6222 : ItemId lp;
6223 : HeapTupleData tp;
6224 : Page page;
6225 : BlockNumber block;
6226 : Buffer buffer;
6227 :
6228 : Assert(ItemPointerIsValid(tid));
6229 :
6230 32 : block = ItemPointerGetBlockNumber(tid);
6231 32 : buffer = ReadBuffer(relation, block);
6232 32 : page = BufferGetPage(buffer);
6233 :
6234 32 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6235 :
6236 : /*
6237 : * Page can't be all visible, we just inserted into it, and are still
6238 : * running.
6239 : */
6240 : Assert(!PageIsAllVisible(page));
6241 :
6242 32 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
6243 : Assert(ItemIdIsNormal(lp));
6244 :
6245 32 : tp.t_tableOid = RelationGetRelid(relation);
6246 32 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
6247 32 : tp.t_len = ItemIdGetLength(lp);
6248 32 : tp.t_self = *tid;
6249 :
6250 : /*
6251 : * Sanity check that the tuple really is a speculatively inserted tuple,
6252 : * inserted by us.
6253 : */
6254 32 : if (tp.t_data->t_choice.t_heap.t_xmin != xid)
6255 0 : elog(ERROR, "attempted to kill a tuple inserted by another transaction");
6256 32 : if (!(IsToastRelation(relation) || HeapTupleHeaderIsSpeculative(tp.t_data)))
6257 0 : elog(ERROR, "attempted to kill a non-speculative tuple");
6258 : Assert(!HeapTupleHeaderIsHeapOnly(tp.t_data));
6259 :
6260 : /*
6261 : * No need to check for serializable conflicts here. There is never a
6262 : * need for a combo CID, either. No need to extract replica identity, or
6263 : * do anything special with infomask bits.
6264 : */
6265 :
6266 32 : START_CRIT_SECTION();
6267 :
6268 : /*
6269 : * The tuple will become DEAD immediately. Flag that this page is a
6270 : * candidate for pruning by setting xmin to TransactionXmin. While not
6271 : * immediately prunable, it is the oldest xid we can cheaply determine
6272 : * that's safe against wraparound / being older than the table's
6273 : * relfrozenxid. To defend against the unlikely case of a new relation
6274 : * having a newer relfrozenxid than our TransactionXmin, use relfrozenxid
6275 : * if so (vacuum can't subsequently move relfrozenxid to beyond
6276 : * TransactionXmin, so there's no race here).
6277 : */
6278 : Assert(TransactionIdIsValid(TransactionXmin));
6279 : {
6280 32 : TransactionId relfrozenxid = relation->rd_rel->relfrozenxid;
6281 : TransactionId prune_xid;
6282 :
6283 32 : if (TransactionIdPrecedes(TransactionXmin, relfrozenxid))
6284 0 : prune_xid = relfrozenxid;
6285 : else
6286 32 : prune_xid = TransactionXmin;
6287 32 : PageSetPrunable(page, prune_xid);
6288 : }
6289 :
6290 : /* store transaction information of xact deleting the tuple */
6291 32 : tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
6292 32 : tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
6293 :
6294 : /*
6295 : * Set the tuple header xmin to InvalidTransactionId. This makes the
6296 : * tuple immediately invisible everyone. (In particular, to any
6297 : * transactions waiting on the speculative token, woken up later.)
6298 : */
6299 32 : HeapTupleHeaderSetXmin(tp.t_data, InvalidTransactionId);
6300 :
6301 : /* Clear the speculative insertion token too */
6302 32 : tp.t_data->t_ctid = tp.t_self;
6303 :
6304 32 : MarkBufferDirty(buffer);
6305 :
6306 : /*
6307 : * XLOG stuff
6308 : *
6309 : * The WAL records generated here match heap_delete(). The same recovery
6310 : * routines are used.
6311 : */
6312 32 : if (RelationNeedsWAL(relation))
6313 : {
6314 : xl_heap_delete xlrec;
6315 : XLogRecPtr recptr;
6316 :
6317 24 : xlrec.flags = XLH_DELETE_IS_SUPER;
6318 48 : xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
6319 24 : tp.t_data->t_infomask2);
6320 24 : xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
6321 24 : xlrec.xmax = xid;
6322 :
6323 24 : XLogBeginInsert();
6324 24 : XLogRegisterData(&xlrec, SizeOfHeapDelete);
6325 24 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
6326 :
6327 : /* No replica identity & replication origin logged */
6328 :
6329 24 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
6330 :
6331 24 : PageSetLSN(page, recptr);
6332 : }
6333 :
6334 32 : END_CRIT_SECTION();
6335 :
6336 32 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6337 :
6338 32 : if (HeapTupleHasExternal(&tp))
6339 : {
6340 : Assert(!IsToastRelation(relation));
6341 2 : heap_toast_delete(relation, &tp, true);
6342 : }
6343 :
6344 : /*
6345 : * Never need to mark tuple for invalidation, since catalogs don't support
6346 : * speculative insertion
6347 : */
6348 :
6349 : /* Now we can release the buffer */
6350 32 : ReleaseBuffer(buffer);
6351 :
6352 : /* count deletion, as we counted the insertion too */
6353 32 : pgstat_count_heap_delete(relation);
6354 32 : }
6355 :
6356 : /*
6357 : * heap_inplace_lock - protect inplace update from concurrent heap_update()
6358 : *
6359 : * Evaluate whether the tuple's state is compatible with a no-key update.
6360 : * Current transaction rowmarks are fine, as is KEY SHARE from any
6361 : * transaction. If compatible, return true with the buffer exclusive-locked,
6362 : * and the caller must release that by calling
6363 : * heap_inplace_update_and_unlock(), calling heap_inplace_unlock(), or raising
6364 : * an error. Otherwise, call release_callback(arg), wait for blocking
6365 : * transactions to end, and return false.
6366 : *
6367 : * Since this is intended for system catalogs and SERIALIZABLE doesn't cover
6368 : * DDL, this doesn't guarantee any particular predicate locking.
6369 : *
6370 : * heap_delete() is a rarer source of blocking transactions (xwait). We'll
6371 : * wait for such a transaction just like for the normal heap_update() case.
6372 : * Normal concurrent DROP commands won't cause that, because all inplace
6373 : * updaters take some lock that conflicts with DROP. An explicit SQL "DELETE
6374 : * FROM pg_class" can cause it. By waiting, if the concurrent transaction
6375 : * executed both "DELETE FROM pg_class" and "INSERT INTO pg_class", our caller
6376 : * can find the successor tuple.
6377 : *
6378 : * Readers of inplace-updated fields expect changes to those fields are
6379 : * durable. For example, vac_truncate_clog() reads datfrozenxid from
6380 : * pg_database tuples via catalog snapshots. A future snapshot must not
6381 : * return a lower datfrozenxid for the same database OID (lower in the
6382 : * FullTransactionIdPrecedes() sense). We achieve that since no update of a
6383 : * tuple can start while we hold a lock on its buffer. In cases like
6384 : * BEGIN;GRANT;CREATE INDEX;COMMIT we're inplace-updating a tuple visible only
6385 : * to this transaction. ROLLBACK then is one case where it's okay to lose
6386 : * inplace updates. (Restoring relhasindex=false on ROLLBACK is fine, since
6387 : * any concurrent CREATE INDEX would have blocked, then inplace-updated the
6388 : * committed tuple.)
6389 : *
6390 : * In principle, we could avoid waiting by overwriting every tuple in the
6391 : * updated tuple chain. Reader expectations permit updating a tuple only if
6392 : * it's aborted, is the tail of the chain, or we already updated the tuple
6393 : * referenced in its t_ctid. Hence, we would need to overwrite the tuples in
6394 : * order from tail to head. That would imply either (a) mutating all tuples
6395 : * in one critical section or (b) accepting a chance of partial completion.
6396 : * Partial completion of a relfrozenxid update would have the weird
6397 : * consequence that the table's next VACUUM could see the table's relfrozenxid
6398 : * move forward between vacuum_get_cutoffs() and finishing.
6399 : */
6400 : bool
6401 323646 : heap_inplace_lock(Relation relation,
6402 : HeapTuple oldtup_ptr, Buffer buffer,
6403 : void (*release_callback) (void *), void *arg)
6404 : {
6405 323646 : HeapTupleData oldtup = *oldtup_ptr; /* minimize diff vs. heap_update() */
6406 : TM_Result result;
6407 : bool ret;
6408 :
6409 : #ifdef USE_ASSERT_CHECKING
6410 : if (RelationGetRelid(relation) == RelationRelationId)
6411 : check_inplace_rel_lock(oldtup_ptr);
6412 : #endif
6413 :
6414 : Assert(BufferIsValid(buffer));
6415 :
6416 : /*
6417 : * Register shared cache invals if necessary. Other sessions may finish
6418 : * inplace updates of this tuple between this step and LockTuple(). Since
6419 : * inplace updates don't change cache keys, that's harmless.
6420 : *
6421 : * While it's tempting to register invals only after confirming we can
6422 : * return true, the following obstacle precludes reordering steps that
6423 : * way. Registering invals might reach a CatalogCacheInitializeCache()
6424 : * that locks "buffer". That would hang indefinitely if running after our
6425 : * own LockBuffer(). Hence, we must register invals before LockBuffer().
6426 : */
6427 323646 : CacheInvalidateHeapTupleInplace(relation, oldtup_ptr);
6428 :
6429 323646 : LockTuple(relation, &oldtup.t_self, InplaceUpdateTupleLock);
6430 323646 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6431 :
6432 : /*----------
6433 : * Interpret HeapTupleSatisfiesUpdate() like heap_update() does, except:
6434 : *
6435 : * - wait unconditionally
6436 : * - already locked tuple above, since inplace needs that unconditionally
6437 : * - don't recheck header after wait: simpler to defer to next iteration
6438 : * - don't try to continue even if the updater aborts: likewise
6439 : * - no crosscheck
6440 : */
6441 323646 : result = HeapTupleSatisfiesUpdate(&oldtup, GetCurrentCommandId(false),
6442 : buffer);
6443 :
6444 323646 : if (result == TM_Invisible)
6445 : {
6446 : /* no known way this can happen */
6447 0 : ereport(ERROR,
6448 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6449 : errmsg_internal("attempted to overwrite invisible tuple")));
6450 : }
6451 323646 : else if (result == TM_SelfModified)
6452 : {
6453 : /*
6454 : * CREATE INDEX might reach this if an expression is silly enough to
6455 : * call e.g. SELECT ... FROM pg_class FOR SHARE. C code of other SQL
6456 : * statements might get here after a heap_update() of the same row, in
6457 : * the absence of an intervening CommandCounterIncrement().
6458 : */
6459 0 : ereport(ERROR,
6460 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6461 : errmsg("tuple to be updated was already modified by an operation triggered by the current command")));
6462 : }
6463 323646 : else if (result == TM_BeingModified)
6464 : {
6465 : TransactionId xwait;
6466 : uint16 infomask;
6467 :
6468 88 : xwait = HeapTupleHeaderGetRawXmax(oldtup.t_data);
6469 88 : infomask = oldtup.t_data->t_infomask;
6470 :
6471 88 : if (infomask & HEAP_XMAX_IS_MULTI)
6472 : {
6473 10 : LockTupleMode lockmode = LockTupleNoKeyExclusive;
6474 10 : MultiXactStatus mxact_status = MultiXactStatusNoKeyUpdate;
6475 : int remain;
6476 :
6477 10 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
6478 : lockmode, NULL))
6479 : {
6480 4 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6481 4 : release_callback(arg);
6482 4 : ret = false;
6483 4 : MultiXactIdWait((MultiXactId) xwait, mxact_status, infomask,
6484 : relation, &oldtup.t_self, XLTW_Update,
6485 : &remain);
6486 : }
6487 : else
6488 6 : ret = true;
6489 : }
6490 78 : else if (TransactionIdIsCurrentTransactionId(xwait))
6491 2 : ret = true;
6492 76 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask))
6493 2 : ret = true;
6494 : else
6495 : {
6496 74 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6497 74 : release_callback(arg);
6498 74 : ret = false;
6499 74 : XactLockTableWait(xwait, relation, &oldtup.t_self,
6500 : XLTW_Update);
6501 : }
6502 : }
6503 : else
6504 : {
6505 323558 : ret = (result == TM_Ok);
6506 323558 : if (!ret)
6507 : {
6508 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6509 0 : release_callback(arg);
6510 : }
6511 : }
6512 :
6513 : /*
6514 : * GetCatalogSnapshot() relies on invalidation messages to know when to
6515 : * take a new snapshot. COMMIT of xwait is responsible for sending the
6516 : * invalidation. We're not acquiring heavyweight locks sufficient to
6517 : * block if not yet sent, so we must take a new snapshot to ensure a later
6518 : * attempt has a fair chance. While we don't need this if xwait aborted,
6519 : * don't bother optimizing that.
6520 : */
6521 323646 : if (!ret)
6522 : {
6523 78 : UnlockTuple(relation, &oldtup.t_self, InplaceUpdateTupleLock);
6524 78 : ForgetInplace_Inval();
6525 78 : InvalidateCatalogSnapshot();
6526 : }
6527 323646 : return ret;
6528 : }
6529 :
6530 : /*
6531 : * heap_inplace_update_and_unlock - core of systable_inplace_update_finish
6532 : *
6533 : * The tuple cannot change size, and therefore its header fields and null
6534 : * bitmap (if any) don't change either.
6535 : *
6536 : * Since we hold LOCKTAG_TUPLE, no updater has a local copy of this tuple.
6537 : */
6538 : void
6539 171344 : heap_inplace_update_and_unlock(Relation relation,
6540 : HeapTuple oldtup, HeapTuple tuple,
6541 : Buffer buffer)
6542 : {
6543 171344 : HeapTupleHeader htup = oldtup->t_data;
6544 : uint32 oldlen;
6545 : uint32 newlen;
6546 : char *dst;
6547 : char *src;
6548 171344 : int nmsgs = 0;
6549 171344 : SharedInvalidationMessage *invalMessages = NULL;
6550 171344 : bool RelcacheInitFileInval = false;
6551 :
6552 : Assert(ItemPointerEquals(&oldtup->t_self, &tuple->t_self));
6553 171344 : oldlen = oldtup->t_len - htup->t_hoff;
6554 171344 : newlen = tuple->t_len - tuple->t_data->t_hoff;
6555 171344 : if (oldlen != newlen || htup->t_hoff != tuple->t_data->t_hoff)
6556 0 : elog(ERROR, "wrong tuple length");
6557 :
6558 171344 : dst = (char *) htup + htup->t_hoff;
6559 171344 : src = (char *) tuple->t_data + tuple->t_data->t_hoff;
6560 :
6561 : /* Like RecordTransactionCommit(), log only if needed */
6562 171344 : if (XLogStandbyInfoActive())
6563 117116 : nmsgs = inplaceGetInvalidationMessages(&invalMessages,
6564 : &RelcacheInitFileInval);
6565 :
6566 : /*
6567 : * Unlink relcache init files as needed. If unlinking, acquire
6568 : * RelCacheInitLock until after associated invalidations. By doing this
6569 : * in advance, if we checkpoint and then crash between inplace
6570 : * XLogInsert() and inval, we don't rely on StartupXLOG() ->
6571 : * RelationCacheInitFileRemove(). That uses elevel==LOG, so replay would
6572 : * neglect to PANIC on EIO.
6573 : */
6574 171344 : PreInplace_Inval();
6575 :
6576 : /*----------
6577 : * NO EREPORT(ERROR) from here till changes are complete
6578 : *
6579 : * Our buffer lock won't stop a reader having already pinned and checked
6580 : * visibility for this tuple. Hence, we write WAL first, then mutate the
6581 : * buffer. Like in MarkBufferDirtyHint() or RecordTransactionCommit(),
6582 : * checkpoint delay makes that acceptable. With the usual order of
6583 : * changes, a crash after memcpy() and before XLogInsert() could allow
6584 : * datfrozenxid to overtake relfrozenxid:
6585 : *
6586 : * ["D" is a VACUUM (ONLY_DATABASE_STATS)]
6587 : * ["R" is a VACUUM tbl]
6588 : * D: vac_update_datfrozenxid() -> systable_beginscan(pg_class)
6589 : * D: systable_getnext() returns pg_class tuple of tbl
6590 : * R: memcpy() into pg_class tuple of tbl
6591 : * D: raise pg_database.datfrozenxid, XLogInsert(), finish
6592 : * [crash]
6593 : * [recovery restores datfrozenxid w/o relfrozenxid]
6594 : *
6595 : * Mimic MarkBufferDirtyHint() subroutine XLogSaveBufferForHint().
6596 : * Specifically, use DELAY_CHKPT_START, and copy the buffer to the stack.
6597 : * The stack copy facilitates a FPI of the post-mutation block before we
6598 : * accept other sessions seeing it. DELAY_CHKPT_START allows us to
6599 : * XLogInsert() before MarkBufferDirty(). Since XLogSaveBufferForHint()
6600 : * can operate under BUFFER_LOCK_SHARED, it can't avoid DELAY_CHKPT_START.
6601 : * This function, however, likely could avoid it with the following order
6602 : * of operations: MarkBufferDirty(), XLogInsert(), memcpy(). Opt to use
6603 : * DELAY_CHKPT_START here, too, as a way to have fewer distinct code
6604 : * patterns to analyze. Inplace update isn't so frequent that it should
6605 : * pursue the small optimization of skipping DELAY_CHKPT_START.
