Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * heapam.c
4 : * heap access method code
5 : *
6 : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/access/heap/heapam.c
12 : *
13 : *
14 : * INTERFACE ROUTINES
15 : * heap_beginscan - begin relation scan
16 : * heap_rescan - restart a relation scan
17 : * heap_endscan - end relation scan
18 : * heap_getnext - retrieve next tuple in scan
19 : * heap_fetch - retrieve tuple with given tid
20 : * heap_insert - insert tuple into a relation
21 : * heap_multi_insert - insert multiple tuples into a relation
22 : * heap_delete - delete a tuple from a relation
23 : * heap_update - replace a tuple in a relation with another tuple
24 : *
25 : * NOTES
26 : * This file contains the heap_ routines which implement
27 : * the POSTGRES heap access method used for all POSTGRES
28 : * relations.
29 : *
30 : *-------------------------------------------------------------------------
31 : */
32 : #include "postgres.h"
33 :
34 : #include "access/heapam.h"
35 : #include "access/heaptoast.h"
36 : #include "access/hio.h"
37 : #include "access/multixact.h"
38 : #include "access/subtrans.h"
39 : #include "access/syncscan.h"
40 : #include "access/valid.h"
41 : #include "access/visibilitymap.h"
42 : #include "access/xloginsert.h"
43 : #include "catalog/pg_database.h"
44 : #include "catalog/pg_database_d.h"
45 : #include "commands/vacuum.h"
46 : #include "pgstat.h"
47 : #include "port/pg_bitutils.h"
48 : #include "storage/lmgr.h"
49 : #include "storage/predicate.h"
50 : #include "storage/proc.h"
51 : #include "storage/procarray.h"
52 : #include "utils/datum.h"
53 : #include "utils/injection_point.h"
54 : #include "utils/inval.h"
55 : #include "utils/spccache.h"
56 : #include "utils/syscache.h"
57 :
58 :
59 : static HeapTuple heap_prepare_insert(Relation relation, HeapTuple tup,
60 : TransactionId xid, CommandId cid, int options);
61 : static XLogRecPtr log_heap_update(Relation reln, Buffer oldbuf,
62 : Buffer newbuf, HeapTuple oldtup,
63 : HeapTuple newtup, HeapTuple old_key_tuple,
64 : bool all_visible_cleared, bool new_all_visible_cleared);
65 : #ifdef USE_ASSERT_CHECKING
66 : static void check_lock_if_inplace_updateable_rel(Relation relation,
67 : const ItemPointerData *otid,
68 : HeapTuple newtup);
69 : static void check_inplace_rel_lock(HeapTuple oldtup);
70 : #endif
71 : static Bitmapset *HeapDetermineColumnsInfo(Relation relation,
72 : Bitmapset *interesting_cols,
73 : Bitmapset *external_cols,
74 : HeapTuple oldtup, HeapTuple newtup,
75 : bool *has_external);
76 : static bool heap_acquire_tuplock(Relation relation, const ItemPointerData *tid,
77 : LockTupleMode mode, LockWaitPolicy wait_policy,
78 : bool *have_tuple_lock);
79 : static inline BlockNumber heapgettup_advance_block(HeapScanDesc scan,
80 : BlockNumber block,
81 : ScanDirection dir);
82 : static pg_noinline BlockNumber heapgettup_initial_block(HeapScanDesc scan,
83 : ScanDirection dir);
84 : static void compute_new_xmax_infomask(TransactionId xmax, uint16 old_infomask,
85 : uint16 old_infomask2, TransactionId add_to_xmax,
86 : LockTupleMode mode, bool is_update,
87 : TransactionId *result_xmax, uint16 *result_infomask,
88 : uint16 *result_infomask2);
89 : static TM_Result heap_lock_updated_tuple(Relation rel,
90 : uint16 prior_infomask,
91 : TransactionId prior_raw_xmax,
92 : const ItemPointerData *prior_ctid,
93 : TransactionId xid,
94 : LockTupleMode mode);
95 : static void GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
96 : uint16 *new_infomask2);
97 : static TransactionId MultiXactIdGetUpdateXid(TransactionId xmax,
98 : uint16 t_infomask);
99 : static bool DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
100 : LockTupleMode lockmode, bool *current_is_member);
101 : static void MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
102 : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
103 : int *remaining);
104 : static bool ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status,
105 : uint16 infomask, Relation rel, int *remaining,
106 : bool logLockFailure);
107 : static void index_delete_sort(TM_IndexDeleteOp *delstate);
108 : static int bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate);
109 : static XLogRecPtr log_heap_new_cid(Relation relation, HeapTuple tup);
110 : static HeapTuple ExtractReplicaIdentity(Relation relation, HeapTuple tp, bool key_required,
111 : bool *copy);
112 :
113 :
114 : /*
115 : * This table lists the heavyweight lock mode that corresponds to each tuple
116 : * lock mode, as well as one or two corresponding MultiXactStatus values:
117 : * .lockstatus to merely lock tuples, and .updstatus to update them. The
118 : * latter is set to -1 if the corresponding tuple lock mode does not allow
119 : * updating tuples -- see get_mxact_status_for_lock().
120 : *
121 : * These interact with InplaceUpdateTupleLock, an alias for ExclusiveLock.
122 : *
123 : * Don't look at lockstatus/updstatus directly! Use get_mxact_status_for_lock
124 : * instead.
125 : */
126 : static const struct
127 : {
128 : LOCKMODE hwlock;
129 : int lockstatus;
130 : int updstatus;
131 : } tupleLockExtraInfo[] =
132 :
133 : {
134 : [LockTupleKeyShare] = {
135 : .hwlock = AccessShareLock,
136 : .lockstatus = MultiXactStatusForKeyShare,
137 : /* KeyShare does not allow updating tuples */
138 : .updstatus = -1
139 : },
140 : [LockTupleShare] = {
141 : .hwlock = RowShareLock,
142 : .lockstatus = MultiXactStatusForShare,
143 : /* Share does not allow updating tuples */
144 : .updstatus = -1
145 : },
146 : [LockTupleNoKeyExclusive] = {
147 : .hwlock = ExclusiveLock,
148 : .lockstatus = MultiXactStatusForNoKeyUpdate,
149 : .updstatus = MultiXactStatusNoKeyUpdate
150 : },
151 : [LockTupleExclusive] = {
152 : .hwlock = AccessExclusiveLock,
153 : .lockstatus = MultiXactStatusForUpdate,
154 : .updstatus = MultiXactStatusUpdate
155 : }
156 : };
157 :
158 : /* Get the LOCKMODE for a given MultiXactStatus */
159 : #define LOCKMODE_from_mxstatus(status) \
160 : (tupleLockExtraInfo[TUPLOCK_from_mxstatus((status))].hwlock)
161 :
162 : /*
163 : * Acquire heavyweight locks on tuples, using a LockTupleMode strength value.
164 : * This is more readable than having every caller translate it to lock.h's
165 : * LOCKMODE.
166 : */
167 : #define LockTupleTuplock(rel, tup, mode) \
168 : LockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
169 : #define UnlockTupleTuplock(rel, tup, mode) \
170 : UnlockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
171 : #define ConditionalLockTupleTuplock(rel, tup, mode, log) \
172 : ConditionalLockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock, (log))
173 :
174 : #ifdef USE_PREFETCH
175 : /*
176 : * heap_index_delete_tuples and index_delete_prefetch_buffer use this
177 : * structure to coordinate prefetching activity
178 : */
179 : typedef struct
180 : {
181 : BlockNumber cur_hblkno;
182 : int next_item;
183 : int ndeltids;
184 : TM_IndexDelete *deltids;
185 : } IndexDeletePrefetchState;
186 : #endif
187 :
188 : /* heap_index_delete_tuples bottom-up index deletion costing constants */
189 : #define BOTTOMUP_MAX_NBLOCKS 6
190 : #define BOTTOMUP_TOLERANCE_NBLOCKS 3
191 :
192 : /*
193 : * heap_index_delete_tuples uses this when determining which heap blocks it
194 : * must visit to help its bottom-up index deletion caller
195 : */
196 : typedef struct IndexDeleteCounts
197 : {
198 : int16 npromisingtids; /* Number of "promising" TIDs in group */
199 : int16 ntids; /* Number of TIDs in group */
200 : int16 ifirsttid; /* Offset to group's first deltid */
201 : } IndexDeleteCounts;
202 :
203 : /*
204 : * This table maps tuple lock strength values for each particular
205 : * MultiXactStatus value.
206 : */
207 : static const int MultiXactStatusLock[MaxMultiXactStatus + 1] =
208 : {
209 : LockTupleKeyShare, /* ForKeyShare */
210 : LockTupleShare, /* ForShare */
211 : LockTupleNoKeyExclusive, /* ForNoKeyUpdate */
212 : LockTupleExclusive, /* ForUpdate */
213 : LockTupleNoKeyExclusive, /* NoKeyUpdate */
214 : LockTupleExclusive /* Update */
215 : };
216 :
217 : /* Get the LockTupleMode for a given MultiXactStatus */
218 : #define TUPLOCK_from_mxstatus(status) \
219 : (MultiXactStatusLock[(status)])
220 :
221 : /*
222 : * Check that we have a valid snapshot if we might need TOAST access.
223 : */
224 : static inline void
225 10604512 : AssertHasSnapshotForToast(Relation rel)
226 : {
227 : #ifdef USE_ASSERT_CHECKING
228 :
229 : /* bootstrap mode in particular breaks this rule */
230 : if (!IsNormalProcessingMode())
231 : return;
232 :
233 : /* if the relation doesn't have a TOAST table, we are good */
234 : if (!OidIsValid(rel->rd_rel->reltoastrelid))
235 : return;
236 :
237 : Assert(HaveRegisteredOrActiveSnapshot());
238 :
239 : #endif /* USE_ASSERT_CHECKING */
240 10604512 : }
241 :
242 : /* ----------------------------------------------------------------
243 : * heap support routines
244 : * ----------------------------------------------------------------
245 : */
246 :
247 : /*
248 : * Streaming read API callback for parallel sequential scans. Returns the next
249 : * block the caller wants from the read stream or InvalidBlockNumber when done.
250 : */
251 : static BlockNumber
252 101761 : heap_scan_stream_read_next_parallel(ReadStream *stream,
253 : void *callback_private_data,
254 : void *per_buffer_data)
255 : {
256 101761 : HeapScanDesc scan = (HeapScanDesc) callback_private_data;
257 :
258 : Assert(ScanDirectionIsForward(scan->rs_dir));
259 : Assert(scan->rs_base.rs_parallel);
260 :
261 101761 : if (unlikely(!scan->rs_inited))
262 : {
263 : /* parallel scan */
264 1698 : table_block_parallelscan_startblock_init(scan->rs_base.rs_rd,
265 1698 : scan->rs_parallelworkerdata,
266 1698 : (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel,
267 : scan->rs_startblock,
268 : scan->rs_numblocks);
269 :
270 : /* may return InvalidBlockNumber if there are no more blocks */
271 3396 : scan->rs_prefetch_block = table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
272 1698 : scan->rs_parallelworkerdata,
273 1698 : (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel);
274 1698 : scan->rs_inited = true;
275 : }
276 : else
277 : {
278 100063 : scan->rs_prefetch_block = table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
279 100063 : scan->rs_parallelworkerdata, (ParallelBlockTableScanDesc)
280 100063 : scan->rs_base.rs_parallel);
281 : }
282 :
283 101761 : return scan->rs_prefetch_block;
284 : }
285 :
286 : /*
287 : * Streaming read API callback for serial sequential and TID range scans.
288 : * Returns the next block the caller wants from the read stream or
289 : * InvalidBlockNumber when done.
290 : */
291 : static BlockNumber
292 4088702 : heap_scan_stream_read_next_serial(ReadStream *stream,
293 : void *callback_private_data,
294 : void *per_buffer_data)
295 : {
296 4088702 : HeapScanDesc scan = (HeapScanDesc) callback_private_data;
297 :
298 4088702 : if (unlikely(!scan->rs_inited))
299 : {
300 1203516 : scan->rs_prefetch_block = heapgettup_initial_block(scan, scan->rs_dir);
301 1203516 : scan->rs_inited = true;
302 : }
303 : else
304 2885186 : scan->rs_prefetch_block = heapgettup_advance_block(scan,
305 : scan->rs_prefetch_block,
306 : scan->rs_dir);
307 :
308 4088702 : return scan->rs_prefetch_block;
309 : }
310 :
311 : /*
312 : * Read stream API callback for bitmap heap scans.
313 : * Returns the next block the caller wants from the read stream or
314 : * InvalidBlockNumber when done.
315 : */
316 : static BlockNumber
317 218245 : bitmapheap_stream_read_next(ReadStream *pgsr, void *private_data,
318 : void *per_buffer_data)
319 : {
320 218245 : TBMIterateResult *tbmres = per_buffer_data;
321 218245 : BitmapHeapScanDesc bscan = (BitmapHeapScanDesc) private_data;
322 218245 : HeapScanDesc hscan = (HeapScanDesc) bscan;
323 218245 : TableScanDesc sscan = &hscan->rs_base;
324 :
325 : for (;;)
326 : {
327 218245 : CHECK_FOR_INTERRUPTS();
328 :
329 : /* no more entries in the bitmap */
330 218245 : if (!tbm_iterate(&sscan->st.rs_tbmiterator, tbmres))
331 13262 : return InvalidBlockNumber;
332 :
333 : /*
334 : * Ignore any claimed entries past what we think is the end of the
335 : * relation. It may have been extended after the start of our scan (we
336 : * only hold an AccessShareLock, and it could be inserts from this
337 : * backend). We don't take this optimization in SERIALIZABLE
338 : * isolation though, as we need to examine all invisible tuples
339 : * reachable by the index.
340 : */
341 204983 : if (!IsolationIsSerializable() &&
342 204874 : tbmres->blockno >= hscan->rs_nblocks)
343 0 : continue;
344 :
345 204983 : return tbmres->blockno;
346 : }
347 :
348 : /* not reachable */
349 : Assert(false);
350 : }
351 :
352 : /* ----------------
353 : * initscan - scan code common to heap_beginscan and heap_rescan
354 : * ----------------
355 : */
356 : static void
357 1228580 : initscan(HeapScanDesc scan, ScanKey key, bool keep_startblock)
358 : {
359 1228580 : ParallelBlockTableScanDesc bpscan = NULL;
360 : bool allow_strat;
361 : bool allow_sync;
362 :
363 : /*
364 : * Determine the number of blocks we have to scan.
365 : *
366 : * It is sufficient to do this once at scan start, since any tuples added
367 : * while the scan is in progress will be invisible to my snapshot anyway.
368 : * (That is not true when using a non-MVCC snapshot. However, we couldn't
369 : * guarantee to return tuples added after scan start anyway, since they
370 : * might go into pages we already scanned. To guarantee consistent
371 : * results for a non-MVCC snapshot, the caller must hold some higher-level
372 : * lock that ensures the interesting tuple(s) won't change.)
373 : */
374 1228580 : if (scan->rs_base.rs_parallel != NULL)
375 : {
376 2233 : bpscan = (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel;
377 2233 : scan->rs_nblocks = bpscan->phs_nblocks;
378 : }
379 : else
380 1226347 : scan->rs_nblocks = RelationGetNumberOfBlocks(scan->rs_base.rs_rd);
381 :
382 : /*
383 : * If the table is large relative to NBuffers, use a bulk-read access
384 : * strategy and enable synchronized scanning (see syncscan.c). Although
385 : * the thresholds for these features could be different, we make them the
386 : * same so that there are only two behaviors to tune rather than four.
387 : * (However, some callers need to be able to disable one or both of these
388 : * behaviors, independently of the size of the table; also there is a GUC
389 : * variable that can disable synchronized scanning.)
390 : *
391 : * Note that table_block_parallelscan_initialize has a very similar test;
392 : * if you change this, consider changing that one, too.
393 : */
394 1228578 : if (!RelationUsesLocalBuffers(scan->rs_base.rs_rd) &&
395 1221243 : scan->rs_nblocks > NBuffers / 4)
396 : {
397 13772 : allow_strat = (scan->rs_base.rs_flags & SO_ALLOW_STRAT) != 0;
398 13772 : allow_sync = (scan->rs_base.rs_flags & SO_ALLOW_SYNC) != 0;
399 : }
400 : else
401 1214806 : allow_strat = allow_sync = false;
402 :
403 1228578 : if (allow_strat)
404 : {
405 : /* During a rescan, keep the previous strategy object. */
406 12447 : if (scan->rs_strategy == NULL)
407 12264 : scan->rs_strategy = GetAccessStrategy(BAS_BULKREAD);
408 : }
409 : else
410 : {
411 1216131 : if (scan->rs_strategy != NULL)
412 0 : FreeAccessStrategy(scan->rs_strategy);
413 1216131 : scan->rs_strategy = NULL;
414 : }
415 :
416 1228578 : if (scan->rs_base.rs_parallel != NULL)
417 : {
418 : /* For parallel scan, believe whatever ParallelTableScanDesc says. */
419 2233 : if (scan->rs_base.rs_parallel->phs_syncscan)
420 2 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
421 : else
422 2231 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
423 :
424 : /*
425 : * If not rescanning, initialize the startblock. Finding the actual
426 : * start location is done in table_block_parallelscan_startblock_init,
427 : * based on whether an alternative start location has been set with
428 : * heap_setscanlimits, or using the syncscan location, when syncscan
429 : * is enabled.
430 : */
431 2233 : if (!keep_startblock)
432 2119 : scan->rs_startblock = InvalidBlockNumber;
433 : }
434 : else
435 : {
436 1226345 : if (keep_startblock)
437 : {
438 : /*
439 : * When rescanning, we want to keep the previous startblock
440 : * setting, so that rewinding a cursor doesn't generate surprising
441 : * results. Reset the active syncscan setting, though.
442 : */
443 823303 : if (allow_sync && synchronize_seqscans)
444 50 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
445 : else
446 823253 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
447 : }
448 403042 : else if (allow_sync && synchronize_seqscans)
449 : {
450 69 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
451 69 : scan->rs_startblock = ss_get_location(scan->rs_base.rs_rd, scan->rs_nblocks);
452 : }
453 : else
454 : {
455 402973 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
456 402973 : scan->rs_startblock = 0;
457 : }
458 : }
459 :
460 1228578 : scan->rs_numblocks = InvalidBlockNumber;
461 1228578 : scan->rs_inited = false;
462 1228578 : scan->rs_ctup.t_data = NULL;
463 1228578 : ItemPointerSetInvalid(&scan->rs_ctup.t_self);
464 1228578 : scan->rs_cbuf = InvalidBuffer;
465 1228578 : scan->rs_cblock = InvalidBlockNumber;
466 1228578 : scan->rs_ntuples = 0;
467 1228578 : scan->rs_cindex = 0;
468 :
469 : /*
470 : * Initialize to ForwardScanDirection because it is most common and
471 : * because heap scans go forward before going backward (e.g. CURSORs).
472 : */
473 1228578 : scan->rs_dir = ForwardScanDirection;
474 1228578 : scan->rs_prefetch_block = InvalidBlockNumber;
475 :
476 : /* page-at-a-time fields are always invalid when not rs_inited */
477 :
478 : /*
479 : * copy the scan key, if appropriate
480 : */
481 1228578 : if (key != NULL && scan->rs_base.rs_nkeys > 0)
482 229536 : memcpy(scan->rs_base.rs_key, key, scan->rs_base.rs_nkeys * sizeof(ScanKeyData));
483 :
484 : /*
485 : * Currently, we only have a stats counter for sequential heap scans (but
486 : * e.g for bitmap scans the underlying bitmap index scans will be counted,
487 : * and for sample scans we update stats for tuple fetches).
488 : */
489 1228578 : if (scan->rs_base.rs_flags & SO_TYPE_SEQSCAN)
490 1204777 : pgstat_count_heap_scan(scan->rs_base.rs_rd);
491 1228578 : }
492 :
493 : /*
494 : * heap_setscanlimits - restrict range of a heapscan
495 : *
496 : * startBlk is the page to start at
497 : * numBlks is number of pages to scan (InvalidBlockNumber means "all")
498 : */
499 : void
500 2881 : heap_setscanlimits(TableScanDesc sscan, BlockNumber startBlk, BlockNumber numBlks)
501 : {
502 2881 : HeapScanDesc scan = (HeapScanDesc) sscan;
503 :
504 : Assert(!scan->rs_inited); /* else too late to change */
505 : /* else rs_startblock is significant */
506 : Assert(!(scan->rs_base.rs_flags & SO_ALLOW_SYNC));
507 :
508 : /* Check startBlk is valid (but allow case of zero blocks...) */
509 : Assert(startBlk == 0 || startBlk < scan->rs_nblocks);
510 :
511 2881 : scan->rs_startblock = startBlk;
512 2881 : scan->rs_numblocks = numBlks;
513 2881 : }
514 :
515 : /*
516 : * Per-tuple loop for heap_prepare_pagescan(). Pulled out so it can be called
517 : * multiple times, with constant arguments for all_visible,
518 : * check_serializable.
519 : */
520 : pg_attribute_always_inline
521 : static int
522 2874562 : page_collect_tuples(HeapScanDesc scan, Snapshot snapshot,
523 : Page page, Buffer buffer,
524 : BlockNumber block, int lines,
525 : bool all_visible, bool check_serializable)
526 : {
527 2874562 : Oid relid = RelationGetRelid(scan->rs_base.rs_rd);
528 2874562 : int ntup = 0;
529 2874562 : int nvis = 0;
530 : BatchMVCCState batchmvcc;
531 :
532 : /* page at a time should have been disabled otherwise */
533 : Assert(IsMVCCSnapshot(snapshot));
534 :
535 : /* first find all tuples on the page */
536 146553167 : for (OffsetNumber lineoff = FirstOffsetNumber; lineoff <= lines; lineoff++)
537 : {
538 143678605 : ItemId lpp = PageGetItemId(page, lineoff);
539 : HeapTuple tup;
540 :
541 143678605 : if (unlikely(!ItemIdIsNormal(lpp)))
542 30394695 : continue;
543 :
544 : /*
545 : * If the page is not all-visible or we need to check serializability,
546 : * maintain enough state to be able to refind the tuple efficiently,
547 : * without again first needing to fetch the item and then via that the
548 : * tuple.
549 : */
550 113283910 : if (!all_visible || check_serializable)
551 : {
552 67077683 : tup = &batchmvcc.tuples[ntup];
553 :
554 67077683 : tup->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
555 67077683 : tup->t_len = ItemIdGetLength(lpp);
556 67077683 : tup->t_tableOid = relid;
557 67077683 : ItemPointerSet(&(tup->t_self), block, lineoff);
558 : }
559 :
560 : /*
561 : * If the page is all visible, these fields otherwise won't be
562 : * populated in loop below.
563 : */
564 113283910 : if (all_visible)
565 : {
566 46206227 : if (check_serializable)
567 : {
568 0 : batchmvcc.visible[ntup] = true;
569 : }
570 46206227 : scan->rs_vistuples[ntup] = lineoff;
571 : }
572 :
573 113283910 : ntup++;
574 : }
575 :
576 : Assert(ntup <= MaxHeapTuplesPerPage);
577 :
578 : /*
579 : * Unless the page is all visible, test visibility for all tuples one go.
580 : * That is considerably more efficient than calling
581 : * HeapTupleSatisfiesMVCC() one-by-one.
582 : */
583 2874562 : if (all_visible)
584 1092183 : nvis = ntup;
585 : else
586 1782379 : nvis = HeapTupleSatisfiesMVCCBatch(snapshot, buffer,
587 : ntup,
588 : &batchmvcc,
589 1782379 : scan->rs_vistuples);
590 :
591 : /*
592 : * So far we don't have batch API for testing serializabilty, so do so
593 : * one-by-one.
594 : */
595 2874562 : if (check_serializable)
596 : {
597 2045 : for (int i = 0; i < ntup; i++)
598 : {
599 1421 : HeapCheckForSerializableConflictOut(batchmvcc.visible[i],
600 : scan->rs_base.rs_rd,
601 : &batchmvcc.tuples[i],
602 : buffer, snapshot);
603 : }
604 : }
605 :
606 2874554 : return nvis;
607 : }
608 :
609 : /*
610 : * heap_prepare_pagescan - Prepare current scan page to be scanned in pagemode
611 : *
612 : * Preparation currently consists of 1. prune the scan's rs_cbuf page, and 2.
613 : * fill the rs_vistuples[] array with the OffsetNumbers of visible tuples.
614 : */
615 : void
616 2874562 : heap_prepare_pagescan(TableScanDesc sscan)
617 : {
618 2874562 : HeapScanDesc scan = (HeapScanDesc) sscan;
619 2874562 : Buffer buffer = scan->rs_cbuf;
620 2874562 : BlockNumber block = scan->rs_cblock;
621 : Snapshot snapshot;
622 : Page page;
623 : int lines;
624 : bool all_visible;
625 : bool check_serializable;
626 :
627 : Assert(BufferGetBlockNumber(buffer) == block);
628 :
629 : /* ensure we're not accidentally being used when not in pagemode */
630 : Assert(scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE);
631 2874562 : snapshot = scan->rs_base.rs_snapshot;
632 :
633 : /*
634 : * Prune and repair fragmentation for the whole page, if possible.
635 : */
636 2874562 : heap_page_prune_opt(scan->rs_base.rs_rd, buffer);
637 :
638 : /*
639 : * We must hold share lock on the buffer content while examining tuple
640 : * visibility. Afterwards, however, the tuples we have found to be
641 : * visible are guaranteed good as long as we hold the buffer pin.
642 : */
643 2874562 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
644 :
645 2874562 : page = BufferGetPage(buffer);
646 2874562 : lines = PageGetMaxOffsetNumber(page);
647 :
648 : /*
649 : * If the all-visible flag indicates that all tuples on the page are
650 : * visible to everyone, we can skip the per-tuple visibility tests.
651 : *
652 : * Note: In hot standby, a tuple that's already visible to all
653 : * transactions on the primary might still be invisible to a read-only
654 : * transaction in the standby. We partly handle this problem by tracking
655 : * the minimum xmin of visible tuples as the cut-off XID while marking a
656 : * page all-visible on the primary and WAL log that along with the
657 : * visibility map SET operation. In hot standby, we wait for (or abort)
658 : * all transactions that can potentially may not see one or more tuples on
659 : * the page. That's how index-only scans work fine in hot standby. A
660 : * crucial difference between index-only scans and heap scans is that the
661 : * index-only scan completely relies on the visibility map where as heap
662 : * scan looks at the page-level PD_ALL_VISIBLE flag. We are not sure if
663 : * the page-level flag can be trusted in the same way, because it might
664 : * get propagated somehow without being explicitly WAL-logged, e.g. via a
665 : * full page write. Until we can prove that beyond doubt, let's check each
666 : * tuple for visibility the hard way.
667 : */
668 2874562 : all_visible = PageIsAllVisible(page) && !snapshot->takenDuringRecovery;
669 : check_serializable =
670 2874562 : CheckForSerializableConflictOutNeeded(scan->rs_base.rs_rd, snapshot);
671 :
672 : /*
673 : * We call page_collect_tuples() with constant arguments, to get the
674 : * compiler to constant fold the constant arguments. Separate calls with
675 : * constant arguments, rather than variables, are needed on several
676 : * compilers to actually perform constant folding.
677 : */
678 2874562 : if (likely(all_visible))
679 : {
680 1092183 : if (likely(!check_serializable))
681 1092183 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
682 : block, lines, true, false);
683 : else
684 0 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
685 : block, lines, true, true);
686 : }
687 : else
688 : {
689 1782379 : if (likely(!check_serializable))
690 1781747 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
691 : block, lines, false, false);
692 : else
693 632 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
694 : block, lines, false, true);
695 : }
696 :
697 2874554 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
698 2874554 : }
699 :
700 : /*
701 : * heap_fetch_next_buffer - read and pin the next block from MAIN_FORKNUM.
702 : *
703 : * Read the next block of the scan relation from the read stream and save it
704 : * in the scan descriptor. It is already pinned.
705 : */
706 : static inline void
707 3998321 : heap_fetch_next_buffer(HeapScanDesc scan, ScanDirection dir)
708 : {
709 : Assert(scan->rs_read_stream);
710 :
711 : /* release previous scan buffer, if any */
712 3998321 : if (BufferIsValid(scan->rs_cbuf))
713 : {
714 2793106 : ReleaseBuffer(scan->rs_cbuf);
715 2793106 : scan->rs_cbuf = InvalidBuffer;
716 : }
717 :
718 : /*
719 : * Be sure to check for interrupts at least once per page. Checks at
720 : * higher code levels won't be able to stop a seqscan that encounters many
721 : * pages' worth of consecutive dead tuples.
722 : */
723 3998321 : CHECK_FOR_INTERRUPTS();
724 :
725 : /*
726 : * If the scan direction is changing, reset the prefetch block to the
727 : * current block. Otherwise, we will incorrectly prefetch the blocks
728 : * between the prefetch block and the current block again before
729 : * prefetching blocks in the new, correct scan direction.
730 : */
731 3998319 : if (unlikely(scan->rs_dir != dir))
732 : {
733 77 : scan->rs_prefetch_block = scan->rs_cblock;
734 77 : read_stream_reset(scan->rs_read_stream);
735 : }
736 :
737 3998319 : scan->rs_dir = dir;
738 :
739 3998319 : scan->rs_cbuf = read_stream_next_buffer(scan->rs_read_stream, NULL);
740 3998293 : if (BufferIsValid(scan->rs_cbuf))
741 2962937 : scan->rs_cblock = BufferGetBlockNumber(scan->rs_cbuf);
742 3998293 : }
743 :
744 : /*
745 : * heapgettup_initial_block - return the first BlockNumber to scan
746 : *
747 : * Returns InvalidBlockNumber when there are no blocks to scan. This can
748 : * occur with empty tables and in parallel scans when parallel workers get all
749 : * of the pages before we can get a chance to get our first page.
750 : */
751 : static pg_noinline BlockNumber
752 1203516 : heapgettup_initial_block(HeapScanDesc scan, ScanDirection dir)
753 : {
754 : Assert(!scan->rs_inited);
755 : Assert(scan->rs_base.rs_parallel == NULL);
756 :
757 : /* When there are no pages to scan, return InvalidBlockNumber */
758 1203516 : if (scan->rs_nblocks == 0 || scan->rs_numblocks == 0)
759 706894 : return InvalidBlockNumber;
760 :
761 496622 : if (ScanDirectionIsForward(dir))
762 : {
763 496590 : return scan->rs_startblock;
764 : }
765 : else
766 : {
767 : /*
768 : * Disable reporting to syncscan logic in a backwards scan; it's not
769 : * very likely anyone else is doing the same thing at the same time,
770 : * and much more likely that we'll just bollix things for forward
771 : * scanners.
772 : */
773 32 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
774 :
775 : /*
776 : * Start from last page of the scan. Ensure we take into account
777 : * rs_numblocks if it's been adjusted by heap_setscanlimits().
778 : */
779 32 : if (scan->rs_numblocks != InvalidBlockNumber)
780 3 : return (scan->rs_startblock + scan->rs_numblocks - 1) % scan->rs_nblocks;
781 :
782 29 : if (scan->rs_startblock > 0)
783 0 : return scan->rs_startblock - 1;
784 :
785 29 : return scan->rs_nblocks - 1;
786 : }
787 : }
788 :
789 :
790 : /*
791 : * heapgettup_start_page - helper function for heapgettup()
792 : *
793 : * Return the next page to scan based on the scan->rs_cbuf and set *linesleft
794 : * to the number of tuples on this page. Also set *lineoff to the first
795 : * offset to scan with forward scans getting the first offset and backward
796 : * getting the final offset on the page.
797 : */
798 : static Page
799 92650 : heapgettup_start_page(HeapScanDesc scan, ScanDirection dir, int *linesleft,
800 : OffsetNumber *lineoff)
801 : {
802 : Page page;
803 :
804 : Assert(scan->rs_inited);
805 : Assert(BufferIsValid(scan->rs_cbuf));
806 :
807 : /* Caller is responsible for ensuring buffer is locked if needed */
808 92650 : page = BufferGetPage(scan->rs_cbuf);
809 :
810 92650 : *linesleft = PageGetMaxOffsetNumber(page) - FirstOffsetNumber + 1;
811 :
812 92650 : if (ScanDirectionIsForward(dir))
813 92650 : *lineoff = FirstOffsetNumber;
814 : else
815 0 : *lineoff = (OffsetNumber) (*linesleft);
816 :
817 : /* lineoff now references the physically previous or next tid */
818 92650 : return page;
819 : }
820 :
821 :
822 : /*
823 : * heapgettup_continue_page - helper function for heapgettup()
824 : *
825 : * Return the next page to scan based on the scan->rs_cbuf and set *linesleft
826 : * to the number of tuples left to scan on this page. Also set *lineoff to
827 : * the next offset to scan according to the ScanDirection in 'dir'.
828 : */
829 : static inline Page
830 7486857 : heapgettup_continue_page(HeapScanDesc scan, ScanDirection dir, int *linesleft,
831 : OffsetNumber *lineoff)
832 : {
833 : Page page;
834 :
835 : Assert(scan->rs_inited);
836 : Assert(BufferIsValid(scan->rs_cbuf));
837 :
838 : /* Caller is responsible for ensuring buffer is locked if needed */
839 7486857 : page = BufferGetPage(scan->rs_cbuf);
840 :
841 7486857 : if (ScanDirectionIsForward(dir))
842 : {
843 7486857 : *lineoff = OffsetNumberNext(scan->rs_coffset);
844 7486857 : *linesleft = PageGetMaxOffsetNumber(page) - (*lineoff) + 1;
845 : }
846 : else
847 : {
848 : /*
849 : * The previous returned tuple may have been vacuumed since the
850 : * previous scan when we use a non-MVCC snapshot, so we must
851 : * re-establish the lineoff <= PageGetMaxOffsetNumber(page) invariant
852 : */
853 0 : *lineoff = Min(PageGetMaxOffsetNumber(page), OffsetNumberPrev(scan->rs_coffset));
854 0 : *linesleft = *lineoff;
855 : }
856 :
857 : /* lineoff now references the physically previous or next tid */
858 7486857 : return page;
859 : }
860 :
861 : /*
862 : * heapgettup_advance_block - helper for heap_fetch_next_buffer()
863 : *
864 : * Given the current block number, the scan direction, and various information
865 : * contained in the scan descriptor, calculate the BlockNumber to scan next
866 : * and return it. If there are no further blocks to scan, return
867 : * InvalidBlockNumber to indicate this fact to the caller.
868 : *
869 : * This should not be called to determine the initial block number -- only for
870 : * subsequent blocks.
871 : *
872 : * This also adjusts rs_numblocks when a limit has been imposed by
873 : * heap_setscanlimits().
874 : */
875 : static inline BlockNumber
876 2885186 : heapgettup_advance_block(HeapScanDesc scan, BlockNumber block, ScanDirection dir)
877 : {
878 : Assert(scan->rs_base.rs_parallel == NULL);
879 :
880 2885186 : if (likely(ScanDirectionIsForward(dir)))
881 : {
882 2885127 : block++;
883 :
884 : /* wrap back to the start of the heap */
885 2885127 : if (block >= scan->rs_nblocks)
886 391453 : block = 0;
887 :
888 : /*
889 : * Report our new scan position for synchronization purposes. We don't
890 : * do that when moving backwards, however. That would just mess up any
891 : * other forward-moving scanners.
892 : *
893 : * Note: we do this before checking for end of scan so that the final
894 : * state of the position hint is back at the start of the rel. That's
895 : * not strictly necessary, but otherwise when you run the same query
896 : * multiple times the starting position would shift a little bit
897 : * backwards on every invocation, which is confusing. We don't
898 : * guarantee any specific ordering in general, though.
899 : */
900 2885127 : if (scan->rs_base.rs_flags & SO_ALLOW_SYNC)
901 11015 : ss_report_location(scan->rs_base.rs_rd, block);
902 :
903 : /* we're done if we're back at where we started */
904 2885127 : if (block == scan->rs_startblock)
905 391412 : return InvalidBlockNumber;
906 :
907 : /* check if the limit imposed by heap_setscanlimits() is met */
908 2493715 : if (scan->rs_numblocks != InvalidBlockNumber)
909 : {
910 2486 : if (--scan->rs_numblocks == 0)
911 1548 : return InvalidBlockNumber;
912 : }
913 :
914 2492167 : return block;
915 : }
916 : else
917 : {
918 : /* we're done if the last block is the start position */
919 59 : if (block == scan->rs_startblock)
920 59 : return InvalidBlockNumber;
921 :
922 : /* check if the limit imposed by heap_setscanlimits() is met */
923 0 : if (scan->rs_numblocks != InvalidBlockNumber)
924 : {
925 0 : if (--scan->rs_numblocks == 0)
926 0 : return InvalidBlockNumber;
927 : }
928 :
929 : /* wrap to the end of the heap when the last page was page 0 */
930 0 : if (block == 0)
931 0 : block = scan->rs_nblocks;
932 :
933 0 : block--;
934 :
935 0 : return block;
936 : }
937 : }
938 :
939 : /* ----------------
940 : * heapgettup - fetch next heap tuple
941 : *
942 : * Initialize the scan if not already done; then advance to the next
943 : * tuple as indicated by "dir"; return the next tuple in scan->rs_ctup,
944 : * or set scan->rs_ctup.t_data = NULL if no more tuples.
945 : *
946 : * Note: the reason nkeys/key are passed separately, even though they are
947 : * kept in the scan descriptor, is that the caller may not want us to check
948 : * the scankeys.
949 : *
950 : * Note: when we fall off the end of the scan in either direction, we
951 : * reset rs_inited. This means that a further request with the same
952 : * scan direction will restart the scan, which is a bit odd, but a
953 : * request with the opposite scan direction will start a fresh scan
954 : * in the proper direction. The latter is required behavior for cursors,
955 : * while the former case is generally undefined behavior in Postgres
956 : * so we don't care too much.
957 : * ----------------
958 : */
959 : static void
960 7507934 : heapgettup(HeapScanDesc scan,
961 : ScanDirection dir,
962 : int nkeys,
963 : ScanKey key)
964 : {
965 7507934 : HeapTuple tuple = &(scan->rs_ctup);
966 : Page page;
967 : OffsetNumber lineoff;
968 : int linesleft;
969 :
970 7507934 : if (likely(scan->rs_inited))
971 : {
972 : /* continue from previously returned page/tuple */
973 7486857 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
974 7486857 : page = heapgettup_continue_page(scan, dir, &linesleft, &lineoff);
975 7486857 : goto continue_page;
976 : }
977 :
978 : /*
979 : * advance the scan until we find a qualifying tuple or run out of stuff
980 : * to scan
981 : */
982 : while (true)
983 : {
984 113579 : heap_fetch_next_buffer(scan, dir);
985 :
986 : /* did we run out of blocks to scan? */
987 113579 : if (!BufferIsValid(scan->rs_cbuf))
988 20929 : break;
989 :
990 : Assert(BufferGetBlockNumber(scan->rs_cbuf) == scan->rs_cblock);
991 :
992 92650 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
993 92650 : page = heapgettup_start_page(scan, dir, &linesleft, &lineoff);
994 7579507 : continue_page:
995 :
996 : /*
997 : * Only continue scanning the page while we have lines left.
998 : *
999 : * Note that this protects us from accessing line pointers past
1000 : * PageGetMaxOffsetNumber(); both for forward scans when we resume the
1001 : * table scan, and for when we start scanning a new page.
1002 : */
1003 7619646 : for (; linesleft > 0; linesleft--, lineoff += dir)
1004 : {
1005 : bool visible;
1006 7527144 : ItemId lpp = PageGetItemId(page, lineoff);
1007 :
1008 7527144 : if (!ItemIdIsNormal(lpp))
1009 34921 : continue;
1010 :
1011 7492223 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
1012 7492223 : tuple->t_len = ItemIdGetLength(lpp);
1013 7492223 : ItemPointerSet(&(tuple->t_self), scan->rs_cblock, lineoff);
1014 :
1015 7492223 : visible = HeapTupleSatisfiesVisibility(tuple,
1016 : scan->rs_base.rs_snapshot,
1017 : scan->rs_cbuf);
1018 :
1019 7492223 : HeapCheckForSerializableConflictOut(visible, scan->rs_base.rs_rd,
1020 : tuple, scan->rs_cbuf,
1021 : scan->rs_base.rs_snapshot);
1022 :
1023 : /* skip tuples not visible to this snapshot */
1024 7492223 : if (!visible)
1025 5218 : continue;
1026 :
1027 : /* skip any tuples that don't match the scan key */
1028 7487005 : if (key != NULL &&
1029 0 : !HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
1030 : nkeys, key))
1031 0 : continue;
1032 :
1033 7487005 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1034 7487005 : scan->rs_coffset = lineoff;
1035 7487005 : return;
1036 : }
1037 :
1038 : /*
1039 : * if we get here, it means we've exhausted the items on this page and
1040 : * it's time to move to the next.
1041 : */
1042 92502 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1043 : }
1044 :
1045 : /* end of scan */
1046 20929 : if (BufferIsValid(scan->rs_cbuf))
1047 0 : ReleaseBuffer(scan->rs_cbuf);
1048 :
1049 20929 : scan->rs_cbuf = InvalidBuffer;
1050 20929 : scan->rs_cblock = InvalidBlockNumber;
1051 20929 : scan->rs_prefetch_block = InvalidBlockNumber;
1052 20929 : tuple->t_data = NULL;
1053 20929 : scan->rs_inited = false;
1054 : }
1055 :
1056 : /* ----------------
1057 : * heapgettup_pagemode - fetch next heap tuple in page-at-a-time mode
1058 : *
1059 : * Same API as heapgettup, but used in page-at-a-time mode
1060 : *
1061 : * The internal logic is much the same as heapgettup's too, but there are some
1062 : * differences: we do not take the buffer content lock (that only needs to
1063 : * happen inside heap_prepare_pagescan), and we iterate through just the
1064 : * tuples listed in rs_vistuples[] rather than all tuples on the page. Notice
1065 : * that lineindex is 0-based, where the corresponding loop variable lineoff in
1066 : * heapgettup is 1-based.
1067 : * ----------------
1068 : */
1069 : static void
1070 52262697 : heapgettup_pagemode(HeapScanDesc scan,
1071 : ScanDirection dir,
1072 : int nkeys,
1073 : ScanKey key)
1074 : {
1075 52262697 : HeapTuple tuple = &(scan->rs_ctup);
1076 : Page page;
1077 : uint32 lineindex;
1078 : uint32 linesleft;
1079 :
1080 52262697 : if (likely(scan->rs_inited))
1081 : {
1082 : /* continue from previously returned page/tuple */
1083 51078559 : page = BufferGetPage(scan->rs_cbuf);
1084 :
1085 51078559 : lineindex = scan->rs_cindex + dir;
1086 51078559 : if (ScanDirectionIsForward(dir))
1087 51078230 : linesleft = scan->rs_ntuples - lineindex;
1088 : else
1089 329 : linesleft = scan->rs_cindex;
1090 : /* lineindex now references the next or previous visible tid */
1091 :
1092 51078559 : goto continue_page;
1093 : }
1094 :
1095 : /*
1096 : * advance the scan until we find a qualifying tuple or run out of stuff
1097 : * to scan
1098 : */
1099 : while (true)
1100 : {
1101 3884742 : heap_fetch_next_buffer(scan, dir);
1102 :
1103 : /* did we run out of blocks to scan? */
1104 3884714 : if (!BufferIsValid(scan->rs_cbuf))
1105 1014427 : break;
1106 :
1107 : Assert(BufferGetBlockNumber(scan->rs_cbuf) == scan->rs_cblock);
1108 :
1109 : /* prune the page and determine visible tuple offsets */
1110 2870287 : heap_prepare_pagescan((TableScanDesc) scan);
1111 2870279 : page = BufferGetPage(scan->rs_cbuf);
1112 2870279 : linesleft = scan->rs_ntuples;
1113 2870279 : lineindex = ScanDirectionIsForward(dir) ? 0 : linesleft - 1;
1114 :
1115 : /* block is the same for all tuples, set it once outside the loop */
1116 2870279 : ItemPointerSetBlockNumber(&tuple->t_self, scan->rs_cblock);
1117 :
1118 : /* lineindex now references the next or previous visible tid */
1119 53948838 : continue_page:
1120 :
1121 104531195 : for (; linesleft > 0; linesleft--, lineindex += dir)
1122 : {
1123 : ItemId lpp;
1124 : OffsetNumber lineoff;
1125 :
1126 : Assert(lineindex < scan->rs_ntuples);
1127 101830591 : lineoff = scan->rs_vistuples[lineindex];
1128 101830591 : lpp = PageGetItemId(page, lineoff);
1129 : Assert(ItemIdIsNormal(lpp));
1130 :
1131 101830591 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
1132 101830591 : tuple->t_len = ItemIdGetLength(lpp);
1133 101830591 : ItemPointerSetOffsetNumber(&tuple->t_self, lineoff);
1134 :
1135 : /* skip any tuples that don't match the scan key */
1136 101830591 : if (key != NULL &&
1137 50996971 : !HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
1138 : nkeys, key))
1139 50582357 : continue;
1140 :
1141 51248234 : scan->rs_cindex = lineindex;
1142 51248234 : return;
1143 : }
1144 : }
1145 :
1146 : /* end of scan */
1147 1014427 : if (BufferIsValid(scan->rs_cbuf))
1148 0 : ReleaseBuffer(scan->rs_cbuf);
1149 1014427 : scan->rs_cbuf = InvalidBuffer;
1150 1014427 : scan->rs_cblock = InvalidBlockNumber;
1151 1014427 : scan->rs_prefetch_block = InvalidBlockNumber;
1152 1014427 : tuple->t_data = NULL;
1153 1014427 : scan->rs_inited = false;
1154 : }
1155 :
1156 :
1157 : /* ----------------------------------------------------------------
1158 : * heap access method interface
1159 : * ----------------------------------------------------------------
1160 : */
1161 :
1162 :
1163 : TableScanDesc
1164 405163 : heap_beginscan(Relation relation, Snapshot snapshot,
1165 : int nkeys, ScanKey key,
1166 : ParallelTableScanDesc parallel_scan,
1167 : uint32 flags)
1168 : {
1169 : HeapScanDesc scan;
1170 :
1171 : /*
1172 : * increment relation ref count while scanning relation
1173 : *
1174 : * This is just to make really sure the relcache entry won't go away while
1175 : * the scan has a pointer to it. Caller should be holding the rel open
1176 : * anyway, so this is redundant in all normal scenarios...
1177 : */
1178 405163 : RelationIncrementReferenceCount(relation);
1179 :
1180 : /*
1181 : * allocate and initialize scan descriptor
1182 : */
1183 405163 : if (flags & SO_TYPE_BITMAPSCAN)
1184 : {
1185 11170 : BitmapHeapScanDesc bscan = palloc_object(BitmapHeapScanDescData);
1186 :
1187 : /*
1188 : * Bitmap Heap scans do not have any fields that a normal Heap Scan
1189 : * does not have, so no special initializations required here.
1190 : */
1191 11170 : scan = (HeapScanDesc) bscan;
1192 : }
1193 : else
1194 393993 : scan = (HeapScanDesc) palloc_object(HeapScanDescData);
1195 :
1196 405163 : scan->rs_base.rs_rd = relation;
1197 405163 : scan->rs_base.rs_snapshot = snapshot;
1198 405163 : scan->rs_base.rs_nkeys = nkeys;
1199 405163 : scan->rs_base.rs_flags = flags;
1200 405163 : scan->rs_base.rs_parallel = parallel_scan;
1201 405163 : scan->rs_strategy = NULL; /* set in initscan */
1202 405163 : scan->rs_cbuf = InvalidBuffer;
1203 :
1204 : /*
1205 : * Disable page-at-a-time mode if it's not a MVCC-safe snapshot.
1206 : */
1207 405163 : if (!(snapshot && IsMVCCSnapshot(snapshot)))
1208 30017 : scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
1209 :
1210 : /* Check that a historic snapshot is not used for non-catalog tables */
1211 405163 : if (snapshot &&
1212 396223 : IsHistoricMVCCSnapshot(snapshot) &&
1213 669 : !RelationIsAccessibleInLogicalDecoding(relation))
1214 : {
1215 0 : ereport(ERROR,
1216 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
1217 : errmsg("cannot query non-catalog table \"%s\" during logical decoding",
1218 : RelationGetRelationName(relation))));
1219 : }
1220 :
1221 : /*
1222 : * For seqscan and sample scans in a serializable transaction, acquire a
1223 : * predicate lock on the entire relation. This is required not only to
1224 : * lock all the matching tuples, but also to conflict with new insertions
1225 : * into the table. In an indexscan, we take page locks on the index pages
1226 : * covering the range specified in the scan qual, but in a heap scan there
1227 : * is nothing more fine-grained to lock. A bitmap scan is a different
1228 : * story, there we have already scanned the index and locked the index
1229 : * pages covering the predicate. But in that case we still have to lock
1230 : * any matching heap tuples. For sample scan we could optimize the locking
1231 : * to be at least page-level granularity, but we'd need to add per-tuple
1232 : * locking for that.
1233 : */
1234 405163 : if (scan->rs_base.rs_flags & (SO_TYPE_SEQSCAN | SO_TYPE_SAMPLESCAN))
1235 : {
1236 : /*
1237 : * Ensure a missing snapshot is noticed reliably, even if the
1238 : * isolation mode means predicate locking isn't performed (and
1239 : * therefore the snapshot isn't used here).
1240 : */
1241 : Assert(snapshot);
1242 383670 : PredicateLockRelation(relation, snapshot);
1243 : }
1244 :
1245 : /* we only need to set this up once */
1246 405163 : scan->rs_ctup.t_tableOid = RelationGetRelid(relation);
1247 :
1248 : /*
1249 : * Allocate memory to keep track of page allocation for parallel workers
1250 : * when doing a parallel scan.
1251 : */
1252 405163 : if (parallel_scan != NULL)
1253 2119 : scan->rs_parallelworkerdata = palloc_object(ParallelBlockTableScanWorkerData);
1254 : else
1255 403044 : scan->rs_parallelworkerdata = NULL;
1256 :
1257 : /*
1258 : * we do this here instead of in initscan() because heap_rescan also calls
1259 : * initscan() and we don't want to allocate memory again
1260 : */
1261 405163 : if (nkeys > 0)
1262 229536 : scan->rs_base.rs_key = palloc_array(ScanKeyData, nkeys);
1263 : else
1264 175627 : scan->rs_base.rs_key = NULL;
1265 :
1266 405163 : initscan(scan, key, false);
1267 :
1268 405161 : scan->rs_read_stream = NULL;
1269 :
1270 : /*
1271 : * Set up a read stream for sequential scans and TID range scans. This
1272 : * should be done after initscan() because initscan() allocates the
1273 : * BufferAccessStrategy object passed to the read stream API.
1274 : */
1275 405161 : if (scan->rs_base.rs_flags & SO_TYPE_SEQSCAN ||
1276 21566 : scan->rs_base.rs_flags & SO_TYPE_TIDRANGESCAN)
1277 384585 : {
1278 : ReadStreamBlockNumberCB cb;
1279 :
1280 384585 : if (scan->rs_base.rs_parallel)
1281 2119 : cb = heap_scan_stream_read_next_parallel;
1282 : else
1283 382466 : cb = heap_scan_stream_read_next_serial;
1284 :
1285 : /* ---
1286 : * It is safe to use batchmode as the only locks taken by `cb`
1287 : * are never taken while waiting for IO:
1288 : * - SyncScanLock is used in the non-parallel case
1289 : * - in the parallel case, only spinlocks and atomics are used
1290 : * ---
1291 : */
1292 384585 : scan->rs_read_stream = read_stream_begin_relation(READ_STREAM_SEQUENTIAL |
1293 : READ_STREAM_USE_BATCHING,
1294 : scan->rs_strategy,
1295 : scan->rs_base.rs_rd,
1296 : MAIN_FORKNUM,
1297 : cb,
1298 : scan,
1299 : 0);
1300 : }
1301 20576 : else if (scan->rs_base.rs_flags & SO_TYPE_BITMAPSCAN)
1302 : {
1303 11170 : scan->rs_read_stream = read_stream_begin_relation(READ_STREAM_DEFAULT |
1304 : READ_STREAM_USE_BATCHING,
1305 : scan->rs_strategy,
1306 : scan->rs_base.rs_rd,
1307 : MAIN_FORKNUM,
1308 : bitmapheap_stream_read_next,
1309 : scan,
1310 : sizeof(TBMIterateResult));
1311 : }
1312 :
1313 :
1314 405161 : return (TableScanDesc) scan;
1315 : }
1316 :
1317 : void
1318 823417 : heap_rescan(TableScanDesc sscan, ScanKey key, bool set_params,
1319 : bool allow_strat, bool allow_sync, bool allow_pagemode)
1320 : {
1321 823417 : HeapScanDesc scan = (HeapScanDesc) sscan;
1322 :
1323 823417 : if (set_params)
1324 : {
1325 15 : if (allow_strat)
1326 15 : scan->rs_base.rs_flags |= SO_ALLOW_STRAT;
1327 : else
1328 0 : scan->rs_base.rs_flags &= ~SO_ALLOW_STRAT;
1329 :
1330 15 : if (allow_sync)
1331 6 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
1332 : else
1333 9 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
1334 :
1335 15 : if (allow_pagemode && scan->rs_base.rs_snapshot &&
1336 15 : IsMVCCSnapshot(scan->rs_base.rs_snapshot))
1337 15 : scan->rs_base.rs_flags |= SO_ALLOW_PAGEMODE;
1338 : else
1339 0 : scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
1340 : }
1341 :
1342 : /*
1343 : * unpin scan buffers
1344 : */
1345 823417 : if (BufferIsValid(scan->rs_cbuf))
1346 : {
1347 1732 : ReleaseBuffer(scan->rs_cbuf);
1348 1732 : scan->rs_cbuf = InvalidBuffer;
1349 : }
1350 :
1351 : /*
1352 : * SO_TYPE_BITMAPSCAN would be cleaned up here, but it does not hold any
1353 : * additional data vs a normal HeapScan
1354 : */
1355 :
1356 : /*
1357 : * The read stream is reset on rescan. This must be done before
1358 : * initscan(), as some state referred to by read_stream_reset() is reset
1359 : * in initscan().
1360 : */
1361 823417 : if (scan->rs_read_stream)
1362 823399 : read_stream_reset(scan->rs_read_stream);
1363 :
1364 : /*
1365 : * reinitialize scan descriptor
1366 : */
1367 823417 : initscan(scan, key, true);
1368 823417 : }
1369 :
1370 : void
1371 402741 : heap_endscan(TableScanDesc sscan)
1372 : {
1373 402741 : HeapScanDesc scan = (HeapScanDesc) sscan;
1374 :
1375 : /* Note: no locking manipulations needed */
1376 :
1377 : /*
1378 : * unpin scan buffers
1379 : */
1380 402741 : if (BufferIsValid(scan->rs_cbuf))
1381 166433 : ReleaseBuffer(scan->rs_cbuf);
1382 :
1383 : /*
1384 : * Must free the read stream before freeing the BufferAccessStrategy.
1385 : */
1386 402741 : if (scan->rs_read_stream)
1387 393388 : read_stream_end(scan->rs_read_stream);
1388 :
1389 : /*
1390 : * decrement relation reference count and free scan descriptor storage
1391 : */
1392 402741 : RelationDecrementReferenceCount(scan->rs_base.rs_rd);
1393 :
1394 402741 : if (scan->rs_base.rs_key)
1395 229506 : pfree(scan->rs_base.rs_key);
1396 :
1397 402741 : if (scan->rs_strategy != NULL)
1398 12256 : FreeAccessStrategy(scan->rs_strategy);
1399 :
1400 402741 : if (scan->rs_parallelworkerdata != NULL)
1401 2119 : pfree(scan->rs_parallelworkerdata);
1402 :
1403 402741 : if (scan->rs_base.rs_flags & SO_TEMP_SNAPSHOT)
1404 41194 : UnregisterSnapshot(scan->rs_base.rs_snapshot);
1405 :
1406 402741 : pfree(scan);
1407 402741 : }
1408 :
1409 : HeapTuple
1410 10001168 : heap_getnext(TableScanDesc sscan, ScanDirection direction)
1411 : {
1412 10001168 : HeapScanDesc scan = (HeapScanDesc) sscan;
1413 :
1414 : /*
1415 : * This is still widely used directly, without going through table AM, so
1416 : * add a safety check. It's possible we should, at a later point,
1417 : * downgrade this to an assert. The reason for checking the AM routine,
1418 : * rather than the AM oid, is that this allows to write regression tests
1419 : * that create another AM reusing the heap handler.
1420 : */
1421 10001168 : if (unlikely(sscan->rs_rd->rd_tableam != GetHeapamTableAmRoutine()))
1422 0 : ereport(ERROR,
1423 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1424 : errmsg_internal("only heap AM is supported")));
1425 :
1426 : /* Note: no locking manipulations needed */
1427 :
1428 10001168 : if (scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE)
1429 2977942 : heapgettup_pagemode(scan, direction,
1430 2977942 : scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1431 : else
1432 7023226 : heapgettup(scan, direction,
1433 7023226 : scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1434 :
1435 10001168 : if (scan->rs_ctup.t_data == NULL)
1436 68155 : return NULL;
1437 :
1438 : /*
1439 : * if we get here it means we have a new current scan tuple, so point to
1440 : * the proper return buffer and return the tuple.
1441 : */
1442 :
1443 9933013 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1444 :
1445 9933013 : return &scan->rs_ctup;
1446 : }
1447 :
1448 : bool
1449 49763734 : heap_getnextslot(TableScanDesc sscan, ScanDirection direction, TupleTableSlot *slot)
1450 : {
1451 49763734 : HeapScanDesc scan = (HeapScanDesc) sscan;
1452 :
1453 : /* Note: no locking manipulations needed */
1454 :
1455 49763734 : if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1456 49279026 : heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1457 : else
1458 484708 : heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1459 :
1460 49763706 : if (scan->rs_ctup.t_data == NULL)
1461 : {
1462 967097 : ExecClearTuple(slot);
1463 967097 : return false;
1464 : }
1465 :
1466 : /*
1467 : * if we get here it means we have a new current scan tuple, so point to
1468 : * the proper return buffer and return the tuple.
1469 : */
1470 :
1471 48796609 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1472 :
1473 48796609 : ExecStoreBufferHeapTuple(&scan->rs_ctup, slot,
1474 : scan->rs_cbuf);
1475 48796609 : return true;
1476 : }
1477 :
1478 : void
1479 1035 : heap_set_tidrange(TableScanDesc sscan, ItemPointer mintid,
1480 : ItemPointer maxtid)
1481 : {
1482 1035 : HeapScanDesc scan = (HeapScanDesc) sscan;
1483 : BlockNumber startBlk;
1484 : BlockNumber numBlks;
1485 : ItemPointerData highestItem;
1486 : ItemPointerData lowestItem;
1487 :
1488 : /*
1489 : * For relations without any pages, we can simply leave the TID range
1490 : * unset. There will be no tuples to scan, therefore no tuples outside
1491 : * the given TID range.
1492 : */
1493 1035 : if (scan->rs_nblocks == 0)
1494 24 : return;
1495 :
1496 : /*
1497 : * Set up some ItemPointers which point to the first and last possible
1498 : * tuples in the heap.
1499 : */
1500 1029 : ItemPointerSet(&highestItem, scan->rs_nblocks - 1, MaxOffsetNumber);
1501 1029 : ItemPointerSet(&lowestItem, 0, FirstOffsetNumber);
1502 :
1503 : /*
1504 : * If the given maximum TID is below the highest possible TID in the
1505 : * relation, then restrict the range to that, otherwise we scan to the end
1506 : * of the relation.
1507 : */
1508 1029 : if (ItemPointerCompare(maxtid, &highestItem) < 0)
1509 130 : ItemPointerCopy(maxtid, &highestItem);
1510 :
1511 : /*
1512 : * If the given minimum TID is above the lowest possible TID in the
1513 : * relation, then restrict the range to only scan for TIDs above that.
1514 : */
1515 1029 : if (ItemPointerCompare(mintid, &lowestItem) > 0)
1516 911 : ItemPointerCopy(mintid, &lowestItem);
1517 :
1518 : /*
1519 : * Check for an empty range and protect from would be negative results
1520 : * from the numBlks calculation below.
1521 : */
1522 1029 : if (ItemPointerCompare(&highestItem, &lowestItem) < 0)
1523 : {
1524 : /* Set an empty range of blocks to scan */
1525 18 : heap_setscanlimits(sscan, 0, 0);
1526 18 : return;
1527 : }
1528 :
1529 : /*
1530 : * Calculate the first block and the number of blocks we must scan. We
1531 : * could be more aggressive here and perform some more validation to try
1532 : * and further narrow the scope of blocks to scan by checking if the
1533 : * lowestItem has an offset above MaxOffsetNumber. In this case, we could
1534 : * advance startBlk by one. Likewise, if highestItem has an offset of 0
1535 : * we could scan one fewer blocks. However, such an optimization does not
1536 : * seem worth troubling over, currently.