6606 : */
6607 : Assert((MyProc->delayChkptFlags & DELAY_CHKPT_START) == 0);
6608 171344 : START_CRIT_SECTION();
6609 171344 : MyProc->delayChkptFlags |= DELAY_CHKPT_START;
6610 :
6611 : /* XLOG stuff */
6612 171344 : if (RelationNeedsWAL(relation))
6613 : {
6614 : xl_heap_inplace xlrec;
6615 : PGAlignedBlock copied_buffer;
6616 171336 : char *origdata = (char *) BufferGetBlock(buffer);
6617 171336 : Page page = BufferGetPage(buffer);
6618 171336 : uint16 lower = ((PageHeader) page)->pd_lower;
6619 171336 : uint16 upper = ((PageHeader) page)->pd_upper;
6620 : uintptr_t dst_offset_in_block;
6621 : RelFileLocator rlocator;
6622 : ForkNumber forkno;
6623 : BlockNumber blkno;
6624 : XLogRecPtr recptr;
6625 :
6626 171336 : xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
6627 171336 : xlrec.dbId = MyDatabaseId;
6628 171336 : xlrec.tsId = MyDatabaseTableSpace;
6629 171336 : xlrec.relcacheInitFileInval = RelcacheInitFileInval;
6630 171336 : xlrec.nmsgs = nmsgs;
6631 :
6632 171336 : XLogBeginInsert();
6633 171336 : XLogRegisterData(&xlrec, MinSizeOfHeapInplace);
6634 171336 : if (nmsgs != 0)
6635 87230 : XLogRegisterData(invalMessages,
6636 : nmsgs * sizeof(SharedInvalidationMessage));
6637 :
6638 : /* register block matching what buffer will look like after changes */
6639 171336 : memcpy(copied_buffer.data, origdata, lower);
6640 171336 : memcpy(copied_buffer.data + upper, origdata + upper, BLCKSZ - upper);
6641 171336 : dst_offset_in_block = dst - origdata;
6642 171336 : memcpy(copied_buffer.data + dst_offset_in_block, src, newlen);
6643 171336 : BufferGetTag(buffer, &rlocator, &forkno, &blkno);
6644 : Assert(forkno == MAIN_FORKNUM);
6645 171336 : XLogRegisterBlock(0, &rlocator, forkno, blkno, copied_buffer.data,
6646 : REGBUF_STANDARD);
6647 171336 : XLogRegisterBufData(0, src, newlen);
6648 :
6649 : /* inplace updates aren't decoded atm, don't log the origin */
6650 :
6651 171336 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_INPLACE);
6652 :
6653 171336 : PageSetLSN(page, recptr);
6654 : }
6655 :
6656 171344 : memcpy(dst, src, newlen);
6657 :
6658 171344 : MarkBufferDirty(buffer);
6659 :
6660 171344 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6661 :
6662 : /*
6663 : * Send invalidations to shared queue. SearchSysCacheLocked1() assumes we
6664 : * do this before UnlockTuple().
6665 : */
6666 171344 : AtInplace_Inval();
6667 :
6668 171344 : MyProc->delayChkptFlags &= ~DELAY_CHKPT_START;
6669 171344 : END_CRIT_SECTION();
6670 171344 : UnlockTuple(relation, &tuple->t_self, InplaceUpdateTupleLock);
6671 :
6672 171344 : AcceptInvalidationMessages(); /* local processing of just-sent inval */
6673 :
6674 : /*
6675 : * Queue a transactional inval, for logical decoding and for third-party
6676 : * code that might have been relying on it since long before inplace
6677 : * update adopted immediate invalidation. See README.tuplock section
6678 : * "Reading inplace-updated columns" for logical decoding details.
6679 : */
6680 171344 : if (!IsBootstrapProcessingMode())
6681 141458 : CacheInvalidateHeapTuple(relation, tuple, NULL);
6682 171344 : }
6683 :
6684 : /*
6685 : * heap_inplace_unlock - reverse of heap_inplace_lock
6686 : */
6687 : void
6688 152224 : heap_inplace_unlock(Relation relation,
6689 : HeapTuple oldtup, Buffer buffer)
6690 : {
6691 152224 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6692 152224 : UnlockTuple(relation, &oldtup->t_self, InplaceUpdateTupleLock);
6693 152224 : ForgetInplace_Inval();
6694 152224 : }
6695 :
6696 : #define FRM_NOOP 0x0001
6697 : #define FRM_INVALIDATE_XMAX 0x0002
6698 : #define FRM_RETURN_IS_XID 0x0004
6699 : #define FRM_RETURN_IS_MULTI 0x0008
6700 : #define FRM_MARK_COMMITTED 0x0010
6701 :
6702 : /*
6703 : * FreezeMultiXactId
6704 : * Determine what to do during freezing when a tuple is marked by a
6705 : * MultiXactId.
6706 : *
6707 : * "flags" is an output value; it's used to tell caller what to do on return.
6708 : * "pagefrz" is an input/output value, used to manage page level freezing.
6709 : *
6710 : * Possible values that we can set in "flags":
6711 : * FRM_NOOP
6712 : * don't do anything -- keep existing Xmax
6713 : * FRM_INVALIDATE_XMAX
6714 : * mark Xmax as InvalidTransactionId and set XMAX_INVALID flag.
6715 : * FRM_RETURN_IS_XID
6716 : * The Xid return value is a single update Xid to set as xmax.
6717 : * FRM_MARK_COMMITTED
6718 : * Xmax can be marked as HEAP_XMAX_COMMITTED
6719 : * FRM_RETURN_IS_MULTI
6720 : * The return value is a new MultiXactId to set as new Xmax.
6721 : * (caller must obtain proper infomask bits using GetMultiXactIdHintBits)
6722 : *
6723 : * Caller delegates control of page freezing to us. In practice we always
6724 : * force freezing of caller's page unless FRM_NOOP processing is indicated.
6725 : * We help caller ensure that XIDs < FreezeLimit and MXIDs < MultiXactCutoff
6726 : * can never be left behind. We freely choose when and how to process each
6727 : * Multi, without ever violating the cutoff postconditions for freezing.
6728 : *
6729 : * It's useful to remove Multis on a proactive timeline (relative to freezing
6730 : * XIDs) to keep MultiXact member SLRU buffer misses to a minimum. It can also
6731 : * be cheaper in the short run, for us, since we too can avoid SLRU buffer
6732 : * misses through eager processing.
6733 : *
6734 : * NB: Creates a _new_ MultiXactId when FRM_RETURN_IS_MULTI is set, though only
6735 : * when FreezeLimit and/or MultiXactCutoff cutoffs leave us with no choice.
6736 : * This can usually be put off, which is usually enough to avoid it altogether.
6737 : * Allocating new multis during VACUUM should be avoided on general principle;
6738 : * only VACUUM can advance relminmxid, so allocating new Multis here comes with
6739 : * its own special risks.
6740 : *
6741 : * NB: Caller must maintain "no freeze" NewRelfrozenXid/NewRelminMxid trackers
6742 : * using heap_tuple_should_freeze when we haven't forced page-level freezing.
6743 : *
6744 : * NB: Caller should avoid needlessly calling heap_tuple_should_freeze when we
6745 : * have already forced page-level freezing, since that might incur the same
6746 : * SLRU buffer misses that we specifically intended to avoid by freezing.
6747 : */
6748 : static TransactionId
6749 12 : FreezeMultiXactId(MultiXactId multi, uint16 t_infomask,
6750 : const struct VacuumCutoffs *cutoffs, uint16 *flags,
6751 : HeapPageFreeze *pagefrz)
6752 : {
6753 : TransactionId newxmax;
6754 : MultiXactMember *members;
6755 : int nmembers;
6756 : bool need_replace;
6757 : int nnewmembers;
6758 : MultiXactMember *newmembers;
6759 : bool has_lockers;
6760 : TransactionId update_xid;
6761 : bool update_committed;
6762 : TransactionId FreezePageRelfrozenXid;
6763 :
6764 12 : *flags = 0;
6765 :
6766 : /* We should only be called in Multis */
6767 : Assert(t_infomask & HEAP_XMAX_IS_MULTI);
6768 :
6769 24 : if (!MultiXactIdIsValid(multi) ||
6770 12 : HEAP_LOCKED_UPGRADED(t_infomask))
6771 : {
6772 0 : *flags |= FRM_INVALIDATE_XMAX;
6773 0 : pagefrz->freeze_required = true;
6774 0 : return InvalidTransactionId;
6775 : }
6776 12 : else if (MultiXactIdPrecedes(multi, cutoffs->relminmxid))
6777 0 : ereport(ERROR,
6778 : (errcode(ERRCODE_DATA_CORRUPTED),
6779 : errmsg_internal("found multixact %u from before relminmxid %u",
6780 : multi, cutoffs->relminmxid)));
6781 12 : else if (MultiXactIdPrecedes(multi, cutoffs->OldestMxact))
6782 : {
6783 : TransactionId update_xact;
6784 :
6785 : /*
6786 : * This old multi cannot possibly have members still running, but
6787 : * verify just in case. If it was a locker only, it can be removed
6788 : * without any further consideration; but if it contained an update,
6789 : * we might need to preserve it.
6790 : */
6791 8 : if (MultiXactIdIsRunning(multi,
6792 8 : HEAP_XMAX_IS_LOCKED_ONLY(t_infomask)))
6793 0 : ereport(ERROR,
6794 : (errcode(ERRCODE_DATA_CORRUPTED),
6795 : errmsg_internal("multixact %u from before multi freeze cutoff %u found to be still running",
6796 : multi, cutoffs->OldestMxact)));
6797 :
6798 8 : if (HEAP_XMAX_IS_LOCKED_ONLY(t_infomask))
6799 : {
6800 8 : *flags |= FRM_INVALIDATE_XMAX;
6801 8 : pagefrz->freeze_required = true;
6802 8 : return InvalidTransactionId;
6803 : }
6804 :
6805 : /* replace multi with single XID for its updater? */
6806 0 : update_xact = MultiXactIdGetUpdateXid(multi, t_infomask);
6807 0 : if (TransactionIdPrecedes(update_xact, cutoffs->relfrozenxid))
6808 0 : ereport(ERROR,
6809 : (errcode(ERRCODE_DATA_CORRUPTED),
6810 : errmsg_internal("multixact %u contains update XID %u from before relfrozenxid %u",
6811 : multi, update_xact,
6812 : cutoffs->relfrozenxid)));
6813 0 : else if (TransactionIdPrecedes(update_xact, cutoffs->OldestXmin))
6814 : {
6815 : /*
6816 : * Updater XID has to have aborted (otherwise the tuple would have
6817 : * been pruned away instead, since updater XID is < OldestXmin).
6818 : * Just remove xmax.
6819 : */
6820 0 : if (TransactionIdDidCommit(update_xact))
6821 0 : ereport(ERROR,
6822 : (errcode(ERRCODE_DATA_CORRUPTED),
6823 : errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
6824 : multi, update_xact,
6825 : cutoffs->OldestXmin)));
6826 0 : *flags |= FRM_INVALIDATE_XMAX;
6827 0 : pagefrz->freeze_required = true;
6828 0 : return InvalidTransactionId;
6829 : }
6830 :
6831 : /* Have to keep updater XID as new xmax */
6832 0 : *flags |= FRM_RETURN_IS_XID;
6833 0 : pagefrz->freeze_required = true;
6834 0 : return update_xact;
6835 : }
6836 :
6837 : /*
6838 : * Some member(s) of this Multi may be below FreezeLimit xid cutoff, so we
6839 : * need to walk the whole members array to figure out what to do, if
6840 : * anything.
6841 : */
6842 : nmembers =
6843 4 : GetMultiXactIdMembers(multi, &members, false,
6844 4 : HEAP_XMAX_IS_LOCKED_ONLY(t_infomask));
6845 4 : if (nmembers <= 0)
6846 : {
6847 : /* Nothing worth keeping */
6848 0 : *flags |= FRM_INVALIDATE_XMAX;
6849 0 : pagefrz->freeze_required = true;
6850 0 : return InvalidTransactionId;
6851 : }
6852 :
6853 : /*
6854 : * The FRM_NOOP case is the only case where we might need to ratchet back
6855 : * FreezePageRelfrozenXid or FreezePageRelminMxid. It is also the only
6856 : * case where our caller might ratchet back its NoFreezePageRelfrozenXid
6857 : * or NoFreezePageRelminMxid "no freeze" trackers to deal with a multi.
6858 : * FRM_NOOP handling should result in the NewRelfrozenXid/NewRelminMxid
6859 : * trackers managed by VACUUM being ratcheting back by xmax to the degree
6860 : * required to make it safe to leave xmax undisturbed, independent of
6861 : * whether or not page freezing is triggered somewhere else.
6862 : *
6863 : * Our policy is to force freezing in every case other than FRM_NOOP,
6864 : * which obviates the need to maintain either set of trackers, anywhere.
6865 : * Every other case will reliably execute a freeze plan for xmax that
6866 : * either replaces xmax with an XID/MXID >= OldestXmin/OldestMxact, or
6867 : * sets xmax to an InvalidTransactionId XID, rendering xmax fully frozen.
6868 : * (VACUUM's NewRelfrozenXid/NewRelminMxid trackers are initialized with
6869 : * OldestXmin/OldestMxact, so later values never need to be tracked here.)
6870 : */
6871 4 : need_replace = false;
6872 4 : FreezePageRelfrozenXid = pagefrz->FreezePageRelfrozenXid;
6873 8 : for (int i = 0; i < nmembers; i++)
6874 : {
6875 6 : TransactionId xid = members[i].xid;
6876 :
6877 : Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
6878 :
6879 6 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
6880 : {
6881 : /* Can't violate the FreezeLimit postcondition */
6882 2 : need_replace = true;
6883 2 : break;
6884 : }
6885 4 : if (TransactionIdPrecedes(xid, FreezePageRelfrozenXid))
6886 0 : FreezePageRelfrozenXid = xid;
6887 : }
6888 :
6889 : /* Can't violate the MultiXactCutoff postcondition, either */
6890 4 : if (!need_replace)
6891 2 : need_replace = MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff);
6892 :
6893 4 : if (!need_replace)
6894 : {
6895 : /*
6896 : * vacuumlazy.c might ratchet back NewRelminMxid, NewRelfrozenXid, or
6897 : * both together to make it safe to retain this particular multi after
6898 : * freezing its page
6899 : */
6900 2 : *flags |= FRM_NOOP;
6901 2 : pagefrz->FreezePageRelfrozenXid = FreezePageRelfrozenXid;
6902 2 : if (MultiXactIdPrecedes(multi, pagefrz->FreezePageRelminMxid))
6903 0 : pagefrz->FreezePageRelminMxid = multi;
6904 2 : pfree(members);
6905 2 : return multi;
6906 : }
6907 :
6908 : /*
6909 : * Do a more thorough second pass over the multi to figure out which
6910 : * member XIDs actually need to be kept. Checking the precise status of
6911 : * individual members might even show that we don't need to keep anything.
6912 : * That is quite possible even though the Multi must be >= OldestMxact,
6913 : * since our second pass only keeps member XIDs when it's truly necessary;
6914 : * even member XIDs >= OldestXmin often won't be kept by second pass.
6915 : */
6916 2 : nnewmembers = 0;
6917 2 : newmembers = palloc_array(MultiXactMember, nmembers);
6918 2 : has_lockers = false;
6919 2 : update_xid = InvalidTransactionId;
6920 2 : update_committed = false;
6921 :
6922 : /*
6923 : * Determine whether to keep each member xid, or to ignore it instead
6924 : */
6925 6 : for (int i = 0; i < nmembers; i++)
6926 : {
6927 4 : TransactionId xid = members[i].xid;
6928 4 : MultiXactStatus mstatus = members[i].status;
6929 :
6930 : Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
6931 :
6932 4 : if (!ISUPDATE_from_mxstatus(mstatus))
6933 : {
6934 : /*
6935 : * Locker XID (not updater XID). We only keep lockers that are
6936 : * still running.
6937 : */
6938 8 : if (TransactionIdIsCurrentTransactionId(xid) ||
6939 4 : TransactionIdIsInProgress(xid))
6940 : {
6941 2 : if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
6942 0 : ereport(ERROR,
6943 : (errcode(ERRCODE_DATA_CORRUPTED),
6944 : errmsg_internal("multixact %u contains running locker XID %u from before removable cutoff %u",
6945 : multi, xid,
6946 : cutoffs->OldestXmin)));
6947 2 : newmembers[nnewmembers++] = members[i];
6948 2 : has_lockers = true;
6949 : }
6950 :
6951 4 : continue;
6952 : }
6953 :
6954 : /*
6955 : * Updater XID (not locker XID). Should we keep it?
6956 : *
6957 : * Since the tuple wasn't totally removed when vacuum pruned, the
6958 : * update Xid cannot possibly be older than OldestXmin cutoff unless
6959 : * the updater XID aborted. If the updater transaction is known
6960 : * aborted or crashed then it's okay to ignore it, otherwise not.
6961 : *
6962 : * In any case the Multi should never contain two updaters, whatever
6963 : * their individual commit status. Check for that first, in passing.
6964 : */
6965 0 : if (TransactionIdIsValid(update_xid))
6966 0 : ereport(ERROR,
6967 : (errcode(ERRCODE_DATA_CORRUPTED),
6968 : errmsg_internal("multixact %u has two or more updating members",
6969 : multi),
6970 : errdetail_internal("First updater XID=%u second updater XID=%u.",
6971 : update_xid, xid)));
6972 :
6973 : /*
6974 : * As with all tuple visibility routines, it's critical to test
6975 : * TransactionIdIsInProgress before TransactionIdDidCommit, because of
6976 : * race conditions explained in detail in heapam_visibility.c.
6977 : */
6978 0 : if (TransactionIdIsCurrentTransactionId(xid) ||
6979 0 : TransactionIdIsInProgress(xid))
6980 0 : update_xid = xid;
6981 0 : else if (TransactionIdDidCommit(xid))
6982 : {
6983 : /*
6984 : * The transaction committed, so we can tell caller to set
6985 : * HEAP_XMAX_COMMITTED. (We can only do this because we know the
6986 : * transaction is not running.)