1537 : */
1538 1011 : startBlk = ItemPointerGetBlockNumberNoCheck(&lowestItem);
1539 :
1540 1011 : numBlks = ItemPointerGetBlockNumberNoCheck(&highestItem) -
1541 1011 : ItemPointerGetBlockNumberNoCheck(&lowestItem) + 1;
1542 :
1543 : /* Set the start block and number of blocks to scan */
1544 1011 : heap_setscanlimits(sscan, startBlk, numBlks);
1545 :
1546 : /* Finally, set the TID range in sscan */
1547 1011 : ItemPointerCopy(&lowestItem, &sscan->st.tidrange.rs_mintid);
1548 1011 : ItemPointerCopy(&highestItem, &sscan->st.tidrange.rs_maxtid);
1549 : }
1550 :
1551 : bool
1552 5636 : heap_getnextslot_tidrange(TableScanDesc sscan, ScanDirection direction,
1553 : TupleTableSlot *slot)
1554 : {
1555 5636 : HeapScanDesc scan = (HeapScanDesc) sscan;
1556 5636 : ItemPointer mintid = &sscan->st.tidrange.rs_mintid;
1557 5636 : ItemPointer maxtid = &sscan->st.tidrange.rs_maxtid;
1558 :
1559 : /* Note: no locking manipulations needed */
1560 : for (;;)
1561 : {
1562 5729 : if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1563 5729 : heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1564 : else
1565 0 : heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1566 :
1567 5721 : if (scan->rs_ctup.t_data == NULL)
1568 : {
1569 104 : ExecClearTuple(slot);
1570 104 : return false;
1571 : }
1572 :
1573 : /*
1574 : * heap_set_tidrange will have used heap_setscanlimits to limit the
1575 : * range of pages we scan to only ones that can contain the TID range
1576 : * we're scanning for. Here we must filter out any tuples from these
1577 : * pages that are outside of that range.
1578 : */
1579 5617 : if (ItemPointerCompare(&scan->rs_ctup.t_self, mintid) < 0)
1580 : {
1581 93 : ExecClearTuple(slot);
1582 :
1583 : /*
1584 : * When scanning backwards, the TIDs will be in descending order.
1585 : * Future tuples in this direction will be lower still, so we can
1586 : * just return false to indicate there will be no more tuples.
1587 : */
1588 93 : if (ScanDirectionIsBackward(direction))
1589 0 : return false;
1590 :
1591 93 : continue;
1592 : }
1593 :
1594 : /*
1595 : * Likewise for the final page, we must filter out TIDs greater than
1596 : * maxtid.
1597 : */
1598 5524 : if (ItemPointerCompare(&scan->rs_ctup.t_self, maxtid) > 0)
1599 : {
1600 56 : ExecClearTuple(slot);
1601 :
1602 : /*
1603 : * When scanning forward, the TIDs will be in ascending order.
1604 : * Future tuples in this direction will be higher still, so we can
1605 : * just return false to indicate there will be no more tuples.
1606 : */
1607 56 : if (ScanDirectionIsForward(direction))
1608 56 : return false;
1609 0 : continue;
1610 : }
1611 :
1612 5468 : break;
1613 : }
1614 :
1615 : /*
1616 : * if we get here it means we have a new current scan tuple, so point to
1617 : * the proper return buffer and return the tuple.
1618 : */
1619 5468 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1620 :
1621 5468 : ExecStoreBufferHeapTuple(&scan->rs_ctup, slot, scan->rs_cbuf);
1622 5468 : return true;
1623 : }
1624 :
1625 : /*
1626 : * heap_fetch - retrieve tuple with given tid
1627 : *
1628 : * On entry, tuple->t_self is the TID to fetch. We pin the buffer holding
1629 : * the tuple, fill in the remaining fields of *tuple, and check the tuple
1630 : * against the specified snapshot.
1631 : *
1632 : * If successful (tuple found and passes snapshot time qual), then *userbuf
1633 : * is set to the buffer holding the tuple and true is returned. The caller
1634 : * must unpin the buffer when done with the tuple.
1635 : *
1636 : * If the tuple is not found (ie, item number references a deleted slot),
1637 : * then tuple->t_data is set to NULL, *userbuf is set to InvalidBuffer,
1638 : * and false is returned.
1639 : *
1640 : * If the tuple is found but fails the time qual check, then the behavior
1641 : * depends on the keep_buf parameter. If keep_buf is false, the results
1642 : * are the same as for the tuple-not-found case. If keep_buf is true,
1643 : * then tuple->t_data and *userbuf are returned as for the success case,
1644 : * and again the caller must unpin the buffer; but false is returned.
1645 : *
1646 : * heap_fetch does not follow HOT chains: only the exact TID requested will
1647 : * be fetched.
1648 : *
1649 : * It is somewhat inconsistent that we ereport() on invalid block number but
1650 : * return false on invalid item number. There are a couple of reasons though.
1651 : * One is that the caller can relatively easily check the block number for
1652 : * validity, but cannot check the item number without reading the page
1653 : * himself. Another is that when we are following a t_ctid link, we can be
1654 : * reasonably confident that the page number is valid (since VACUUM shouldn't
1655 : * truncate off the destination page without having killed the referencing
1656 : * tuple first), but the item number might well not be good.
1657 : */
1658 : bool
1659 181413 : heap_fetch(Relation relation,
1660 : Snapshot snapshot,
1661 : HeapTuple tuple,
1662 : Buffer *userbuf,
1663 : bool keep_buf)
1664 : {
1665 181413 : ItemPointer tid = &(tuple->t_self);
1666 : ItemId lp;
1667 : Buffer buffer;
1668 : Page page;
1669 : OffsetNumber offnum;
1670 : bool valid;
1671 :
1672 : /*
1673 : * Fetch and pin the appropriate page of the relation.
1674 : */
1675 181413 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
1676 :
1677 : /*
1678 : * Need share lock on buffer to examine tuple commit status.
1679 : */
1680 181405 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
1681 181405 : page = BufferGetPage(buffer);
1682 :
1683 : /*
1684 : * We'd better check for out-of-range offnum in case of VACUUM since the
1685 : * TID was obtained.
1686 : */
1687 181405 : offnum = ItemPointerGetOffsetNumber(tid);
1688 181405 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1689 : {
1690 3 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1691 3 : ReleaseBuffer(buffer);
1692 3 : *userbuf = InvalidBuffer;
1693 3 : tuple->t_data = NULL;
1694 3 : return false;
1695 : }
1696 :
1697 : /*
1698 : * get the item line pointer corresponding to the requested tid
1699 : */
1700 181402 : lp = PageGetItemId(page, offnum);
1701 :
1702 : /*
1703 : * Must check for deleted tuple.
1704 : */
1705 181402 : if (!ItemIdIsNormal(lp))
1706 : {
1707 331 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1708 331 : ReleaseBuffer(buffer);
1709 331 : *userbuf = InvalidBuffer;
1710 331 : tuple->t_data = NULL;
1711 331 : return false;
1712 : }
1713 :
1714 : /*
1715 : * fill in *tuple fields
1716 : */
1717 181071 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
1718 181071 : tuple->t_len = ItemIdGetLength(lp);
1719 181071 : tuple->t_tableOid = RelationGetRelid(relation);
1720 :
1721 : /*
1722 : * check tuple visibility, then release lock
1723 : */
1724 181071 : valid = HeapTupleSatisfiesVisibility(tuple, snapshot, buffer);
1725 :
1726 181071 : if (valid)
1727 181018 : PredicateLockTID(relation, &(tuple->t_self), snapshot,
1728 181018 : HeapTupleHeaderGetXmin(tuple->t_data));
1729 :
1730 181071 : HeapCheckForSerializableConflictOut(valid, relation, tuple, buffer, snapshot);
1731 :
1732 181071 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1733 :
1734 181071 : if (valid)
1735 : {
1736 : /*
1737 : * All checks passed, so return the tuple as valid. Caller is now
1738 : * responsible for releasing the buffer.
1739 : */
1740 181018 : *userbuf = buffer;
1741 :
1742 181018 : return true;
1743 : }
1744 :
1745 : /* Tuple failed time qual, but maybe caller wants to see it anyway. */
1746 53 : if (keep_buf)
1747 34 : *userbuf = buffer;
1748 : else
1749 : {
1750 19 : ReleaseBuffer(buffer);
1751 19 : *userbuf = InvalidBuffer;
1752 19 : tuple->t_data = NULL;
1753 : }
1754 :
1755 53 : return false;
1756 : }
1757 :
1758 : /*
1759 : * heap_hot_search_buffer - search HOT chain for tuple satisfying snapshot
1760 : *
1761 : * On entry, *tid is the TID of a tuple (either a simple tuple, or the root
1762 : * of a HOT chain), and buffer is the buffer holding this tuple. We search
1763 : * for the first chain member satisfying the given snapshot. If one is
1764 : * found, we update *tid to reference that tuple's offset number, and
1765 : * return true. If no match, return false without modifying *tid.
1766 : *
1767 : * heapTuple is a caller-supplied buffer. When a match is found, we return
1768 : * the tuple here, in addition to updating *tid. If no match is found, the
1769 : * contents of this buffer on return are undefined.
1770 : *
1771 : * If all_dead is not NULL, we check non-visible tuples to see if they are
1772 : * globally dead; *all_dead is set true if all members of the HOT chain
1773 : * are vacuumable, false if not.
1774 : *
1775 : * Unlike heap_fetch, the caller must already have pin and (at least) share
1776 : * lock on the buffer; it is still pinned/locked at exit.
1777 : */
1778 : bool
1779 23415186 : heap_hot_search_buffer(ItemPointer tid, Relation relation, Buffer buffer,
1780 : Snapshot snapshot, HeapTuple heapTuple,
1781 : bool *all_dead, bool first_call)
1782 : {
1783 23415186 : Page page = BufferGetPage(buffer);
1784 23415186 : TransactionId prev_xmax = InvalidTransactionId;
1785 : BlockNumber blkno;
1786 : OffsetNumber offnum;
1787 : bool at_chain_start;
1788 : bool valid;
1789 : bool skip;
1790 23415186 : GlobalVisState *vistest = NULL;
1791 :
1792 : /* If this is not the first call, previous call returned a (live!) tuple */
1793 23415186 : if (all_dead)
1794 20039019 : *all_dead = first_call;
1795 :
1796 23415186 : blkno = ItemPointerGetBlockNumber(tid);
1797 23415186 : offnum = ItemPointerGetOffsetNumber(tid);
1798 23415186 : at_chain_start = first_call;
1799 23415186 : skip = !first_call;
1800 :
1801 : /* XXX: we should assert that a snapshot is pushed or registered */
1802 : Assert(TransactionIdIsValid(RecentXmin));
1803 : Assert(BufferGetBlockNumber(buffer) == blkno);
1804 :
1805 : /* Scan through possible multiple members of HOT-chain */
1806 : for (;;)
1807 1744249 : {
1808 : ItemId lp;
1809 :
1810 : /* check for bogus TID */
1811 25159435 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1812 : break;
1813 :
1814 25159435 : lp = PageGetItemId(page, offnum);
1815 :
1816 : /* check for unused, dead, or redirected items */
1817 25159435 : if (!ItemIdIsNormal(lp))
1818 : {
1819 : /* We should only see a redirect at start of chain */
1820 907402 : if (ItemIdIsRedirected(lp) && at_chain_start)
1821 : {
1822 : /* Follow the redirect */
1823 537281 : offnum = ItemIdGetRedirect(lp);
1824 537281 : at_chain_start = false;
1825 537281 : continue;
1826 : }
1827 : /* else must be end of chain */
1828 370121 : break;
1829 : }
1830 :
1831 : /*
1832 : * Update heapTuple to point to the element of the HOT chain we're
1833 : * currently investigating. Having t_self set correctly is important
1834 : * because the SSI checks and the *Satisfies routine for historical
1835 : * MVCC snapshots need the correct tid to decide about the visibility.
1836 : */
1837 24252033 : heapTuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
1838 24252033 : heapTuple->t_len = ItemIdGetLength(lp);
1839 24252033 : heapTuple->t_tableOid = RelationGetRelid(relation);
1840 24252033 : ItemPointerSet(&heapTuple->t_self, blkno, offnum);
1841 :
1842 : /*
1843 : * Shouldn't see a HEAP_ONLY tuple at chain start.
1844 : */
1845 24252033 : if (at_chain_start && HeapTupleIsHeapOnly(heapTuple))
1846 0 : break;
1847 :
1848 : /*
1849 : * The xmin should match the previous xmax value, else chain is
1850 : * broken.
1851 : */
1852 25459001 : if (TransactionIdIsValid(prev_xmax) &&
1853 1206968 : !TransactionIdEquals(prev_xmax,
1854 : HeapTupleHeaderGetXmin(heapTuple->t_data)))
1855 0 : break;
1856 :
1857 : /*
1858 : * When first_call is true (and thus, skip is initially false) we'll
1859 : * return the first tuple we find. But on later passes, heapTuple
1860 : * will initially be pointing to the tuple we returned last time.
1861 : * Returning it again would be incorrect (and would loop forever), so
1862 : * we skip it and return the next match we find.
1863 : */
1864 24252033 : if (!skip)
1865 : {
1866 : /* If it's visible per the snapshot, we must return it */
1867 24163051 : valid = HeapTupleSatisfiesVisibility(heapTuple, snapshot, buffer);
1868 24163051 : HeapCheckForSerializableConflictOut(valid, relation, heapTuple,
1869 : buffer, snapshot);
1870 :
1871 24163046 : if (valid)
1872 : {
1873 16803032 : ItemPointerSetOffsetNumber(tid, offnum);
1874 16803032 : PredicateLockTID(relation, &heapTuple->t_self, snapshot,
1875 16803032 : HeapTupleHeaderGetXmin(heapTuple->t_data));
1876 16803032 : if (all_dead)
1877 13720895 : *all_dead = false;
1878 16803032 : return true;
1879 : }
1880 : }
1881 7448996 : skip = false;
1882 :
1883 : /*
1884 : * If we can't see it, maybe no one else can either. At caller
1885 : * request, check whether all chain members are dead to all
1886 : * transactions.
1887 : *
1888 : * Note: if you change the criterion here for what is "dead", fix the
1889 : * planner's get_actual_variable_range() function to match.
1890 : */
1891 7448996 : if (all_dead && *all_dead)
1892 : {
1893 6484094 : if (!vistest)
1894 6362362 : vistest = GlobalVisTestFor(relation);
1895 :
1896 6484094 : if (!HeapTupleIsSurelyDead(heapTuple, vistest))
1897 6125168 : *all_dead = false;
1898 : }
1899 :
1900 : /*
1901 : * Check to see if HOT chain continues past this tuple; if so fetch
1902 : * the next offnum and loop around.
1903 : */
1904 7448996 : if (HeapTupleIsHotUpdated(heapTuple))
1905 : {
1906 : Assert(ItemPointerGetBlockNumber(&heapTuple->t_data->t_ctid) ==
1907 : blkno);
1908 1206968 : offnum = ItemPointerGetOffsetNumber(&heapTuple->t_data->t_ctid);
1909 1206968 : at_chain_start = false;
1910 1206968 : prev_xmax = HeapTupleHeaderGetUpdateXid(heapTuple->t_data);
1911 : }
1912 : else
1913 6242028 : break; /* end of chain */
1914 : }
1915 :
1916 6612149 : return false;
1917 : }
1918 :
1919 : /*
1920 : * heap_get_latest_tid - get the latest tid of a specified tuple
1921 : *
1922 : * Actually, this gets the latest version that is visible according to the
1923 : * scan's snapshot. Create a scan using SnapshotDirty to get the very latest,
1924 : * possibly uncommitted version.
1925 : *
1926 : * *tid is both an input and an output parameter: it is updated to
1927 : * show the latest version of the row. Note that it will not be changed
1928 : * if no version of the row passes the snapshot test.
1929 : */
1930 : void
1931 150 : heap_get_latest_tid(TableScanDesc sscan,
1932 : ItemPointer tid)
1933 : {
1934 150 : Relation relation = sscan->rs_rd;
1935 150 : Snapshot snapshot = sscan->rs_snapshot;
1936 : ItemPointerData ctid;
1937 : TransactionId priorXmax;
1938 :
1939 : /*
1940 : * table_tuple_get_latest_tid() verified that the passed in tid is valid.
1941 : * Assume that t_ctid links are valid however - there shouldn't be invalid
1942 : * ones in the table.
1943 : */
1944 : Assert(ItemPointerIsValid(tid));
1945 :
1946 : /*
1947 : * Loop to chase down t_ctid links. At top of loop, ctid is the tuple we
1948 : * need to examine, and *tid is the TID we will return if ctid turns out
1949 : * to be bogus.
1950 : *
1951 : * Note that we will loop until we reach the end of the t_ctid chain.
1952 : * Depending on the snapshot passed, there might be at most one visible
1953 : * version of the row, but we don't try to optimize for that.
1954 : */
1955 150 : ctid = *tid;
1956 150 : priorXmax = InvalidTransactionId; /* cannot check first XMIN */
1957 : for (;;)
1958 45 : {
1959 : Buffer buffer;
1960 : Page page;
1961 : OffsetNumber offnum;
1962 : ItemId lp;
1963 : HeapTupleData tp;
1964 : bool valid;
1965 :
1966 : /*
1967 : * Read, pin, and lock the page.
1968 : */
1969 195 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(&ctid));
1970 195 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
1971 195 : page = BufferGetPage(buffer);
1972 :
1973 : /*
1974 : * Check for bogus item number. This is not treated as an error
1975 : * condition because it can happen while following a t_ctid link. We
1976 : * just assume that the prior tid is OK and return it unchanged.
1977 : */
1978 195 : offnum = ItemPointerGetOffsetNumber(&ctid);
1979 195 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1980 : {
1981 0 : UnlockReleaseBuffer(buffer);
1982 0 : break;
1983 : }
1984 195 : lp = PageGetItemId(page, offnum);
1985 195 : if (!ItemIdIsNormal(lp))
1986 : {
1987 0 : UnlockReleaseBuffer(buffer);
1988 0 : break;
1989 : }
1990 :
1991 : /* OK to access the tuple */
1992 195 : tp.t_self = ctid;
1993 195 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
1994 195 : tp.t_len = ItemIdGetLength(lp);
1995 195 : tp.t_tableOid = RelationGetRelid(relation);
1996 :
1997 : /*
1998 : * After following a t_ctid link, we might arrive at an unrelated
1999 : * tuple. Check for XMIN match.
2000 : */
2001 240 : if (TransactionIdIsValid(priorXmax) &&
2002 45 : !TransactionIdEquals(priorXmax, HeapTupleHeaderGetXmin(tp.t_data)))
2003 : {
2004 0 : UnlockReleaseBuffer(buffer);
2005 0 : break;
2006 : }
2007 :
2008 : /*
2009 : * Check tuple visibility; if visible, set it as the new result
2010 : * candidate.
2011 : */
2012 195 : valid = HeapTupleSatisfiesVisibility(&tp, snapshot, buffer);
2013 195 : HeapCheckForSerializableConflictOut(valid, relation, &tp, buffer, snapshot);
2014 195 : if (valid)
2015 138 : *tid = ctid;
2016 :
2017 : /*
2018 : * If there's a valid t_ctid link, follow it, else we're done.
2019 : */
2020 276 : if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
2021 138 : HeapTupleHeaderIsOnlyLocked(tp.t_data) ||
2022 114 : HeapTupleHeaderIndicatesMovedPartitions(tp.t_data) ||
2023 57 : ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
2024 : {
2025 150 : UnlockReleaseBuffer(buffer);
2026 150 : break;
2027 : }
2028 :
2029 45 : ctid = tp.t_data->t_ctid;
2030 45 : priorXmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
2031 45 : UnlockReleaseBuffer(buffer);
2032 : } /* end of loop */
2033 150 : }
2034 :
2035 :
2036 : /*
2037 : * UpdateXmaxHintBits - update tuple hint bits after xmax transaction ends
2038 : *
2039 : * This is called after we have waited for the XMAX transaction to terminate.
2040 : * If the transaction aborted, we guarantee the XMAX_INVALID hint bit will
2041 : * be set on exit. If the transaction committed, we set the XMAX_COMMITTED
2042 : * hint bit if possible --- but beware that that may not yet be possible,
2043 : * if the transaction committed asynchronously.
2044 : *
2045 : * Note that if the transaction was a locker only, we set HEAP_XMAX_INVALID
2046 : * even if it commits.
2047 : *
2048 : * Hence callers should look only at XMAX_INVALID.
2049 : *
2050 : * Note this is not allowed for tuples whose xmax is a multixact.
2051 : */
2052 : static void
2053 227 : UpdateXmaxHintBits(HeapTupleHeader tuple, Buffer buffer, TransactionId xid)
2054 : {
2055 : Assert(TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple), xid));
2056 : Assert(!(tuple->t_infomask & HEAP_XMAX_IS_MULTI));
2057 :
2058 227 : if (!(tuple->t_infomask & (HEAP_XMAX_COMMITTED | HEAP_XMAX_INVALID)))
2059 : {
2060 402 : if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask) &&
2061 176 : TransactionIdDidCommit(xid))
2062 149 : HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_COMMITTED,
2063 : xid);
2064 : else
2065 77 : HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_INVALID,
2066 : InvalidTransactionId);
2067 : }
2068 227 : }
2069 :
2070 :
2071 : /*
2072 : * GetBulkInsertState - prepare status object for a bulk insert
2073 : */
2074 : BulkInsertState
2075 2712 : GetBulkInsertState(void)
2076 : {
2077 : BulkInsertState bistate;
2078 :
2079 2712 : bistate = (BulkInsertState) palloc_object(BulkInsertStateData);
2080 2712 : bistate->strategy = GetAccessStrategy(BAS_BULKWRITE);
2081 2712 : bistate->current_buf = InvalidBuffer;
2082 2712 : bistate->next_free = InvalidBlockNumber;
2083 2712 : bistate->last_free = InvalidBlockNumber;
2084 2712 : bistate->already_extended_by = 0;
2085 2712 : return bistate;
2086 : }
2087 :
2088 : /*
2089 : * FreeBulkInsertState - clean up after finishing a bulk insert
2090 : */
2091 : void
2092 2567 : FreeBulkInsertState(BulkInsertState bistate)
2093 : {
2094 2567 : if (bistate->current_buf != InvalidBuffer)
2095 2013 : ReleaseBuffer(bistate->current_buf);
2096 2567 : FreeAccessStrategy(bistate->strategy);
2097 2567 : pfree(bistate);
2098 2567 : }
2099 :
2100 : /*
2101 : * ReleaseBulkInsertStatePin - release a buffer currently held in bistate
2102 : */
2103 : void
2104 80758 : ReleaseBulkInsertStatePin(BulkInsertState bistate)
2105 : {
2106 80758 : if (bistate->current_buf != InvalidBuffer)
2107 30021 : ReleaseBuffer(bistate->current_buf);
2108 80758 : bistate->current_buf = InvalidBuffer;
2109 :
2110 : /*
2111 : * Despite the name, we also reset bulk relation extension state.
2112 : * Otherwise we can end up erroring out due to looking for free space in
2113 : * ->next_free of one partition, even though ->next_free was set when
2114 : * extending another partition. It could obviously also be bad for
2115 : * efficiency to look at existing blocks at offsets from another
2116 : * partition, even if we don't error out.
2117 : */
2118 80758 : bistate->next_free = InvalidBlockNumber;
2119 80758 : bistate->last_free = InvalidBlockNumber;
2120 80758 : }
2121 :
2122 :
2123 : /*
2124 : * heap_insert - insert tuple into a heap
2125 : *
2126 : * The new tuple is stamped with current transaction ID and the specified
2127 : * command ID.
2128 : *
2129 : * See table_tuple_insert for comments about most of the input flags, except
2130 : * that this routine directly takes a tuple rather than a slot.
2131 : *
2132 : * There's corresponding HEAP_INSERT_ options to all the TABLE_INSERT_
2133 : * options, and there additionally is HEAP_INSERT_SPECULATIVE which is used to
2134 : * implement table_tuple_insert_speculative().
2135 : *
2136 : * On return the header fields of *tup are updated to match the stored tuple;
2137 : * in particular tup->t_self receives the actual TID where the tuple was
2138 : * stored. But note that any toasting of fields within the tuple data is NOT
2139 : * reflected into *tup.
2140 : */
2141 : void
2142 8391325 : heap_insert(Relation relation, HeapTuple tup, CommandId cid,
2143 : int options, BulkInsertState bistate)
2144 : {
2145 8391325 : TransactionId xid = GetCurrentTransactionId();
2146 : HeapTuple heaptup;
2147 : Buffer buffer;
2148 8391317 : Buffer vmbuffer = InvalidBuffer;
2149 8391317 : bool all_visible_cleared = false;
2150 :
2151 : /* Cheap, simplistic check that the tuple matches the rel's rowtype. */
2152 : Assert(HeapTupleHeaderGetNatts(tup->t_data) <=
2153 : RelationGetNumberOfAttributes(relation));
2154 :
2155 8391317 : AssertHasSnapshotForToast(relation);
2156 :
2157 : /*
2158 : * Fill in tuple header fields and toast the tuple if necessary.
2159 : *
2160 : * Note: below this point, heaptup is the data we actually intend to store
2161 : * into the relation; tup is the caller's original untoasted data.
2162 : */
2163 8391317 : heaptup = heap_prepare_insert(relation, tup, xid, cid, options);
2164 :
2165 : /*
2166 : * Find buffer to insert this tuple into. If the page is all visible,
2167 : * this will also pin the requisite visibility map page.
2168 : */
2169 8391317 : buffer = RelationGetBufferForTuple(relation, heaptup->t_len,
2170 : InvalidBuffer, options, bistate,
2171 : &vmbuffer, NULL,
2172 : 0);
2173 :
2174 : /*
2175 : * We're about to do the actual insert -- but check for conflict first, to
2176 : * avoid possibly having to roll back work we've just done.
2177 : *
2178 : * This is safe without a recheck as long as there is no possibility of
2179 : * another process scanning the page between this check and the insert
2180 : * being visible to the scan (i.e., an exclusive buffer content lock is
2181 : * continuously held from this point until the tuple insert is visible).
2182 : *
2183 : * For a heap insert, we only need to check for table-level SSI locks. Our
2184 : * new tuple can't possibly conflict with existing tuple locks, and heap
2185 : * page locks are only consolidated versions of tuple locks; they do not
2186 : * lock "gaps" as index page locks do. So we don't need to specify a
2187 : * buffer when making the call, which makes for a faster check.
2188 : */
2189 8391317 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2190 :
2191 : /* NO EREPORT(ERROR) from here till changes are logged */
2192 8391305 : START_CRIT_SECTION();
2193 :
2194 8391305 : RelationPutHeapTuple(relation, buffer, heaptup,
2195 8391305 : (options & HEAP_INSERT_SPECULATIVE) != 0);
2196 :
2197 8391305 : if (PageIsAllVisible(BufferGetPage(buffer)))
2198 : {
2199 7590 : all_visible_cleared = true;
2200 7590 : PageClearAllVisible(BufferGetPage(buffer));
2201 7590 : visibilitymap_clear(relation,
2202 7590 : ItemPointerGetBlockNumber(&(heaptup->t_self)),
2203 : vmbuffer, VISIBILITYMAP_VALID_BITS);
2204 : }
2205 :
2206 : /*
2207 : * XXX Should we set PageSetPrunable on this page ?
2208 : *
2209 : * The inserting transaction may eventually abort thus making this tuple
2210 : * DEAD and hence available for pruning. Though we don't want to optimize
2211 : * for aborts, if no other tuple in this page is UPDATEd/DELETEd, the
2212 : * aborted tuple will never be pruned until next vacuum is triggered.
2213 : *
2214 : * If you do add PageSetPrunable here, add it in heap_xlog_insert too.
2215 : */
2216 :
2217 8391305 : MarkBufferDirty(buffer);
2218 :
2219 : /* XLOG stuff */
2220 8391305 : if (RelationNeedsWAL(relation))
2221 : {
2222 : xl_heap_insert xlrec;
2223 : xl_heap_header xlhdr;
2224 : XLogRecPtr recptr;
2225 7501599 : Page page = BufferGetPage(buffer);
2226 7501599 : uint8 info = XLOG_HEAP_INSERT;
2227 7501599 : int bufflags = 0;
2228 :
2229 : /*
2230 : * If this is a catalog, we need to transmit combo CIDs to properly
2231 : * decode, so log that as well.
2232 : */
2233 7501599 : if (RelationIsAccessibleInLogicalDecoding(relation))
2234 3433 : log_heap_new_cid(relation, heaptup);
2235 :
2236 : /*
2237 : * If this is the single and first tuple on page, we can reinit the
2238 : * page instead of restoring the whole thing. Set flag, and hide
2239 : * buffer references from XLogInsert.
2240 : */
2241 7598527 : if (ItemPointerGetOffsetNumber(&(heaptup->t_self)) == FirstOffsetNumber &&
2242 96928 : PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
2243 : {
2244 95883 : info |= XLOG_HEAP_INIT_PAGE;
2245 95883 : bufflags |= REGBUF_WILL_INIT;
2246 : }
2247 :
2248 7501599 : xlrec.offnum = ItemPointerGetOffsetNumber(&heaptup->t_self);
2249 7501599 : xlrec.flags = 0;
2250 7501599 : if (all_visible_cleared)
2251 7587 : xlrec.flags |= XLH_INSERT_ALL_VISIBLE_CLEARED;
2252 7501599 : if (options & HEAP_INSERT_SPECULATIVE)
2253 2112 : xlrec.flags |= XLH_INSERT_IS_SPECULATIVE;
2254 : Assert(ItemPointerGetBlockNumber(&heaptup->t_self) == BufferGetBlockNumber(buffer));
2255 :
2256 : /*
2257 : * For logical decoding, we need the tuple even if we're doing a full
2258 : * page write, so make sure it's included even if we take a full-page
2259 : * image. (XXX We could alternatively store a pointer into the FPW).
2260 : */
2261 7501599 : if (RelationIsLogicallyLogged(relation) &&
2262 250516 : !(options & HEAP_INSERT_NO_LOGICAL))
2263 : {
2264 250489 : xlrec.flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
2265 250489 : bufflags |= REGBUF_KEEP_DATA;
2266 :
2267 250489 : if (IsToastRelation(relation))
2268 1786 : xlrec.flags |= XLH_INSERT_ON_TOAST_RELATION;
2269 : }
2270 :
2271 7501599 : XLogBeginInsert();
2272 7501599 : XLogRegisterData(&xlrec, SizeOfHeapInsert);
2273 :
2274 7501599 : xlhdr.t_infomask2 = heaptup->t_data->t_infomask2;
2275 7501599 : xlhdr.t_infomask = heaptup->t_data->t_infomask;
2276 7501599 : xlhdr.t_hoff = heaptup->t_data->t_hoff;
2277 :
2278 : /*
2279 : * note we mark xlhdr as belonging to buffer; if XLogInsert decides to
2280 : * write the whole page to the xlog, we don't need to store
2281 : * xl_heap_header in the xlog.