6987 : */
6988 0 : update_committed = true;
6989 0 : update_xid = xid;
6990 : }
6991 : else
6992 : {
6993 : /*
6994 : * Not in progress, not committed -- must be aborted or crashed;
6995 : * we can ignore it.
6996 : */
6997 0 : continue;
6998 : }
6999 :
7000 : /*
7001 : * We determined that updater must be kept -- add it to pending new
7002 : * members list
7003 : */
7004 0 : if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
7005 0 : ereport(ERROR,
7006 : (errcode(ERRCODE_DATA_CORRUPTED),
7007 : errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
7008 : multi, xid, cutoffs->OldestXmin)));
7009 0 : newmembers[nnewmembers++] = members[i];
7010 : }
7011 :
7012 2 : pfree(members);
7013 :
7014 : /*
7015 : * Determine what to do with caller's multi based on information gathered
7016 : * during our second pass
7017 : */
7018 2 : if (nnewmembers == 0)
7019 : {
7020 : /* Nothing worth keeping */
7021 0 : *flags |= FRM_INVALIDATE_XMAX;
7022 0 : newxmax = InvalidTransactionId;
7023 : }
7024 2 : else if (TransactionIdIsValid(update_xid) && !has_lockers)
7025 : {
7026 : /*
7027 : * If there's a single member and it's an update, pass it back alone
7028 : * without creating a new Multi. (XXX we could do this when there's a
7029 : * single remaining locker, too, but that would complicate the API too
7030 : * much; moreover, the case with the single updater is more
7031 : * interesting, because those are longer-lived.)
7032 : */
7033 : Assert(nnewmembers == 1);
7034 0 : *flags |= FRM_RETURN_IS_XID;
7035 0 : if (update_committed)
7036 0 : *flags |= FRM_MARK_COMMITTED;
7037 0 : newxmax = update_xid;
7038 : }
7039 : else
7040 : {
7041 : /*
7042 : * Create a new multixact with the surviving members of the previous
7043 : * one, to set as new Xmax in the tuple
7044 : */
7045 2 : newxmax = MultiXactIdCreateFromMembers(nnewmembers, newmembers);
7046 2 : *flags |= FRM_RETURN_IS_MULTI;
7047 : }
7048 :
7049 2 : pfree(newmembers);
7050 :
7051 2 : pagefrz->freeze_required = true;
7052 2 : return newxmax;
7053 : }
7054 :
7055 : /*
7056 : * heap_prepare_freeze_tuple
7057 : *
7058 : * Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
7059 : * are older than the OldestXmin and/or OldestMxact freeze cutoffs. If so,
7060 : * setup enough state (in the *frz output argument) to enable caller to
7061 : * process this tuple as part of freezing its page, and return true. Return
7062 : * false if nothing can be changed about the tuple right now.
7063 : *
7064 : * Also sets *totally_frozen to true if the tuple will be totally frozen once
7065 : * caller executes returned freeze plan (or if the tuple was already totally
7066 : * frozen by an earlier VACUUM). This indicates that there are no remaining
7067 : * XIDs or MultiXactIds that will need to be processed by a future VACUUM.
7068 : *
7069 : * VACUUM caller must assemble HeapTupleFreeze freeze plan entries for every
7070 : * tuple that we returned true for, and then execute freezing. Caller must
7071 : * initialize pagefrz fields for page as a whole before first call here for
7072 : * each heap page.
7073 : *
7074 : * VACUUM caller decides on whether or not to freeze the page as a whole.
7075 : * We'll often prepare freeze plans for a page that caller just discards.
7076 : * However, VACUUM doesn't always get to make a choice; it must freeze when
7077 : * pagefrz.freeze_required is set, to ensure that any XIDs < FreezeLimit (and
7078 : * MXIDs < MultiXactCutoff) can never be left behind. We help to make sure
7079 : * that VACUUM always follows that rule.
7080 : *
7081 : * We sometimes force freezing of xmax MultiXactId values long before it is
7082 : * strictly necessary to do so just to ensure the FreezeLimit postcondition.
7083 : * It's worth processing MultiXactIds proactively when it is cheap to do so,
7084 : * and it's convenient to make that happen by piggy-backing it on the "force
7085 : * freezing" mechanism. Conversely, we sometimes delay freezing MultiXactIds
7086 : * because it is expensive right now (though only when it's still possible to
7087 : * do so without violating the FreezeLimit/MultiXactCutoff postcondition).
7088 : *
7089 : * It is assumed that the caller has checked the tuple with
7090 : * HeapTupleSatisfiesVacuum() and determined that it is not HEAPTUPLE_DEAD
7091 : * (else we should be removing the tuple, not freezing it).
7092 : *
7093 : * NB: This function has side effects: it might allocate a new MultiXactId.
7094 : * It will be set as tuple's new xmax when our *frz output is processed within
7095 : * heap_execute_freeze_tuple later on. If the tuple is in a shared buffer
7096 : * then caller had better have an exclusive lock on it already.
7097 : */
7098 : bool
7099 29729470 : heap_prepare_freeze_tuple(HeapTupleHeader tuple,
7100 : const struct VacuumCutoffs *cutoffs,
7101 : HeapPageFreeze *pagefrz,
7102 : HeapTupleFreeze *frz, bool *totally_frozen)
7103 : {
7104 29729470 : bool xmin_already_frozen = false,
7105 29729470 : xmax_already_frozen = false;
7106 29729470 : bool freeze_xmin = false,
7107 29729470 : replace_xvac = false,
7108 29729470 : replace_xmax = false,
7109 29729470 : freeze_xmax = false;
7110 : TransactionId xid;
7111 :
7112 29729470 : frz->xmax = HeapTupleHeaderGetRawXmax(tuple);
7113 29729470 : frz->t_infomask2 = tuple->t_infomask2;
7114 29729470 : frz->t_infomask = tuple->t_infomask;
7115 29729470 : frz->frzflags = 0;
7116 29729470 : frz->checkflags = 0;
7117 :
7118 : /*
7119 : * Process xmin, while keeping track of whether it's already frozen, or
7120 : * will become frozen iff our freeze plan is executed by caller (could be
7121 : * neither).
7122 : */
7123 29729470 : xid = HeapTupleHeaderGetXmin(tuple);
7124 29729470 : if (!TransactionIdIsNormal(xid))
7125 23547824 : xmin_already_frozen = true;
7126 : else
7127 : {
7128 6181646 : if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
7129 0 : ereport(ERROR,
7130 : (errcode(ERRCODE_DATA_CORRUPTED),
7131 : errmsg_internal("found xmin %u from before relfrozenxid %u",
7132 : xid, cutoffs->relfrozenxid)));
7133 :
7134 : /* Will set freeze_xmin flags in freeze plan below */
7135 6181646 : freeze_xmin = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
7136 :
7137 : /* Verify that xmin committed if and when freeze plan is executed */
7138 6181646 : if (freeze_xmin)
7139 4972512 : frz->checkflags |= HEAP_FREEZE_CHECK_XMIN_COMMITTED;
7140 : }
7141 :
7142 : /*
7143 : * Old-style VACUUM FULL is gone, but we have to process xvac for as long
7144 : * as we support having MOVED_OFF/MOVED_IN tuples in the database
7145 : */
7146 29729470 : xid = HeapTupleHeaderGetXvac(tuple);
7147 29729470 : if (TransactionIdIsNormal(xid))
7148 : {
7149 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7150 : Assert(TransactionIdPrecedes(xid, cutoffs->OldestXmin));
7151 :
7152 : /*
7153 : * For Xvac, we always freeze proactively. This allows totally_frozen
7154 : * tracking to ignore xvac.
7155 : */
7156 0 : replace_xvac = pagefrz->freeze_required = true;
7157 :
7158 : /* Will set replace_xvac flags in freeze plan below */
7159 : }
7160 :
7161 : /* Now process xmax */
7162 29729470 : xid = frz->xmax;
7163 29729470 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7164 : {
7165 : /* Raw xmax is a MultiXactId */
7166 : TransactionId newxmax;
7167 : uint16 flags;
7168 :
7169 : /*
7170 : * We will either remove xmax completely (in the "freeze_xmax" path),
7171 : * process xmax by replacing it (in the "replace_xmax" path), or
7172 : * perform no-op xmax processing. The only constraint is that the
7173 : * FreezeLimit/MultiXactCutoff postcondition must never be violated.
7174 : */
7175 12 : newxmax = FreezeMultiXactId(xid, tuple->t_infomask, cutoffs,
7176 : &flags, pagefrz);
7177 :
7178 12 : if (flags & FRM_NOOP)
7179 : {
7180 : /*
7181 : * xmax is a MultiXactId, and nothing about it changes for now.
7182 : * This is the only case where 'freeze_required' won't have been
7183 : * set for us by FreezeMultiXactId, as well as the only case where
7184 : * neither freeze_xmax nor replace_xmax are set (given a multi).
7185 : *
7186 : * This is a no-op, but the call to FreezeMultiXactId might have
7187 : * ratcheted back NewRelfrozenXid and/or NewRelminMxid trackers
7188 : * for us (the "freeze page" variants, specifically). That'll
7189 : * make it safe for our caller to freeze the page later on, while
7190 : * leaving this particular xmax undisturbed.
7191 : *
7192 : * FreezeMultiXactId is _not_ responsible for the "no freeze"
7193 : * NewRelfrozenXid/NewRelminMxid trackers, though -- that's our
7194 : * job. A call to heap_tuple_should_freeze for this same tuple
7195 : * will take place below if 'freeze_required' isn't set already.
7196 : * (This repeats work from FreezeMultiXactId, but allows "no
7197 : * freeze" tracker maintenance to happen in only one place.)
7198 : */
7199 : Assert(!MultiXactIdPrecedes(newxmax, cutoffs->MultiXactCutoff));
7200 : Assert(MultiXactIdIsValid(newxmax) && xid == newxmax);
7201 : }
7202 10 : else if (flags & FRM_RETURN_IS_XID)
7203 : {
7204 : /*
7205 : * xmax will become an updater Xid (original MultiXact's updater
7206 : * member Xid will be carried forward as a simple Xid in Xmax).
7207 : */
7208 : Assert(!TransactionIdPrecedes(newxmax, cutoffs->OldestXmin));
7209 :
7210 : /*
7211 : * NB -- some of these transformations are only valid because we
7212 : * know the return Xid is a tuple updater (i.e. not merely a
7213 : * locker.) Also note that the only reason we don't explicitly
7214 : * worry about HEAP_KEYS_UPDATED is because it lives in
7215 : * t_infomask2 rather than t_infomask.
7216 : */
7217 0 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7218 0 : frz->xmax = newxmax;
7219 0 : if (flags & FRM_MARK_COMMITTED)
7220 0 : frz->t_infomask |= HEAP_XMAX_COMMITTED;
7221 0 : replace_xmax = true;
7222 : }
7223 10 : else if (flags & FRM_RETURN_IS_MULTI)
7224 : {
7225 : uint16 newbits;
7226 : uint16 newbits2;
7227 :
7228 : /*
7229 : * xmax is an old MultiXactId that we have to replace with a new
7230 : * MultiXactId, to carry forward two or more original member XIDs.
7231 : */
7232 : Assert(!MultiXactIdPrecedes(newxmax, cutoffs->OldestMxact));
7233 :
7234 : /*
7235 : * We can't use GetMultiXactIdHintBits directly on the new multi
7236 : * here; that routine initializes the masks to all zeroes, which
7237 : * would lose other bits we need. Doing it this way ensures all
7238 : * unrelated bits remain untouched.
7239 : */
7240 2 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7241 2 : frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
7242 2 : GetMultiXactIdHintBits(newxmax, &newbits, &newbits2);
7243 2 : frz->t_infomask |= newbits;
7244 2 : frz->t_infomask2 |= newbits2;
7245 2 : frz->xmax = newxmax;
7246 2 : replace_xmax = true;
7247 : }
7248 : else
7249 : {
7250 : /*
7251 : * Freeze plan for tuple "freezes xmax" in the strictest sense:
7252 : * it'll leave nothing in xmax (neither an Xid nor a MultiXactId).
7253 : */
7254 : Assert(flags & FRM_INVALIDATE_XMAX);
7255 : Assert(!TransactionIdIsValid(newxmax));
7256 :
7257 : /* Will set freeze_xmax flags in freeze plan below */
7258 8 : freeze_xmax = true;
7259 : }
7260 :
7261 : /* MultiXactId processing forces freezing (barring FRM_NOOP case) */
7262 : Assert(pagefrz->freeze_required || (!freeze_xmax && !replace_xmax));
7263 : }
7264 29729458 : else if (TransactionIdIsNormal(xid))
7265 : {
7266 : /* Raw xmax is normal XID */
7267 6944476 : if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
7268 0 : ereport(ERROR,
7269 : (errcode(ERRCODE_DATA_CORRUPTED),
7270 : errmsg_internal("found xmax %u from before relfrozenxid %u",
7271 : xid, cutoffs->relfrozenxid)));
7272 :
7273 : /* Will set freeze_xmax flags in freeze plan below */
7274 6944476 : freeze_xmax = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
7275 :
7276 : /*
7277 : * Verify that xmax aborted if and when freeze plan is executed,
7278 : * provided it's from an update. (A lock-only xmax can be removed
7279 : * independent of this, since the lock is released at xact end.)
7280 : */
7281 6944476 : if (freeze_xmax && !HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask))
7282 2472 : frz->checkflags |= HEAP_FREEZE_CHECK_XMAX_ABORTED;
7283 : }
7284 22784982 : else if (!TransactionIdIsValid(xid))
7285 : {
7286 : /* Raw xmax is InvalidTransactionId XID */
7287 : Assert((tuple->t_infomask & HEAP_XMAX_IS_MULTI) == 0);
7288 22784982 : xmax_already_frozen = true;
7289 : }
7290 : else
7291 0 : ereport(ERROR,
7292 : (errcode(ERRCODE_DATA_CORRUPTED),
7293 : errmsg_internal("found raw xmax %u (infomask 0x%04x) not invalid and not multi",
7294 : xid, tuple->t_infomask)));
7295 :
7296 29729470 : if (freeze_xmin)
7297 : {
7298 : Assert(!xmin_already_frozen);
7299 :
7300 4972512 : frz->t_infomask |= HEAP_XMIN_FROZEN;
7301 : }
7302 29729470 : if (replace_xvac)
7303 : {
7304 : /*
7305 : * If a MOVED_OFF tuple is not dead, the xvac transaction must have
7306 : * failed; whereas a non-dead MOVED_IN tuple must mean the xvac
7307 : * transaction succeeded.
7308 : */
7309 : Assert(pagefrz->freeze_required);
7310 0 : if (tuple->t_infomask & HEAP_MOVED_OFF)
7311 0 : frz->frzflags |= XLH_INVALID_XVAC;
7312 : else
7313 0 : frz->frzflags |= XLH_FREEZE_XVAC;
7314 : }
7315 : if (replace_xmax)
7316 : {
7317 : Assert(!xmax_already_frozen && !freeze_xmax);
7318 : Assert(pagefrz->freeze_required);
7319 :
7320 : /* Already set replace_xmax flags in freeze plan earlier */
7321 : }
7322 29729470 : if (freeze_xmax)
7323 : {
7324 : Assert(!xmax_already_frozen && !replace_xmax);
7325 :
7326 4432 : frz->xmax = InvalidTransactionId;
7327 :
7328 : /*
7329 : * The tuple might be marked either XMAX_INVALID or XMAX_COMMITTED +
7330 : * LOCKED. Normalize to INVALID just to be sure no one gets confused.
7331 : * Also get rid of the HEAP_KEYS_UPDATED bit.
7332 : */
7333 4432 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7334 4432 : frz->t_infomask |= HEAP_XMAX_INVALID;
7335 4432 : frz->t_infomask2 &= ~HEAP_HOT_UPDATED;
7336 4432 : frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
7337 : }
7338 :
7339 : /*
7340 : * Determine if this tuple is already totally frozen, or will become
7341 : * totally frozen (provided caller executes freeze plans for the page)
7342 : */
7343 58245374 : *totally_frozen = ((freeze_xmin || xmin_already_frozen) &&
7344 28515904 : (freeze_xmax || xmax_already_frozen));
7345 :
7346 29729470 : if (!pagefrz->freeze_required && !(xmin_already_frozen &&
7347 : xmax_already_frozen))
7348 : {
7349 : /*
7350 : * So far no previous tuple from the page made freezing mandatory.
7351 : * Does this tuple force caller to freeze the entire page?
7352 : */
7353 10080468 : pagefrz->freeze_required =
7354 10080468 : heap_tuple_should_freeze(tuple, cutoffs,
7355 : &pagefrz->NoFreezePageRelfrozenXid,
7356 : &pagefrz->NoFreezePageRelminMxid);
7357 : }
7358 :
7359 : /* Tell caller if this tuple has a usable freeze plan set in *frz */
7360 29729470 : return freeze_xmin || replace_xvac || replace_xmax || freeze_xmax;
7361 : }
7362 :
7363 : /*
7364 : * Perform xmin/xmax XID status sanity checks before actually executing freeze
7365 : * plans.
7366 : *
7367 : * heap_prepare_freeze_tuple doesn't perform these checks directly because
7368 : * pg_xact lookups are relatively expensive. They shouldn't be repeated by
7369 : * successive VACUUMs that each decide against freezing the same page.
7370 : */
7371 : void
7372 46472 : heap_pre_freeze_checks(Buffer buffer,
7373 : HeapTupleFreeze *tuples, int ntuples)
7374 : {
7375 46472 : Page page = BufferGetPage(buffer);
7376 :
7377 2010212 : for (int i = 0; i < ntuples; i++)
7378 : {
7379 1963740 : HeapTupleFreeze *frz = tuples + i;
7380 1963740 : ItemId itemid = PageGetItemId(page, frz->offset);
7381 : HeapTupleHeader htup;
7382 :
7383 1963740 : htup = (HeapTupleHeader) PageGetItem(page, itemid);
7384 :
7385 : /* Deliberately avoid relying on tuple hint bits here */
7386 1963740 : if (frz->checkflags & HEAP_FREEZE_CHECK_XMIN_COMMITTED)
7387 : {
7388 1963738 : TransactionId xmin = HeapTupleHeaderGetRawXmin(htup);
7389 :
7390 : Assert(!HeapTupleHeaderXminFrozen(htup));
7391 1963738 : if (unlikely(!TransactionIdDidCommit(xmin)))
7392 0 : ereport(ERROR,
7393 : (errcode(ERRCODE_DATA_CORRUPTED),
7394 : errmsg_internal("uncommitted xmin %u needs to be frozen",
7395 : xmin)));
7396 : }
7397 :
7398 : /*
7399 : * TransactionIdDidAbort won't work reliably in the presence of XIDs
7400 : * left behind by transactions that were in progress during a crash,
7401 : * so we can only check that xmax didn't commit
7402 : */
7403 1963740 : if (frz->checkflags & HEAP_FREEZE_CHECK_XMAX_ABORTED)
7404 : {
7405 954 : TransactionId xmax = HeapTupleHeaderGetRawXmax(htup);
7406 :
7407 : Assert(TransactionIdIsNormal(xmax));
7408 954 : if (unlikely(TransactionIdDidCommit(xmax)))
7409 0 : ereport(ERROR,
7410 : (errcode(ERRCODE_DATA_CORRUPTED),
7411 : errmsg_internal("cannot freeze committed xmax %u",
7412 : xmax)));
7413 : }
7414 : }
7415 46472 : }
7416 :
7417 : /*
7418 : * Helper which executes freezing of one or more heap tuples on a page on
7419 : * behalf of caller. Caller passes an array of tuple plans from
7420 : * heap_prepare_freeze_tuple. Caller must set 'offset' in each plan for us.
7421 : * Must be called in a critical section that also marks the buffer dirty and,
7422 : * if needed, emits WAL.
7423 : */
7424 : void
7425 46472 : heap_freeze_prepared_tuples(Buffer buffer, HeapTupleFreeze *tuples, int ntuples)
7426 : {
7427 46472 : Page page = BufferGetPage(buffer);
7428 :
7429 2010212 : for (int i = 0; i < ntuples; i++)
7430 : {
7431 1963740 : HeapTupleFreeze *frz = tuples + i;
7432 1963740 : ItemId itemid = PageGetItemId(page, frz->offset);
7433 : HeapTupleHeader htup;
7434 :
7435 1963740 : htup = (HeapTupleHeader) PageGetItem(page, itemid);
7436 1963740 : heap_execute_freeze_tuple(htup, frz);
7437 : }
7438 46472 : }
7439 :
7440 : /*
7441 : * heap_freeze_tuple
7442 : * Freeze tuple in place, without WAL logging.
7443 : *
7444 : * Useful for callers like CLUSTER that perform their own WAL logging.
7445 : */
7446 : bool
7447 732758 : heap_freeze_tuple(HeapTupleHeader tuple,
7448 : TransactionId relfrozenxid, TransactionId relminmxid,
7449 : TransactionId FreezeLimit, TransactionId MultiXactCutoff)
7450 : {
7451 : HeapTupleFreeze frz;
7452 : bool do_freeze;
7453 : bool totally_frozen;
7454 : struct VacuumCutoffs cutoffs;
7455 : HeapPageFreeze pagefrz;
7456 :
7457 732758 : cutoffs.relfrozenxid = relfrozenxid;
7458 732758 : cutoffs.relminmxid = relminmxid;
7459 732758 : cutoffs.OldestXmin = FreezeLimit;
7460 732758 : cutoffs.OldestMxact = MultiXactCutoff;
7461 732758 : cutoffs.FreezeLimit = FreezeLimit;
7462 732758 : cutoffs.MultiXactCutoff = MultiXactCutoff;
7463 :
7464 732758 : pagefrz.freeze_required = true;
7465 732758 : pagefrz.FreezePageRelfrozenXid = FreezeLimit;
7466 732758 : pagefrz.FreezePageRelminMxid = MultiXactCutoff;
7467 732758 : pagefrz.NoFreezePageRelfrozenXid = FreezeLimit;
7468 732758 : pagefrz.NoFreezePageRelminMxid = MultiXactCutoff;
7469 :
7470 732758 : do_freeze = heap_prepare_freeze_tuple(tuple, &cutoffs,
7471 : &pagefrz, &frz, &totally_frozen);
7472 :
7473 : /*
7474 : * Note that because this is not a WAL-logged operation, we don't need to
7475 : * fill in the offset in the freeze record.
7476 : */
7477 :
7478 732758 : if (do_freeze)
7479 530004 : heap_execute_freeze_tuple(tuple, &frz);
7480 732758 : return do_freeze;
7481 : }
7482 :
7483 : /*
7484 : * For a given MultiXactId, return the hint bits that should be set in the
7485 : * tuple's infomask.
7486 : *
7487 : * Normally this should be called for a multixact that was just created, and
7488 : * so is on our local cache, so the GetMembers call is fast.
7489 : */
7490 : static void
7491 153588 : GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
7492 : uint16 *new_infomask2)
7493 : {
7494 : int nmembers;
7495 : MultiXactMember *members;
7496 : int i;
7497 153588 : uint16 bits = HEAP_XMAX_IS_MULTI;
7498 153588 : uint16 bits2 = 0;
7499 153588 : bool has_update = false;
7500 153588 : LockTupleMode strongest = LockTupleKeyShare;
7501 :
7502 : /*
7503 : * We only use this in multis we just created, so they cannot be values
7504 : * pre-pg_upgrade.
7505 : */
7506 153588 : nmembers = GetMultiXactIdMembers(multi, &members, false, false);
7507 :
7508 2945190 : for (i = 0; i < nmembers; i++)
7509 : {
7510 : LockTupleMode mode;
7511 :
7512 : /*
7513 : * Remember the strongest lock mode held by any member of the
7514 : * multixact.
7515 : */
7516 2791602 : mode = TUPLOCK_from_mxstatus(members[i].status);
7517 2791602 : if (mode > strongest)
7518 5776 : strongest = mode;
7519 :
7520 : /* See what other bits we need */
7521 2791602 : switch (members[i].status)
7522 : {
7523 2786766 : case MultiXactStatusForKeyShare:
7524 : case MultiXactStatusForShare:
7525 : case MultiXactStatusForNoKeyUpdate:
7526 2786766 : break;
7527 :
7528 104 : case MultiXactStatusForUpdate:
7529 104 : bits2 |= HEAP_KEYS_UPDATED;
7530 104 : break;
7531 :
7532 4712 : case MultiXactStatusNoKeyUpdate:
7533 4712 : has_update = true;
7534 4712 : break;
7535 :
7536 20 : case MultiXactStatusUpdate:
7537 20 : bits2 |= HEAP_KEYS_UPDATED;
7538 20 : has_update = true;
7539 20 : break;
7540 : }
7541 : }
7542 :
7543 153588 : if (strongest == LockTupleExclusive ||
7544 : strongest == LockTupleNoKeyExclusive)
7545 4892 : bits |= HEAP_XMAX_EXCL_LOCK;
7546 148696 : else if (strongest == LockTupleShare)
7547 878 : bits |= HEAP_XMAX_SHR_LOCK;
7548 147818 : else if (strongest == LockTupleKeyShare)
7549 147818 : bits |= HEAP_XMAX_KEYSHR_LOCK;
7550 :
7551 153588 : if (!has_update)
7552 148856 : bits |= HEAP_XMAX_LOCK_ONLY;
7553 :
7554 153588 : if (nmembers > 0)
7555 153588 : pfree(members);
7556 :
7557 153588 : *new_infomask = bits;
7558 153588 : *new_infomask2 = bits2;
7559 153588 : }
7560 :
7561 : /*
7562 : * MultiXactIdGetUpdateXid
7563 : *
7564 : * Given a multixact Xmax and corresponding infomask, which does not have the
7565 : * HEAP_XMAX_LOCK_ONLY bit set, obtain and return the Xid of the updating
7566 : * transaction.
7567 : *
7568 : * Caller is expected to check the status of the updating transaction, if
7569 : * necessary.
7570 : */
7571 : static TransactionId
7572 324036 : MultiXactIdGetUpdateXid(TransactionId xmax, uint16 t_infomask)
7573 : {
7574 324036 : TransactionId update_xact = InvalidTransactionId;
7575 : MultiXactMember *members;
7576 : int nmembers;
7577 :
7578 : Assert(!(t_infomask & HEAP_XMAX_LOCK_ONLY));
7579 : Assert(t_infomask & HEAP_XMAX_IS_MULTI);
7580 :
7581 : /*
7582 : * Since we know the LOCK_ONLY bit is not set, this cannot be a multi from
7583 : * pre-pg_upgrade.
7584 : */
7585 324036 : nmembers = GetMultiXactIdMembers(xmax, &members, false, false);
7586 :
7587 324036 : if (nmembers > 0)
7588 : {
7589 : int i;
7590 :
7591 491556 : for (i = 0; i < nmembers; i++)
7592 : {
7593 : /* Ignore lockers */
7594 491556 : if (!ISUPDATE_from_mxstatus(members[i].status))
7595 167520 : continue;
7596 :
7597 : /* there can be at most one updater */
7598 : Assert(update_xact == InvalidTransactionId);
7599 324036 : update_xact = members[i].xid;
7600 : #ifndef USE_ASSERT_CHECKING
7601 :
7602 : /*
7603 : * in an assert-enabled build, walk the whole array to ensure
7604 : * there's no other updater.
7605 : */
7606 324036 : break;
7607 : #endif
7608 : }
7609 :
7610 324036 : pfree(members);
7611 : }
7612 :
7613 324036 : return update_xact;
7614 : }
7615 :
7616 : /*
7617 : * HeapTupleGetUpdateXid
7618 : * As above, but use a HeapTupleHeader
7619 : *
7620 : * See also HeapTupleHeaderGetUpdateXid, which can be used without previously
7621 : * checking the hint bits.
7622 : */
7623 : TransactionId
7624 319760 : HeapTupleGetUpdateXid(const HeapTupleHeaderData *tup)
7625 : {
7626 319760 : return MultiXactIdGetUpdateXid(HeapTupleHeaderGetRawXmax(tup),
7627 319760 : tup->t_infomask);
7628 : }
7629 :
7630 : /*
7631 : * Does the given multixact conflict with the current transaction grabbing a
7632 : * tuple lock of the given strength?
7633 : *
7634 : * The passed infomask pairs up with the given multixact in the tuple header.
7635 : *
7636 : * If current_is_member is not NULL, it is set to 'true' if the current
7637 : * transaction is a member of the given multixact.
7638 : */
7639 : static bool
7640 436 : DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
7641 : LockTupleMode lockmode, bool *current_is_member)
7642 : {
7643 : int nmembers;
7644 : MultiXactMember *members;
7645 436 : bool result = false;
7646 436 : LOCKMODE wanted = tupleLockExtraInfo[lockmode].hwlock;
7647 :
7648 436 : if (HEAP_LOCKED_UPGRADED(infomask))
7649 0 : return false;
7650 :
7651 436 : nmembers = GetMultiXactIdMembers(multi, &members, false,
7652 436 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
7653 436 : if (nmembers >= 0)
7654 : {
7655 : int i;
7656 :
7657 5364 : for (i = 0; i < nmembers; i++)
7658 : {
7659 : TransactionId memxid;
7660 : LOCKMODE memlockmode;
7661 :
7662 4942 : if (result && (current_is_member == NULL || *current_is_member))
7663 : break;
7664 :
7665 4928 : memlockmode = LOCKMODE_from_mxstatus(members[i].status);
7666 :
7667 : /* ignore members from current xact (but track their presence) */
7668 4928 : memxid = members[i].xid;
7669 4928 : if (TransactionIdIsCurrentTransactionId(memxid))
7670 : {
7671 184 : if (current_is_member != NULL)
7672 156 : *current_is_member = true;
7673 184 : continue;
7674 : }
7675 4744 : else if (result)
7676 16 : continue;
7677 :
7678 : /* ignore members that don't conflict with the lock we want */
7679 4728 : if (!DoLockModesConflict(memlockmode, wanted))
7680 4650 : continue;
7681 :
7682 78 : if (ISUPDATE_from_mxstatus(members[i].status))
7683 : {
7684 : /* ignore aborted updaters */
7685 34 : if (TransactionIdDidAbort(memxid))
7686 2 : continue;
7687 : }
7688 : else
7689 : {
7690 : /* ignore lockers-only that are no longer in progress */
7691 44 : if (!TransactionIdIsInProgress(memxid))
7692 14 : continue;
7693 : }
7694 :
7695 : /*
7696 : * Whatever remains are either live lockers that conflict with our
7697 : * wanted lock, and updaters that are not aborted. Those conflict
7698 : * with what we want. Set up to return true, but keep going to
7699 : * look for the current transaction among the multixact members,
7700 : * if needed.
7701 : */
7702 62 : result = true;
7703 : }
7704 436 : pfree(members);
7705 : }
7706 :
7707 436 : return result;
7708 : }
7709 :
7710 : /*
7711 : * Do_MultiXactIdWait
7712 : * Actual implementation for the two functions below.
7713 : *
7714 : * 'multi', 'status' and 'infomask' indicate what to sleep on (the status is
7715 : * needed to ensure we only sleep on conflicting members, and the infomask is
7716 : * used to optimize multixact access in case it's a lock-only multi); 'nowait'
7717 : * indicates whether to use conditional lock acquisition, to allow callers to
7718 : * fail if lock is unavailable. 'rel', 'ctid' and 'oper' are used to set up
7719 : * context information for error messages. 'remaining', if not NULL, receives
7720 : * the number of members that are still running, including any (non-aborted)
7721 : * subtransactions of our own transaction. 'logLockFailure' indicates whether
7722 : * to log details when a lock acquisition fails with 'nowait' enabled.
7723 : *
7724 : * We do this by sleeping on each member using XactLockTableWait. Any
7725 : * members that belong to the current backend are *not* waited for, however;
7726 : * this would not merely be useless but would lead to Assert failure inside
7727 : * XactLockTableWait. By the time this returns, it is certain that all
7728 : * transactions *of other backends* that were members of the MultiXactId
7729 : * that conflict with the requested status are dead (and no new ones can have
7730 : * been added, since it is not legal to add members to an existing
7731 : * MultiXactId).
7732 : *
7733 : * But by the time we finish sleeping, someone else may have changed the Xmax
7734 : * of the containing tuple, so the caller needs to iterate on us somehow.
7735 : *
7736 : * Note that in case we return false, the number of remaining members is
7737 : * not to be trusted.
7738 : */
7739 : static bool
7740 116 : Do_MultiXactIdWait(MultiXactId multi, MultiXactStatus status,
7741 : uint16 infomask, bool nowait,
7742 : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
7743 : int *remaining, bool logLockFailure)
7744 : {
7745 116 : bool result = true;
7746 : MultiXactMember *members;
7747 : int nmembers;
7748 116 : int remain = 0;
7749 :
7750 : /* for pre-pg_upgrade tuples, no need to sleep at all */
7751 116 : nmembers = HEAP_LOCKED_UPGRADED(infomask) ? -1 :
7752 116 : GetMultiXactIdMembers(multi, &members, false,
7753 116 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
7754 :
7755 116 : if (nmembers >= 0)
7756 : {
7757 : int i;
7758 :
7759 374 : for (i = 0; i < nmembers; i++)
7760 : {
7761 266 : TransactionId memxid = members[i].xid;
7762 266 : MultiXactStatus memstatus = members[i].status;
7763 :
7764 266 : if (TransactionIdIsCurrentTransactionId(memxid))
7765 : {
7766 48 : remain++;
7767 48 : continue;
7768 : }
7769 :
7770 218 : if (!DoLockModesConflict(LOCKMODE_from_mxstatus(memstatus),
7771 218 : LOCKMODE_from_mxstatus(status)))
7772 : {
7773 44 : if (remaining && TransactionIdIsInProgress(memxid))
7774 16 : remain++;
7775 44 : continue;
7776 : }
7777 :
7778 : /*
7779 : * This member conflicts with our multi, so we have to sleep (or
7780 : * return failure, if asked to avoid waiting.)
7781 : *
7782 : * Note that we don't set up an error context callback ourselves,
7783 : * but instead we pass the info down to XactLockTableWait. This
7784 : * might seem a bit wasteful because the context is set up and
7785 : * tore down for each member of the multixact, but in reality it
7786 : * should be barely noticeable, and it avoids duplicate code.
7787 : */
7788 174 : if (nowait)
7789 : {
7790 8 : result = ConditionalXactLockTableWait(memxid, logLockFailure);
7791 8 : if (!result)
7792 8 : break;
7793 : }
7794 : else
7795 166 : XactLockTableWait(memxid, rel, ctid, oper);
7796 : }
7797 :
7798 116 : pfree(members);
7799 : }
7800 :
7801 116 : if (remaining)
7802 20 : *remaining = remain;
7803 :
7804 116 : return result;
7805 : }
7806 :
7807 : /*
7808 : * MultiXactIdWait
7809 : * Sleep on a MultiXactId.
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 : * We return (in *remaining, if not NULL) the number of members that are still
7815 : * running, including any (non-aborted) subtransactions of our own transaction.