2282 : */
2283 7501599 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD | bufflags);
2284 7501599 : XLogRegisterBufData(0, &xlhdr, SizeOfHeapHeader);
2285 : /* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
2286 7501599 : XLogRegisterBufData(0,
2287 7501599 : (char *) heaptup->t_data + SizeofHeapTupleHeader,
2288 7501599 : heaptup->t_len - SizeofHeapTupleHeader);
2289 :
2290 : /* filtering by origin on a row level is much more efficient */
2291 7501599 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
2292 :
2293 7501599 : recptr = XLogInsert(RM_HEAP_ID, info);
2294 :
2295 7501599 : PageSetLSN(page, recptr);
2296 : }
2297 :
2298 8391305 : END_CRIT_SECTION();
2299 :
2300 8391305 : UnlockReleaseBuffer(buffer);
2301 8391305 : if (vmbuffer != InvalidBuffer)
2302 7871 : ReleaseBuffer(vmbuffer);
2303 :
2304 : /*
2305 : * If tuple is cacheable, mark it for invalidation from the caches in case
2306 : * we abort. Note it is OK to do this after releasing the buffer, because
2307 : * the heaptup data structure is all in local memory, not in the shared
2308 : * buffer.
2309 : */
2310 8391305 : CacheInvalidateHeapTuple(relation, heaptup, NULL);
2311 :
2312 : /* Note: speculative insertions are counted too, even if aborted later */
2313 8391305 : pgstat_count_heap_insert(relation, 1);
2314 :
2315 : /*
2316 : * If heaptup is a private copy, release it. Don't forget to copy t_self
2317 : * back to the caller's image, too.
2318 : */
2319 8391305 : if (heaptup != tup)
2320 : {
2321 18470 : tup->t_self = heaptup->t_self;
2322 18470 : heap_freetuple(heaptup);
2323 : }
2324 8391305 : }
2325 :
2326 : /*
2327 : * Subroutine for heap_insert(). Prepares a tuple for insertion. This sets the
2328 : * tuple header fields and toasts the tuple if necessary. Returns a toasted
2329 : * version of the tuple if it was toasted, or the original tuple if not. Note
2330 : * that in any case, the header fields are also set in the original tuple.
2331 : */
2332 : static HeapTuple
2333 9899341 : heap_prepare_insert(Relation relation, HeapTuple tup, TransactionId xid,
2334 : CommandId cid, int options)
2335 : {
2336 : /*
2337 : * To allow parallel inserts, we need to ensure that they are safe to be
2338 : * performed in workers. We have the infrastructure to allow parallel
2339 : * inserts in general except for the cases where inserts generate a new
2340 : * CommandId (eg. inserts into a table having a foreign key column).
2341 : */
2342 9899341 : if (IsParallelWorker())
2343 0 : ereport(ERROR,
2344 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
2345 : errmsg("cannot insert tuples in a parallel worker")));
2346 :
2347 9899341 : tup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
2348 9899341 : tup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
2349 9899341 : tup->t_data->t_infomask |= HEAP_XMAX_INVALID;
2350 9899341 : HeapTupleHeaderSetXmin(tup->t_data, xid);
2351 9899341 : if (options & HEAP_INSERT_FROZEN)
2352 102088 : HeapTupleHeaderSetXminFrozen(tup->t_data);
2353 :
2354 9899341 : HeapTupleHeaderSetCmin(tup->t_data, cid);
2355 9899341 : HeapTupleHeaderSetXmax(tup->t_data, 0); /* for cleanliness */
2356 9899341 : tup->t_tableOid = RelationGetRelid(relation);
2357 :
2358 : /*
2359 : * If the new tuple is too big for storage or contains already toasted
2360 : * out-of-line attributes from some other relation, invoke the toaster.
2361 : */
2362 9899341 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
2363 31494 : relation->rd_rel->relkind != RELKIND_MATVIEW)
2364 : {
2365 : /* toast table entries should never be recursively toasted */
2366 : Assert(!HeapTupleHasExternal(tup));
2367 31446 : return tup;
2368 : }
2369 9867895 : else if (HeapTupleHasExternal(tup) || tup->t_len > TOAST_TUPLE_THRESHOLD)
2370 18522 : return heap_toast_insert_or_update(relation, tup, NULL, options);
2371 : else
2372 9849373 : return tup;
2373 : }
2374 :
2375 : /*
2376 : * Helper for heap_multi_insert() that computes the number of entire pages
2377 : * that inserting the remaining heaptuples requires. Used to determine how
2378 : * much the relation needs to be extended by.
2379 : */
2380 : static int
2381 381437 : heap_multi_insert_pages(HeapTuple *heaptuples, int done, int ntuples, Size saveFreeSpace)
2382 : {
2383 381437 : size_t page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
2384 381437 : int npages = 1;
2385 :
2386 2483268 : for (int i = done; i < ntuples; i++)
2387 : {
2388 2101831 : size_t tup_sz = sizeof(ItemIdData) + MAXALIGN(heaptuples[i]->t_len);
2389 :
2390 2101831 : if (page_avail < tup_sz)
2391 : {
2392 15637 : npages++;
2393 15637 : page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
2394 : }
2395 2101831 : page_avail -= tup_sz;
2396 : }
2397 :
2398 381437 : return npages;
2399 : }
2400 :
2401 : /*
2402 : * heap_multi_insert - insert multiple tuples into a heap
2403 : *
2404 : * This is like heap_insert(), but inserts multiple tuples in one operation.
2405 : * That's faster than calling heap_insert() in a loop, because when multiple
2406 : * tuples can be inserted on a single page, we can write just a single WAL
2407 : * record covering all of them, and only need to lock/unlock the page once.
2408 : *
2409 : * Note: this leaks memory into the current memory context. You can create a
2410 : * temporary context before calling this, if that's a problem.
2411 : */
2412 : void
2413 374842 : heap_multi_insert(Relation relation, TupleTableSlot **slots, int ntuples,
2414 : CommandId cid, int options, BulkInsertState bistate)
2415 : {
2416 374842 : TransactionId xid = GetCurrentTransactionId();
2417 : HeapTuple *heaptuples;
2418 : int i;
2419 : int ndone;
2420 : PGAlignedBlock scratch;
2421 : Page page;
2422 374842 : Buffer vmbuffer = InvalidBuffer;
2423 : bool needwal;
2424 : Size saveFreeSpace;
2425 374842 : bool need_tuple_data = RelationIsLogicallyLogged(relation);
2426 374842 : bool need_cids = RelationIsAccessibleInLogicalDecoding(relation);
2427 374842 : bool starting_with_empty_page = false;
2428 374842 : int npages = 0;
2429 374842 : int npages_used = 0;
2430 :
2431 : /* currently not needed (thus unsupported) for heap_multi_insert() */
2432 : Assert(!(options & HEAP_INSERT_NO_LOGICAL));
2433 :
2434 374842 : AssertHasSnapshotForToast(relation);
2435 :
2436 374842 : needwal = RelationNeedsWAL(relation);
2437 374842 : saveFreeSpace = RelationGetTargetPageFreeSpace(relation,
2438 : HEAP_DEFAULT_FILLFACTOR);
2439 :
2440 : /* Toast and set header data in all the slots */
2441 374842 : heaptuples = palloc(ntuples * sizeof(HeapTuple));
2442 1882866 : for (i = 0; i < ntuples; i++)
2443 : {
2444 : HeapTuple tuple;
2445 :
2446 1508024 : tuple = ExecFetchSlotHeapTuple(slots[i], true, NULL);
2447 1508024 : slots[i]->tts_tableOid = RelationGetRelid(relation);
2448 1508024 : tuple->t_tableOid = slots[i]->tts_tableOid;
2449 1508024 : heaptuples[i] = heap_prepare_insert(relation, tuple, xid, cid,
2450 : options);
2451 : }
2452 :
2453 : /*
2454 : * We're about to do the actual inserts -- but check for conflict first,
2455 : * to minimize the possibility of having to roll back work we've just
2456 : * done.
2457 : *
2458 : * A check here does not definitively prevent a serialization anomaly;
2459 : * that check MUST be done at least past the point of acquiring an
2460 : * exclusive buffer content lock on every buffer that will be affected,
2461 : * and MAY be done after all inserts are reflected in the buffers and
2462 : * those locks are released; otherwise there is a race condition. Since
2463 : * multiple buffers can be locked and unlocked in the loop below, and it
2464 : * would not be feasible to identify and lock all of those buffers before
2465 : * the loop, we must do a final check at the end.
2466 : *
2467 : * The check here could be omitted with no loss of correctness; it is
2468 : * present strictly as an optimization.
2469 : *
2470 : * For heap inserts, we only need to check for table-level SSI locks. Our
2471 : * new tuples can't possibly conflict with existing tuple locks, and heap
2472 : * page locks are only consolidated versions of tuple locks; they do not
2473 : * lock "gaps" as index page locks do. So we don't need to specify a
2474 : * buffer when making the call, which makes for a faster check.
2475 : */
2476 374842 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2477 :
2478 374842 : ndone = 0;
2479 764485 : while (ndone < ntuples)
2480 : {
2481 : Buffer buffer;
2482 389643 : bool all_visible_cleared = false;
2483 389643 : bool all_frozen_set = false;
2484 : int nthispage;
2485 :
2486 389643 : CHECK_FOR_INTERRUPTS();
2487 :
2488 : /*
2489 : * Compute number of pages needed to fit the to-be-inserted tuples in
2490 : * the worst case. This will be used to determine how much to extend
2491 : * the relation by in RelationGetBufferForTuple(), if needed. If we
2492 : * filled a prior page from scratch, we can just update our last
2493 : * computation, but if we started with a partially filled page,
2494 : * recompute from scratch, the number of potentially required pages
2495 : * can vary due to tuples needing to fit onto the page, page headers
2496 : * etc.
2497 : */
2498 389643 : if (ndone == 0 || !starting_with_empty_page)
2499 : {
2500 381437 : npages = heap_multi_insert_pages(heaptuples, ndone, ntuples,
2501 : saveFreeSpace);
2502 381437 : npages_used = 0;
2503 : }
2504 : else
2505 8206 : npages_used++;
2506 :
2507 : /*
2508 : * Find buffer where at least the next tuple will fit. If the page is
2509 : * all-visible, this will also pin the requisite visibility map page.
2510 : *
2511 : * Also pin visibility map page if COPY FREEZE inserts tuples into an
2512 : * empty page. See all_frozen_set below.
2513 : */
2514 389643 : buffer = RelationGetBufferForTuple(relation, heaptuples[ndone]->t_len,
2515 : InvalidBuffer, options, bistate,
2516 : &vmbuffer, NULL,
2517 : npages - npages_used);
2518 389643 : page = BufferGetPage(buffer);
2519 :
2520 389643 : starting_with_empty_page = PageGetMaxOffsetNumber(page) == 0;
2521 :
2522 389643 : if (starting_with_empty_page && (options & HEAP_INSERT_FROZEN))
2523 : {
2524 1661 : all_frozen_set = true;
2525 : /* Lock the vmbuffer before entering the critical section */
2526 1661 : LockBuffer(vmbuffer, BUFFER_LOCK_EXCLUSIVE);
2527 : }
2528 :
2529 : /* NO EREPORT(ERROR) from here till changes are logged */
2530 389643 : START_CRIT_SECTION();
2531 :
2532 : /*
2533 : * RelationGetBufferForTuple has ensured that the first tuple fits.
2534 : * Put that on the page, and then as many other tuples as fit.
2535 : */
2536 389643 : RelationPutHeapTuple(relation, buffer, heaptuples[ndone], false);
2537 :
2538 : /*
2539 : * For logical decoding we need combo CIDs to properly decode the
2540 : * catalog.
2541 : */
2542 389643 : if (needwal && need_cids)
2543 5063 : log_heap_new_cid(relation, heaptuples[ndone]);
2544 :
2545 1508024 : for (nthispage = 1; ndone + nthispage < ntuples; nthispage++)
2546 : {
2547 1133182 : HeapTuple heaptup = heaptuples[ndone + nthispage];
2548 :
2549 1133182 : if (PageGetHeapFreeSpace(page) < MAXALIGN(heaptup->t_len) + saveFreeSpace)
2550 14801 : break;
2551 :
2552 1118381 : RelationPutHeapTuple(relation, buffer, heaptup, false);
2553 :
2554 : /*
2555 : * For logical decoding we need combo CIDs to properly decode the
2556 : * catalog.
2557 : */
2558 1118381 : if (needwal && need_cids)
2559 4738 : log_heap_new_cid(relation, heaptup);
2560 : }
2561 :
2562 : /*
2563 : * If the page is all visible, need to clear that, unless we're only
2564 : * going to add further frozen rows to it.
2565 : *
2566 : * If we're only adding already frozen rows to a previously empty
2567 : * page, mark it as all-frozen and update the visibility map. We're
2568 : * already holding a pin on the vmbuffer.
2569 : */
2570 389643 : if (PageIsAllVisible(page) && !(options & HEAP_INSERT_FROZEN))
2571 : {
2572 3610 : all_visible_cleared = true;
2573 3610 : PageClearAllVisible(page);
2574 3610 : visibilitymap_clear(relation,
2575 : BufferGetBlockNumber(buffer),
2576 : vmbuffer, VISIBILITYMAP_VALID_BITS);
2577 : }
2578 386033 : else if (all_frozen_set)
2579 : {
2580 1661 : PageSetAllVisible(page);
2581 1661 : visibilitymap_set_vmbits(BufferGetBlockNumber(buffer),
2582 : vmbuffer,
2583 : VISIBILITYMAP_ALL_VISIBLE |
2584 : VISIBILITYMAP_ALL_FROZEN,
2585 : relation->rd_locator);
2586 : }
2587 :
2588 : /*
2589 : * XXX Should we set PageSetPrunable on this page ? See heap_insert()
2590 : */
2591 :
2592 389643 : MarkBufferDirty(buffer);
2593 :
2594 : /* XLOG stuff */
2595 389643 : if (needwal)
2596 : {
2597 : XLogRecPtr recptr;
2598 : xl_heap_multi_insert *xlrec;
2599 385820 : uint8 info = XLOG_HEAP2_MULTI_INSERT;
2600 : char *tupledata;
2601 : int totaldatalen;
2602 385820 : char *scratchptr = scratch.data;
2603 : bool init;
2604 385820 : int bufflags = 0;
2605 :
2606 : /*
2607 : * If the page was previously empty, we can reinit the page
2608 : * instead of restoring the whole thing.
2609 : */
2610 385820 : init = starting_with_empty_page;
2611 :
2612 : /* allocate xl_heap_multi_insert struct from the scratch area */
2613 385820 : xlrec = (xl_heap_multi_insert *) scratchptr;
2614 385820 : scratchptr += SizeOfHeapMultiInsert;
2615 :
2616 : /*
2617 : * Allocate offsets array. Unless we're reinitializing the page,
2618 : * in that case the tuples are stored in order starting at
2619 : * FirstOffsetNumber and we don't need to store the offsets
2620 : * explicitly.
2621 : */
2622 385820 : if (!init)
2623 372762 : scratchptr += nthispage * sizeof(OffsetNumber);
2624 :
2625 : /* the rest of the scratch space is used for tuple data */
2626 385820 : tupledata = scratchptr;
2627 :
2628 : /* check that the mutually exclusive flags are not both set */
2629 : Assert(!(all_visible_cleared && all_frozen_set));
2630 :
2631 385820 : xlrec->flags = 0;
2632 385820 : if (all_visible_cleared)
2633 3610 : xlrec->flags = XLH_INSERT_ALL_VISIBLE_CLEARED;
2634 :
2635 : /*
2636 : * We don't have to worry about including a conflict xid in the
2637 : * WAL record, as HEAP_INSERT_FROZEN intentionally violates
2638 : * visibility rules.
2639 : */
2640 385820 : if (all_frozen_set)
2641 17 : xlrec->flags = XLH_INSERT_ALL_FROZEN_SET;
2642 :
2643 385820 : xlrec->ntuples = nthispage;
2644 :
2645 : /*
2646 : * Write out an xl_multi_insert_tuple and the tuple data itself
2647 : * for each tuple.
2648 : */
2649 1688394 : for (i = 0; i < nthispage; i++)
2650 : {
2651 1302574 : HeapTuple heaptup = heaptuples[ndone + i];
2652 : xl_multi_insert_tuple *tuphdr;
2653 : int datalen;
2654 :
2655 1302574 : if (!init)
2656 777737 : xlrec->offsets[i] = ItemPointerGetOffsetNumber(&heaptup->t_self);
2657 : /* xl_multi_insert_tuple needs two-byte alignment. */
2658 1302574 : tuphdr = (xl_multi_insert_tuple *) SHORTALIGN(scratchptr);
2659 1302574 : scratchptr = ((char *) tuphdr) + SizeOfMultiInsertTuple;
2660 :
2661 1302574 : tuphdr->t_infomask2 = heaptup->t_data->t_infomask2;
2662 1302574 : tuphdr->t_infomask = heaptup->t_data->t_infomask;
2663 1302574 : tuphdr->t_hoff = heaptup->t_data->t_hoff;
2664 :
2665 : /* write bitmap [+ padding] [+ oid] + data */
2666 1302574 : datalen = heaptup->t_len - SizeofHeapTupleHeader;
2667 1302574 : memcpy(scratchptr,
2668 1302574 : (char *) heaptup->t_data + SizeofHeapTupleHeader,
2669 : datalen);
2670 1302574 : tuphdr->datalen = datalen;
2671 1302574 : scratchptr += datalen;
2672 : }
2673 385820 : totaldatalen = scratchptr - tupledata;
2674 : Assert((scratchptr - scratch.data) < BLCKSZ);
2675 :
2676 385820 : if (need_tuple_data)
2677 72 : xlrec->flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
2678 :
2679 : /*
2680 : * Signal that this is the last xl_heap_multi_insert record
2681 : * emitted by this call to heap_multi_insert(). Needed for logical
2682 : * decoding so it knows when to cleanup temporary data.
2683 : */
2684 385820 : if (ndone + nthispage == ntuples)
2685 374428 : xlrec->flags |= XLH_INSERT_LAST_IN_MULTI;
2686 :
2687 385820 : if (init)
2688 : {
2689 13058 : info |= XLOG_HEAP_INIT_PAGE;
2690 13058 : bufflags |= REGBUF_WILL_INIT;
2691 : }
2692 :
2693 : /*
2694 : * If we're doing logical decoding, include the new tuple data
2695 : * even if we take a full-page image of the page.
2696 : */
2697 385820 : if (need_tuple_data)
2698 72 : bufflags |= REGBUF_KEEP_DATA;
2699 :
2700 385820 : XLogBeginInsert();
2701 385820 : XLogRegisterData(xlrec, tupledata - scratch.data);
2702 385820 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD | bufflags);
2703 385820 : if (all_frozen_set)
2704 17 : XLogRegisterBuffer(1, vmbuffer, 0);
2705 :
2706 385820 : XLogRegisterBufData(0, tupledata, totaldatalen);
2707 :
2708 : /* filtering by origin on a row level is much more efficient */
2709 385820 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
2710 :
2711 385820 : recptr = XLogInsert(RM_HEAP2_ID, info);
2712 :
2713 385820 : PageSetLSN(page, recptr);
2714 385820 : if (all_frozen_set)
2715 : {
2716 : Assert(BufferIsDirty(vmbuffer));
2717 17 : PageSetLSN(BufferGetPage(vmbuffer), recptr);
2718 : }
2719 : }
2720 :
2721 389643 : END_CRIT_SECTION();
2722 :
2723 389643 : if (all_frozen_set)
2724 1661 : LockBuffer(vmbuffer, BUFFER_LOCK_UNLOCK);
2725 :
2726 389643 : UnlockReleaseBuffer(buffer);
2727 389643 : ndone += nthispage;
2728 :
2729 : /*
2730 : * NB: Only release vmbuffer after inserting all tuples - it's fairly
2731 : * likely that we'll insert into subsequent heap pages that are likely
2732 : * to use the same vm page.
2733 : */
2734 : }
2735 :
2736 : /* We're done with inserting all tuples, so release the last vmbuffer. */
2737 374842 : if (vmbuffer != InvalidBuffer)
2738 3689 : ReleaseBuffer(vmbuffer);
2739 :
2740 : /*
2741 : * We're done with the actual inserts. Check for conflicts again, to
2742 : * ensure that all rw-conflicts in to these inserts are detected. Without
2743 : * this final check, a sequential scan of the heap may have locked the
2744 : * table after the "before" check, missing one opportunity to detect the
2745 : * conflict, and then scanned the table before the new tuples were there,
2746 : * missing the other chance to detect the conflict.
2747 : *
2748 : * For heap inserts, we only need to check for table-level SSI locks. Our
2749 : * new tuples can't possibly conflict with existing tuple locks, and heap
2750 : * page locks are only consolidated versions of tuple locks; they do not
2751 : * lock "gaps" as index page locks do. So we don't need to specify a
2752 : * buffer when making the call.
2753 : */
2754 374842 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2755 :
2756 : /*
2757 : * If tuples are cacheable, mark them for invalidation from the caches in
2758 : * case we abort. Note it is OK to do this after releasing the buffer,
2759 : * because the heaptuples data structure is all in local memory, not in
2760 : * the shared buffer.
2761 : */
2762 374842 : if (IsCatalogRelation(relation))
2763 : {
2764 1279245 : for (i = 0; i < ntuples; i++)
2765 905633 : CacheInvalidateHeapTuple(relation, heaptuples[i], NULL);
2766 : }
2767 :
2768 : /* copy t_self fields back to the caller's slots */
2769 1882866 : for (i = 0; i < ntuples; i++)
2770 1508024 : slots[i]->tts_tid = heaptuples[i]->t_self;
2771 :
2772 374842 : pgstat_count_heap_insert(relation, ntuples);
2773 374842 : }
2774 :
2775 : /*
2776 : * simple_heap_insert - insert a tuple
2777 : *
2778 : * Currently, this routine differs from heap_insert only in supplying
2779 : * a default command ID and not allowing access to the speedup options.
2780 : *
2781 : * This should be used rather than using heap_insert directly in most places
2782 : * where we are modifying system catalogs.
2783 : */
2784 : void
2785 934869 : simple_heap_insert(Relation relation, HeapTuple tup)
2786 : {
2787 934869 : heap_insert(relation, tup, GetCurrentCommandId(true), 0, NULL);
2788 934869 : }
2789 :
2790 : /*
2791 : * Given infomask/infomask2, compute the bits that must be saved in the
2792 : * "infobits" field of xl_heap_delete, xl_heap_update, xl_heap_lock,
2793 : * xl_heap_lock_updated WAL records.
2794 : *
2795 : * See fix_infomask_from_infobits.
2796 : */
2797 : static uint8
2798 2063286 : compute_infobits(uint16 infomask, uint16 infomask2)
2799 : {
2800 : return
2801 2063286 : ((infomask & HEAP_XMAX_IS_MULTI) != 0 ? XLHL_XMAX_IS_MULTI : 0) |
2802 2063286 : ((infomask & HEAP_XMAX_LOCK_ONLY) != 0 ? XLHL_XMAX_LOCK_ONLY : 0) |
2803 2063286 : ((infomask & HEAP_XMAX_EXCL_LOCK) != 0 ? XLHL_XMAX_EXCL_LOCK : 0) |
2804 : /* note we ignore HEAP_XMAX_SHR_LOCK here */
2805 4126572 : ((infomask & HEAP_XMAX_KEYSHR_LOCK) != 0 ? XLHL_XMAX_KEYSHR_LOCK : 0) |
2806 : ((infomask2 & HEAP_KEYS_UPDATED) != 0 ?
2807 2063286 : XLHL_KEYS_UPDATED : 0);
2808 : }
2809 :
2810 : /*
2811 : * Given two versions of the same t_infomask for a tuple, compare them and
2812 : * return whether the relevant status for a tuple Xmax has changed. This is
2813 : * used after a buffer lock has been released and reacquired: we want to ensure
2814 : * that the tuple state continues to be the same it was when we previously
2815 : * examined it.
2816 : *
2817 : * Note the Xmax field itself must be compared separately.
2818 : */
2819 : static inline bool
2820 5386 : xmax_infomask_changed(uint16 new_infomask, uint16 old_infomask)
2821 : {
2822 5386 : const uint16 interesting =
2823 : HEAP_XMAX_IS_MULTI | HEAP_XMAX_LOCK_ONLY | HEAP_LOCK_MASK;
2824 :
2825 5386 : if ((new_infomask & interesting) != (old_infomask & interesting))
2826 16 : return true;
2827 :
2828 5370 : return false;
2829 : }
2830 :
2831 : /*
2832 : * heap_delete - delete a tuple
2833 : *
2834 : * See table_tuple_delete() for an explanation of the parameters, except that
2835 : * this routine directly takes a tuple rather than a slot.
2836 : *
2837 : * In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
2838 : * t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
2839 : * only for TM_SelfModified, since we cannot obtain cmax from a combo CID
2840 : * generated by another transaction).
2841 : */
2842 : TM_Result
2843 1527443 : heap_delete(Relation relation, const ItemPointerData *tid,
2844 : CommandId cid, Snapshot crosscheck, bool wait,
2845 : TM_FailureData *tmfd, bool changingPart)
2846 : {
2847 : TM_Result result;
2848 1527443 : TransactionId xid = GetCurrentTransactionId();
2849 : ItemId lp;
2850 : HeapTupleData tp;
2851 : Page page;
2852 : BlockNumber block;
2853 : Buffer buffer;
2854 1527443 : Buffer vmbuffer = InvalidBuffer;
2855 : TransactionId new_xmax;
2856 : uint16 new_infomask,
2857 : new_infomask2;
2858 1527443 : bool have_tuple_lock = false;
2859 : bool iscombo;
2860 1527443 : bool all_visible_cleared = false;
2861 1527443 : HeapTuple old_key_tuple = NULL; /* replica identity of the tuple */
2862 1527443 : bool old_key_copied = false;
2863 :
2864 : Assert(ItemPointerIsValid(tid));
2865 :
2866 1527443 : AssertHasSnapshotForToast(relation);
2867 :
2868 : /*
2869 : * Forbid this during a parallel operation, lest it allocate a combo CID.
2870 : * Other workers might need that combo CID for visibility checks, and we
2871 : * have no provision for broadcasting it to them.
2872 : */
2873 1527443 : if (IsInParallelMode())
2874 0 : ereport(ERROR,
2875 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
2876 : errmsg("cannot delete tuples during a parallel operation")));
2877 :
2878 1527443 : block = ItemPointerGetBlockNumber(tid);
2879 1527443 : buffer = ReadBuffer(relation, block);
2880 1527443 : page = BufferGetPage(buffer);
2881 :
2882 : /*
2883 : * Before locking the buffer, pin the visibility map page if it appears to
2884 : * be necessary. Since we haven't got the lock yet, someone else might be
2885 : * in the middle of changing this, so we'll need to recheck after we have
2886 : * the lock.
2887 : */
2888 1527443 : if (PageIsAllVisible(page))
2889 243 : visibilitymap_pin(relation, block, &vmbuffer);
2890 :
2891 1527443 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2892 :
2893 1527443 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
2894 : Assert(ItemIdIsNormal(lp));
2895 :
2896 1527443 : tp.t_tableOid = RelationGetRelid(relation);
2897 1527443 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
2898 1527443 : tp.t_len = ItemIdGetLength(lp);
2899 1527443 : tp.t_self = *tid;
2900 :
2901 1 : l1:
2902 :
2903 : /*
2904 : * If we didn't pin the visibility map page and the page has become all
2905 : * visible while we were busy locking the buffer, we'll have to unlock and
2906 : * re-lock, to avoid holding the buffer lock across an I/O. That's a bit
2907 : * unfortunate, but hopefully shouldn't happen often.
2908 : */
2909 1527444 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
2910 : {
2911 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2912 0 : visibilitymap_pin(relation, block, &vmbuffer);
2913 0 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2914 : }
2915 :
2916 1527444 : result = HeapTupleSatisfiesUpdate(&tp, cid, buffer);
2917 :
2918 1527444 : if (result == TM_Invisible)
2919 : {
2920 0 : UnlockReleaseBuffer(buffer);
2921 0 : ereport(ERROR,
2922 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2923 : errmsg("attempted to delete invisible tuple")));
2924 : }
2925 1527444 : else if (result == TM_BeingModified && wait)
2926 : {
2927 : TransactionId xwait;
2928 : uint16 infomask;
2929 :
2930 : /* must copy state data before unlocking buffer */
2931 40561 : xwait = HeapTupleHeaderGetRawXmax(tp.t_data);
2932 40561 : infomask = tp.t_data->t_infomask;
2933 :
2934 : /*
2935 : * Sleep until concurrent transaction ends -- except when there's a
2936 : * single locker and it's our own transaction. Note we don't care
2937 : * which lock mode the locker has, because we need the strongest one.
2938 : *
2939 : * Before sleeping, we need to acquire tuple lock to establish our
2940 : * priority for the tuple (see heap_lock_tuple). LockTuple will
2941 : * release us when we are next-in-line for the tuple.
2942 : *
2943 : * If we are forced to "start over" below, we keep the tuple lock;
2944 : * this arranges that we stay at the head of the line while rechecking
2945 : * tuple state.
2946 : */
2947 40561 : if (infomask & HEAP_XMAX_IS_MULTI)
2948 : {
2949 8 : bool current_is_member = false;
2950 :
2951 8 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
2952 : LockTupleExclusive, ¤t_is_member))
2953 : {
2954 8 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2955 :
2956 : /*
2957 : * Acquire the lock, if necessary (but skip it when we're
2958 : * requesting a lock and already have one; avoids deadlock).
2959 : */
2960 8 : if (!current_is_member)
2961 6 : heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
2962 : LockWaitBlock, &have_tuple_lock);
2963 :
2964 : /* wait for multixact */
2965 8 : MultiXactIdWait((MultiXactId) xwait, MultiXactStatusUpdate, infomask,
2966 : relation, &(tp.t_self), XLTW_Delete,
2967 : NULL);
2968 8 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2969 :
2970 : /*
2971 : * If xwait had just locked the tuple then some other xact
2972 : * could update this tuple before we get to this point. Check
2973 : * for xmax change, and start over if so.
2974 : *
2975 : * We also must start over if we didn't pin the VM page, and
2976 : * the page has become all visible.
2977 : */
2978 16 : if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
2979 16 : xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
2980 8 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
2981 : xwait))
2982 0 : goto l1;
2983 : }
2984 :
2985 : /*
2986 : * You might think the multixact is necessarily done here, but not
2987 : * so: it could have surviving members, namely our own xact or
2988 : * other subxacts of this backend. It is legal for us to delete
2989 : * the tuple in either case, however (the latter case is
2990 : * essentially a situation of upgrading our former shared lock to
2991 : * exclusive). We don't bother changing the on-disk hint bits
2992 : * since we are about to overwrite the xmax altogether.
2993 : */
2994 : }
2995 40553 : else if (!TransactionIdIsCurrentTransactionId(xwait))
2996 : {
2997 : /*
2998 : * Wait for regular transaction to end; but first, acquire tuple
2999 : * lock.
3000 : */
3001 52 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3002 52 : heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
3003 : LockWaitBlock, &have_tuple_lock);
3004 52 : XactLockTableWait(xwait, relation, &(tp.t_self), XLTW_Delete);
3005 48 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3006 :
3007 : /*
3008 : * xwait is done, but if xwait had just locked the tuple then some
3009 : * other xact could update this tuple before we get to this point.
3010 : * Check for xmax change, and start over if so.
3011 : *
3012 : * We also must start over if we didn't pin the VM page, and the
3013 : * page has become all visible.
3014 : */
3015 96 : if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
3016 95 : xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
3017 47 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
3018 : xwait))
3019 1 : goto l1;
3020 :
3021 : /* Otherwise check if it committed or aborted */
3022 47 : UpdateXmaxHintBits(tp.t_data, buffer, xwait);
3023 : }
3024 :
3025 : /*
3026 : * We may overwrite if previous xmax aborted, or if it committed but
3027 : * only locked the tuple without updating it.
3028 : */
3029 81092 : if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
3030 40567 : HEAP_XMAX_IS_LOCKED_ONLY(tp.t_data->t_infomask) ||
3031 31 : HeapTupleHeaderIsOnlyLocked(tp.t_data))
3032 40529 : result = TM_Ok;
3033 27 : else if (!ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
3034 23 : result = TM_Updated;
3035 : else
3036 4 : result = TM_Deleted;
3037 : }
3038 :
3039 : /* sanity check the result HeapTupleSatisfiesUpdate() and the logic above */
3040 : if (result != TM_Ok)
3041 : {
3042 : Assert(result == TM_SelfModified ||
3043 : result == TM_Updated ||
3044 : result == TM_Deleted ||
3045 : result == TM_BeingModified);
3046 : Assert(!(tp.t_data->t_infomask & HEAP_XMAX_INVALID));
3047 : Assert(result != TM_Updated ||
3048 : !ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid));
3049 : }
3050 :
3051 1527439 : if (crosscheck != InvalidSnapshot && result == TM_Ok)
3052 : {
3053 : /* Perform additional check for transaction-snapshot mode RI updates */
3054 1 : if (!HeapTupleSatisfiesVisibility(&tp, crosscheck, buffer))
3055 1 : result = TM_Updated;
3056 : }
3057 :
3058 1527439 : if (result != TM_Ok)
3059 : {
3060 62 : tmfd->ctid = tp.t_data->t_ctid;
3061 62 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
3062 62 : if (result == TM_SelfModified)
3063 21 : tmfd->cmax = HeapTupleHeaderGetCmax(tp.t_data);
3064 : else
3065 41 : tmfd->cmax = InvalidCommandId;
3066 62 : UnlockReleaseBuffer(buffer);
3067 62 : if (have_tuple_lock)
3068 27 : UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
3069 62 : if (vmbuffer != InvalidBuffer)
3070 0 : ReleaseBuffer(vmbuffer);
3071 62 : return result;
3072 : }
3073 :
3074 : /*
3075 : * We're about to do the actual delete -- check for conflict first, to
3076 : * avoid possibly having to roll back work we've just done.
3077 : *
3078 : * This is safe without a recheck as long as there is no possibility of
3079 : * another process scanning the page between this check and the delete
3080 : * being visible to the scan (i.e., an exclusive buffer content lock is
3081 : * continuously held from this point until the tuple delete is visible).
3082 : */
3083 1527377 : CheckForSerializableConflictIn(relation, tid, BufferGetBlockNumber(buffer));
3084 :
3085 : /* replace cid with a combo CID if necessary */
3086 1527363 : HeapTupleHeaderAdjustCmax(tp.t_data, &cid, &iscombo);
3087 :
3088 : /*
3089 : * Compute replica identity tuple before entering the critical section so
3090 : * we don't PANIC upon a memory allocation failure.
3091 : */
3092 1527363 : old_key_tuple = ExtractReplicaIdentity(relation, &tp, true, &old_key_copied);
3093 :
3094 : /*
3095 : * If this is the first possibly-multixact-able operation in the current
3096 : * transaction, set my per-backend OldestMemberMXactId setting. We can be
3097 : * certain that the transaction will never become a member of any older
3098 : * MultiXactIds than that. (We have to do this even if we end up just
3099 : * using our own TransactionId below, since some other backend could
3100 : * incorporate our XID into a MultiXact immediately afterwards.)
3101 : */
3102 1527363 : MultiXactIdSetOldestMember();
3103 :
3104 1527363 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(tp.t_data),
3105 1527363 : tp.t_data->t_infomask, tp.t_data->t_infomask2,
3106 : xid, LockTupleExclusive, true,
3107 : &new_xmax, &new_infomask, &new_infomask2);
3108 :
3109 1527363 : START_CRIT_SECTION();
3110 :
3111 : /*
3112 : * If this transaction commits, the tuple will become DEAD sooner or
3113 : * later. Set flag that this page is a candidate for pruning once our xid
3114 : * falls below the OldestXmin horizon. If the transaction finally aborts,
3115 : * the subsequent page pruning will be a no-op and the hint will be
3116 : * cleared.
3117 : */
3118 1527363 : PageSetPrunable(page, xid);
3119 :
3120 1527363 : if (PageIsAllVisible(page))
3121 : {
3122 243 : all_visible_cleared = true;
3123 243 : PageClearAllVisible(page);
3124 243 : visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
3125 : vmbuffer, VISIBILITYMAP_VALID_BITS);
3126 : }
3127 :
3128 : /* store transaction information of xact deleting the tuple */
3129 1527363 : tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
3130 1527363 : tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
3131 1527363 : tp.t_data->t_infomask |= new_infomask;
3132 1527363 : tp.t_data->t_infomask2 |= new_infomask2;
3133 1527363 : HeapTupleHeaderClearHotUpdated(tp.t_data);
3134 1527363 : HeapTupleHeaderSetXmax(tp.t_data, new_xmax);
3135 1527363 : HeapTupleHeaderSetCmax(tp.t_data, cid, iscombo);
3136 : /* Make sure there is no forward chain link in t_ctid */
3137 1527363 : tp.t_data->t_ctid = tp.t_self;
3138 :
3139 : /* Signal that this is actually a move into another partition */
3140 1527363 : if (changingPart)
3141 493 : HeapTupleHeaderSetMovedPartitions(tp.t_data);
3142 :
3143 1527363 : MarkBufferDirty(buffer);
3144 :
3145 : /*
3146 : * XLOG stuff
3147 : *
3148 : * NB: heap_abort_speculative() uses the same xlog record and replay
3149 : * routines.
3150 : */
3151 1527363 : if (RelationNeedsWAL(relation))
3152 : {
3153 : xl_heap_delete xlrec;
3154 : xl_heap_header xlhdr;
3155 : XLogRecPtr recptr;
3156 :
3157 : /*
3158 : * For logical decode we need combo CIDs to properly decode the
3159 : * catalog
3160 : */
3161 1464769 : if (RelationIsAccessibleInLogicalDecoding(relation))
3162 6291 : log_heap_new_cid(relation, &tp);
3163 :
3164 1464769 : xlrec.flags = 0;
3165 1464769 : if (all_visible_cleared)
3166 243 : xlrec.flags |= XLH_DELETE_ALL_VISIBLE_CLEARED;
3167 1464769 : if (changingPart)
3168 493 : xlrec.flags |= XLH_DELETE_IS_PARTITION_MOVE;
3169 2929538 : xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
3170 1464769 : tp.t_data->t_infomask2);
3171 1464769 : xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
3172 1464769 : xlrec.xmax = new_xmax;
3173 :
3174 1464769 : if (old_key_tuple != NULL)
3175 : {
3176 47018 : if (relation->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
3177 132 : xlrec.flags |= XLH_DELETE_CONTAINS_OLD_TUPLE;
3178 : else
3179 46886 : xlrec.flags |= XLH_DELETE_CONTAINS_OLD_KEY;
3180 : }
3181 :
3182 1464769 : XLogBeginInsert();
3183 1464769 : XLogRegisterData(&xlrec, SizeOfHeapDelete);
3184 :
3185 1464769 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
3186 :
3187 : /*
3188 : * Log replica identity of the deleted tuple if there is one
3189 : */
3190 1464769 : if (old_key_tuple != NULL)
3191 : {
3192 47018 : xlhdr.t_infomask2 = old_key_tuple->t_data->t_infomask2;
3193 47018 : xlhdr.t_infomask = old_key_tuple->t_data->t_infomask;
3194 47018 : xlhdr.t_hoff = old_key_tuple->t_data->t_hoff;
3195 :
3196 47018 : XLogRegisterData(&xlhdr, SizeOfHeapHeader);
3197 47018 : XLogRegisterData((char *) old_key_tuple->t_data
3198 : + SizeofHeapTupleHeader,
3199 47018 : old_key_tuple->t_len
3200 : - SizeofHeapTupleHeader);
3201 : }
3202 :
3203 : /* filtering by origin on a row level is much more efficient */
3204 1464769 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
3205 :
3206 1464769 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
3207 :
3208 1464769 : PageSetLSN(page, recptr);
3209 : }
3210 :
3211 1527363 : END_CRIT_SECTION();
3212 :
3213 1527363 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3214 :
3215 1527363 : if (vmbuffer != InvalidBuffer)
3216 243 : ReleaseBuffer(vmbuffer);
3217 :
3218 : /*
3219 : * If the tuple has toasted out-of-line attributes, we need to delete
3220 : * those items too. We have to do this before releasing the buffer
3221 : * because we need to look at the contents of the tuple, but it's OK to
3222 : * release the content lock on the buffer first.
3223 : */
3224 1529946 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
3225 2593 : relation->rd_rel->relkind != RELKIND_MATVIEW)
3226 : {
3227 : /* toast table entries should never be recursively toasted */
3228 : Assert(!HeapTupleHasExternal(&tp));
3229 : }
3230 1524780 : else if (HeapTupleHasExternal(&tp))
3231 294 : heap_toast_delete(relation, &tp, false);
3232 :
3233 : /*
3234 : * Mark tuple for invalidation from system caches at next command
3235 : * boundary. We have to do this before releasing the buffer because we
3236 : * need to look at the contents of the tuple.
3237 : */
3238 1527363 : CacheInvalidateHeapTuple(relation, &tp, NULL);
3239 :
3240 : /* Now we can release the buffer */
3241 1527363 : ReleaseBuffer(buffer);
3242 :
3243 : /*
3244 : * Release the lmgr tuple lock, if we had it.
3245 : */
3246 1527363 : if (have_tuple_lock)
3247 26 : UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
3248 :
3249 1527363 : pgstat_count_heap_delete(relation);
3250 :
3251 1527363 : if (old_key_tuple != NULL && old_key_copied)
3252 46887 : heap_freetuple(old_key_tuple);
3253 :
3254 1527363 : return TM_Ok;
3255 : }
3256 :
3257 : /*
3258 : * simple_heap_delete - delete a tuple
3259 : *
3260 : * This routine may be used to delete a tuple when concurrent updates of
3261 : * the target tuple are not expected (for example, because we have a lock
3262 : * on the relation associated with the tuple). Any failure is reported
3263 : * via ereport().
3264 : */
3265 : void
3266 661464 : simple_heap_delete(Relation relation, const ItemPointerData *tid)
3267 : {
3268 : TM_Result result;
3269 : TM_FailureData tmfd;
3270 :
3271 661464 : result = heap_delete(relation, tid,
3272 : GetCurrentCommandId(true), InvalidSnapshot,
3273 : true /* wait for commit */ ,
3274 : &tmfd, false /* changingPart */ );
3275 661464 : switch (result)
3276 : {
3277 0 : case TM_SelfModified:
3278 : /* Tuple was already updated in current command? */
3279 0 : elog(ERROR, "tuple already updated by self");
3280 : break;
3281 :
3282 661464 : case TM_Ok:
3283 : /* done successfully */
3284 661464 : break;
3285 :
3286 0 : case TM_Updated:
3287 0 : elog(ERROR, "tuple concurrently updated");
3288 : break;
3289 :
3290 0 : case TM_Deleted:
3291 0 : elog(ERROR, "tuple concurrently deleted");
3292 : break;
3293 :
3294 0 : default:
3295 0 : elog(ERROR, "unrecognized heap_delete status: %u", result);
3296 : break;
3297 : }
3298 661464 : }
3299 :
3300 : /*
3301 : * heap_update - replace a tuple
3302 : *
3303 : * See table_tuple_update() for an explanation of the parameters, except that
3304 : * this routine directly takes a tuple rather than a slot.
3305 : *
3306 : * In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
3307 : * t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
3308 : * only for TM_SelfModified, since we cannot obtain cmax from a combo CID
3309 : * generated by another transaction).
3310 : */
3311 : TM_Result
3312 310910 : heap_update(Relation relation, const ItemPointerData *otid, HeapTuple newtup,
3313 : CommandId cid, Snapshot crosscheck, bool wait,
3314 : TM_FailureData *tmfd, LockTupleMode *lockmode,
3315 : TU_UpdateIndexes *update_indexes)
3316 : {
3317 : TM_Result result;
3318 310910 : TransactionId xid = GetCurrentTransactionId();
3319 : Bitmapset *hot_attrs;
3320 : Bitmapset *sum_attrs;
3321 : Bitmapset *key_attrs;
3322 : Bitmapset *id_attrs;
3323 : Bitmapset *interesting_attrs;
3324 : Bitmapset *modified_attrs;
3325 : ItemId lp;
3326 : HeapTupleData oldtup;
3327 : HeapTuple heaptup;
3328 310910 : HeapTuple old_key_tuple = NULL;
3329 310910 : bool old_key_copied = false;
3330 : Page page;
3331 : BlockNumber block;
3332 : MultiXactStatus mxact_status;
3333 : Buffer buffer,
3334 : newbuf,
3335 310910 : vmbuffer = InvalidBuffer,
3336 310910 : vmbuffer_new = InvalidBuffer;
3337 : bool need_toast;
3338 : Size newtupsize,
3339 : pagefree;
3340 310910 : bool have_tuple_lock = false;
3341 : bool iscombo;
3342 310910 : bool use_hot_update = false;
3343 310910 : bool summarized_update = false;
3344 : bool key_intact;
3345 310910 : bool all_visible_cleared = false;
3346 310910 : bool all_visible_cleared_new = false;
3347 : bool checked_lockers;
3348 : bool locker_remains;
3349 310910 : bool id_has_external = false;
3350 : TransactionId xmax_new_tuple,
3351 : xmax_old_tuple;
3352 : uint16 infomask_old_tuple,
3353 : infomask2_old_tuple,
3354 : infomask_new_tuple,
3355 : infomask2_new_tuple;
3356 :
3357 : Assert(ItemPointerIsValid(otid));
3358 :
3359 : /* Cheap, simplistic check that the tuple matches the rel's rowtype. */
3360 : Assert(HeapTupleHeaderGetNatts(newtup->t_data) <=
3361 : RelationGetNumberOfAttributes(relation));
3362 :
3363 310910 : AssertHasSnapshotForToast(relation);
3364 :
3365 : /*
3366 : * Forbid this during a parallel operation, lest it allocate a combo CID.
3367 : * Other workers might need that combo CID for visibility checks, and we
3368 : * have no provision for broadcasting it to them.
3369 : */
3370 310910 : if (IsInParallelMode())
3371 0 : ereport(ERROR,
3372 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
3373 : errmsg("cannot update tuples during a parallel operation")));
3374 :
3375 : #ifdef USE_ASSERT_CHECKING
3376 : check_lock_if_inplace_updateable_rel(relation, otid, newtup);
3377 : #endif
3378 :
3379 : /*
3380 : * Fetch the list of attributes to be checked for various operations.
3381 : *
3382 : * For HOT considerations, this is wasted effort if we fail to update or
3383 : * have to put the new tuple on a different page. But we must compute the
3384 : * list before obtaining buffer lock --- in the worst case, if we are
3385 : * doing an update on one of the relevant system catalogs, we could
3386 : * deadlock if we try to fetch the list later. In any case, the relcache
3387 : * caches the data so this is usually pretty cheap.
3388 : *
3389 : * We also need columns used by the replica identity and columns that are
3390 : * considered the "key" of rows in the table.
3391 : *
3392 : * Note that we get copies of each bitmap, so we need not worry about
3393 : * relcache flush happening midway through.
3394 : */
3395 310910 : hot_attrs = RelationGetIndexAttrBitmap(relation,
3396 : INDEX_ATTR_BITMAP_HOT_BLOCKING);
3397 310910 : sum_attrs = RelationGetIndexAttrBitmap(relation,
3398 : INDEX_ATTR_BITMAP_SUMMARIZED);
3399 310910 : key_attrs = RelationGetIndexAttrBitmap(relation, INDEX_ATTR_BITMAP_KEY);
3400 310910 : id_attrs = RelationGetIndexAttrBitmap(relation,
3401 : INDEX_ATTR_BITMAP_IDENTITY_KEY);
3402 310910 : interesting_attrs = NULL;
3403 310910 : interesting_attrs = bms_add_members(interesting_attrs, hot_attrs);
3404 310910 : interesting_attrs = bms_add_members(interesting_attrs, sum_attrs);
3405 310910 : interesting_attrs = bms_add_members(interesting_attrs, key_attrs);
3406 310910 : interesting_attrs = bms_add_members(interesting_attrs, id_attrs);
3407 :
3408 310910 : block = ItemPointerGetBlockNumber(otid);
3409 310910 : INJECTION_POINT("heap_update-before-pin", NULL);
3410 310910 : buffer = ReadBuffer(relation, block);
3411 310910 : page = BufferGetPage(buffer);
3412 :
3413 : /*
3414 : * Before locking the buffer, pin the visibility map page if it appears to
3415 : * be necessary. Since we haven't got the lock yet, someone else might be
3416 : * in the middle of changing this, so we'll need to recheck after we have
3417 : * the lock.
3418 : */
3419 310910 : if (PageIsAllVisible(page))
3420 1604 : visibilitymap_pin(relation, block, &vmbuffer);
3421 :
3422 310910 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3423 :
3424 310910 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(otid));
3425 :
3426 : /*
3427 : * Usually, a buffer pin and/or snapshot blocks pruning of otid, ensuring
3428 : * we see LP_NORMAL here. When the otid origin is a syscache, we may have
3429 : * neither a pin nor a snapshot. Hence, we may see other LP_ states, each
3430 : * of which indicates concurrent pruning.
3431 : *
3432 : * Failing with TM_Updated would be most accurate. However, unlike other
3433 : * TM_Updated scenarios, we don't know the successor ctid in LP_UNUSED and
3434 : * LP_DEAD cases. While the distinction between TM_Updated and TM_Deleted
3435 : * does matter to SQL statements UPDATE and MERGE, those SQL statements
3436 : * hold a snapshot that ensures LP_NORMAL. Hence, the choice between
3437 : * TM_Updated and TM_Deleted affects only the wording of error messages.
3438 : * Settle on TM_Deleted, for two reasons. First, it avoids complicating
3439 : * the specification of when tmfd->ctid is valid. Second, it creates
3440 : * error log evidence that we took this branch.
3441 : *
3442 : * Since it's possible to see LP_UNUSED at otid, it's also possible to see
3443 : * LP_NORMAL for a tuple that replaced LP_UNUSED. If it's a tuple for an
3444 : * unrelated row, we'll fail with "duplicate key value violates unique".
3445 : * XXX if otid is the live, newer version of the newtup row, we'll discard
3446 : * changes originating in versions of this catalog row after the version
3447 : * the caller got from syscache. See syscache-update-pruned.spec.
3448 : */
3449 310910 : if (!ItemIdIsNormal(lp))
3450 : {
3451 : Assert(RelationSupportsSysCache(RelationGetRelid(relation)));
3452 :
3453 1 : UnlockReleaseBuffer(buffer);
3454 : Assert(!have_tuple_lock);
3455 1 : if (vmbuffer != InvalidBuffer)
3456 1 : ReleaseBuffer(vmbuffer);
3457 1 : tmfd->ctid = *otid;
3458 1 : tmfd->xmax = InvalidTransactionId;
3459 1 : tmfd->cmax = InvalidCommandId;
3460 1 : *update_indexes = TU_None;
3461 :
3462 1 : bms_free(hot_attrs);
3463 1 : bms_free(sum_attrs);
3464 1 : bms_free(key_attrs);
3465 1 : bms_free(id_attrs);
3466 : /* modified_attrs not yet initialized */
3467 1 : bms_free(interesting_attrs);
3468 1 : return TM_Deleted;
3469 : }
3470 :
3471 : /*
3472 : * Fill in enough data in oldtup for HeapDetermineColumnsInfo to work
3473 : * properly.
3474 : */
3475 310909 : oldtup.t_tableOid = RelationGetRelid(relation);
3476 310909 : oldtup.t_data = (HeapTupleHeader) PageGetItem(page, lp);
3477 310909 : oldtup.t_len = ItemIdGetLength(lp);
3478 310909 : oldtup.t_self = *otid;
3479 :
3480 : /* the new tuple is ready, except for this: */
3481 310909 : newtup->t_tableOid = RelationGetRelid(relation);
3482 :
3483 : /*
3484 : * Determine columns modified by the update. Additionally, identify
3485 : * whether any of the unmodified replica identity key attributes in the
3486 : * old tuple is externally stored or not. This is required because for
3487 : * such attributes the flattened value won't be WAL logged as part of the
3488 : * new tuple so we must include it as part of the old_key_tuple. See
3489 : * ExtractReplicaIdentity.
3490 : */
3491 310909 : modified_attrs = HeapDetermineColumnsInfo(relation, interesting_attrs,
3492 : id_attrs, &oldtup,
3493 : newtup, &id_has_external);
3494 :
3495 : /*
3496 : * If we're not updating any "key" column, we can grab a weaker lock type.
3497 : * This allows for more concurrency when we are running simultaneously
3498 : * with foreign key checks.
3499 : *
3500 : * Note that if a column gets detoasted while executing the update, but
3501 : * the value ends up being the same, this test will fail and we will use
3502 : * the stronger lock. This is acceptable; the important case to optimize
3503 : * is updates that don't manipulate key columns, not those that
3504 : * serendipitously arrive at the same key values.
3505 : */
3506 310909 : if (!bms_overlap(modified_attrs, key_attrs))
3507 : {
3508 306586 : *lockmode = LockTupleNoKeyExclusive;
3509 306586 : mxact_status = MultiXactStatusNoKeyUpdate;
3510 306586 : key_intact = true;
3511 :
3512 : /*
3513 : * If this is the first possibly-multixact-able operation in the
3514 : * current transaction, set my per-backend OldestMemberMXactId
3515 : * setting. We can be certain that the transaction will never become a
3516 : * member of any older MultiXactIds than that. (We have to do this
3517 : * even if we end up just using our own TransactionId below, since
3518 : * some other backend could incorporate our XID into a MultiXact
3519 : * immediately afterwards.)
3520 : */
3521 306586 : MultiXactIdSetOldestMember();
3522 : }
3523 : else
3524 : {
3525 4323 : *lockmode = LockTupleExclusive;
3526 4323 : mxact_status = MultiXactStatusUpdate;
3527 4323 : key_intact = false;
3528 : }
3529 :
3530 : /*
3531 : * Note: beyond this point, use oldtup not otid to refer to old tuple.
3532 : * otid may very well point at newtup->t_self, which we will overwrite
3533 : * with the new tuple's location, so there's great risk of confusion if we
3534 : * use otid anymore.
3535 : */
3536 :
3537 1 : l2:
3538 310910 : checked_lockers = false;
3539 310910 : locker_remains = false;
3540 310910 : result = HeapTupleSatisfiesUpdate(&oldtup, cid, buffer);
3541 :
3542 : /* see below about the "no wait" case */
3543 : Assert(result != TM_BeingModified || wait);
3544 :
3545 310910 : if (result == TM_Invisible)
3546 : {
3547 0 : UnlockReleaseBuffer(buffer);
3548 0 : ereport(ERROR,
3549 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3550 : errmsg("attempted to update invisible tuple")));
3551 : }
3552 310910 : else if (result == TM_BeingModified && wait)
3553 : {
3554 : TransactionId xwait;
3555 : uint16 infomask;
3556 36111 : bool can_continue = false;
3557 :
3558 : /*
3559 : * XXX note that we don't consider the "no wait" case here. This
3560 : * isn't a problem currently because no caller uses that case, but it
3561 : * should be fixed if such a caller is introduced. It wasn't a
3562 : * problem previously because this code would always wait, but now
3563 : * that some tuple locks do not conflict with one of the lock modes we
3564 : * use, it is possible that this case is interesting to handle
3565 : * specially.
3566 : *
3567 : * This may cause failures with third-party code that calls
3568 : * heap_update directly.
3569 : */
3570 :
3571 : /* must copy state data before unlocking buffer */
3572 36111 : xwait = HeapTupleHeaderGetRawXmax(oldtup.t_data);
3573 36111 : infomask = oldtup.t_data->t_infomask;
3574 :
3575 : /*
3576 : * Now we have to do something about the existing locker. If it's a
3577 : * multi, sleep on it; we might be awakened before it is completely
3578 : * gone (or even not sleep at all in some cases); we need to preserve
3579 : * it as locker, unless it is gone completely.
3580 : *
3581 : * If it's not a multi, we need to check for sleeping conditions
3582 : * before actually going to sleep. If the update doesn't conflict
3583 : * with the locks, we just continue without sleeping (but making sure
3584 : * it is preserved).
3585 : *
3586 : * Before sleeping, we need to acquire tuple lock to establish our
3587 : * priority for the tuple (see heap_lock_tuple). LockTuple will
3588 : * release us when we are next-in-line for the tuple. Note we must
3589 : * not acquire the tuple lock until we're sure we're going to sleep;
3590 : * otherwise we're open for race conditions with other transactions
3591 : * holding the tuple lock which sleep on us.
3592 : *
3593 : * If we are forced to "start over" below, we keep the tuple lock;
3594 : * this arranges that we stay at the head of the line while rechecking
3595 : * tuple state.
3596 : */
3597 36111 : if (infomask & HEAP_XMAX_IS_MULTI)
3598 : {
3599 : TransactionId update_xact;
3600 : int remain;
3601 179 : bool current_is_member = false;
3602 :
3603 179 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
3604 : *lockmode, ¤t_is_member))
3605 : {
3606 8 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3607 :
3608 : /*
3609 : * Acquire the lock, if necessary (but skip it when we're
3610 : * requesting a lock and already have one; avoids deadlock).
3611 : */
3612 8 : if (!current_is_member)
3613 0 : heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3614 : LockWaitBlock, &have_tuple_lock);
3615 :
3616 : /* wait for multixact */
3617 8 : MultiXactIdWait((MultiXactId) xwait, mxact_status, infomask,
3618 : relation, &oldtup.t_self, XLTW_Update,
3619 : &remain);
3620 8 : checked_lockers = true;
3621 8 : locker_remains = remain != 0;
3622 8 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3623 :
3624 : /*
3625 : * If xwait had just locked the tuple then some other xact
3626 : * could update this tuple before we get to this point. Check
3627 : * for xmax change, and start over if so.
3628 : */
3629 8 : if (xmax_infomask_changed(oldtup.t_data->t_infomask,
3630 8 : infomask) ||
3631 8 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3632 : xwait))
3633 0 : goto l2;
3634 : }
3635 :
3636 : /*
3637 : * Note that the multixact may not be done by now. It could have
3638 : * surviving members; our own xact or other subxacts of this
3639 : * backend, and also any other concurrent transaction that locked
3640 : * the tuple with LockTupleKeyShare if we only got
3641 : * LockTupleNoKeyExclusive. If this is the case, we have to be
3642 : * careful to mark the updated tuple with the surviving members in
3643 : * Xmax.
3644 : *
3645 : * Note that there could have been another update in the
3646 : * MultiXact. In that case, we need to check whether it committed
3647 : * or aborted. If it aborted we are safe to update it again;
3648 : * otherwise there is an update conflict, and we have to return
3649 : * TableTuple{Deleted, Updated} below.
3650 : *
3651 : * In the LockTupleExclusive case, we still need to preserve the
3652 : * surviving members: those would include the tuple locks we had
3653 : * before this one, which are important to keep in case this
3654 : * subxact aborts.
3655 : */
3656 179 : if (!HEAP_XMAX_IS_LOCKED_ONLY(oldtup.t_data->t_infomask))
3657 8 : update_xact = HeapTupleGetUpdateXid(oldtup.t_data);
3658 : else
3659 171 : update_xact = InvalidTransactionId;
3660 :
3661 : /*
3662 : * There was no UPDATE in the MultiXact; or it aborted. No
3663 : * TransactionIdIsInProgress() call needed here, since we called
3664 : * MultiXactIdWait() above.
3665 : */
3666 187 : if (!TransactionIdIsValid(update_xact) ||
3667 8 : TransactionIdDidAbort(update_xact))
3668 172 : can_continue = true;
3669 : }
3670 35932 : else if (TransactionIdIsCurrentTransactionId(xwait))
3671 : {
3672 : /*
3673 : * The only locker is ourselves; we can avoid grabbing the tuple
3674 : * lock here, but must preserve our locking information.
3675 : */
3676 35823 : checked_lockers = true;
3677 35823 : locker_remains = true;
3678 35823 : can_continue = true;
3679 : }
3680 109 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) && key_intact)
3681 : {
3682 : /*
3683 : * If it's just a key-share locker, and we're not changing the key
3684 : * columns, we don't need to wait for it to end; but we need to
3685 : * preserve it as locker.
3686 : */
3687 29 : checked_lockers = true;
3688 29 : locker_remains = true;
3689 29 : can_continue = true;
3690 : }
3691 : else
3692 : {
3693 : /*
3694 : * Wait for regular transaction to end; but first, acquire tuple
3695 : * lock.