7816 : */
7817 : static void
7818 108 : MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
7819 : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
7820 : int *remaining)
7821 : {
7822 108 : (void) Do_MultiXactIdWait(multi, status, infomask, false,
7823 : rel, ctid, oper, remaining, false);
7824 108 : }
7825 :
7826 : /*
7827 : * ConditionalMultiXactIdWait
7828 : * As above, but only lock if we can get the lock without blocking.
7829 : *
7830 : * By the time we finish sleeping, someone else may have changed the Xmax
7831 : * of the containing tuple, so the caller needs to iterate on us somehow.
7832 : *
7833 : * If the multixact is now all gone, return true. Returns false if some
7834 : * transactions might still be running.
7835 : *
7836 : * We return (in *remaining, if not NULL) the number of members that are still
7837 : * running, including any (non-aborted) subtransactions of our own transaction.
7838 : */
7839 : static bool
7840 8 : ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status,
7841 : uint16 infomask, Relation rel, int *remaining,
7842 : bool logLockFailure)
7843 : {
7844 8 : return Do_MultiXactIdWait(multi, status, infomask, true,
7845 : rel, NULL, XLTW_None, remaining, logLockFailure);
7846 : }
7847 :
7848 : /*
7849 : * heap_tuple_needs_eventual_freeze
7850 : *
7851 : * Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
7852 : * will eventually require freezing (if tuple isn't removed by pruning first).
7853 : */
7854 : bool
7855 309550 : heap_tuple_needs_eventual_freeze(HeapTupleHeader tuple)
7856 : {
7857 : TransactionId xid;
7858 :
7859 : /*
7860 : * If xmin is a normal transaction ID, this tuple is definitely not
7861 : * frozen.
7862 : */
7863 309550 : xid = HeapTupleHeaderGetXmin(tuple);
7864 309550 : if (TransactionIdIsNormal(xid))
7865 5722 : return true;
7866 :
7867 : /*
7868 : * If xmax is a valid xact or multixact, this tuple is also not frozen.
7869 : */
7870 303828 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7871 : {
7872 : MultiXactId multi;
7873 :
7874 0 : multi = HeapTupleHeaderGetRawXmax(tuple);
7875 0 : if (MultiXactIdIsValid(multi))
7876 0 : return true;
7877 : }
7878 : else
7879 : {
7880 303828 : xid = HeapTupleHeaderGetRawXmax(tuple);
7881 303828 : if (TransactionIdIsNormal(xid))
7882 16 : return true;
7883 : }
7884 :
7885 303812 : if (tuple->t_infomask & HEAP_MOVED)
7886 : {
7887 0 : xid = HeapTupleHeaderGetXvac(tuple);
7888 0 : if (TransactionIdIsNormal(xid))
7889 0 : return true;
7890 : }
7891 :
7892 303812 : return false;
7893 : }
7894 :
7895 : /*
7896 : * heap_tuple_should_freeze
7897 : *
7898 : * Return value indicates if heap_prepare_freeze_tuple sibling function would
7899 : * (or should) force freezing of the heap page that contains caller's tuple.
7900 : * Tuple header XIDs/MXIDs < FreezeLimit/MultiXactCutoff trigger freezing.
7901 : * This includes (xmin, xmax, xvac) fields, as well as MultiXact member XIDs.
7902 : *
7903 : * The *NoFreezePageRelfrozenXid and *NoFreezePageRelminMxid input/output
7904 : * arguments help VACUUM track the oldest extant XID/MXID remaining in rel.
7905 : * Our working assumption is that caller won't decide to freeze this tuple.
7906 : * It's up to caller to only ratchet back its own top-level trackers after the
7907 : * point that it fully commits to not freezing the tuple/page in question.
7908 : */
7909 : bool
7910 10084062 : heap_tuple_should_freeze(HeapTupleHeader tuple,
7911 : const struct VacuumCutoffs *cutoffs,
7912 : TransactionId *NoFreezePageRelfrozenXid,
7913 : MultiXactId *NoFreezePageRelminMxid)
7914 : {
7915 : TransactionId xid;
7916 : MultiXactId multi;
7917 10084062 : bool freeze = false;
7918 :
7919 : /* First deal with xmin */
7920 10084062 : xid = HeapTupleHeaderGetXmin(tuple);
7921 10084062 : if (TransactionIdIsNormal(xid))
7922 : {
7923 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7924 3796396 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7925 46736 : *NoFreezePageRelfrozenXid = xid;
7926 3796396 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7927 42652 : freeze = true;
7928 : }
7929 :
7930 : /* Now deal with xmax */
7931 10084062 : xid = InvalidTransactionId;
7932 10084062 : multi = InvalidMultiXactId;
7933 10084062 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7934 4 : multi = HeapTupleHeaderGetRawXmax(tuple);
7935 : else
7936 10084058 : xid = HeapTupleHeaderGetRawXmax(tuple);
7937 :
7938 10084062 : if (TransactionIdIsNormal(xid))
7939 : {
7940 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7941 : /* xmax is a non-permanent XID */
7942 6796358 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7943 10 : *NoFreezePageRelfrozenXid = xid;
7944 6796358 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7945 52 : freeze = true;
7946 : }
7947 3287704 : else if (!MultiXactIdIsValid(multi))
7948 : {
7949 : /* xmax is a permanent XID or invalid MultiXactId/XID */
7950 : }
7951 4 : else if (HEAP_LOCKED_UPGRADED(tuple->t_infomask))
7952 : {
7953 : /* xmax is a pg_upgrade'd MultiXact, which can't have updater XID */
7954 0 : if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
7955 0 : *NoFreezePageRelminMxid = multi;
7956 : /* heap_prepare_freeze_tuple always freezes pg_upgrade'd xmax */
7957 0 : freeze = true;
7958 : }
7959 : else
7960 : {
7961 : /* xmax is a MultiXactId that may have an updater XID */
7962 : MultiXactMember *members;
7963 : int nmembers;
7964 :
7965 : Assert(MultiXactIdPrecedesOrEquals(cutoffs->relminmxid, multi));
7966 4 : if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
7967 4 : *NoFreezePageRelminMxid = multi;
7968 4 : if (MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff))
7969 4 : freeze = true;
7970 :
7971 : /* need to check whether any member of the mxact is old */
7972 4 : nmembers = GetMultiXactIdMembers(multi, &members, false,
7973 4 : HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask));
7974 :
7975 10 : for (int i = 0; i < nmembers; i++)
7976 : {
7977 6 : xid = members[i].xid;
7978 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7979 6 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7980 0 : *NoFreezePageRelfrozenXid = xid;
7981 6 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7982 0 : freeze = true;
7983 : }
7984 4 : if (nmembers > 0)
7985 2 : pfree(members);
7986 : }
7987 :
7988 10084062 : if (tuple->t_infomask & HEAP_MOVED)
7989 : {
7990 0 : xid = HeapTupleHeaderGetXvac(tuple);
7991 0 : if (TransactionIdIsNormal(xid))
7992 : {
7993 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7994 0 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7995 0 : *NoFreezePageRelfrozenXid = xid;
7996 : /* heap_prepare_freeze_tuple forces xvac freezing */
7997 0 : freeze = true;
7998 : }
7999 : }
8000 :
8001 10084062 : return freeze;
8002 : }
8003 :
8004 : /*
8005 : * Maintain snapshotConflictHorizon for caller by ratcheting forward its value
8006 : * using any committed XIDs contained in 'tuple', an obsolescent heap tuple
8007 : * that caller is in the process of physically removing, e.g. via HOT pruning
8008 : * or index deletion.
8009 : *
8010 : * Caller must initialize its value to InvalidTransactionId, which is
8011 : * generally interpreted as "definitely no need for a recovery conflict".
8012 : * Final value must reflect all heap tuples that caller will physically remove
8013 : * (or remove TID references to) via its ongoing pruning/deletion operation.
8014 : * ResolveRecoveryConflictWithSnapshot() is passed the final value (taken from
8015 : * caller's WAL record) by REDO routine when it replays caller's operation.
8016 : */
8017 : void
8018 3126056 : HeapTupleHeaderAdvanceConflictHorizon(HeapTupleHeader tuple,
8019 : TransactionId *snapshotConflictHorizon)
8020 : {
8021 3126056 : TransactionId xmin = HeapTupleHeaderGetXmin(tuple);
8022 3126056 : TransactionId xmax = HeapTupleHeaderGetUpdateXid(tuple);
8023 3126056 : TransactionId xvac = HeapTupleHeaderGetXvac(tuple);
8024 :
8025 3126056 : if (tuple->t_infomask & HEAP_MOVED)
8026 : {
8027 0 : if (TransactionIdPrecedes(*snapshotConflictHorizon, xvac))
8028 0 : *snapshotConflictHorizon = xvac;
8029 : }
8030 :
8031 : /*
8032 : * Ignore tuples inserted by an aborted transaction or if the tuple was
8033 : * updated/deleted by the inserting transaction.
8034 : *
8035 : * Look for a committed hint bit, or if no xmin bit is set, check clog.
8036 : */
8037 3126056 : if (HeapTupleHeaderXminCommitted(tuple) ||
8038 214284 : (!HeapTupleHeaderXminInvalid(tuple) && TransactionIdDidCommit(xmin)))
8039 : {
8040 5618066 : if (xmax != xmin &&
8041 2649536 : TransactionIdFollows(xmax, *snapshotConflictHorizon))
8042 203000 : *snapshotConflictHorizon = xmax;
8043 : }
8044 3126056 : }
8045 :
8046 : #ifdef USE_PREFETCH
8047 : /*
8048 : * Helper function for heap_index_delete_tuples. Issues prefetch requests for
8049 : * prefetch_count buffers. The prefetch_state keeps track of all the buffers
8050 : * we can prefetch, and which have already been prefetched; each call to this
8051 : * function picks up where the previous call left off.
8052 : *
8053 : * Note: we expect the deltids array to be sorted in an order that groups TIDs
8054 : * by heap block, with all TIDs for each block appearing together in exactly
8055 : * one group.
8056 : */
8057 : static void
8058 38888 : index_delete_prefetch_buffer(Relation rel,
8059 : IndexDeletePrefetchState *prefetch_state,
8060 : int prefetch_count)
8061 : {
8062 38888 : BlockNumber cur_hblkno = prefetch_state->cur_hblkno;
8063 38888 : int count = 0;
8064 : int i;
8065 38888 : int ndeltids = prefetch_state->ndeltids;
8066 38888 : TM_IndexDelete *deltids = prefetch_state->deltids;
8067 :
8068 38888 : for (i = prefetch_state->next_item;
8069 1347604 : i < ndeltids && count < prefetch_count;
8070 1308716 : i++)
8071 : {
8072 1308716 : ItemPointer htid = &deltids[i].tid;
8073 :
8074 2605922 : if (cur_hblkno == InvalidBlockNumber ||
8075 1297206 : ItemPointerGetBlockNumber(htid) != cur_hblkno)
8076 : {
8077 35216 : cur_hblkno = ItemPointerGetBlockNumber(htid);
8078 35216 : PrefetchBuffer(rel, MAIN_FORKNUM, cur_hblkno);
8079 35216 : count++;
8080 : }
8081 : }
8082 :
8083 : /*
8084 : * Save the prefetch position so that next time we can continue from that
8085 : * position.
8086 : */
8087 38888 : prefetch_state->next_item = i;
8088 38888 : prefetch_state->cur_hblkno = cur_hblkno;
8089 38888 : }
8090 : #endif
8091 :
8092 : /*
8093 : * Helper function for heap_index_delete_tuples. Checks for index corruption
8094 : * involving an invalid TID in index AM caller's index page.
8095 : *
8096 : * This is an ideal place for these checks. The index AM must hold a buffer
8097 : * lock on the index page containing the TIDs we examine here, so we don't
8098 : * have to worry about concurrent VACUUMs at all. We can be sure that the
8099 : * index is corrupt when htid points directly to an LP_UNUSED item or
8100 : * heap-only tuple, which is not the case during standard index scans.
8101 : */
8102 : static inline void
8103 1088266 : index_delete_check_htid(TM_IndexDeleteOp *delstate,
8104 : Page page, OffsetNumber maxoff,
8105 : const ItemPointerData *htid, TM_IndexStatus *istatus)
8106 : {
8107 1088266 : OffsetNumber indexpagehoffnum = ItemPointerGetOffsetNumber(htid);
8108 : ItemId iid;
8109 :
8110 : Assert(OffsetNumberIsValid(istatus->idxoffnum));
8111 :
8112 1088266 : if (unlikely(indexpagehoffnum > maxoff))
8113 0 : ereport(ERROR,
8114 : (errcode(ERRCODE_INDEX_CORRUPTED),
8115 : 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\"",
8116 : ItemPointerGetBlockNumber(htid),
8117 : indexpagehoffnum,
8118 : istatus->idxoffnum, delstate->iblknum,
8119 : RelationGetRelationName(delstate->irel))));
8120 :
8121 1088266 : iid = PageGetItemId(page, indexpagehoffnum);
8122 1088266 : if (unlikely(!ItemIdIsUsed(iid)))
8123 0 : ereport(ERROR,
8124 : (errcode(ERRCODE_INDEX_CORRUPTED),
8125 : errmsg_internal("heap tid from index tuple (%u,%u) points to unused heap page item at offset %u of block %u in index \"%s\"",
8126 : ItemPointerGetBlockNumber(htid),
8127 : indexpagehoffnum,
8128 : istatus->idxoffnum, delstate->iblknum,
8129 : RelationGetRelationName(delstate->irel))));
8130 :
8131 1088266 : if (ItemIdHasStorage(iid))
8132 : {
8133 : HeapTupleHeader htup;
8134 :
8135 : Assert(ItemIdIsNormal(iid));
8136 643180 : htup = (HeapTupleHeader) PageGetItem(page, iid);
8137 :
8138 643180 : if (unlikely(HeapTupleHeaderIsHeapOnly(htup)))
8139 0 : ereport(ERROR,
8140 : (errcode(ERRCODE_INDEX_CORRUPTED),
8141 : errmsg_internal("heap tid from index tuple (%u,%u) points to heap-only tuple at offset %u of block %u in index \"%s\"",
8142 : ItemPointerGetBlockNumber(htid),
8143 : indexpagehoffnum,
8144 : istatus->idxoffnum, delstate->iblknum,
8145 : RelationGetRelationName(delstate->irel))));
8146 : }
8147 1088266 : }
8148 :
8149 : /*
8150 : * heapam implementation of tableam's index_delete_tuples interface.
8151 : *
8152 : * This helper function is called by index AMs during index tuple deletion.
8153 : * See tableam header comments for an explanation of the interface implemented
8154 : * here and a general theory of operation. Note that each call here is either
8155 : * a simple index deletion call, or a bottom-up index deletion call.
8156 : *
8157 : * It's possible for this to generate a fair amount of I/O, since we may be
8158 : * deleting hundreds of tuples from a single index block. To amortize that
8159 : * cost to some degree, this uses prefetching and combines repeat accesses to
8160 : * the same heap block.
8161 : */
8162 : TransactionId
8163 11510 : heap_index_delete_tuples(Relation rel, TM_IndexDeleteOp *delstate)
8164 : {
8165 : /* Initial assumption is that earlier pruning took care of conflict */
8166 11510 : TransactionId snapshotConflictHorizon = InvalidTransactionId;
8167 11510 : BlockNumber blkno = InvalidBlockNumber;
8168 11510 : Buffer buf = InvalidBuffer;
8169 11510 : Page page = NULL;
8170 11510 : OffsetNumber maxoff = InvalidOffsetNumber;
8171 : TransactionId priorXmax;
8172 : #ifdef USE_PREFETCH
8173 : IndexDeletePrefetchState prefetch_state;
8174 : int prefetch_distance;
8175 : #endif
8176 : SnapshotData SnapshotNonVacuumable;
8177 11510 : int finalndeltids = 0,
8178 11510 : nblocksaccessed = 0;
8179 :
8180 : /* State that's only used in bottom-up index deletion case */
8181 11510 : int nblocksfavorable = 0;
8182 11510 : int curtargetfreespace = delstate->bottomupfreespace,
8183 11510 : lastfreespace = 0,
8184 11510 : actualfreespace = 0;
8185 11510 : bool bottomup_final_block = false;
8186 :
8187 11510 : InitNonVacuumableSnapshot(SnapshotNonVacuumable, GlobalVisTestFor(rel));
8188 :
8189 : /* Sort caller's deltids array by TID for further processing */
8190 11510 : index_delete_sort(delstate);
8191 :
8192 : /*
8193 : * Bottom-up case: resort deltids array in an order attuned to where the
8194 : * greatest number of promising TIDs are to be found, and determine how
8195 : * many blocks from the start of sorted array should be considered
8196 : * favorable. This will also shrink the deltids array in order to
8197 : * eliminate completely unfavorable blocks up front.
8198 : */
8199 11510 : if (delstate->bottomup)
8200 3844 : nblocksfavorable = bottomup_sort_and_shrink(delstate);
8201 :
8202 : #ifdef USE_PREFETCH
8203 : /* Initialize prefetch state. */
8204 11510 : prefetch_state.cur_hblkno = InvalidBlockNumber;
8205 11510 : prefetch_state.next_item = 0;
8206 11510 : prefetch_state.ndeltids = delstate->ndeltids;
8207 11510 : prefetch_state.deltids = delstate->deltids;
8208 :
8209 : /*
8210 : * Determine the prefetch distance that we will attempt to maintain.
8211 : *
8212 : * Since the caller holds a buffer lock somewhere in rel, we'd better make
8213 : * sure that isn't a catalog relation before we call code that does
8214 : * syscache lookups, to avoid risk of deadlock.