3696 : */
3697 80 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3698 80 : heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3699 : LockWaitBlock, &have_tuple_lock);
3700 80 : XactLockTableWait(xwait, relation, &oldtup.t_self,
3701 : XLTW_Update);
3702 80 : checked_lockers = true;
3703 80 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3704 :
3705 : /*
3706 : * xwait is done, but if xwait had just locked the tuple then some
3707 : * other xact could update this tuple before we get to this point.
3708 : * Check for xmax change, and start over if so.
3709 : */
3710 159 : if (xmax_infomask_changed(oldtup.t_data->t_infomask, infomask) ||
3711 79 : !TransactionIdEquals(xwait,
3712 : HeapTupleHeaderGetRawXmax(oldtup.t_data)))
3713 1 : goto l2;
3714 :
3715 : /* Otherwise check if it committed or aborted */
3716 79 : UpdateXmaxHintBits(oldtup.t_data, buffer, xwait);
3717 79 : if (oldtup.t_data->t_infomask & HEAP_XMAX_INVALID)
3718 22 : can_continue = true;
3719 : }
3720 :
3721 36110 : if (can_continue)
3722 36046 : result = TM_Ok;
3723 64 : else if (!ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid))
3724 59 : result = TM_Updated;
3725 : else
3726 5 : result = TM_Deleted;
3727 : }
3728 :
3729 : /* Sanity check the result HeapTupleSatisfiesUpdate() and the logic above */
3730 : if (result != TM_Ok)
3731 : {
3732 : Assert(result == TM_SelfModified ||
3733 : result == TM_Updated ||
3734 : result == TM_Deleted ||
3735 : result == TM_BeingModified);
3736 : Assert(!(oldtup.t_data->t_infomask & HEAP_XMAX_INVALID));
3737 : Assert(result != TM_Updated ||
3738 : !ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid));
3739 : }
3740 :
3741 310909 : if (crosscheck != InvalidSnapshot && result == TM_Ok)
3742 : {
3743 : /* Perform additional check for transaction-snapshot mode RI updates */
3744 1 : if (!HeapTupleSatisfiesVisibility(&oldtup, crosscheck, buffer))
3745 1 : result = TM_Updated;
3746 : }
3747 :
3748 310909 : if (result != TM_Ok)
3749 : {
3750 162 : tmfd->ctid = oldtup.t_data->t_ctid;
3751 162 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(oldtup.t_data);
3752 162 : if (result == TM_SelfModified)
3753 52 : tmfd->cmax = HeapTupleHeaderGetCmax(oldtup.t_data);
3754 : else
3755 110 : tmfd->cmax = InvalidCommandId;
3756 162 : UnlockReleaseBuffer(buffer);
3757 162 : if (have_tuple_lock)
3758 57 : UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
3759 162 : if (vmbuffer != InvalidBuffer)
3760 0 : ReleaseBuffer(vmbuffer);
3761 162 : *update_indexes = TU_None;
3762 :
3763 162 : bms_free(hot_attrs);
3764 162 : bms_free(sum_attrs);
3765 162 : bms_free(key_attrs);
3766 162 : bms_free(id_attrs);
3767 162 : bms_free(modified_attrs);
3768 162 : bms_free(interesting_attrs);
3769 162 : return result;
3770 : }
3771 :
3772 : /*
3773 : * If we didn't pin the visibility map page and the page has become all
3774 : * visible while we were busy locking the buffer, or during some
3775 : * subsequent window during which we had it unlocked, we'll have to unlock
3776 : * and re-lock, to avoid holding the buffer lock across an I/O. That's a
3777 : * bit unfortunate, especially since we'll now have to recheck whether the
3778 : * tuple has been locked or updated under us, but hopefully it won't
3779 : * happen very often.
3780 : */
3781 310747 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
3782 : {
3783 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3784 0 : visibilitymap_pin(relation, block, &vmbuffer);
3785 0 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3786 0 : goto l2;
3787 : }
3788 :
3789 : /* Fill in transaction status data */
3790 :
3791 : /*
3792 : * If the tuple we're updating is locked, we need to preserve the locking
3793 : * info in the old tuple's Xmax. Prepare a new Xmax value for this.
3794 : */
3795 310747 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3796 310747 : oldtup.t_data->t_infomask,
3797 310747 : oldtup.t_data->t_infomask2,
3798 : xid, *lockmode, true,
3799 : &xmax_old_tuple, &infomask_old_tuple,
3800 : &infomask2_old_tuple);
3801 :
3802 : /*
3803 : * And also prepare an Xmax value for the new copy of the tuple. If there
3804 : * was no xmax previously, or there was one but all lockers are now gone,
3805 : * then use InvalidTransactionId; otherwise, get the xmax from the old
3806 : * tuple. (In rare cases that might also be InvalidTransactionId and yet
3807 : * not have the HEAP_XMAX_INVALID bit set; that's fine.)
3808 : */
3809 346771 : if ((oldtup.t_data->t_infomask & HEAP_XMAX_INVALID) ||
3810 72048 : HEAP_LOCKED_UPGRADED(oldtup.t_data->t_infomask) ||
3811 35853 : (checked_lockers && !locker_remains))
3812 274723 : xmax_new_tuple = InvalidTransactionId;
3813 : else
3814 36024 : xmax_new_tuple = HeapTupleHeaderGetRawXmax(oldtup.t_data);
3815 :
3816 310747 : if (!TransactionIdIsValid(xmax_new_tuple))
3817 : {
3818 274723 : infomask_new_tuple = HEAP_XMAX_INVALID;
3819 274723 : infomask2_new_tuple = 0;
3820 : }
3821 : else
3822 : {
3823 : /*
3824 : * If we found a valid Xmax for the new tuple, then the infomask bits
3825 : * to use on the new tuple depend on what was there on the old one.
3826 : * Note that since we're doing an update, the only possibility is that
3827 : * the lockers had FOR KEY SHARE lock.
3828 : */
3829 36024 : if (oldtup.t_data->t_infomask & HEAP_XMAX_IS_MULTI)
3830 : {
3831 172 : GetMultiXactIdHintBits(xmax_new_tuple, &infomask_new_tuple,
3832 : &infomask2_new_tuple);
3833 : }
3834 : else
3835 : {
3836 35852 : infomask_new_tuple = HEAP_XMAX_KEYSHR_LOCK | HEAP_XMAX_LOCK_ONLY;
3837 35852 : infomask2_new_tuple = 0;
3838 : }
3839 : }
3840 :
3841 : /*
3842 : * Prepare the new tuple with the appropriate initial values of Xmin and
3843 : * Xmax, as well as initial infomask bits as computed above.
3844 : */
3845 310747 : newtup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
3846 310747 : newtup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
3847 310747 : HeapTupleHeaderSetXmin(newtup->t_data, xid);
3848 310747 : HeapTupleHeaderSetCmin(newtup->t_data, cid);
3849 310747 : newtup->t_data->t_infomask |= HEAP_UPDATED | infomask_new_tuple;
3850 310747 : newtup->t_data->t_infomask2 |= infomask2_new_tuple;
3851 310747 : HeapTupleHeaderSetXmax(newtup->t_data, xmax_new_tuple);
3852 :
3853 : /*
3854 : * Replace cid with a combo CID if necessary. Note that we already put
3855 : * the plain cid into the new tuple.
3856 : */
3857 310747 : HeapTupleHeaderAdjustCmax(oldtup.t_data, &cid, &iscombo);
3858 :
3859 : /*
3860 : * If the toaster needs to be activated, OR if the new tuple will not fit
3861 : * on the same page as the old, then we need to release the content lock
3862 : * (but not the pin!) on the old tuple's buffer while we are off doing
3863 : * TOAST and/or table-file-extension work. We must mark the old tuple to
3864 : * show that it's locked, else other processes may try to update it
3865 : * themselves.
3866 : *
3867 : * We need to invoke the toaster if there are already any out-of-line
3868 : * toasted values present, or if the new tuple is over-threshold.
3869 : */
3870 310747 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
3871 0 : relation->rd_rel->relkind != RELKIND_MATVIEW)
3872 : {
3873 : /* toast table entries should never be recursively toasted */
3874 : Assert(!HeapTupleHasExternal(&oldtup));
3875 : Assert(!HeapTupleHasExternal(newtup));
3876 0 : need_toast = false;
3877 : }
3878 : else
3879 931852 : need_toast = (HeapTupleHasExternal(&oldtup) ||
3880 621105 : HeapTupleHasExternal(newtup) ||
3881 310334 : newtup->t_len > TOAST_TUPLE_THRESHOLD);
3882 :
3883 310747 : pagefree = PageGetHeapFreeSpace(page);
3884 :
3885 310747 : newtupsize = MAXALIGN(newtup->t_len);
3886 :
3887 310747 : if (need_toast || newtupsize > pagefree)
3888 150781 : {
3889 : TransactionId xmax_lock_old_tuple;
3890 : uint16 infomask_lock_old_tuple,
3891 : infomask2_lock_old_tuple;
3892 150781 : bool cleared_all_frozen = false;
3893 :
3894 : /*
3895 : * To prevent concurrent sessions from updating the tuple, we have to
3896 : * temporarily mark it locked, while we release the page-level lock.
3897 : *
3898 : * To satisfy the rule that any xid potentially appearing in a buffer
3899 : * written out to disk, we unfortunately have to WAL log this
3900 : * temporary modification. We can reuse xl_heap_lock for this
3901 : * purpose. If we crash/error before following through with the
3902 : * actual update, xmax will be of an aborted transaction, allowing
3903 : * other sessions to proceed.
3904 : */
3905 :
3906 : /*
3907 : * Compute xmax / infomask appropriate for locking the tuple. This has
3908 : * to be done separately from the combo that's going to be used for
3909 : * updating, because the potentially created multixact would otherwise
3910 : * be wrong.
3911 : */
3912 150781 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3913 150781 : oldtup.t_data->t_infomask,
3914 150781 : oldtup.t_data->t_infomask2,
3915 : xid, *lockmode, false,
3916 : &xmax_lock_old_tuple, &infomask_lock_old_tuple,
3917 : &infomask2_lock_old_tuple);
3918 :
3919 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(infomask_lock_old_tuple));
3920 :
3921 150781 : START_CRIT_SECTION();
3922 :
3923 : /* Clear obsolete visibility flags ... */
3924 150781 : oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
3925 150781 : oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
3926 150781 : HeapTupleClearHotUpdated(&oldtup);
3927 : /* ... and store info about transaction updating this tuple */
3928 : Assert(TransactionIdIsValid(xmax_lock_old_tuple));
3929 150781 : HeapTupleHeaderSetXmax(oldtup.t_data, xmax_lock_old_tuple);
3930 150781 : oldtup.t_data->t_infomask |= infomask_lock_old_tuple;
3931 150781 : oldtup.t_data->t_infomask2 |= infomask2_lock_old_tuple;
3932 150781 : HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
3933 :
3934 : /* temporarily make it look not-updated, but locked */
3935 150781 : oldtup.t_data->t_ctid = oldtup.t_self;
3936 :
3937 : /*
3938 : * Clear all-frozen bit on visibility map if needed. We could
3939 : * immediately reset ALL_VISIBLE, but given that the WAL logging
3940 : * overhead would be unchanged, that doesn't seem necessarily
3941 : * worthwhile.
3942 : */
3943 151760 : if (PageIsAllVisible(page) &&
3944 979 : visibilitymap_clear(relation, block, vmbuffer,
3945 : VISIBILITYMAP_ALL_FROZEN))
3946 815 : cleared_all_frozen = true;
3947 :
3948 150781 : MarkBufferDirty(buffer);
3949 :
3950 150781 : if (RelationNeedsWAL(relation))
3951 : {
3952 : xl_heap_lock xlrec;
3953 : XLogRecPtr recptr;
3954 :
3955 140652 : XLogBeginInsert();
3956 140652 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
3957 :
3958 140652 : xlrec.offnum = ItemPointerGetOffsetNumber(&oldtup.t_self);
3959 140652 : xlrec.xmax = xmax_lock_old_tuple;
3960 281304 : xlrec.infobits_set = compute_infobits(oldtup.t_data->t_infomask,
3961 140652 : oldtup.t_data->t_infomask2);
3962 140652 : xlrec.flags =
3963 140652 : cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
3964 140652 : XLogRegisterData(&xlrec, SizeOfHeapLock);
3965 140652 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
3966 140652 : PageSetLSN(page, recptr);
3967 : }
3968 :
3969 150781 : END_CRIT_SECTION();
3970 :
3971 150781 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3972 :
3973 : /*
3974 : * Let the toaster do its thing, if needed.
3975 : *
3976 : * Note: below this point, heaptup is the data we actually intend to
3977 : * store into the relation; newtup is the caller's original untoasted
3978 : * data.
3979 : */
3980 150781 : if (need_toast)
3981 : {
3982 : /* Note we always use WAL and FSM during updates */
3983 1710 : heaptup = heap_toast_insert_or_update(relation, newtup, &oldtup, 0);
3984 1710 : newtupsize = MAXALIGN(heaptup->t_len);
3985 : }
3986 : else
3987 149071 : heaptup = newtup;
3988 :
3989 : /*
3990 : * Now, do we need a new page for the tuple, or not? This is a bit
3991 : * tricky since someone else could have added tuples to the page while
3992 : * we weren't looking. We have to recheck the available space after
3993 : * reacquiring the buffer lock. But don't bother to do that if the
3994 : * former amount of free space is still not enough; it's unlikely
3995 : * there's more free now than before.
3996 : *
3997 : * What's more, if we need to get a new page, we will need to acquire
3998 : * buffer locks on both old and new pages. To avoid deadlock against
3999 : * some other backend trying to get the same two locks in the other
4000 : * order, we must be consistent about the order we get the locks in.
4001 : * We use the rule "lock the lower-numbered page of the relation
4002 : * first". To implement this, we must do RelationGetBufferForTuple
4003 : * while not holding the lock on the old page, and we must rely on it
4004 : * to get the locks on both pages in the correct order.
4005 : *
4006 : * Another consideration is that we need visibility map page pin(s) if
4007 : * we will have to clear the all-visible flag on either page. If we
4008 : * call RelationGetBufferForTuple, we rely on it to acquire any such
4009 : * pins; but if we don't, we have to handle that here. Hence we need
4010 : * a loop.
4011 : */
4012 : for (;;)
4013 : {
4014 150781 : if (newtupsize > pagefree)
4015 : {
4016 : /* It doesn't fit, must use RelationGetBufferForTuple. */
4017 150217 : newbuf = RelationGetBufferForTuple(relation, heaptup->t_len,
4018 : buffer, 0, NULL,
4019 : &vmbuffer_new, &vmbuffer,
4020 : 0);
4021 : /* We're all done. */
4022 150217 : break;
4023 : }
4024 : /* Acquire VM page pin if needed and we don't have it. */
4025 564 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
4026 0 : visibilitymap_pin(relation, block, &vmbuffer);
4027 : /* Re-acquire the lock on the old tuple's page. */
4028 564 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
4029 : /* Re-check using the up-to-date free space */
4030 564 : pagefree = PageGetHeapFreeSpace(page);
4031 564 : if (newtupsize > pagefree ||
4032 564 : (vmbuffer == InvalidBuffer && PageIsAllVisible(page)))
4033 : {
4034 : /*
4035 : * Rats, it doesn't fit anymore, or somebody just now set the
4036 : * all-visible flag. We must now unlock and loop to avoid
4037 : * deadlock. Fortunately, this path should seldom be taken.
4038 : */
4039 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
4040 : }
4041 : else
4042 : {
4043 : /* We're all done. */
4044 564 : newbuf = buffer;
4045 564 : break;
4046 : }
4047 : }
4048 : }
4049 : else
4050 : {
4051 : /* No TOAST work needed, and it'll fit on same page */
4052 159966 : newbuf = buffer;
4053 159966 : heaptup = newtup;
4054 : }
4055 :
4056 : /*
4057 : * We're about to do the actual update -- check for conflict first, to
4058 : * avoid possibly having to roll back work we've just done.
4059 : *
4060 : * This is safe without a recheck as long as there is no possibility of
4061 : * another process scanning the pages between this check and the update
4062 : * being visible to the scan (i.e., exclusive buffer content lock(s) are
4063 : * continuously held from this point until the tuple update is visible).
4064 : *
4065 : * For the new tuple the only check needed is at the relation level, but
4066 : * since both tuples are in the same relation and the check for oldtup
4067 : * will include checking the relation level, there is no benefit to a
4068 : * separate check for the new tuple.
4069 : */
4070 310747 : CheckForSerializableConflictIn(relation, &oldtup.t_self,
4071 : BufferGetBlockNumber(buffer));
4072 :
4073 : /*
4074 : * At this point newbuf and buffer are both pinned and locked, and newbuf
4075 : * has enough space for the new tuple. If they are the same buffer, only
4076 : * one pin is held.
4077 : */
4078 :
4079 310735 : if (newbuf == buffer)
4080 : {
4081 : /*
4082 : * Since the new tuple is going into the same page, we might be able
4083 : * to do a HOT update. Check if any of the index columns have been
4084 : * changed.
4085 : */
4086 160518 : if (!bms_overlap(modified_attrs, hot_attrs))
4087 : {
4088 148123 : use_hot_update = true;
4089 :
4090 : /*
4091 : * If none of the columns that are used in hot-blocking indexes
4092 : * were updated, we can apply HOT, but we do still need to check
4093 : * if we need to update the summarizing indexes, and update those
4094 : * indexes if the columns were updated, or we may fail to detect
4095 : * e.g. value bound changes in BRIN minmax indexes.
4096 : */
4097 148123 : if (bms_overlap(modified_attrs, sum_attrs))
4098 1641 : summarized_update = true;
4099 : }
4100 : }
4101 : else
4102 : {
4103 : /* Set a hint that the old page could use prune/defrag */
4104 150217 : PageSetFull(page);
4105 : }
4106 :
4107 : /*
4108 : * Compute replica identity tuple before entering the critical section so
4109 : * we don't PANIC upon a memory allocation failure.
4110 : * ExtractReplicaIdentity() will return NULL if nothing needs to be
4111 : * logged. Pass old key required as true only if the replica identity key
4112 : * columns are modified or it has external data.
4113 : */
4114 310735 : old_key_tuple = ExtractReplicaIdentity(relation, &oldtup,
4115 310735 : bms_overlap(modified_attrs, id_attrs) ||
4116 : id_has_external,
4117 310735 : &old_key_copied);
4118 :
4119 : /* NO EREPORT(ERROR) from here till changes are logged */
4120 310735 : START_CRIT_SECTION();
4121 :
4122 : /*
4123 : * If this transaction commits, the old tuple will become DEAD sooner or
4124 : * later. Set flag that this page is a candidate for pruning once our xid
4125 : * falls below the OldestXmin horizon. If the transaction finally aborts,
4126 : * the subsequent page pruning will be a no-op and the hint will be
4127 : * cleared.
4128 : *
4129 : * XXX Should we set hint on newbuf as well? If the transaction aborts,
4130 : * there would be a prunable tuple in the newbuf; but for now we choose
4131 : * not to optimize for aborts. Note that heap_xlog_update must be kept in
4132 : * sync if this decision changes.
4133 : */
4134 310735 : PageSetPrunable(page, xid);
4135 :
4136 310735 : if (use_hot_update)
4137 : {
4138 : /* Mark the old tuple as HOT-updated */
4139 148123 : HeapTupleSetHotUpdated(&oldtup);
4140 : /* And mark the new tuple as heap-only */
4141 148123 : HeapTupleSetHeapOnly(heaptup);
4142 : /* Mark the caller's copy too, in case different from heaptup */
4143 148123 : HeapTupleSetHeapOnly(newtup);
4144 : }
4145 : else
4146 : {
4147 : /* Make sure tuples are correctly marked as not-HOT */
4148 162612 : HeapTupleClearHotUpdated(&oldtup);
4149 162612 : HeapTupleClearHeapOnly(heaptup);
4150 162612 : HeapTupleClearHeapOnly(newtup);
4151 : }
4152 :
4153 310735 : RelationPutHeapTuple(relation, newbuf, heaptup, false); /* insert new tuple */
4154 :
4155 :
4156 : /* Clear obsolete visibility flags, possibly set by ourselves above... */
4157 310735 : oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
4158 310735 : oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
4159 : /* ... and store info about transaction updating this tuple */
4160 : Assert(TransactionIdIsValid(xmax_old_tuple));
4161 310735 : HeapTupleHeaderSetXmax(oldtup.t_data, xmax_old_tuple);
4162 310735 : oldtup.t_data->t_infomask |= infomask_old_tuple;
4163 310735 : oldtup.t_data->t_infomask2 |= infomask2_old_tuple;
4164 310735 : HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
4165 :
4166 : /* record address of new tuple in t_ctid of old one */
4167 310735 : oldtup.t_data->t_ctid = heaptup->t_self;
4168 :
4169 : /* clear PD_ALL_VISIBLE flags, reset all visibilitymap bits */
4170 310735 : if (PageIsAllVisible(BufferGetPage(buffer)))
4171 : {
4172 1603 : all_visible_cleared = true;
4173 1603 : PageClearAllVisible(BufferGetPage(buffer));
4174 1603 : visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
4175 : vmbuffer, VISIBILITYMAP_VALID_BITS);
4176 : }
4177 310735 : if (newbuf != buffer && PageIsAllVisible(BufferGetPage(newbuf)))
4178 : {
4179 926 : all_visible_cleared_new = true;
4180 926 : PageClearAllVisible(BufferGetPage(newbuf));
4181 926 : visibilitymap_clear(relation, BufferGetBlockNumber(newbuf),
4182 : vmbuffer_new, VISIBILITYMAP_VALID_BITS);
4183 : }
4184 :
4185 310735 : if (newbuf != buffer)
4186 150217 : MarkBufferDirty(newbuf);
4187 310735 : MarkBufferDirty(buffer);
4188 :
4189 : /* XLOG stuff */
4190 310735 : if (RelationNeedsWAL(relation))
4191 : {
4192 : XLogRecPtr recptr;
4193 :
4194 : /*
4195 : * For logical decoding we need combo CIDs to properly decode the
4196 : * catalog.
4197 : */
4198 299371 : if (RelationIsAccessibleInLogicalDecoding(relation))
4199 : {
4200 2568 : log_heap_new_cid(relation, &oldtup);
4201 2568 : log_heap_new_cid(relation, heaptup);
4202 : }
4203 :
4204 299371 : recptr = log_heap_update(relation, buffer,
4205 : newbuf, &oldtup, heaptup,
4206 : old_key_tuple,
4207 : all_visible_cleared,
4208 : all_visible_cleared_new);
4209 299371 : if (newbuf != buffer)
4210 : {
4211 140094 : PageSetLSN(BufferGetPage(newbuf), recptr);
4212 : }
4213 299371 : PageSetLSN(BufferGetPage(buffer), recptr);
4214 : }
4215 :
4216 310735 : END_CRIT_SECTION();
4217 :
4218 310735 : if (newbuf != buffer)
4219 150217 : LockBuffer(newbuf, BUFFER_LOCK_UNLOCK);
4220 310735 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
4221 :
4222 : /*
4223 : * Mark old tuple for invalidation from system caches at next command
4224 : * boundary, and mark the new tuple for invalidation in case we abort. We
4225 : * have to do this before releasing the buffer because oldtup is in the
4226 : * buffer. (heaptup is all in local memory, but it's necessary to process
4227 : * both tuple versions in one call to inval.c so we can avoid redundant
4228 : * sinval messages.)
4229 : */
4230 310735 : CacheInvalidateHeapTuple(relation, &oldtup, heaptup);
4231 :
4232 : /* Now we can release the buffer(s) */
4233 310735 : if (newbuf != buffer)
4234 150217 : ReleaseBuffer(newbuf);
4235 310735 : ReleaseBuffer(buffer);
4236 310735 : if (BufferIsValid(vmbuffer_new))
4237 926 : ReleaseBuffer(vmbuffer_new);
4238 310735 : if (BufferIsValid(vmbuffer))
4239 1603 : ReleaseBuffer(vmbuffer);
4240 :
4241 : /*
4242 : * Release the lmgr tuple lock, if we had it.
4243 : */
4244 310735 : if (have_tuple_lock)
4245 22 : UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
4246 :
4247 310735 : pgstat_count_heap_update(relation, use_hot_update, newbuf != buffer);
4248 :
4249 : /*
4250 : * If heaptup is a private copy, release it. Don't forget to copy t_self
4251 : * back to the caller's image, too.
4252 : */
4253 310735 : if (heaptup != newtup)
4254 : {
4255 1662 : newtup->t_self = heaptup->t_self;
4256 1662 : heap_freetuple(heaptup);
4257 : }
4258 :
4259 : /*
4260 : * If it is a HOT update, the update may still need to update summarized
4261 : * indexes, lest we fail to update those summaries and get incorrect
4262 : * results (for example, minmax bounds of the block may change with this
4263 : * update).
4264 : */
4265 310735 : if (use_hot_update)
4266 : {
4267 148123 : if (summarized_update)
4268 1641 : *update_indexes = TU_Summarizing;
4269 : else
4270 146482 : *update_indexes = TU_None;
4271 : }
4272 : else
4273 162612 : *update_indexes = TU_All;
4274 :
4275 310735 : if (old_key_tuple != NULL && old_key_copied)
4276 84 : heap_freetuple(old_key_tuple);
4277 :
4278 310735 : bms_free(hot_attrs);
4279 310735 : bms_free(sum_attrs);
4280 310735 : bms_free(key_attrs);
4281 310735 : bms_free(id_attrs);
4282 310735 : bms_free(modified_attrs);
4283 310735 : bms_free(interesting_attrs);
4284 :
4285 310735 : return TM_Ok;
4286 : }
4287 :
4288 : #ifdef USE_ASSERT_CHECKING
4289 : /*
4290 : * Confirm adequate lock held during heap_update(), per rules from
4291 : * README.tuplock section "Locking to write inplace-updated tables".
4292 : */
4293 : static void
4294 : check_lock_if_inplace_updateable_rel(Relation relation,
4295 : const ItemPointerData *otid,
4296 : HeapTuple newtup)
4297 : {
4298 : /* LOCKTAG_TUPLE acceptable for any catalog */
4299 : switch (RelationGetRelid(relation))
4300 : {
4301 : case RelationRelationId:
4302 : case DatabaseRelationId:
4303 : {
4304 : LOCKTAG tuptag;
4305 :
4306 : SET_LOCKTAG_TUPLE(tuptag,
4307 : relation->rd_lockInfo.lockRelId.dbId,
4308 : relation->rd_lockInfo.lockRelId.relId,
4309 : ItemPointerGetBlockNumber(otid),
4310 : ItemPointerGetOffsetNumber(otid));
4311 : if (LockHeldByMe(&tuptag, InplaceUpdateTupleLock, false))
4312 : return;
4313 : }
4314 : break;
4315 : default:
4316 : Assert(!IsInplaceUpdateRelation(relation));
4317 : return;
4318 : }
4319 :
4320 : switch (RelationGetRelid(relation))
4321 : {
4322 : case RelationRelationId:
4323 : {
4324 : /* LOCKTAG_TUPLE or LOCKTAG_RELATION ok */
4325 : Form_pg_class classForm = (Form_pg_class) GETSTRUCT(newtup);
4326 : Oid relid = classForm->oid;
4327 : Oid dbid;
4328 : LOCKTAG tag;
4329 :
4330 : if (IsSharedRelation(relid))
4331 : dbid = InvalidOid;
4332 : else
4333 : dbid = MyDatabaseId;
4334 :
4335 : if (classForm->relkind == RELKIND_INDEX)
4336 : {
4337 : Relation irel = index_open(relid, AccessShareLock);
4338 :
4339 : SET_LOCKTAG_RELATION(tag, dbid, irel->rd_index->indrelid);
4340 : index_close(irel, AccessShareLock);
4341 : }
4342 : else
4343 : SET_LOCKTAG_RELATION(tag, dbid, relid);
4344 :
4345 : if (!LockHeldByMe(&tag, ShareUpdateExclusiveLock, false) &&
4346 : !LockHeldByMe(&tag, ShareRowExclusiveLock, true))
4347 : elog(WARNING,
4348 : "missing lock for relation \"%s\" (OID %u, relkind %c) @ TID (%u,%u)",
4349 : NameStr(classForm->relname),
4350 : relid,
4351 : classForm->relkind,
4352 : ItemPointerGetBlockNumber(otid),
4353 : ItemPointerGetOffsetNumber(otid));
4354 : }
4355 : break;
4356 : case DatabaseRelationId:
4357 : {
4358 : /* LOCKTAG_TUPLE required */
4359 : Form_pg_database dbForm = (Form_pg_database) GETSTRUCT(newtup);
4360 :
4361 : elog(WARNING,
4362 : "missing lock on database \"%s\" (OID %u) @ TID (%u,%u)",
4363 : NameStr(dbForm->datname),
4364 : dbForm->oid,
4365 : ItemPointerGetBlockNumber(otid),
4366 : ItemPointerGetOffsetNumber(otid));
4367 : }
4368 : break;
4369 : }
4370 : }
4371 :
4372 : /*
4373 : * Confirm adequate relation lock held, per rules from README.tuplock section
4374 : * "Locking to write inplace-updated tables".
4375 : */
4376 : static void
4377 : check_inplace_rel_lock(HeapTuple oldtup)
4378 : {
4379 : Form_pg_class classForm = (Form_pg_class) GETSTRUCT(oldtup);
4380 : Oid relid = classForm->oid;
4381 : Oid dbid;
4382 : LOCKTAG tag;
4383 :
4384 : if (IsSharedRelation(relid))
4385 : dbid = InvalidOid;
4386 : else
4387 : dbid = MyDatabaseId;
4388 :
4389 : if (classForm->relkind == RELKIND_INDEX)
4390 : {
4391 : Relation irel = index_open(relid, AccessShareLock);
4392 :
4393 : SET_LOCKTAG_RELATION(tag, dbid, irel->rd_index->indrelid);
4394 : index_close(irel, AccessShareLock);
4395 : }
4396 : else
4397 : SET_LOCKTAG_RELATION(tag, dbid, relid);
4398 :
4399 : if (!LockHeldByMe(&tag, ShareUpdateExclusiveLock, true))
4400 : elog(WARNING,
4401 : "missing lock for relation \"%s\" (OID %u, relkind %c) @ TID (%u,%u)",
4402 : NameStr(classForm->relname),
4403 : relid,
4404 : classForm->relkind,
4405 : ItemPointerGetBlockNumber(&oldtup->t_self),
4406 : ItemPointerGetOffsetNumber(&oldtup->t_self));
4407 : }
4408 : #endif
4409 :
4410 : /*
4411 : * Check if the specified attribute's values are the same. Subroutine for
4412 : * HeapDetermineColumnsInfo.