8215 : */
8216 11510 : if (IsCatalogRelation(rel))
8217 8324 : prefetch_distance = maintenance_io_concurrency;
8218 : else
8219 : prefetch_distance =
8220 3186 : get_tablespace_maintenance_io_concurrency(rel->rd_rel->reltablespace);
8221 :
8222 : /* Cap initial prefetch distance for bottom-up deletion caller */
8223 11510 : if (delstate->bottomup)
8224 : {
8225 : Assert(nblocksfavorable >= 1);
8226 : Assert(nblocksfavorable <= BOTTOMUP_MAX_NBLOCKS);
8227 3844 : prefetch_distance = Min(prefetch_distance, nblocksfavorable);
8228 : }
8229 :
8230 : /* Start prefetching. */
8231 11510 : index_delete_prefetch_buffer(rel, &prefetch_state, prefetch_distance);
8232 : #endif
8233 :
8234 : /* Iterate over deltids, determine which to delete, check their horizon */
8235 : Assert(delstate->ndeltids > 0);
8236 1099776 : for (int i = 0; i < delstate->ndeltids; i++)
8237 : {
8238 1092110 : TM_IndexDelete *ideltid = &delstate->deltids[i];
8239 1092110 : TM_IndexStatus *istatus = delstate->status + ideltid->id;
8240 1092110 : ItemPointer htid = &ideltid->tid;
8241 : OffsetNumber offnum;
8242 :
8243 : /*
8244 : * Read buffer, and perform required extra steps each time a new block
8245 : * is encountered. Avoid refetching if it's the same block as the one
8246 : * from the last htid.
8247 : */
8248 2172710 : if (blkno == InvalidBlockNumber ||
8249 1080600 : ItemPointerGetBlockNumber(htid) != blkno)
8250 : {
8251 : /*
8252 : * Consider giving up early for bottom-up index deletion caller
8253 : * first. (Only prefetch next-next block afterwards, when it
8254 : * becomes clear that we're at least going to access the next
8255 : * block in line.)
8256 : *
8257 : * Sometimes the first block frees so much space for bottom-up
8258 : * caller that the deletion process can end without accessing any
8259 : * more blocks. It is usually necessary to access 2 or 3 blocks
8260 : * per bottom-up deletion operation, though.
8261 : */
8262 31222 : if (delstate->bottomup)
8263 : {
8264 : /*
8265 : * We often allow caller to delete a few additional items
8266 : * whose entries we reached after the point that space target
8267 : * from caller was satisfied. The cost of accessing the page
8268 : * was already paid at that point, so it made sense to finish
8269 : * it off. When that happened, we finalize everything here
8270 : * (by finishing off the whole bottom-up deletion operation
8271 : * without needlessly paying the cost of accessing any more
8272 : * blocks).
8273 : */
8274 8520 : if (bottomup_final_block)
8275 290 : break;
8276 :
8277 : /*
8278 : * Give up when we didn't enable our caller to free any
8279 : * additional space as a result of processing the page that we
8280 : * just finished up with. This rule is the main way in which
8281 : * we keep the cost of bottom-up deletion under control.
8282 : */
8283 8230 : if (nblocksaccessed >= 1 && actualfreespace == lastfreespace)
8284 3554 : break;
8285 4676 : lastfreespace = actualfreespace; /* for next time */
8286 :
8287 : /*
8288 : * Deletion operation (which is bottom-up) will definitely
8289 : * access the next block in line. Prepare for that now.
8290 : *
8291 : * Decay target free space so that we don't hang on for too
8292 : * long with a marginal case. (Space target is only truly
8293 : * helpful when it allows us to recognize that we don't need
8294 : * to access more than 1 or 2 blocks to satisfy caller due to
8295 : * agreeable workload characteristics.)
8296 : *
8297 : * We are a bit more patient when we encounter contiguous
8298 : * blocks, though: these are treated as favorable blocks. The
8299 : * decay process is only applied when the next block in line
8300 : * is not a favorable/contiguous block. This is not an
8301 : * exception to the general rule; we still insist on finding
8302 : * at least one deletable item per block accessed. See
8303 : * bottomup_nblocksfavorable() for full details of the theory
8304 : * behind favorable blocks and heap block locality in general.
8305 : *
8306 : * Note: The first block in line is always treated as a
8307 : * favorable block, so the earliest possible point that the
8308 : * decay can be applied is just before we access the second
8309 : * block in line. The Assert() verifies this for us.
8310 : */
8311 : Assert(nblocksaccessed > 0 || nblocksfavorable > 0);
8312 4676 : if (nblocksfavorable > 0)
8313 4192 : nblocksfavorable--;
8314 : else
8315 484 : curtargetfreespace /= 2;
8316 : }
8317 :
8318 : /* release old buffer */
8319 27378 : if (BufferIsValid(buf))
8320 15868 : UnlockReleaseBuffer(buf);
8321 :
8322 27378 : blkno = ItemPointerGetBlockNumber(htid);
8323 27378 : buf = ReadBuffer(rel, blkno);
8324 27378 : nblocksaccessed++;
8325 : Assert(!delstate->bottomup ||
8326 : nblocksaccessed <= BOTTOMUP_MAX_NBLOCKS);
8327 :
8328 : #ifdef USE_PREFETCH
8329 :
8330 : /*
8331 : * To maintain the prefetch distance, prefetch one more page for
8332 : * each page we read.
8333 : */
8334 27378 : index_delete_prefetch_buffer(rel, &prefetch_state, 1);
8335 : #endif
8336 :
8337 27378 : LockBuffer(buf, BUFFER_LOCK_SHARE);
8338 :
8339 27378 : page = BufferGetPage(buf);
8340 27378 : maxoff = PageGetMaxOffsetNumber(page);
8341 : }
8342 :
8343 : /*
8344 : * In passing, detect index corruption involving an index page with a
8345 : * TID that points to a location in the heap that couldn't possibly be
8346 : * correct. We only do this with actual TIDs from caller's index page
8347 : * (not items reached by traversing through a HOT chain).
8348 : */
8349 1088266 : index_delete_check_htid(delstate, page, maxoff, htid, istatus);
8350 :
8351 1088266 : if (istatus->knowndeletable)
8352 : Assert(!delstate->bottomup && !istatus->promising);
8353 : else
8354 : {
8355 819442 : ItemPointerData tmp = *htid;
8356 : HeapTupleData heapTuple;
8357 :
8358 : /* Are any tuples from this HOT chain non-vacuumable? */
8359 819442 : if (heap_hot_search_buffer(&tmp, rel, buf, &SnapshotNonVacuumable,
8360 : &heapTuple, NULL, true))
8361 488372 : continue; /* can't delete entry */
8362 :
8363 : /* Caller will delete, since whole HOT chain is vacuumable */
8364 331070 : istatus->knowndeletable = true;
8365 :
8366 : /* Maintain index free space info for bottom-up deletion case */
8367 331070 : if (delstate->bottomup)
8368 : {
8369 : Assert(istatus->freespace > 0);
8370 16938 : actualfreespace += istatus->freespace;
8371 16938 : if (actualfreespace >= curtargetfreespace)
8372 4390 : bottomup_final_block = true;
8373 : }
8374 : }
8375 :
8376 : /*
8377 : * Maintain snapshotConflictHorizon value for deletion operation as a
8378 : * whole by advancing current value using heap tuple headers. This is
8379 : * loosely based on the logic for pruning a HOT chain.
8380 : */
8381 599894 : offnum = ItemPointerGetOffsetNumber(htid);
8382 599894 : priorXmax = InvalidTransactionId; /* cannot check first XMIN */
8383 : for (;;)
8384 41214 : {
8385 : ItemId lp;
8386 : HeapTupleHeader htup;
8387 :
8388 : /* Sanity check (pure paranoia) */
8389 641108 : if (offnum < FirstOffsetNumber)
8390 0 : break;
8391 :
8392 : /*
8393 : * An offset past the end of page's line pointer array is possible
8394 : * when the array was truncated
8395 : */
8396 641108 : if (offnum > maxoff)
8397 0 : break;
8398 :
8399 641108 : lp = PageGetItemId(page, offnum);
8400 641108 : if (ItemIdIsRedirected(lp))
8401 : {
8402 18554 : offnum = ItemIdGetRedirect(lp);
8403 18554 : continue;
8404 : }
8405 :
8406 : /*
8407 : * We'll often encounter LP_DEAD line pointers (especially with an
8408 : * entry marked knowndeletable by our caller up front). No heap
8409 : * tuple headers get examined for an htid that leads us to an
8410 : * LP_DEAD item. This is okay because the earlier pruning
8411 : * operation that made the line pointer LP_DEAD in the first place
8412 : * must have considered the original tuple header as part of
8413 : * generating its own snapshotConflictHorizon value.
8414 : *
8415 : * Relying on XLOG_HEAP2_PRUNE_VACUUM_SCAN records like this is
8416 : * the same strategy that index vacuuming uses in all cases. Index
8417 : * VACUUM WAL records don't even have a snapshotConflictHorizon
8418 : * field of their own for this reason.
8419 : */
8420 622554 : if (!ItemIdIsNormal(lp))
8421 395982 : break;
8422 :
8423 226572 : htup = (HeapTupleHeader) PageGetItem(page, lp);
8424 :
8425 : /*
8426 : * Check the tuple XMIN against prior XMAX, if any
8427 : */
8428 249232 : if (TransactionIdIsValid(priorXmax) &&
8429 22660 : !TransactionIdEquals(HeapTupleHeaderGetXmin(htup), priorXmax))
8430 0 : break;
8431 :
8432 226572 : HeapTupleHeaderAdvanceConflictHorizon(htup,
8433 : &snapshotConflictHorizon);
8434 :
8435 : /*
8436 : * If the tuple is not HOT-updated, then we are at the end of this
8437 : * HOT-chain. No need to visit later tuples from the same update
8438 : * chain (they get their own index entries) -- just move on to
8439 : * next htid from index AM caller.
8440 : */
8441 226572 : if (!HeapTupleHeaderIsHotUpdated(htup))
8442 203912 : break;
8443 :
8444 : /* Advance to next HOT chain member */
8445 : Assert(ItemPointerGetBlockNumber(&htup->t_ctid) == blkno);
8446 22660 : offnum = ItemPointerGetOffsetNumber(&htup->t_ctid);
8447 22660 : priorXmax = HeapTupleHeaderGetUpdateXid(htup);
8448 : }
8449 :
8450 : /* Enable further/final shrinking of deltids for caller */
8451 599894 : finalndeltids = i + 1;
8452 : }
8453 :
8454 11510 : UnlockReleaseBuffer(buf);
8455 :
8456 : /*
8457 : * Shrink deltids array to exclude non-deletable entries at the end. This
8458 : * is not just a minor optimization. Final deltids array size might be
8459 : * zero for a bottom-up caller. Index AM is explicitly allowed to rely on
8460 : * ndeltids being zero in all cases with zero total deletable entries.
8461 : */
8462 : Assert(finalndeltids > 0 || delstate->bottomup);
8463 11510 : delstate->ndeltids = finalndeltids;
8464 :
8465 11510 : return snapshotConflictHorizon;
8466 : }
8467 :
8468 : /*
8469 : * Specialized inlineable comparison function for index_delete_sort()
8470 : */
8471 : static inline int
8472 25355578 : index_delete_sort_cmp(TM_IndexDelete *deltid1, TM_IndexDelete *deltid2)
8473 : {
8474 25355578 : ItemPointer tid1 = &deltid1->tid;
8475 25355578 : ItemPointer tid2 = &deltid2->tid;
8476 :
8477 : {
8478 25355578 : BlockNumber blk1 = ItemPointerGetBlockNumber(tid1);
8479 25355578 : BlockNumber blk2 = ItemPointerGetBlockNumber(tid2);
8480 :
8481 25355578 : if (blk1 != blk2)
8482 10293090 : return (blk1 < blk2) ? -1 : 1;
8483 : }
8484 : {
8485 15062488 : OffsetNumber pos1 = ItemPointerGetOffsetNumber(tid1);
8486 15062488 : OffsetNumber pos2 = ItemPointerGetOffsetNumber(tid2);
8487 :
8488 15062488 : if (pos1 != pos2)
8489 15062488 : return (pos1 < pos2) ? -1 : 1;
8490 : }
8491 :
8492 : Assert(false);
8493 :
8494 0 : return 0;
8495 : }
8496 :
8497 : /*
8498 : * Sort deltids array from delstate by TID. This prepares it for further
8499 : * processing by heap_index_delete_tuples().
8500 : *
8501 : * This operation becomes a noticeable consumer of CPU cycles with some
8502 : * workloads, so we go to the trouble of specialization/micro optimization.
8503 : * We use shellsort for this because it's easy to specialize, compiles to
8504 : * relatively few instructions, and is adaptive to presorted inputs/subsets
8505 : * (which are typical here).
8506 : */
8507 : static void
8508 11510 : index_delete_sort(TM_IndexDeleteOp *delstate)
8509 : {
8510 11510 : TM_IndexDelete *deltids = delstate->deltids;
8511 11510 : int ndeltids = delstate->ndeltids;
8512 :
8513 : /*
8514 : * Shellsort gap sequence (taken from Sedgewick-Incerpi paper).
8515 : *
8516 : * This implementation is fast with array sizes up to ~4500. This covers
8517 : * all supported BLCKSZ values.
8518 : */
8519 11510 : const int gaps[9] = {1968, 861, 336, 112, 48, 21, 7, 3, 1};
8520 :
8521 : /* Think carefully before changing anything here -- keep swaps cheap */
8522 : StaticAssertDecl(sizeof(TM_IndexDelete) <= 8,
8523 : "element size exceeds 8 bytes");
8524 :
8525 115100 : for (int g = 0; g < lengthof(gaps); g++)
8526 : {
8527 15117170 : for (int hi = gaps[g], i = hi; i < ndeltids; i++)
8528 : {
8529 15013580 : TM_IndexDelete d = deltids[i];
8530 15013580 : int j = i;
8531 :
8532 26090326 : while (j >= hi && index_delete_sort_cmp(&deltids[j - hi], &d) >= 0)
8533 : {
8534 11076746 : deltids[j] = deltids[j - hi];
8535 11076746 : j -= hi;
8536 : }
8537 15013580 : deltids[j] = d;
8538 : }
8539 : }
8540 11510 : }
8541 :
8542 : /*
8543 : * Returns how many blocks should be considered favorable/contiguous for a
8544 : * bottom-up index deletion pass. This is a number of heap blocks that starts
8545 : * from and includes the first block in line.
8546 : *
8547 : * There is always at least one favorable block during bottom-up index
8548 : * deletion. In the worst case (i.e. with totally random heap blocks) the
8549 : * first block in line (the only favorable block) can be thought of as a
8550 : * degenerate array of contiguous blocks that consists of a single block.
8551 : * heap_index_delete_tuples() will expect this.
8552 : *
8553 : * Caller passes blockgroups, a description of the final order that deltids
8554 : * will be sorted in for heap_index_delete_tuples() bottom-up index deletion
8555 : * processing. Note that deltids need not actually be sorted just yet (caller
8556 : * only passes deltids to us so that we can interpret blockgroups).
8557 : *
8558 : * You might guess that the existence of contiguous blocks cannot matter much,
8559 : * since in general the main factor that determines which blocks we visit is
8560 : * the number of promising TIDs, which is a fixed hint from the index AM.
8561 : * We're not really targeting the general case, though -- the actual goal is
8562 : * to adapt our behavior to a wide variety of naturally occurring conditions.
8563 : * The effects of most of the heuristics we apply are only noticeable in the
8564 : * aggregate, over time and across many _related_ bottom-up index deletion
8565 : * passes.
8566 : *
8567 : * Deeming certain blocks favorable allows heapam to recognize and adapt to
8568 : * workloads where heap blocks visited during bottom-up index deletion can be
8569 : * accessed contiguously, in the sense that each newly visited block is the
8570 : * neighbor of the block that bottom-up deletion just finished processing (or
8571 : * close enough to it). It will likely be cheaper to access more favorable
8572 : * blocks sooner rather than later (e.g. in this pass, not across a series of
8573 : * related bottom-up passes). Either way it is probably only a matter of time
8574 : * (or a matter of further correlated version churn) before all blocks that
8575 : * appear together as a single large batch of favorable blocks get accessed by
8576 : * _some_ bottom-up pass. Large batches of favorable blocks tend to either
8577 : * appear almost constantly or not even once (it all depends on per-index
8578 : * workload characteristics).
8579 : *
8580 : * Note that the blockgroups sort order applies a power-of-two bucketing
8581 : * scheme that creates opportunities for contiguous groups of blocks to get
8582 : * batched together, at least with workloads that are naturally amenable to
8583 : * being driven by heap block locality. This doesn't just enhance the spatial
8584 : * locality of bottom-up heap block processing in the obvious way. It also
8585 : * enables temporal locality of access, since sorting by heap block number
8586 : * naturally tends to make the bottom-up processing order deterministic.
8587 : *
8588 : * Consider the following example to get a sense of how temporal locality
8589 : * might matter: There is a heap relation with several indexes, each of which
8590 : * is low to medium cardinality. It is subject to constant non-HOT updates.
8591 : * The updates are skewed (in one part of the primary key, perhaps). None of
8592 : * the indexes are logically modified by the UPDATE statements (if they were
8593 : * then bottom-up index deletion would not be triggered in the first place).
8594 : * Naturally, each new round of index tuples (for each heap tuple that gets a
8595 : * heap_update() call) will have the same heap TID in each and every index.
8596 : * Since these indexes are low cardinality and never get logically modified,
8597 : * heapam processing during bottom-up deletion passes will access heap blocks
8598 : * in approximately sequential order. Temporal locality of access occurs due
8599 : * to bottom-up deletion passes behaving very similarly across each of the
8600 : * indexes at any given moment. This keeps the number of buffer misses needed
8601 : * to visit heap blocks to a minimum.
8602 : */
8603 : static int
8604 3844 : bottomup_nblocksfavorable(IndexDeleteCounts *blockgroups, int nblockgroups,
8605 : TM_IndexDelete *deltids)
8606 : {
8607 3844 : int64 lastblock = -1;
8608 3844 : int nblocksfavorable = 0;
8609 :
8610 : Assert(nblockgroups >= 1);
8611 : Assert(nblockgroups <= BOTTOMUP_MAX_NBLOCKS);
8612 :
8613 : /*
8614 : * We tolerate heap blocks that will be accessed only slightly out of
8615 : * physical order. Small blips occur when a pair of almost-contiguous
8616 : * blocks happen to fall into different buckets (perhaps due only to a
8617 : * small difference in npromisingtids that the bucketing scheme didn't
8618 : * quite manage to ignore). We effectively ignore these blips by applying
8619 : * a small tolerance. The precise tolerance we use is a little arbitrary,
8620 : * but it works well enough in practice.
8621 : */
8622 12274 : for (int b = 0; b < nblockgroups; b++)
8623 : {
8624 11720 : IndexDeleteCounts *group = blockgroups + b;
8625 11720 : TM_IndexDelete *firstdtid = deltids + group->ifirsttid;
8626 11720 : BlockNumber block = ItemPointerGetBlockNumber(&firstdtid->tid);
8627 :
8628 11720 : if (lastblock != -1 &&
8629 7876 : ((int64) block < lastblock - BOTTOMUP_TOLERANCE_NBLOCKS ||
8630 6930 : (int64) block > lastblock + BOTTOMUP_TOLERANCE_NBLOCKS))
8631 : break;
8632 :
8633 8430 : nblocksfavorable++;
8634 8430 : lastblock = block;
8635 : }
8636 :
8637 : /* Always indicate that there is at least 1 favorable block */
8638 : Assert(nblocksfavorable >= 1);
8639 :
8640 3844 : return nblocksfavorable;
8641 : }
8642 :
8643 : /*
8644 : * qsort comparison function for bottomup_sort_and_shrink()
8645 : */
8646 : static int
8647 373080 : bottomup_sort_and_shrink_cmp(const void *arg1, const void *arg2)
8648 : {
8649 373080 : const IndexDeleteCounts *group1 = (const IndexDeleteCounts *) arg1;
8650 373080 : const IndexDeleteCounts *group2 = (const IndexDeleteCounts *) arg2;
8651 :
8652 : /*
8653 : * Most significant field is npromisingtids (which we invert the order of
8654 : * so as to sort in desc order).
8655 : *
8656 : * Caller should have already normalized npromisingtids fields into
8657 : * power-of-two values (buckets).
8658 : */
8659 373080 : if (group1->npromisingtids > group2->npromisingtids)
8660 18668 : return -1;
8661 354412 : if (group1->npromisingtids < group2->npromisingtids)
8662 21112 : return 1;
8663 :
8664 : /*
8665 : * Tiebreak: desc ntids sort order.
8666 : *
8667 : * We cannot expect power-of-two values for ntids fields. We should
8668 : * behave as if they were already rounded up for us instead.
8669 : */
8670 333300 : if (group1->ntids != group2->ntids)
8671 : {
8672 235032 : uint32 ntids1 = pg_nextpower2_32((uint32) group1->ntids);
8673 235032 : uint32 ntids2 = pg_nextpower2_32((uint32) group2->ntids);
8674 :
8675 235032 : if (ntids1 > ntids2)
8676 33640 : return -1;
8677 201392 : if (ntids1 < ntids2)
8678 42784 : return 1;
8679 : }
8680 :
8681 : /*
8682 : * Tiebreak: asc offset-into-deltids-for-block (offset to first TID for
8683 : * block in deltids array) order.
8684 : *
8685 : * This is equivalent to sorting in ascending heap block number order
8686 : * (among otherwise equal subsets of the array). This approach allows us
8687 : * to avoid accessing the out-of-line TID. (We rely on the assumption
8688 : * that the deltids array was sorted in ascending heap TID order when
8689 : * these offsets to the first TID from each heap block group were formed.)
8690 : */
8691 256876 : if (group1->ifirsttid > group2->ifirsttid)
8692 126654 : return 1;
8693 130222 : if (group1->ifirsttid < group2->ifirsttid)
8694 130222 : return -1;
8695 :
8696 0 : pg_unreachable();
8697 :
8698 : return 0;
8699 : }
8700 :
8701 : /*
8702 : * heap_index_delete_tuples() helper function for bottom-up deletion callers.
8703 : *
8704 : * Sorts deltids array in the order needed for useful processing by bottom-up
8705 : * deletion. The array should already be sorted in TID order when we're
8706 : * called. The sort process groups heap TIDs from deltids into heap block
8707 : * groupings. Earlier/more-promising groups/blocks are usually those that are
8708 : * known to have the most "promising" TIDs.
8709 : *
8710 : * Sets new size of deltids array (ndeltids) in state. deltids will only have
8711 : * TIDs from the BOTTOMUP_MAX_NBLOCKS most promising heap blocks when we
8712 : * return. This often means that deltids will be shrunk to a small fraction
8713 : * of its original size (we eliminate many heap blocks from consideration for
8714 : * caller up front).
8715 : *
8716 : * Returns the number of "favorable" blocks. See bottomup_nblocksfavorable()
8717 : * for a definition and full details.
8718 : */
8719 : static int
8720 3844 : bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate)
8721 : {
8722 : IndexDeleteCounts *blockgroups;
8723 : TM_IndexDelete *reordereddeltids;
8724 3844 : BlockNumber curblock = InvalidBlockNumber;
8725 3844 : int nblockgroups = 0;
8726 3844 : int ncopied = 0;
8727 3844 : int nblocksfavorable = 0;
8728 :
8729 : Assert(delstate->bottomup);
8730 : Assert(delstate->ndeltids > 0);
8731 :
8732 : /* Calculate per-heap-block count of TIDs */
8733 3844 : blockgroups = palloc_array(IndexDeleteCounts, delstate->ndeltids);
8734 1866410 : for (int i = 0; i < delstate->ndeltids; i++)
8735 : {
8736 1862566 : TM_IndexDelete *ideltid = &delstate->deltids[i];
8737 1862566 : TM_IndexStatus *istatus = delstate->status + ideltid->id;
8738 1862566 : ItemPointer htid = &ideltid->tid;
8739 1862566 : bool promising = istatus->promising;
8740 :
8741 1862566 : if (curblock != ItemPointerGetBlockNumber(htid))
8742 : {
8743 : /* New block group */
8744 73164 : nblockgroups++;
8745 :
8746 : Assert(curblock < ItemPointerGetBlockNumber(htid) ||
8747 : !BlockNumberIsValid(curblock));
8748 :
8749 73164 : curblock = ItemPointerGetBlockNumber(htid);
8750 73164 : blockgroups[nblockgroups - 1].ifirsttid = i;
8751 73164 : blockgroups[nblockgroups - 1].ntids = 1;
8752 73164 : blockgroups[nblockgroups - 1].npromisingtids = 0;
8753 : }
8754 : else
8755 : {
8756 1789402 : blockgroups[nblockgroups - 1].ntids++;
8757 : }
8758 :
8759 1862566 : if (promising)
8760 243070 : blockgroups[nblockgroups - 1].npromisingtids++;
8761 : }
8762 :
8763 : /*
8764 : * We're about ready to sort block groups to determine the optimal order
8765 : * for visiting heap blocks. But before we do, round the number of
8766 : * promising tuples for each block group up to the next power-of-two,
8767 : * unless it is very low (less than 4), in which case we round up to 4.
8768 : * npromisingtids is far too noisy to trust when choosing between a pair
8769 : * of block groups that both have very low values.
8770 : *
8771 : * This scheme divides heap blocks/block groups into buckets. Each bucket
8772 : * contains blocks that have _approximately_ the same number of promising
8773 : * TIDs as each other. The goal is to ignore relatively small differences
8774 : * in the total number of promising entries, so that the whole process can
8775 : * give a little weight to heapam factors (like heap block locality)
8776 : * instead. This isn't a trade-off, really -- we have nothing to lose. It
8777 : * would be foolish to interpret small differences in npromisingtids
8778 : * values as anything more than noise.
8779 : *
8780 : * We tiebreak on nhtids when sorting block group subsets that have the
8781 : * same npromisingtids, but this has the same issues as npromisingtids,
8782 : * and so nhtids is subject to the same power-of-two bucketing scheme. The
8783 : * only reason that we don't fix nhtids in the same way here too is that
8784 : * we'll need accurate nhtids values after the sort. We handle nhtids
8785 : * bucketization dynamically instead (in the sort comparator).
8786 : *
8787 : * See bottomup_nblocksfavorable() for a full explanation of when and how
8788 : * heap locality/favorable blocks can significantly influence when and how
8789 : * heap blocks are accessed.
8790 : */
8791 77008 : for (int b = 0; b < nblockgroups; b++)
8792 : {
8793 73164 : IndexDeleteCounts *group = blockgroups + b;
8794 :
8795 : /* Better off falling back on nhtids with low npromisingtids */
8796 73164 : if (group->npromisingtids <= 4)
8797 62036 : group->npromisingtids = 4;
8798 : else
8799 11128 : group->npromisingtids =
8800 11128 : pg_nextpower2_32((uint32) group->npromisingtids);
8801 : }
8802 :
8803 : /* Sort groups and rearrange caller's deltids array */
8804 3844 : qsort(blockgroups, nblockgroups, sizeof(IndexDeleteCounts),
8805 : bottomup_sort_and_shrink_cmp);
8806 3844 : reordereddeltids = palloc(delstate->ndeltids * sizeof(TM_IndexDelete));
8807 :
8808 3844 : nblockgroups = Min(BOTTOMUP_MAX_NBLOCKS, nblockgroups);
8809 : /* Determine number of favorable blocks at the start of final deltids */
8810 3844 : nblocksfavorable = bottomup_nblocksfavorable(blockgroups, nblockgroups,
8811 : delstate->deltids);
8812 :
8813 25686 : for (int b = 0; b < nblockgroups; b++)
8814 : {
8815 21842 : IndexDeleteCounts *group = blockgroups + b;
8816 21842 : TM_IndexDelete *firstdtid = delstate->deltids + group->ifirsttid;
8817 :
8818 21842 : memcpy(reordereddeltids + ncopied, firstdtid,
8819 21842 : sizeof(TM_IndexDelete) * group->ntids);
8820 21842 : ncopied += group->ntids;
8821 : }
8822 :
8823 : /* Copy final grouped and sorted TIDs back into start of caller's array */
8824 3844 : memcpy(delstate->deltids, reordereddeltids,
8825 : sizeof(TM_IndexDelete) * ncopied);
8826 3844 : delstate->ndeltids = ncopied;
8827 :
8828 3844 : pfree(reordereddeltids);
8829 3844 : pfree(blockgroups);
8830 :
8831 3844 : return nblocksfavorable;
8832 : }
8833 :
8834 : /*
8835 : * Perform XLogInsert for a heap-visible operation. 'block' is the block
8836 : * being marked all-visible, and vm_buffer is the buffer containing the
8837 : * corresponding visibility map block. Both should have already been modified
8838 : * and dirtied.
8839 : *
8840 : * snapshotConflictHorizon comes from the largest xmin on the page being
8841 : * marked all-visible. REDO routine uses it to generate recovery conflicts.
8842 : *
8843 : * If checksums or wal_log_hints are enabled, we may also generate a full-page
8844 : * image of heap_buffer. Otherwise, we optimize away the FPI (by specifying
8845 : * REGBUF_NO_IMAGE for the heap buffer), in which case the caller should *not*
8846 : * update the heap page's LSN.
8847 : */
8848 : XLogRecPtr
8849 70600 : log_heap_visible(Relation rel, Buffer heap_buffer, Buffer vm_buffer,
8850 : TransactionId snapshotConflictHorizon, uint8 vmflags)
8851 : {
8852 : xl_heap_visible xlrec;
8853 : XLogRecPtr recptr;
8854 : uint8 flags;
8855 :
8856 : Assert(BufferIsValid(heap_buffer));
8857 : Assert(BufferIsValid(vm_buffer));
8858 :
8859 70600 : xlrec.snapshotConflictHorizon = snapshotConflictHorizon;
8860 70600 : xlrec.flags = vmflags;
8861 70600 : if (RelationIsAccessibleInLogicalDecoding(rel))
8862 112 : xlrec.flags |= VISIBILITYMAP_XLOG_CATALOG_REL;
8863 70600 : XLogBeginInsert();
8864 70600 : XLogRegisterData(&xlrec, SizeOfHeapVisible);
8865 :
8866 70600 : XLogRegisterBuffer(0, vm_buffer, 0);
8867 :
8868 70600 : flags = REGBUF_STANDARD;
8869 70600 : if (!XLogHintBitIsNeeded())
8870 6158 : flags |= REGBUF_NO_IMAGE;
8871 70600 : XLogRegisterBuffer(1, heap_buffer, flags);
8872 :
8873 70600 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_VISIBLE);
8874 :
8875 70600 : return recptr;
8876 : }
8877 :
8878 : /*
8879 : * Perform XLogInsert for a heap-update operation. Caller must already
8880 : * have modified the buffer(s) and marked them dirty.
8881 : */
8882 : static XLogRecPtr
8883 601236 : log_heap_update(Relation reln, Buffer oldbuf,
8884 : Buffer newbuf, HeapTuple oldtup, HeapTuple newtup,
8885 : HeapTuple old_key_tuple,
8886 : bool all_visible_cleared, bool new_all_visible_cleared)
8887 : {
8888 : xl_heap_update xlrec;
8889 : xl_heap_header xlhdr;
8890 : xl_heap_header xlhdr_idx;
8891 : uint8 info;
8892 : uint16 prefix_suffix[2];
8893 601236 : uint16 prefixlen = 0,
8894 601236 : suffixlen = 0;
8895 : XLogRecPtr recptr;
8896 601236 : Page page = BufferGetPage(newbuf);
8897 601236 : bool need_tuple_data = RelationIsLogicallyLogged(reln);
8898 : bool init;
8899 : int bufflags;
8900 :
8901 : /* Caller should not call me on a non-WAL-logged relation */
8902 : Assert(RelationNeedsWAL(reln));
8903 :
8904 601236 : XLogBeginInsert();
8905 :
8906 601236 : if (HeapTupleIsHeapOnly(newtup))
8907 293734 : info = XLOG_HEAP_HOT_UPDATE;
8908 : else
8909 307502 : info = XLOG_HEAP_UPDATE;
8910 :
8911 : /*
8912 : * If the old and new tuple are on the same page, we only need to log the
8913 : * parts of the new tuple that were changed. That saves on the amount of
8914 : * WAL we need to write. Currently, we just count any unchanged bytes in
8915 : * the beginning and end of the tuple. That's quick to check, and
8916 : * perfectly covers the common case that only one field is updated.
8917 : *
8918 : * We could do this even if the old and new tuple are on different pages,
8919 : * but only if we don't make a full-page image of the old page, which is
8920 : * difficult to know in advance. Also, if the old tuple is corrupt for
8921 : * some reason, it would allow the corruption to propagate the new page,
8922 : * so it seems best to avoid. Under the general assumption that most
8923 : * updates tend to create the new tuple version on the same page, there
8924 : * isn't much to be gained by doing this across pages anyway.
8925 : *
8926 : * Skip this if we're taking a full-page image of the new page, as we
8927 : * don't include the new tuple in the WAL record in that case. Also
8928 : * disable if effective_wal_level='logical', as logical decoding needs to
8929 : * be able to read the new tuple in whole from the WAL record alone.
8930 : */
8931 601236 : if (oldbuf == newbuf && !need_tuple_data &&
8932 294428 : !XLogCheckBufferNeedsBackup(newbuf))
8933 : {
8934 293134 : char *oldp = (char *) oldtup->t_data + oldtup->t_data->t_hoff;
8935 293134 : char *newp = (char *) newtup->t_data + newtup->t_data->t_hoff;
8936 293134 : int oldlen = oldtup->t_len - oldtup->t_data->t_hoff;
8937 293134 : int newlen = newtup->t_len - newtup->t_data->t_hoff;
8938 :
8939 : /* Check for common prefix between old and new tuple */
8940 24529262 : for (prefixlen = 0; prefixlen < Min(oldlen, newlen); prefixlen++)
8941 : {
8942 24477662 : if (newp[prefixlen] != oldp[prefixlen])
8943 241534 : break;
8944 : }
8945 :
8946 : /*
8947 : * Storing the length of the prefix takes 2 bytes, so we need to save
8948 : * at least 3 bytes or there's no point.