4413 : */
4414 : static bool
4415 760531 : heap_attr_equals(TupleDesc tupdesc, int attrnum, Datum value1, Datum value2,
4416 : bool isnull1, bool isnull2)
4417 : {
4418 : /*
4419 : * If one value is NULL and other is not, then they are certainly not
4420 : * equal
4421 : */
4422 760531 : if (isnull1 != isnull2)
4423 45 : return false;
4424 :
4425 : /*
4426 : * If both are NULL, they can be considered equal.
4427 : */
4428 760486 : if (isnull1)
4429 4991 : return true;
4430 :
4431 : /*
4432 : * We do simple binary comparison of the two datums. This may be overly
4433 : * strict because there can be multiple binary representations for the
4434 : * same logical value. But we should be OK as long as there are no false
4435 : * positives. Using a type-specific equality operator is messy because
4436 : * there could be multiple notions of equality in different operator
4437 : * classes; furthermore, we cannot safely invoke user-defined functions
4438 : * while holding exclusive buffer lock.
4439 : */
4440 755495 : if (attrnum <= 0)
4441 : {
4442 : /* The only allowed system columns are OIDs, so do this */
4443 0 : return (DatumGetObjectId(value1) == DatumGetObjectId(value2));
4444 : }
4445 : else
4446 : {
4447 : CompactAttribute *att;
4448 :
4449 : Assert(attrnum <= tupdesc->natts);
4450 755495 : att = TupleDescCompactAttr(tupdesc, attrnum - 1);
4451 755495 : return datumIsEqual(value1, value2, att->attbyval, att->attlen);
4452 : }
4453 : }
4454 :
4455 : /*
4456 : * Check which columns are being updated.
4457 : *
4458 : * Given an updated tuple, determine (and return into the output bitmapset),
4459 : * from those listed as interesting, the set of columns that changed.
4460 : *
4461 : * has_external indicates if any of the unmodified attributes (from those
4462 : * listed as interesting) of the old tuple is a member of external_cols and is
4463 : * stored externally.
4464 : */
4465 : static Bitmapset *
4466 310909 : HeapDetermineColumnsInfo(Relation relation,
4467 : Bitmapset *interesting_cols,
4468 : Bitmapset *external_cols,
4469 : HeapTuple oldtup, HeapTuple newtup,
4470 : bool *has_external)
4471 : {
4472 : int attidx;
4473 310909 : Bitmapset *modified = NULL;
4474 310909 : TupleDesc tupdesc = RelationGetDescr(relation);
4475 :
4476 310909 : attidx = -1;
4477 1071440 : while ((attidx = bms_next_member(interesting_cols, attidx)) >= 0)
4478 : {
4479 : /* attidx is zero-based, attrnum is the normal attribute number */
4480 760531 : AttrNumber attrnum = attidx + FirstLowInvalidHeapAttributeNumber;
4481 : Datum value1,
4482 : value2;
4483 : bool isnull1,
4484 : isnull2;
4485 :
4486 : /*
4487 : * If it's a whole-tuple reference, say "not equal". It's not really
4488 : * worth supporting this case, since it could only succeed after a
4489 : * no-op update, which is hardly a case worth optimizing for.
4490 : */
4491 760531 : if (attrnum == 0)
4492 : {
4493 0 : modified = bms_add_member(modified, attidx);
4494 730521 : continue;
4495 : }
4496 :
4497 : /*
4498 : * Likewise, automatically say "not equal" for any system attribute
4499 : * other than tableOID; we cannot expect these to be consistent in a
4500 : * HOT chain, or even to be set correctly yet in the new tuple.
4501 : */
4502 760531 : if (attrnum < 0)
4503 : {
4504 0 : if (attrnum != TableOidAttributeNumber)
4505 : {
4506 0 : modified = bms_add_member(modified, attidx);
4507 0 : continue;
4508 : }
4509 : }
4510 :
4511 : /*
4512 : * Extract the corresponding values. XXX this is pretty inefficient
4513 : * if there are many indexed columns. Should we do a single
4514 : * heap_deform_tuple call on each tuple, instead? But that doesn't
4515 : * work for system columns ...
4516 : */
4517 760531 : value1 = heap_getattr(oldtup, attrnum, tupdesc, &isnull1);
4518 760531 : value2 = heap_getattr(newtup, attrnum, tupdesc, &isnull2);
4519 :
4520 760531 : if (!heap_attr_equals(tupdesc, attrnum, value1,
4521 : value2, isnull1, isnull2))
4522 : {
4523 27278 : modified = bms_add_member(modified, attidx);
4524 27278 : continue;
4525 : }
4526 :
4527 : /*
4528 : * No need to check attributes that can't be stored externally. Note
4529 : * that system attributes can't be stored externally.
4530 : */
4531 733253 : if (attrnum < 0 || isnull1 ||
4532 728262 : TupleDescCompactAttr(tupdesc, attrnum - 1)->attlen != -1)
4533 703243 : continue;
4534 :
4535 : /*
4536 : * Check if the old tuple's attribute is stored externally and is a
4537 : * member of external_cols.
4538 : */
4539 30015 : if (VARATT_IS_EXTERNAL((varlena *) DatumGetPointer(value1)) &&
4540 5 : bms_is_member(attidx, external_cols))
4541 2 : *has_external = true;
4542 : }
4543 :
4544 310909 : return modified;
4545 : }
4546 :
4547 : /*
4548 : * simple_heap_update - replace a tuple
4549 : *
4550 : * This routine may be used to update a tuple when concurrent updates of
4551 : * the target tuple are not expected (for example, because we have a lock
4552 : * on the relation associated with the tuple). Any failure is reported
4553 : * via ereport().
4554 : */
4555 : void
4556 116165 : simple_heap_update(Relation relation, const ItemPointerData *otid, HeapTuple tup,
4557 : TU_UpdateIndexes *update_indexes)
4558 : {
4559 : TM_Result result;
4560 : TM_FailureData tmfd;
4561 : LockTupleMode lockmode;
4562 :
4563 116165 : result = heap_update(relation, otid, tup,
4564 : GetCurrentCommandId(true), InvalidSnapshot,
4565 : true /* wait for commit */ ,
4566 : &tmfd, &lockmode, update_indexes);
4567 116165 : switch (result)
4568 : {
4569 0 : case TM_SelfModified:
4570 : /* Tuple was already updated in current command? */
4571 0 : elog(ERROR, "tuple already updated by self");
4572 : break;
4573 :
4574 116164 : case TM_Ok:
4575 : /* done successfully */
4576 116164 : break;
4577 :
4578 0 : case TM_Updated:
4579 0 : elog(ERROR, "tuple concurrently updated");
4580 : break;
4581 :
4582 1 : case TM_Deleted:
4583 1 : elog(ERROR, "tuple concurrently deleted");
4584 : break;
4585 :
4586 0 : default:
4587 0 : elog(ERROR, "unrecognized heap_update status: %u", result);
4588 : break;
4589 : }
4590 116164 : }
4591 :
4592 :
4593 : /*
4594 : * Return the MultiXactStatus corresponding to the given tuple lock mode.
4595 : */
4596 : static MultiXactStatus
4597 115442 : get_mxact_status_for_lock(LockTupleMode mode, bool is_update)
4598 : {
4599 : int retval;
4600 :
4601 115442 : if (is_update)
4602 215 : retval = tupleLockExtraInfo[mode].updstatus;
4603 : else
4604 115227 : retval = tupleLockExtraInfo[mode].lockstatus;
4605 :
4606 115442 : if (retval == -1)
4607 0 : elog(ERROR, "invalid lock tuple mode %d/%s", mode,
4608 : is_update ? "true" : "false");
4609 :
4610 115442 : return (MultiXactStatus) retval;
4611 : }
4612 :
4613 : /*
4614 : * heap_lock_tuple - lock a tuple in shared or exclusive mode
4615 : *
4616 : * Note that this acquires a buffer pin, which the caller must release.
4617 : *
4618 : * Input parameters:
4619 : * relation: relation containing tuple (caller must hold suitable lock)
4620 : * cid: current command ID (used for visibility test, and stored into
4621 : * tuple's cmax if lock is successful)
4622 : * mode: indicates if shared or exclusive tuple lock is desired
4623 : * wait_policy: what to do if tuple lock is not available
4624 : * follow_updates: if true, follow the update chain to also lock descendant
4625 : * tuples.
4626 : *
4627 : * Output parameters:
4628 : * *tuple: all fields filled in
4629 : * *buffer: set to buffer holding tuple (pinned but not locked at exit)
4630 : * *tmfd: filled in failure cases (see below)
4631 : *
4632 : * Function results are the same as the ones for table_tuple_lock().
4633 : *
4634 : * In the failure cases other than TM_Invisible, the routine fills
4635 : * *tmfd with the tuple's t_ctid, t_xmax (resolving a possible MultiXact,
4636 : * if necessary), and t_cmax (the last only for TM_SelfModified,
4637 : * since we cannot obtain cmax from a combo CID generated by another
4638 : * transaction).
4639 : * See comments for struct TM_FailureData for additional info.
4640 : *
4641 : * See README.tuplock for a thorough explanation of this mechanism.
4642 : */
4643 : TM_Result
4644 158474 : heap_lock_tuple(Relation relation, HeapTuple tuple,
4645 : CommandId cid, LockTupleMode mode, LockWaitPolicy wait_policy,
4646 : bool follow_updates,
4647 : Buffer *buffer, TM_FailureData *tmfd)
4648 : {
4649 : TM_Result result;
4650 158474 : ItemPointer tid = &(tuple->t_self);
4651 : ItemId lp;
4652 : Page page;
4653 158474 : Buffer vmbuffer = InvalidBuffer;
4654 : BlockNumber block;
4655 : TransactionId xid,
4656 : xmax;
4657 : uint16 old_infomask,
4658 : new_infomask,
4659 : new_infomask2;
4660 158474 : bool first_time = true;
4661 158474 : bool skip_tuple_lock = false;
4662 158474 : bool have_tuple_lock = false;
4663 158474 : bool cleared_all_frozen = false;
4664 :
4665 158474 : *buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
4666 158474 : block = ItemPointerGetBlockNumber(tid);
4667 :
4668 : /*
4669 : * Before locking the buffer, pin the visibility map page if it appears to
4670 : * be necessary. Since we haven't got the lock yet, someone else might be
4671 : * in the middle of changing this, so we'll need to recheck after we have
4672 : * the lock.
4673 : */
4674 158474 : if (PageIsAllVisible(BufferGetPage(*buffer)))
4675 1666 : visibilitymap_pin(relation, block, &vmbuffer);
4676 :
4677 158474 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4678 :
4679 158474 : page = BufferGetPage(*buffer);
4680 158474 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
4681 : Assert(ItemIdIsNormal(lp));
4682 :
4683 158474 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
4684 158474 : tuple->t_len = ItemIdGetLength(lp);
4685 158474 : tuple->t_tableOid = RelationGetRelid(relation);
4686 :
4687 16 : l3:
4688 158490 : result = HeapTupleSatisfiesUpdate(tuple, cid, *buffer);
4689 :
4690 158490 : if (result == TM_Invisible)
4691 : {
4692 : /*
4693 : * This is possible, but only when locking a tuple for ON CONFLICT DO
4694 : * SELECT/UPDATE. We return this value here rather than throwing an
4695 : * error in order to give that case the opportunity to throw a more
4696 : * specific error.
4697 : */
4698 21 : result = TM_Invisible;
4699 21 : goto out_locked;
4700 : }
4701 158469 : else if (result == TM_BeingModified ||
4702 77258 : result == TM_Updated ||
4703 : result == TM_Deleted)
4704 : {
4705 : TransactionId xwait;
4706 : uint16 infomask;
4707 : uint16 infomask2;
4708 : bool require_sleep;
4709 : ItemPointerData t_ctid;
4710 :
4711 : /* must copy state data before unlocking buffer */
4712 81212 : xwait = HeapTupleHeaderGetRawXmax(tuple->t_data);
4713 81212 : infomask = tuple->t_data->t_infomask;
4714 81212 : infomask2 = tuple->t_data->t_infomask2;
4715 81212 : ItemPointerCopy(&tuple->t_data->t_ctid, &t_ctid);
4716 :
4717 81212 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
4718 :
4719 : /*
4720 : * If any subtransaction of the current top transaction already holds
4721 : * a lock as strong as or stronger than what we're requesting, we
4722 : * effectively hold the desired lock already. We *must* succeed
4723 : * without trying to take the tuple lock, else we will deadlock
4724 : * against anyone wanting to acquire a stronger lock.
4725 : *
4726 : * Note we only do this the first time we loop on the HTSU result;
4727 : * there is no point in testing in subsequent passes, because
4728 : * evidently our own transaction cannot have acquired a new lock after
4729 : * the first time we checked.
4730 : */
4731 81212 : if (first_time)
4732 : {
4733 81201 : first_time = false;
4734 :
4735 81201 : if (infomask & HEAP_XMAX_IS_MULTI)
4736 : {
4737 : int i;
4738 : int nmembers;
4739 : MultiXactMember *members;
4740 :
4741 : /*
4742 : * We don't need to allow old multixacts here; if that had
4743 : * been the case, HeapTupleSatisfiesUpdate would have returned
4744 : * MayBeUpdated and we wouldn't be here.
4745 : */
4746 : nmembers =
4747 73302 : GetMultiXactIdMembers(xwait, &members, false,
4748 73302 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
4749 :
4750 1422680 : for (i = 0; i < nmembers; i++)
4751 : {
4752 : /* only consider members of our own transaction */
4753 1349392 : if (!TransactionIdIsCurrentTransactionId(members[i].xid))
4754 1349343 : continue;
4755 :
4756 49 : if (TUPLOCK_from_mxstatus(members[i].status) >= mode)
4757 : {
4758 14 : pfree(members);
4759 14 : result = TM_Ok;
4760 14 : goto out_unlocked;
4761 : }
4762 : else
4763 : {
4764 : /*
4765 : * Disable acquisition of the heavyweight tuple lock.
4766 : * Otherwise, when promoting a weaker lock, we might
4767 : * deadlock with another locker that has acquired the
4768 : * heavyweight tuple lock and is waiting for our
4769 : * transaction to finish.
4770 : *
4771 : * Note that in this case we still need to wait for
4772 : * the multixact if required, to avoid acquiring
4773 : * conflicting locks.
4774 : */
4775 35 : skip_tuple_lock = true;
4776 : }
4777 : }
4778 :
4779 73288 : if (members)
4780 73288 : pfree(members);
4781 : }
4782 7899 : else if (TransactionIdIsCurrentTransactionId(xwait))
4783 : {
4784 6577 : switch (mode)
4785 : {
4786 176 : case LockTupleKeyShare:
4787 : Assert(HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) ||
4788 : HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4789 : HEAP_XMAX_IS_EXCL_LOCKED(infomask));
4790 176 : result = TM_Ok;
4791 176 : goto out_unlocked;
4792 24 : case LockTupleShare:
4793 30 : if (HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4794 6 : HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4795 : {
4796 18 : result = TM_Ok;
4797 18 : goto out_unlocked;
4798 : }
4799 6 : break;
4800 72 : case LockTupleNoKeyExclusive:
4801 72 : if (HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4802 : {
4803 60 : result = TM_Ok;
4804 60 : goto out_unlocked;
4805 : }
4806 12 : break;
4807 6305 : case LockTupleExclusive:
4808 6305 : if (HEAP_XMAX_IS_EXCL_LOCKED(infomask) &&
4809 1265 : infomask2 & HEAP_KEYS_UPDATED)
4810 : {
4811 1244 : result = TM_Ok;
4812 1244 : goto out_unlocked;
4813 : }
4814 5061 : break;
4815 : }
4816 : }
4817 : }
4818 :
4819 : /*
4820 : * Initially assume that we will have to wait for the locking
4821 : * transaction(s) to finish. We check various cases below in which
4822 : * this can be turned off.
4823 : */
4824 79700 : require_sleep = true;
4825 79700 : if (mode == LockTupleKeyShare)
4826 : {
4827 : /*
4828 : * If we're requesting KeyShare, and there's no update present, we
4829 : * don't need to wait. Even if there is an update, we can still
4830 : * continue if the key hasn't been modified.
4831 : *
4832 : * However, if there are updates, we need to walk the update chain
4833 : * to mark future versions of the row as locked, too. That way,
4834 : * if somebody deletes that future version, we're protected
4835 : * against the key going away. This locking of future versions
4836 : * could block momentarily, if a concurrent transaction is
4837 : * deleting a key; or it could return a value to the effect that
4838 : * the transaction deleting the key has already committed. So we
4839 : * do this before re-locking the buffer; otherwise this would be
4840 : * prone to deadlocks.
4841 : *
4842 : * Note that the TID we're locking was grabbed before we unlocked
4843 : * the buffer. For it to change while we're not looking, the
4844 : * other properties we're testing for below after re-locking the
4845 : * buffer would also change, in which case we would restart this
4846 : * loop above.
4847 : */
4848 73863 : if (!(infomask2 & HEAP_KEYS_UPDATED))
4849 : {
4850 : bool updated;
4851 :
4852 73820 : updated = !HEAP_XMAX_IS_LOCKED_ONLY(infomask);
4853 :
4854 : /*
4855 : * If there are updates, follow the update chain; bail out if
4856 : * that cannot be done.
4857 : */
4858 73820 : if (follow_updates && updated &&
4859 2169 : !ItemPointerEquals(&tuple->t_self, &t_ctid))
4860 : {
4861 : TM_Result res;
4862 :
4863 2169 : res = heap_lock_updated_tuple(relation,
4864 : infomask, xwait, &t_ctid,
4865 : GetCurrentTransactionId(),
4866 : mode);
4867 2169 : if (res != TM_Ok)
4868 : {
4869 6 : result = res;
4870 : /* recovery code expects to have buffer lock held */
4871 6 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4872 198 : goto failed;
4873 : }
4874 : }
4875 :
4876 73814 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4877 :
4878 : /*
4879 : * Make sure it's still an appropriate lock, else start over.
4880 : * Also, if it wasn't updated before we released the lock, but
4881 : * is updated now, we start over too; the reason is that we
4882 : * now need to follow the update chain to lock the new
4883 : * versions.
4884 : */
4885 73814 : if (!HeapTupleHeaderIsOnlyLocked(tuple->t_data) &&
4886 2151 : ((tuple->t_data->t_infomask2 & HEAP_KEYS_UPDATED) ||
4887 2151 : !updated))
4888 16 : goto l3;
4889 :
4890 : /* Things look okay, so we can skip sleeping */
4891 73814 : require_sleep = false;
4892 :
4893 : /*
4894 : * Note we allow Xmax to change here; other updaters/lockers
4895 : * could have modified it before we grabbed the buffer lock.
4896 : * However, this is not a problem, because with the recheck we
4897 : * just did we ensure that they still don't conflict with the
4898 : * lock we want.
4899 : */
4900 : }
4901 : }
4902 5837 : else if (mode == LockTupleShare)
4903 : {
4904 : /*
4905 : * If we're requesting Share, we can similarly avoid sleeping if
4906 : * there's no update and no exclusive lock present.
4907 : */
4908 443 : if (HEAP_XMAX_IS_LOCKED_ONLY(infomask) &&
4909 443 : !HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4910 : {
4911 437 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4912 :
4913 : /*
4914 : * Make sure it's still an appropriate lock, else start over.
4915 : * See above about allowing xmax to change.
4916 : */
4917 874 : if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
4918 437 : HEAP_XMAX_IS_EXCL_LOCKED(tuple->t_data->t_infomask))
4919 0 : goto l3;
4920 437 : require_sleep = false;
4921 : }
4922 : }
4923 5394 : else if (mode == LockTupleNoKeyExclusive)
4924 : {
4925 : /*
4926 : * If we're requesting NoKeyExclusive, we might also be able to
4927 : * avoid sleeping; just ensure that there no conflicting lock
4928 : * already acquired.
4929 : */
4930 172 : if (infomask & HEAP_XMAX_IS_MULTI)
4931 : {
4932 26 : if (!DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
4933 : mode, NULL))
4934 : {
4935 : /*
4936 : * No conflict, but if the xmax changed under us in the
4937 : * meantime, start over.
4938 : */
4939 13 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4940 26 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4941 13 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4942 : xwait))
4943 0 : goto l3;
4944 :
4945 : /* otherwise, we're good */
4946 13 : require_sleep = false;
4947 : }
4948 : }
4949 146 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask))
4950 : {
4951 18 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4952 :
4953 : /* if the xmax changed in the meantime, start over */
4954 36 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4955 18 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4956 : xwait))
4957 0 : goto l3;
4958 : /* otherwise, we're good */
4959 18 : require_sleep = false;
4960 : }
4961 : }
4962 :
4963 : /*
4964 : * As a check independent from those above, we can also avoid sleeping
4965 : * if the current transaction is the sole locker of the tuple. Note
4966 : * that the strength of the lock already held is irrelevant; this is
4967 : * not about recording the lock in Xmax (which will be done regardless
4968 : * of this optimization, below). Also, note that the cases where we
4969 : * hold a lock stronger than we are requesting are already handled
4970 : * above by not doing anything.
4971 : *
4972 : * Note we only deal with the non-multixact case here; MultiXactIdWait
4973 : * is well equipped to deal with this situation on its own.
4974 : */
4975 85066 : if (require_sleep && !(infomask & HEAP_XMAX_IS_MULTI) &&
4976 5372 : TransactionIdIsCurrentTransactionId(xwait))
4977 : {
4978 : /* ... but if the xmax changed in the meantime, start over */
4979 5061 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4980 10122 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4981 5061 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4982 : xwait))
4983 0 : goto l3;
4984 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask));
4985 5061 : require_sleep = false;
4986 : }
4987 :
4988 : /*
4989 : * Time to sleep on the other transaction/multixact, if necessary.
4990 : *
4991 : * If the other transaction is an update/delete that's already
4992 : * committed, then sleeping cannot possibly do any good: if we're
4993 : * required to sleep, get out to raise an error instead.
4994 : *
4995 : * By here, we either have already acquired the buffer exclusive lock,
4996 : * or we must wait for the locking transaction or multixact; so below
4997 : * we ensure that we grab buffer lock after the sleep.
4998 : */
4999 79694 : if (require_sleep && (result == TM_Updated || result == TM_Deleted))
5000 : {
5001 154 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5002 154 : goto failed;
5003 : }
5004 79540 : else if (require_sleep)
5005 : {
5006 : /*
5007 : * Acquire tuple lock to establish our priority for the tuple, or
5008 : * die trying. LockTuple will release us when we are next-in-line
5009 : * for the tuple. We must do this even if we are share-locking,
5010 : * but not if we already have a weaker lock on the tuple.
5011 : *
5012 : * If we are forced to "start over" below, we keep the tuple lock;
5013 : * this arranges that we stay at the head of the line while
5014 : * rechecking tuple state.
5015 : */
5016 197 : if (!skip_tuple_lock &&
5017 181 : !heap_acquire_tuplock(relation, tid, mode, wait_policy,
5018 : &have_tuple_lock))
5019 : {
5020 : /*
5021 : * This can only happen if wait_policy is Skip and the lock
5022 : * couldn't be obtained.
5023 : */
5024 1 : result = TM_WouldBlock;
5025 : /* recovery code expects to have buffer lock held */
5026 1 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5027 1 : goto failed;
5028 : }
5029 :
5030 195 : if (infomask & HEAP_XMAX_IS_MULTI)
5031 : {
5032 40 : MultiXactStatus status = get_mxact_status_for_lock(mode, false);
5033 :
5034 : /* We only ever lock tuples, never update them */
5035 40 : if (status >= MultiXactStatusNoKeyUpdate)
5036 0 : elog(ERROR, "invalid lock mode in heap_lock_tuple");
5037 :
5038 : /* wait for multixact to end, or die trying */
5039 40 : switch (wait_policy)
5040 : {
5041 36 : case LockWaitBlock:
5042 36 : MultiXactIdWait((MultiXactId) xwait, status, infomask,
5043 36 : relation, &tuple->t_self, XLTW_Lock, NULL);
5044 36 : break;
5045 2 : case LockWaitSkip:
5046 2 : if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
5047 : status, infomask, relation,
5048 : NULL, false))
5049 : {
5050 2 : result = TM_WouldBlock;
5051 : /* recovery code expects to have buffer lock held */
5052 2 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5053 2 : goto failed;
5054 : }
5055 0 : break;
5056 2 : case LockWaitError:
5057 2 : if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
5058 : status, infomask, relation,
5059 : NULL, log_lock_failures))
5060 2 : ereport(ERROR,
5061 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5062 : errmsg("could not obtain lock on row in relation \"%s\"",
5063 : RelationGetRelationName(relation))));
5064 :
5065 0 : break;
5066 : }
5067 :
5068 : /*
5069 : * Of course, the multixact might not be done here: if we're
5070 : * requesting a light lock mode, other transactions with light
5071 : * locks could still be alive, as well as locks owned by our
5072 : * own xact or other subxacts of this backend. We need to
5073 : * preserve the surviving MultiXact members. Note that it
5074 : * isn't absolutely necessary in the latter case, but doing so
5075 : * is simpler.
5076 : */
5077 : }
5078 : else
5079 : {
5080 : /* wait for regular transaction to end, or die trying */
5081 155 : switch (wait_policy)
5082 : {
5083 116 : case LockWaitBlock:
5084 116 : XactLockTableWait(xwait, relation, &tuple->t_self,
5085 : XLTW_Lock);
5086 116 : break;
5087 33 : case LockWaitSkip:
5088 33 : if (!ConditionalXactLockTableWait(xwait, false))
5089 : {
5090 33 : result = TM_WouldBlock;
5091 : /* recovery code expects to have buffer lock held */
5092 33 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5093 33 : goto failed;
5094 : }
5095 0 : break;
5096 6 : case LockWaitError:
5097 6 : if (!ConditionalXactLockTableWait(xwait, log_lock_failures))
5098 6 : ereport(ERROR,
5099 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5100 : errmsg("could not obtain lock on row in relation \"%s\"",
5101 : RelationGetRelationName(relation))));
5102 0 : break;
5103 : }
5104 : }
5105 :
5106 : /* if there are updates, follow the update chain */
5107 152 : if (follow_updates && !HEAP_XMAX_IS_LOCKED_ONLY(infomask) &&
5108 59 : !ItemPointerEquals(&tuple->t_self, &t_ctid))
5109 : {
5110 : TM_Result res;
5111 :
5112 45 : res = heap_lock_updated_tuple(relation,
5113 : infomask, xwait, &t_ctid,
5114 : GetCurrentTransactionId(),
5115 : mode);
5116 45 : if (res != TM_Ok)
5117 : {
5118 2 : result = res;
5119 : /* recovery code expects to have buffer lock held */
5120 2 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5121 2 : goto failed;
5122 : }
5123 : }
5124 :
5125 150 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5126 :
5127 : /*
5128 : * xwait is done, but if xwait had just locked the tuple then some
5129 : * other xact could update this tuple before we get to this point.
5130 : * Check for xmax change, and start over if so.
5131 : */
5132 286 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
5133 136 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
5134 : xwait))
5135 16 : goto l3;
5136 :
5137 134 : if (!(infomask & HEAP_XMAX_IS_MULTI))
5138 : {
5139 : /*
5140 : * Otherwise check if it committed or aborted. Note we cannot
5141 : * be here if the tuple was only locked by somebody who didn't
5142 : * conflict with us; that would have been handled above. So
5143 : * that transaction must necessarily be gone by now. But
5144 : * don't check for this in the multixact case, because some
5145 : * locker transactions might still be running.
5146 : */
5147 101 : UpdateXmaxHintBits(tuple->t_data, *buffer, xwait);
5148 : }
5149 : }
5150 :
5151 : /* By here, we're certain that we hold buffer exclusive lock again */
5152 :
5153 : /*
5154 : * We may lock if previous xmax aborted, or if it committed but only
5155 : * locked the tuple without updating it; or if we didn't have to wait
5156 : * at all for whatever reason.
5157 : */
5158 79477 : if (!require_sleep ||
5159 232 : (tuple->t_data->t_infomask & HEAP_XMAX_INVALID) ||
5160 180 : HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
5161 82 : HeapTupleHeaderIsOnlyLocked(tuple->t_data))
5162 79401 : result = TM_Ok;
5163 76 : else if (!ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid))
5164 57 : result = TM_Updated;
5165 : else
5166 19 : result = TM_Deleted;
5167 : }
5168 :
5169 77257 : failed:
5170 156932 : if (result != TM_Ok)
5171 : {
5172 : Assert(result == TM_SelfModified || result == TM_Updated ||
5173 : result == TM_Deleted || result == TM_WouldBlock);
5174 :
5175 : /*
5176 : * When locking a tuple under LockWaitSkip semantics and we fail with
5177 : * TM_WouldBlock above, it's possible for concurrent transactions to
5178 : * release the lock and set HEAP_XMAX_INVALID in the meantime. So
5179 : * this assert is slightly different from the equivalent one in
5180 : * heap_delete and heap_update.
5181 : */
5182 : Assert((result == TM_WouldBlock) ||
5183 : !(tuple->t_data->t_infomask & HEAP_XMAX_INVALID));
5184 : Assert(result != TM_Updated ||
5185 : !ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid));
5186 280 : tmfd->ctid = tuple->t_data->t_ctid;
5187 280 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(tuple->t_data);
5188 280 : if (result == TM_SelfModified)
5189 6 : tmfd->cmax = HeapTupleHeaderGetCmax(tuple->t_data);
5190 : else
5191 274 : tmfd->cmax = InvalidCommandId;
5192 280 : goto out_locked;
5193 : }
5194 :
5195 : /*
5196 : * If we didn't pin the visibility map page and the page has become all
5197 : * visible while we were busy locking the buffer, or during some
5198 : * subsequent window during which we had it unlocked, we'll have to unlock
5199 : * and re-lock, to avoid holding the buffer lock across I/O. That's a bit
5200 : * unfortunate, especially since we'll now have to recheck whether the
5201 : * tuple has been locked or updated under us, but hopefully it won't
5202 : * happen very often.
5203 : */
5204 156652 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
5205 : {
5206 0 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
5207 0 : visibilitymap_pin(relation, block, &vmbuffer);
5208 0 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5209 0 : goto l3;
5210 : }
5211 :
5212 156652 : xmax = HeapTupleHeaderGetRawXmax(tuple->t_data);
5213 156652 : old_infomask = tuple->t_data->t_infomask;
5214 :
5215 : /*
5216 : * If this is the first possibly-multixact-able operation in the current
5217 : * transaction, set my per-backend OldestMemberMXactId setting. We can be
5218 : * certain that the transaction will never become a member of any older
5219 : * MultiXactIds than that. (We have to do this even if we end up just
5220 : * using our own TransactionId below, since some other backend could
5221 : * incorporate our XID into a MultiXact immediately afterwards.)