8949 : */
8950 293134 : if (prefixlen < 3)
8951 44012 : prefixlen = 0;
8952 :
8953 : /* Same for suffix */
8954 9622620 : for (suffixlen = 0; suffixlen < Min(oldlen, newlen) - prefixlen; suffixlen++)
8955 : {
8956 9570504 : if (newp[newlen - suffixlen - 1] != oldp[oldlen - suffixlen - 1])
8957 241018 : break;
8958 : }
8959 293134 : if (suffixlen < 3)
8960 73340 : suffixlen = 0;
8961 : }
8962 :
8963 : /* Prepare main WAL data chain */
8964 601236 : xlrec.flags = 0;
8965 601236 : if (all_visible_cleared)
8966 3116 : xlrec.flags |= XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED;
8967 601236 : if (new_all_visible_cleared)
8968 1704 : xlrec.flags |= XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED;
8969 601236 : if (prefixlen > 0)
8970 249122 : xlrec.flags |= XLH_UPDATE_PREFIX_FROM_OLD;
8971 601236 : if (suffixlen > 0)
8972 219794 : xlrec.flags |= XLH_UPDATE_SUFFIX_FROM_OLD;
8973 601236 : if (need_tuple_data)
8974 : {
8975 94048 : xlrec.flags |= XLH_UPDATE_CONTAINS_NEW_TUPLE;
8976 94048 : if (old_key_tuple)
8977 : {
8978 292 : if (reln->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
8979 130 : xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_TUPLE;
8980 : else
8981 162 : xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_KEY;
8982 : }
8983 : }
8984 :
8985 : /* If new tuple is the single and first tuple on page... */
8986 608356 : if (ItemPointerGetOffsetNumber(&(newtup->t_self)) == FirstOffsetNumber &&
8987 7120 : PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
8988 : {
8989 6664 : info |= XLOG_HEAP_INIT_PAGE;
8990 6664 : init = true;
8991 : }
8992 : else
8993 594572 : init = false;
8994 :
8995 : /* Prepare WAL data for the old page */
8996 601236 : xlrec.old_offnum = ItemPointerGetOffsetNumber(&oldtup->t_self);
8997 601236 : xlrec.old_xmax = HeapTupleHeaderGetRawXmax(oldtup->t_data);
8998 1202472 : xlrec.old_infobits_set = compute_infobits(oldtup->t_data->t_infomask,
8999 601236 : oldtup->t_data->t_infomask2);
9000 :
9001 : /* Prepare WAL data for the new page */
9002 601236 : xlrec.new_offnum = ItemPointerGetOffsetNumber(&newtup->t_self);
9003 601236 : xlrec.new_xmax = HeapTupleHeaderGetRawXmax(newtup->t_data);
9004 :
9005 601236 : bufflags = REGBUF_STANDARD;
9006 601236 : if (init)
9007 6664 : bufflags |= REGBUF_WILL_INIT;
9008 601236 : if (need_tuple_data)
9009 94048 : bufflags |= REGBUF_KEEP_DATA;
9010 :
9011 601236 : XLogRegisterBuffer(0, newbuf, bufflags);
9012 601236 : if (oldbuf != newbuf)
9013 282920 : XLogRegisterBuffer(1, oldbuf, REGBUF_STANDARD);
9014 :
9015 601236 : XLogRegisterData(&xlrec, SizeOfHeapUpdate);
9016 :
9017 : /*
9018 : * Prepare WAL data for the new tuple.
9019 : */
9020 601236 : if (prefixlen > 0 || suffixlen > 0)
9021 : {
9022 292208 : if (prefixlen > 0 && suffixlen > 0)
9023 : {
9024 176708 : prefix_suffix[0] = prefixlen;
9025 176708 : prefix_suffix[1] = suffixlen;
9026 176708 : XLogRegisterBufData(0, &prefix_suffix, sizeof(uint16) * 2);
9027 : }
9028 115500 : else if (prefixlen > 0)
9029 : {
9030 72414 : XLogRegisterBufData(0, &prefixlen, sizeof(uint16));
9031 : }
9032 : else
9033 : {
9034 43086 : XLogRegisterBufData(0, &suffixlen, sizeof(uint16));
9035 : }
9036 : }
9037 :
9038 601236 : xlhdr.t_infomask2 = newtup->t_data->t_infomask2;
9039 601236 : xlhdr.t_infomask = newtup->t_data->t_infomask;
9040 601236 : xlhdr.t_hoff = newtup->t_data->t_hoff;
9041 : Assert(SizeofHeapTupleHeader + prefixlen + suffixlen <= newtup->t_len);
9042 :
9043 : /*
9044 : * PG73FORMAT: write bitmap [+ padding] [+ oid] + data
9045 : *
9046 : * The 'data' doesn't include the common prefix or suffix.
9047 : */
9048 601236 : XLogRegisterBufData(0, &xlhdr, SizeOfHeapHeader);
9049 601236 : if (prefixlen == 0)
9050 : {
9051 352114 : XLogRegisterBufData(0,
9052 352114 : (char *) newtup->t_data + SizeofHeapTupleHeader,
9053 352114 : newtup->t_len - SizeofHeapTupleHeader - suffixlen);
9054 : }
9055 : else
9056 : {
9057 : /*
9058 : * Have to write the null bitmap and data after the common prefix as
9059 : * two separate rdata entries.
9060 : */
9061 : /* bitmap [+ padding] [+ oid] */
9062 249122 : if (newtup->t_data->t_hoff - SizeofHeapTupleHeader > 0)
9063 : {
9064 249122 : XLogRegisterBufData(0,
9065 249122 : (char *) newtup->t_data + SizeofHeapTupleHeader,
9066 249122 : newtup->t_data->t_hoff - SizeofHeapTupleHeader);
9067 : }
9068 :
9069 : /* data after common prefix */
9070 249122 : XLogRegisterBufData(0,
9071 249122 : (char *) newtup->t_data + newtup->t_data->t_hoff + prefixlen,
9072 249122 : newtup->t_len - newtup->t_data->t_hoff - prefixlen - suffixlen);
9073 : }
9074 :
9075 : /* We need to log a tuple identity */
9076 601236 : if (need_tuple_data && old_key_tuple)
9077 : {
9078 : /* don't really need this, but its more comfy to decode */
9079 292 : xlhdr_idx.t_infomask2 = old_key_tuple->t_data->t_infomask2;
9080 292 : xlhdr_idx.t_infomask = old_key_tuple->t_data->t_infomask;
9081 292 : xlhdr_idx.t_hoff = old_key_tuple->t_data->t_hoff;
9082 :
9083 292 : XLogRegisterData(&xlhdr_idx, SizeOfHeapHeader);
9084 :
9085 : /* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
9086 292 : XLogRegisterData((char *) old_key_tuple->t_data + SizeofHeapTupleHeader,
9087 292 : old_key_tuple->t_len - SizeofHeapTupleHeader);
9088 : }
9089 :
9090 : /* filtering by origin on a row level is much more efficient */
9091 601236 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
9092 :
9093 601236 : recptr = XLogInsert(RM_HEAP_ID, info);
9094 :
9095 601236 : return recptr;
9096 : }
9097 :
9098 : /*
9099 : * Perform XLogInsert of an XLOG_HEAP2_NEW_CID record
9100 : *
9101 : * This is only used when effective_wal_level is logical, and only for
9102 : * catalog tuples.
9103 : */
9104 : static XLogRecPtr
9105 49208 : log_heap_new_cid(Relation relation, HeapTuple tup)
9106 : {
9107 : xl_heap_new_cid xlrec;
9108 :
9109 : XLogRecPtr recptr;
9110 49208 : HeapTupleHeader hdr = tup->t_data;
9111 :
9112 : Assert(ItemPointerIsValid(&tup->t_self));
9113 : Assert(tup->t_tableOid != InvalidOid);
9114 :
9115 49208 : xlrec.top_xid = GetTopTransactionId();
9116 49208 : xlrec.target_locator = relation->rd_locator;
9117 49208 : xlrec.target_tid = tup->t_self;
9118 :
9119 : /*
9120 : * If the tuple got inserted & deleted in the same TX we definitely have a
9121 : * combo CID, set cmin and cmax.
9122 : */
9123 49208 : if (hdr->t_infomask & HEAP_COMBOCID)
9124 : {
9125 : Assert(!(hdr->t_infomask & HEAP_XMAX_INVALID));
9126 : Assert(!HeapTupleHeaderXminInvalid(hdr));
9127 4048 : xlrec.cmin = HeapTupleHeaderGetCmin(hdr);
9128 4048 : xlrec.cmax = HeapTupleHeaderGetCmax(hdr);
9129 4048 : xlrec.combocid = HeapTupleHeaderGetRawCommandId(hdr);
9130 : }
9131 : /* No combo CID, so only cmin or cmax can be set by this TX */
9132 : else
9133 : {
9134 : /*
9135 : * Tuple inserted.
9136 : *
9137 : * We need to check for LOCK ONLY because multixacts might be
9138 : * transferred to the new tuple in case of FOR KEY SHARE updates in
9139 : * which case there will be an xmax, although the tuple just got
9140 : * inserted.
9141 : */
9142 58772 : if (hdr->t_infomask & HEAP_XMAX_INVALID ||
9143 13612 : HEAP_XMAX_IS_LOCKED_ONLY(hdr->t_infomask))
9144 : {
9145 31550 : xlrec.cmin = HeapTupleHeaderGetRawCommandId(hdr);
9146 31550 : xlrec.cmax = InvalidCommandId;
9147 : }
9148 : /* Tuple from a different tx updated or deleted. */
9149 : else
9150 : {
9151 13610 : xlrec.cmin = InvalidCommandId;
9152 13610 : xlrec.cmax = HeapTupleHeaderGetRawCommandId(hdr);
9153 : }
9154 45160 : xlrec.combocid = InvalidCommandId;
9155 : }
9156 :
9157 : /*
9158 : * Note that we don't need to register the buffer here, because this
9159 : * operation does not modify the page. The insert/update/delete that
9160 : * called us certainly did, but that's WAL-logged separately.
9161 : */
9162 49208 : XLogBeginInsert();
9163 49208 : XLogRegisterData(&xlrec, SizeOfHeapNewCid);
9164 :
9165 : /* will be looked at irrespective of origin */
9166 :
9167 49208 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_NEW_CID);
9168 :
9169 49208 : return recptr;
9170 : }
9171 :
9172 : /*
9173 : * Build a heap tuple representing the configured REPLICA IDENTITY to represent
9174 : * the old tuple in an UPDATE or DELETE.
9175 : *
9176 : * Returns NULL if there's no need to log an identity or if there's no suitable
9177 : * key defined.
9178 : *
9179 : * Pass key_required true if any replica identity columns changed value, or if
9180 : * any of them have any external data. Delete must always pass true.
9181 : *
9182 : * *copy is set to true if the returned tuple is a modified copy rather than
9183 : * the same tuple that was passed in.
9184 : */
9185 : static HeapTuple
9186 3674346 : ExtractReplicaIdentity(Relation relation, HeapTuple tp, bool key_required,
9187 : bool *copy)
9188 : {
9189 3674346 : TupleDesc desc = RelationGetDescr(relation);
9190 3674346 : char replident = relation->rd_rel->relreplident;
9191 : Bitmapset *idattrs;
9192 : HeapTuple key_tuple;
9193 : bool nulls[MaxHeapAttributeNumber];
9194 : Datum values[MaxHeapAttributeNumber];
9195 :
9196 3674346 : *copy = false;
9197 :
9198 3674346 : if (!RelationIsLogicallyLogged(relation))
9199 3473756 : return NULL;
9200 :
9201 200590 : if (replident == REPLICA_IDENTITY_NOTHING)
9202 462 : return NULL;
9203 :
9204 200128 : if (replident == REPLICA_IDENTITY_FULL)
9205 : {
9206 : /*
9207 : * When logging the entire old tuple, it very well could contain
9208 : * toasted columns. If so, force them to be inlined.
9209 : */
9210 394 : if (HeapTupleHasExternal(tp))
9211 : {
9212 8 : *copy = true;
9213 8 : tp = toast_flatten_tuple(tp, desc);
9214 : }
9215 394 : return tp;
9216 : }
9217 :
9218 : /* if the key isn't required and we're only logging the key, we're done */
9219 199734 : if (!key_required)
9220 93756 : return NULL;
9221 :
9222 : /* find out the replica identity columns */
9223 105978 : idattrs = RelationGetIndexAttrBitmap(relation,
9224 : INDEX_ATTR_BITMAP_IDENTITY_KEY);
9225 :
9226 : /*
9227 : * If there's no defined replica identity columns, treat as !key_required.
9228 : * (This case should not be reachable from heap_update, since that should
9229 : * calculate key_required accurately. But heap_delete just passes
9230 : * constant true for key_required, so we can hit this case in deletes.)
9231 : */
9232 105978 : if (bms_is_empty(idattrs))
9233 12042 : return NULL;
9234 :
9235 : /*
9236 : * Construct a new tuple containing only the replica identity columns,
9237 : * with nulls elsewhere. While we're at it, assert that the replica
9238 : * identity columns aren't null.
9239 : */
9240 93936 : heap_deform_tuple(tp, desc, values, nulls);
9241 :
9242 301796 : for (int i = 0; i < desc->natts; i++)
9243 : {
9244 207860 : if (bms_is_member(i + 1 - FirstLowInvalidHeapAttributeNumber,
9245 : idattrs))
9246 : Assert(!nulls[i]);
9247 : else
9248 113900 : nulls[i] = true;
9249 : }
9250 :
9251 93936 : key_tuple = heap_form_tuple(desc, values, nulls);
9252 93936 : *copy = true;
9253 :
9254 93936 : bms_free(idattrs);
9255 :
9256 : /*
9257 : * If the tuple, which by here only contains indexed columns, still has
9258 : * toasted columns, force them to be inlined. This is somewhat unlikely
9259 : * since there's limits on the size of indexed columns, so we don't
9260 : * duplicate toast_flatten_tuple()s functionality in the above loop over
9261 : * the indexed columns, even if it would be more efficient.
9262 : */
9263 93936 : if (HeapTupleHasExternal(key_tuple))
9264 : {
9265 8 : HeapTuple oldtup = key_tuple;
9266 :
9267 8 : key_tuple = toast_flatten_tuple(oldtup, desc);
9268 8 : heap_freetuple(oldtup);
9269 : }
9270 :
9271 93936 : return key_tuple;
9272 : }
9273 :
9274 : /*
9275 : * HeapCheckForSerializableConflictOut
9276 : * We are reading a tuple. If it's not visible, there may be a
9277 : * rw-conflict out with the inserter. Otherwise, if it is visible to us
9278 : * but has been deleted, there may be a rw-conflict out with the deleter.
9279 : *
9280 : * We will determine the top level xid of the writing transaction with which
9281 : * we may be in conflict, and ask CheckForSerializableConflictOut() to check
9282 : * for overlap with our own transaction.
9283 : *
9284 : * This function should be called just about anywhere in heapam.c where a
9285 : * tuple has been read. The caller must hold at least a shared lock on the
9286 : * buffer, because this function might set hint bits on the tuple. There is
9287 : * currently no known reason to call this function from an index AM.
9288 : */
9289 : void
9290 63173938 : HeapCheckForSerializableConflictOut(bool visible, Relation relation,
9291 : HeapTuple tuple, Buffer buffer,
9292 : Snapshot snapshot)
9293 : {
9294 : TransactionId xid;
9295 : HTSV_Result htsvResult;
9296 :
9297 63173938 : if (!CheckForSerializableConflictOutNeeded(relation, snapshot))
9298 63123190 : return;
9299 :
9300 : /*
9301 : * Check to see whether the tuple has been written to by a concurrent
9302 : * transaction, either to create it not visible to us, or to delete it
9303 : * while it is visible to us. The "visible" bool indicates whether the
9304 : * tuple is visible to us, while HeapTupleSatisfiesVacuum checks what else
9305 : * is going on with it.
9306 : *
9307 : * In the event of a concurrently inserted tuple that also happens to have
9308 : * been concurrently updated (by a separate transaction), the xmin of the
9309 : * tuple will be used -- not the updater's xid.
9310 : */
9311 50748 : htsvResult = HeapTupleSatisfiesVacuum(tuple, TransactionXmin, buffer);
9312 50748 : switch (htsvResult)
9313 : {
9314 49122 : case HEAPTUPLE_LIVE:
9315 49122 : if (visible)
9316 49096 : return;
9317 26 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9318 26 : break;
9319 722 : case HEAPTUPLE_RECENTLY_DEAD:
9320 : case HEAPTUPLE_DELETE_IN_PROGRESS:
9321 722 : if (visible)
9322 570 : xid = HeapTupleHeaderGetUpdateXid(tuple->t_data);
9323 : else
9324 152 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9325 :
9326 722 : if (TransactionIdPrecedes(xid, TransactionXmin))
9327 : {
9328 : /* This is like the HEAPTUPLE_DEAD case */
9329 : Assert(!visible);
9330 136 : return;
9331 : }
9332 586 : break;
9333 656 : case HEAPTUPLE_INSERT_IN_PROGRESS:
9334 656 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9335 656 : break;
9336 248 : case HEAPTUPLE_DEAD:
9337 : Assert(!visible);
9338 248 : return;
9339 0 : default:
9340 :
9341 : /*
9342 : * The only way to get to this default clause is if a new value is
9343 : * added to the enum type without adding it to this switch
9344 : * statement. That's a bug, so elog.
9345 : */
9346 0 : elog(ERROR, "unrecognized return value from HeapTupleSatisfiesVacuum: %u", htsvResult);
9347 :
9348 : /*
9349 : * In spite of having all enum values covered and calling elog on
9350 : * this default, some compilers think this is a code path which
9351 : * allows xid to be used below without initialization. Silence
9352 : * that warning.
9353 : */
9354 : xid = InvalidTransactionId;
9355 : }
9356 :
9357 : Assert(TransactionIdIsValid(xid));
9358 : Assert(TransactionIdFollowsOrEquals(xid, TransactionXmin));
9359 :
9360 : /*
9361 : * Find top level xid. Bail out if xid is too early to be a conflict, or
9362 : * if it's our own xid.
9363 : */
9364 1268 : if (TransactionIdEquals(xid, GetTopTransactionIdIfAny()))
9365 128 : return;
9366 1140 : xid = SubTransGetTopmostTransaction(xid);
9367 1140 : if (TransactionIdPrecedes(xid, TransactionXmin))
9368 0 : return;
9369 :
9370 1140 : CheckForSerializableConflictOut(relation, xid, snapshot);
9371 : }
|