5222 : */
5223 156652 : MultiXactIdSetOldestMember();
5224 :
5225 : /*
5226 : * Compute the new xmax and infomask to store into the tuple. Note we do
5227 : * not modify the tuple just yet, because that would leave it in the wrong
5228 : * state if multixact.c elogs.
5229 : */
5230 156652 : compute_new_xmax_infomask(xmax, old_infomask, tuple->t_data->t_infomask2,
5231 : GetCurrentTransactionId(), mode, false,
5232 : &xid, &new_infomask, &new_infomask2);
5233 :
5234 156652 : START_CRIT_SECTION();
5235 :
5236 : /*
5237 : * Store transaction information of xact locking the tuple.
5238 : *
5239 : * Note: Cmax is meaningless in this context, so don't set it; this avoids
5240 : * possibly generating a useless combo CID. Moreover, if we're locking a
5241 : * previously updated tuple, it's important to preserve the Cmax.
5242 : *
5243 : * Also reset the HOT UPDATE bit, but only if there's no update; otherwise
5244 : * we would break the HOT chain.
5245 : */
5246 156652 : tuple->t_data->t_infomask &= ~HEAP_XMAX_BITS;
5247 156652 : tuple->t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
5248 156652 : tuple->t_data->t_infomask |= new_infomask;
5249 156652 : tuple->t_data->t_infomask2 |= new_infomask2;
5250 156652 : if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
5251 154505 : HeapTupleHeaderClearHotUpdated(tuple->t_data);
5252 156652 : HeapTupleHeaderSetXmax(tuple->t_data, xid);
5253 :
5254 : /*
5255 : * Make sure there is no forward chain link in t_ctid. Note that in the
5256 : * cases where the tuple has been updated, we must not overwrite t_ctid,
5257 : * because it was set by the updater. Moreover, if the tuple has been
5258 : * updated, we need to follow the update chain to lock the new versions of
5259 : * the tuple as well.
5260 : */
5261 156652 : if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
5262 154505 : tuple->t_data->t_ctid = *tid;
5263 :
5264 : /* Clear only the all-frozen bit on visibility map if needed */
5265 158318 : if (PageIsAllVisible(page) &&
5266 1666 : visibilitymap_clear(relation, block, vmbuffer,
5267 : VISIBILITYMAP_ALL_FROZEN))
5268 14 : cleared_all_frozen = true;
5269 :
5270 :
5271 156652 : MarkBufferDirty(*buffer);
5272 :
5273 : /*
5274 : * XLOG stuff. You might think that we don't need an XLOG record because
5275 : * there is no state change worth restoring after a crash. You would be
5276 : * wrong however: we have just written either a TransactionId or a
5277 : * MultiXactId that may never have been seen on disk before, and we need
5278 : * to make sure that there are XLOG entries covering those ID numbers.
5279 : * Else the same IDs might be re-used after a crash, which would be
5280 : * disastrous if this page made it to disk before the crash. Essentially
5281 : * we have to enforce the WAL log-before-data rule even in this case.
5282 : * (Also, in a PITR log-shipping or 2PC environment, we have to have XLOG
5283 : * entries for everything anyway.)
5284 : */
5285 156652 : if (RelationNeedsWAL(relation))
5286 : {
5287 : xl_heap_lock xlrec;
5288 : XLogRecPtr recptr;
5289 :
5290 156301 : XLogBeginInsert();
5291 156301 : XLogRegisterBuffer(0, *buffer, REGBUF_STANDARD);
5292 :
5293 156301 : xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
5294 156301 : xlrec.xmax = xid;
5295 312602 : xlrec.infobits_set = compute_infobits(new_infomask,
5296 156301 : tuple->t_data->t_infomask2);
5297 156301 : xlrec.flags = cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
5298 156301 : XLogRegisterData(&xlrec, SizeOfHeapLock);
5299 :
5300 : /* we don't decode row locks atm, so no need to log the origin */
5301 :
5302 156301 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
5303 :
5304 156301 : PageSetLSN(page, recptr);
5305 : }
5306 :
5307 156652 : END_CRIT_SECTION();
5308 :
5309 156652 : result = TM_Ok;
5310 :
5311 156953 : out_locked:
5312 156953 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
5313 :
5314 158465 : out_unlocked:
5315 158465 : if (BufferIsValid(vmbuffer))
5316 1666 : ReleaseBuffer(vmbuffer);
5317 :
5318 : /*
5319 : * Don't update the visibility map here. Locking a tuple doesn't change
5320 : * visibility info.
5321 : */
5322 :
5323 : /*
5324 : * Now that we have successfully marked the tuple as locked, we can
5325 : * release the lmgr tuple lock, if we had it.
5326 : */
5327 158465 : if (have_tuple_lock)
5328 166 : UnlockTupleTuplock(relation, tid, mode);
5329 :
5330 158465 : return result;
5331 : }
5332 :
5333 : /*
5334 : * Acquire heavyweight lock on the given tuple, in preparation for acquiring
5335 : * its normal, Xmax-based tuple lock.
5336 : *
5337 : * have_tuple_lock is an input and output parameter: on input, it indicates
5338 : * whether the lock has previously been acquired (and this function does
5339 : * nothing in that case). If this function returns success, have_tuple_lock
5340 : * has been flipped to true.
5341 : *
5342 : * Returns false if it was unable to obtain the lock; this can only happen if
5343 : * wait_policy is Skip.
5344 : */
5345 : static bool
5346 319 : heap_acquire_tuplock(Relation relation, const ItemPointerData *tid, LockTupleMode mode,
5347 : LockWaitPolicy wait_policy, bool *have_tuple_lock)
5348 : {
5349 319 : if (*have_tuple_lock)
5350 9 : return true;
5351 :
5352 310 : switch (wait_policy)
5353 : {
5354 269 : case LockWaitBlock:
5355 269 : LockTupleTuplock(relation, tid, mode);
5356 269 : break;
5357 :
5358 34 : case LockWaitSkip:
5359 34 : if (!ConditionalLockTupleTuplock(relation, tid, mode, false))
5360 1 : return false;
5361 33 : break;
5362 :
5363 7 : case LockWaitError:
5364 7 : if (!ConditionalLockTupleTuplock(relation, tid, mode, log_lock_failures))
5365 1 : ereport(ERROR,
5366 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5367 : errmsg("could not obtain lock on row in relation \"%s\"",
5368 : RelationGetRelationName(relation))));
5369 6 : break;
5370 : }
5371 308 : *have_tuple_lock = true;
5372 :
5373 308 : return true;
5374 : }
5375 :
5376 : /*
5377 : * Given an original set of Xmax and infomask, and a transaction (identified by
5378 : * add_to_xmax) acquiring a new lock of some mode, compute the new Xmax and
5379 : * corresponding infomasks to use on the tuple.
5380 : *
5381 : * Note that this might have side effects such as creating a new MultiXactId.
5382 : *
5383 : * Most callers will have called HeapTupleSatisfiesUpdate before this function;
5384 : * that will have set the HEAP_XMAX_INVALID bit if the xmax was a MultiXactId
5385 : * but it was not running anymore. There is a race condition, which is that the
5386 : * MultiXactId may have finished since then, but that uncommon case is handled
5387 : * either here, or within MultiXactIdExpand.
5388 : *
5389 : * There is a similar race condition possible when the old xmax was a regular
5390 : * TransactionId. We test TransactionIdIsInProgress again just to narrow the
5391 : * window, but it's still possible to end up creating an unnecessary
5392 : * MultiXactId. Fortunately this is harmless.
5393 : */
5394 : static void
5395 2147724 : compute_new_xmax_infomask(TransactionId xmax, uint16 old_infomask,
5396 : uint16 old_infomask2, TransactionId add_to_xmax,
5397 : LockTupleMode mode, bool is_update,
5398 : TransactionId *result_xmax, uint16 *result_infomask,
5399 : uint16 *result_infomask2)
5400 : {
5401 : TransactionId new_xmax;
5402 : uint16 new_infomask,
5403 : new_infomask2;
5404 :
5405 : Assert(TransactionIdIsCurrentTransactionId(add_to_xmax));
5406 :
5407 103986 : l5:
5408 2251710 : new_infomask = 0;
5409 2251710 : new_infomask2 = 0;
5410 2251710 : if (old_infomask & HEAP_XMAX_INVALID)
5411 : {
5412 : /*
5413 : * No previous locker; we just insert our own TransactionId.
5414 : *
5415 : * Note that it's critical that this case be the first one checked,
5416 : * because there are several blocks below that come back to this one
5417 : * to implement certain optimizations; old_infomask might contain
5418 : * other dirty bits in those cases, but we don't really care.
5419 : */
5420 2071096 : if (is_update)
5421 : {
5422 1837895 : new_xmax = add_to_xmax;
5423 1837895 : if (mode == LockTupleExclusive)
5424 1563725 : new_infomask2 |= HEAP_KEYS_UPDATED;
5425 : }
5426 : else
5427 : {
5428 233201 : new_infomask |= HEAP_XMAX_LOCK_ONLY;
5429 233201 : switch (mode)
5430 : {
5431 2639 : case LockTupleKeyShare:
5432 2639 : new_xmax = add_to_xmax;
5433 2639 : new_infomask |= HEAP_XMAX_KEYSHR_LOCK;
5434 2639 : break;
5435 745 : case LockTupleShare:
5436 745 : new_xmax = add_to_xmax;
5437 745 : new_infomask |= HEAP_XMAX_SHR_LOCK;
5438 745 : break;
5439 134057 : case LockTupleNoKeyExclusive:
5440 134057 : new_xmax = add_to_xmax;
5441 134057 : new_infomask |= HEAP_XMAX_EXCL_LOCK;
5442 134057 : break;
5443 95760 : case LockTupleExclusive:
5444 95760 : new_xmax = add_to_xmax;
5445 95760 : new_infomask |= HEAP_XMAX_EXCL_LOCK;
5446 95760 : new_infomask2 |= HEAP_KEYS_UPDATED;
5447 95760 : break;
5448 0 : default:
5449 0 : new_xmax = InvalidTransactionId; /* silence compiler */
5450 0 : elog(ERROR, "invalid lock mode");
5451 : }
5452 : }
5453 : }
5454 180614 : else if (old_infomask & HEAP_XMAX_IS_MULTI)
5455 : {
5456 : MultiXactStatus new_status;
5457 :
5458 : /*
5459 : * Currently we don't allow XMAX_COMMITTED to be set for multis, so
5460 : * cross-check.
5461 : */
5462 : Assert(!(old_infomask & HEAP_XMAX_COMMITTED));
5463 :
5464 : /*
5465 : * A multixact together with LOCK_ONLY set but neither lock bit set
5466 : * (i.e. a pg_upgraded share locked tuple) cannot possibly be running
5467 : * anymore. This check is critical for databases upgraded by
5468 : * pg_upgrade; both MultiXactIdIsRunning and MultiXactIdExpand assume
5469 : * that such multis are never passed.
5470 : */
5471 75566 : if (HEAP_LOCKED_UPGRADED(old_infomask))
5472 : {
5473 0 : old_infomask &= ~HEAP_XMAX_IS_MULTI;
5474 0 : old_infomask |= HEAP_XMAX_INVALID;
5475 0 : goto l5;
5476 : }
5477 :
5478 : /*
5479 : * If the XMAX is already a MultiXactId, then we need to expand it to
5480 : * include add_to_xmax; but if all the members were lockers and are
5481 : * all gone, we can do away with the IS_MULTI bit and just set
5482 : * add_to_xmax as the only locker/updater. If all lockers are gone
5483 : * and we have an updater that aborted, we can also do without a
5484 : * multi.
5485 : *
5486 : * The cost of doing GetMultiXactIdMembers would be paid by
5487 : * MultiXactIdExpand if we weren't to do this, so this check is not
5488 : * incurring extra work anyhow.
5489 : */
5490 75566 : if (!MultiXactIdIsRunning(xmax, HEAP_XMAX_IS_LOCKED_ONLY(old_infomask)))
5491 : {
5492 24 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) ||
5493 8 : !TransactionIdDidCommit(MultiXactIdGetUpdateXid(xmax,
5494 : old_infomask)))
5495 : {
5496 : /*
5497 : * Reset these bits and restart; otherwise fall through to
5498 : * create a new multi below.
5499 : */
5500 24 : old_infomask &= ~HEAP_XMAX_IS_MULTI;
5501 24 : old_infomask |= HEAP_XMAX_INVALID;
5502 24 : goto l5;
5503 : }
5504 : }
5505 :
5506 75542 : new_status = get_mxact_status_for_lock(mode, is_update);
5507 :
5508 75542 : new_xmax = MultiXactIdExpand((MultiXactId) xmax, add_to_xmax,
5509 : new_status);
5510 75542 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5511 : }
5512 105048 : else if (old_infomask & HEAP_XMAX_COMMITTED)
5513 : {
5514 : /*
5515 : * It's a committed update, so we need to preserve him as updater of
5516 : * the tuple.
5517 : */
5518 : MultiXactStatus status;
5519 : MultiXactStatus new_status;
5520 :
5521 13 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5522 0 : status = MultiXactStatusUpdate;
5523 : else
5524 13 : status = MultiXactStatusNoKeyUpdate;
5525 :
5526 13 : new_status = get_mxact_status_for_lock(mode, is_update);
5527 :
5528 : /*
5529 : * since it's not running, it's obviously impossible for the old
5530 : * updater to be identical to the current one, so we need not check
5531 : * for that case as we do in the block above.
5532 : */
5533 13 : new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5534 13 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5535 : }
5536 105035 : else if (TransactionIdIsInProgress(xmax))
5537 : {
5538 : /*
5539 : * If the XMAX is a valid, in-progress TransactionId, then we need to
5540 : * create a new MultiXactId that includes both the old locker or
5541 : * updater and our own TransactionId.
5542 : */
5543 : MultiXactStatus new_status;
5544 : MultiXactStatus old_status;
5545 : LockTupleMode old_mode;
5546 :
5547 105026 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
5548 : {
5549 105000 : if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
5550 5673 : old_status = MultiXactStatusForKeyShare;
5551 99327 : else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
5552 433 : old_status = MultiXactStatusForShare;
5553 98894 : else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
5554 : {
5555 98894 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5556 92766 : old_status = MultiXactStatusForUpdate;
5557 : else
5558 6128 : old_status = MultiXactStatusForNoKeyUpdate;
5559 : }
5560 : else
5561 : {
5562 : /*
5563 : * LOCK_ONLY can be present alone only when a page has been
5564 : * upgraded by pg_upgrade. But in that case,
5565 : * TransactionIdIsInProgress() should have returned false. We
5566 : * assume it's no longer locked in this case.
5567 : */
5568 0 : elog(WARNING, "LOCK_ONLY found for Xid in progress %u", xmax);
5569 0 : old_infomask |= HEAP_XMAX_INVALID;
5570 0 : old_infomask &= ~HEAP_XMAX_LOCK_ONLY;
5571 0 : goto l5;
5572 : }
5573 : }
5574 : else
5575 : {
5576 : /* it's an update, but which kind? */
5577 26 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5578 0 : old_status = MultiXactStatusUpdate;
5579 : else
5580 26 : old_status = MultiXactStatusNoKeyUpdate;
5581 : }
5582 :
5583 105026 : old_mode = TUPLOCK_from_mxstatus(old_status);
5584 :
5585 : /*
5586 : * If the lock to be acquired is for the same TransactionId as the
5587 : * existing lock, there's an optimization possible: consider only the
5588 : * strongest of both locks as the only one present, and restart.
5589 : */
5590 105026 : if (xmax == add_to_xmax)
5591 : {
5592 : /*
5593 : * Note that it's not possible for the original tuple to be
5594 : * updated: we wouldn't be here because the tuple would have been
5595 : * invisible and we wouldn't try to update it. As a subtlety,
5596 : * this code can also run when traversing an update chain to lock
5597 : * future versions of a tuple. But we wouldn't be here either,
5598 : * because the add_to_xmax would be different from the original
5599 : * updater.
5600 : */
5601 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
5602 :
5603 : /* acquire the strongest of both */
5604 103954 : if (mode < old_mode)
5605 52200 : mode = old_mode;
5606 : /* mustn't touch is_update */
5607 :
5608 103954 : old_infomask |= HEAP_XMAX_INVALID;
5609 103954 : goto l5;
5610 : }
5611 :
5612 : /* otherwise, just fall back to creating a new multixact */
5613 1072 : new_status = get_mxact_status_for_lock(mode, is_update);
5614 1072 : new_xmax = MultiXactIdCreate(xmax, old_status,
5615 : add_to_xmax, new_status);
5616 1072 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5617 : }
5618 14 : else if (!HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) &&
5619 5 : TransactionIdDidCommit(xmax))
5620 1 : {
5621 : /*
5622 : * It's a committed update, so we gotta preserve him as updater of the
5623 : * tuple.
5624 : */
5625 : MultiXactStatus status;
5626 : MultiXactStatus new_status;
5627 :
5628 1 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5629 0 : status = MultiXactStatusUpdate;
5630 : else
5631 1 : status = MultiXactStatusNoKeyUpdate;
5632 :
5633 1 : new_status = get_mxact_status_for_lock(mode, is_update);
5634 :
5635 : /*
5636 : * since it's not running, it's obviously impossible for the old
5637 : * updater to be identical to the current one, so we need not check
5638 : * for that case as we do in the block above.
5639 : */
5640 1 : new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5641 1 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5642 : }
5643 : else
5644 : {
5645 : /*
5646 : * Can get here iff the locking/updating transaction was running when
5647 : * the infomask was extracted from the tuple, but finished before
5648 : * TransactionIdIsInProgress got to run. Deal with it as if there was
5649 : * no locker at all in the first place.
5650 : */
5651 8 : old_infomask |= HEAP_XMAX_INVALID;
5652 8 : goto l5;
5653 : }
5654 :
5655 2147724 : *result_infomask = new_infomask;
5656 2147724 : *result_infomask2 = new_infomask2;
5657 2147724 : *result_xmax = new_xmax;
5658 2147724 : }
5659 :
5660 : /*
5661 : * Subroutine for heap_lock_updated_tuple_rec.
5662 : *
5663 : * Given a hypothetical multixact status held by the transaction identified
5664 : * with the given xid, does the current transaction need to wait, fail, or can
5665 : * it continue if it wanted to acquire a lock of the given mode? "needwait"
5666 : * is set to true if waiting is necessary; if it can continue, then TM_Ok is
5667 : * returned. If the lock is already held by the current transaction, return
5668 : * TM_SelfModified. In case of a conflict with another transaction, a
5669 : * different HeapTupleSatisfiesUpdate return code is returned.
5670 : *
5671 : * The held status is said to be hypothetical because it might correspond to a
5672 : * lock held by a single Xid, i.e. not a real MultiXactId; we express it this
5673 : * way for simplicity of API.
5674 : */
5675 : static TM_Result
5676 38774 : test_lockmode_for_conflict(MultiXactStatus status, TransactionId xid,
5677 : LockTupleMode mode, HeapTuple tup,
5678 : bool *needwait)
5679 : {
5680 : MultiXactStatus wantedstatus;
5681 :
5682 38774 : *needwait = false;
5683 38774 : wantedstatus = get_mxact_status_for_lock(mode, false);
5684 :
5685 : /*
5686 : * Note: we *must* check TransactionIdIsInProgress before
5687 : * TransactionIdDidAbort/Commit; see comment at top of heapam_visibility.c
5688 : * for an explanation.
5689 : */
5690 38774 : if (TransactionIdIsCurrentTransactionId(xid))
5691 : {
5692 : /*
5693 : * The tuple has already been locked by our own transaction. This is
5694 : * very rare but can happen if multiple transactions are trying to
5695 : * lock an ancient version of the same tuple.
5696 : */
5697 0 : return TM_SelfModified;
5698 : }
5699 38774 : else if (TransactionIdIsInProgress(xid))
5700 : {
5701 : /*
5702 : * If the locking transaction is running, what we do depends on
5703 : * whether the lock modes conflict: if they do, then we must wait for
5704 : * it to finish; otherwise we can fall through to lock this tuple
5705 : * version without waiting.
5706 : */
5707 36539 : if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
5708 36539 : LOCKMODE_from_mxstatus(wantedstatus)))
5709 : {
5710 8 : *needwait = true;
5711 : }
5712 :
5713 : /*
5714 : * If we set needwait above, then this value doesn't matter;
5715 : * otherwise, this value signals to caller that it's okay to proceed.
5716 : */
5717 36539 : return TM_Ok;
5718 : }
5719 2235 : else if (TransactionIdDidAbort(xid))
5720 206 : return TM_Ok;
5721 2029 : else if (TransactionIdDidCommit(xid))
5722 : {
5723 : /*
5724 : * The other transaction committed. If it was only a locker, then the
5725 : * lock is completely gone now and we can return success; but if it
5726 : * was an update, then what we do depends on whether the two lock
5727 : * modes conflict. If they conflict, then we must report error to
5728 : * caller. But if they don't, we can fall through to allow the current
5729 : * transaction to lock the tuple.
5730 : *
5731 : * Note: the reason we worry about ISUPDATE here is because as soon as
5732 : * a transaction ends, all its locks are gone and meaningless, and
5733 : * thus we can ignore them; whereas its updates persist. In the
5734 : * TransactionIdIsInProgress case, above, we don't need to check
5735 : * because we know the lock is still "alive" and thus a conflict needs
5736 : * always be checked.
5737 : */
5738 2029 : if (!ISUPDATE_from_mxstatus(status))
5739 2020 : return TM_Ok;
5740 :
5741 9 : if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
5742 9 : LOCKMODE_from_mxstatus(wantedstatus)))
5743 : {
5744 : /* bummer */
5745 8 : if (!ItemPointerEquals(&tup->t_self, &tup->t_data->t_ctid))
5746 6 : return TM_Updated;
5747 : else
5748 2 : return TM_Deleted;
5749 : }
5750 :
5751 1 : return TM_Ok;
5752 : }
5753 :
5754 : /* Not in progress, not aborted, not committed -- must have crashed */
5755 0 : return TM_Ok;
5756 : }
5757 :
5758 :
5759 : /*
5760 : * Recursive part of heap_lock_updated_tuple
5761 : *
5762 : * Fetch the tuple pointed to by tid in rel, and mark it as locked by the given
5763 : * xid with the given mode; if this tuple is updated, recurse to lock the new
5764 : * version as well.
5765 : */
5766 : static TM_Result
5767 2212 : heap_lock_updated_tuple_rec(Relation rel, TransactionId priorXmax,
5768 : const ItemPointerData *tid, TransactionId xid,
5769 : LockTupleMode mode)
5770 : {
5771 : TM_Result result;
5772 : ItemPointerData tupid;
5773 : HeapTupleData mytup;
5774 : Buffer buf;
5775 : uint16 new_infomask,
5776 : new_infomask2,
5777 : old_infomask,
5778 : old_infomask2;
5779 : TransactionId xmax,
5780 : new_xmax;
5781 2212 : bool cleared_all_frozen = false;
5782 : bool pinned_desired_page;
5783 2212 : Buffer vmbuffer = InvalidBuffer;
5784 : BlockNumber block;
5785 :
5786 2212 : ItemPointerCopy(tid, &tupid);
5787 :
5788 : for (;;)
5789 : {
5790 2215 : new_infomask = 0;
5791 2215 : new_xmax = InvalidTransactionId;
5792 2215 : block = ItemPointerGetBlockNumber(&tupid);
5793 2215 : ItemPointerCopy(&tupid, &(mytup.t_self));
5794 :
5795 2215 : if (!heap_fetch(rel, SnapshotAny, &mytup, &buf, false))
5796 : {
5797 : /*
5798 : * if we fail to find the updated version of the tuple, it's
5799 : * because it was vacuumed/pruned away after its creator
5800 : * transaction aborted. So behave as if we got to the end of the
5801 : * chain, and there's no further tuple to lock: return success to
5802 : * caller.
5803 : */
5804 0 : result = TM_Ok;
5805 0 : goto out_unlocked;
5806 : }
5807 :
5808 2215 : l4:
5809 2223 : CHECK_FOR_INTERRUPTS();
5810 :
5811 : /*
5812 : * Before locking the buffer, pin the visibility map page if it
5813 : * appears to be necessary. Since we haven't got the lock yet,
5814 : * someone else might be in the middle of changing this, so we'll need
5815 : * to recheck after we have the lock.
5816 : */
5817 2223 : if (PageIsAllVisible(BufferGetPage(buf)))
5818 : {
5819 0 : visibilitymap_pin(rel, block, &vmbuffer);
5820 0 : pinned_desired_page = true;
5821 : }
5822 : else
5823 2223 : pinned_desired_page = false;
5824 :
5825 2223 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
5826 :
5827 : /*
5828 : * If we didn't pin the visibility map page and the page has become
5829 : * all visible while we were busy locking the buffer, we'll have to
5830 : * unlock and re-lock, to avoid holding the buffer lock across I/O.
5831 : * That's a bit unfortunate, but hopefully shouldn't happen often.
5832 : *
5833 : * Note: in some paths through this function, we will reach here
5834 : * holding a pin on a vm page that may or may not be the one matching
5835 : * this page. If this page isn't all-visible, we won't use the vm
5836 : * page, but we hold onto such a pin till the end of the function.
5837 : */
5838 2223 : if (!pinned_desired_page && PageIsAllVisible(BufferGetPage(buf)))
5839 : {
5840 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5841 0 : visibilitymap_pin(rel, block, &vmbuffer);
5842 0 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
5843 : }
5844 :
5845 : /*
5846 : * Check the tuple XMIN against prior XMAX, if any. If we reached the
5847 : * end of the chain, we're done, so return success.
5848 : */
5849 4446 : if (TransactionIdIsValid(priorXmax) &&
5850 2223 : !TransactionIdEquals(HeapTupleHeaderGetXmin(mytup.t_data),
5851 : priorXmax))
5852 : {
5853 2 : result = TM_Ok;
5854 2 : goto out_locked;
5855 : }
5856 :
5857 : /*
5858 : * Also check Xmin: if this tuple was created by an aborted
5859 : * (sub)transaction, then we already locked the last live one in the
5860 : * chain, thus we're done, so return success.
5861 : */
5862 2221 : if (TransactionIdDidAbort(HeapTupleHeaderGetXmin(mytup.t_data)))
5863 : {
5864 24 : result = TM_Ok;
5865 24 : goto out_locked;
5866 : }
5867 :
5868 2197 : old_infomask = mytup.t_data->t_infomask;
5869 2197 : old_infomask2 = mytup.t_data->t_infomask2;
5870 2197 : xmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
5871 :
5872 : /*
5873 : * If this tuple version has been updated or locked by some concurrent
5874 : * transaction(s), what we do depends on whether our lock mode
5875 : * conflicts with what those other transactions hold, and also on the
5876 : * status of them.
5877 : */
5878 2197 : if (!(old_infomask & HEAP_XMAX_INVALID))
5879 : {
5880 : TransactionId rawxmax;
5881 : bool needwait;
5882 :
5883 2138 : rawxmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
5884 2138 : if (old_infomask & HEAP_XMAX_IS_MULTI)
5885 : {
5886 : int nmembers;
5887 : int i;
5888 : MultiXactMember *members;
5889 :
5890 : /*
5891 : * We don't need a test for pg_upgrade'd tuples: this is only
5892 : * applied to tuples after the first in an update chain. Said
5893 : * first tuple in the chain may well be locked-in-9.2-and-
5894 : * pg_upgraded, but that one was already locked by our caller,
5895 : * not us; and any subsequent ones cannot be because our
5896 : * caller must necessarily have obtained a snapshot later than
5897 : * the pg_upgrade itself.
5898 : */
5899 : Assert(!HEAP_LOCKED_UPGRADED(mytup.t_data->t_infomask));
5900 :
5901 2109 : nmembers = GetMultiXactIdMembers(rawxmax, &members, false,
5902 2109 : HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
5903 40854 : for (i = 0; i < nmembers; i++)
5904 : {
5905 38745 : result = test_lockmode_for_conflict(members[i].status,
5906 38745 : members[i].xid,
5907 : mode,
5908 : &mytup,
5909 : &needwait);
5910 :
5911 : /*
5912 : * If the tuple was already locked by ourselves in a
5913 : * previous iteration of this (say heap_lock_tuple was
5914 : * forced to restart the locking loop because of a change
5915 : * in xmax), then we hold the lock already on this tuple
5916 : * version and we don't need to do anything; and this is
5917 : * not an error condition either. We just need to skip
5918 : * this tuple and continue locking the next version in the
5919 : * update chain.
5920 : */
5921 38745 : if (result == TM_SelfModified)
5922 : {
5923 0 : pfree(members);
5924 0 : goto next;
5925 : }
5926 :
5927 38745 : if (needwait)
5928 : {
5929 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5930 0 : XactLockTableWait(members[i].xid, rel,
5931 : &mytup.t_self,
5932 : XLTW_LockUpdated);
5933 0 : pfree(members);
5934 0 : goto l4;
5935 : }
5936 38745 : if (result != TM_Ok)
5937 : {
5938 0 : pfree(members);
5939 0 : goto out_locked;
5940 : }
5941 : }
5942 2109 : if (members)
5943 2109 : pfree(members);
5944 : }
5945 : else
5946 : {
5947 : MultiXactStatus status;
5948 :
5949 : /*
5950 : * For a non-multi Xmax, we first need to compute the
5951 : * corresponding MultiXactStatus by using the infomask bits.
5952 : */
5953 29 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
5954 : {
5955 10 : if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
5956 10 : status = MultiXactStatusForKeyShare;
5957 0 : else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
5958 0 : status = MultiXactStatusForShare;
5959 0 : else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
5960 : {
5961 0 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5962 0 : status = MultiXactStatusForUpdate;
5963 : else
5964 0 : status = MultiXactStatusForNoKeyUpdate;
5965 : }
5966 : else
5967 : {
5968 : /*
5969 : * LOCK_ONLY present alone (a pg_upgraded tuple marked
5970 : * as share-locked in the old cluster) shouldn't be
5971 : * seen in the middle of an update chain.
5972 : */
5973 0 : elog(ERROR, "invalid lock status in tuple");
5974 : }
5975 : }
5976 : else
5977 : {
5978 : /* it's an update, but which kind? */
5979 19 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5980 14 : status = MultiXactStatusUpdate;
5981 : else
5982 5 : status = MultiXactStatusNoKeyUpdate;
5983 : }
5984 :
5985 29 : result = test_lockmode_for_conflict(status, rawxmax, mode,
5986 : &mytup, &needwait);
5987 :
5988 : /*
5989 : * If the tuple was already locked by ourselves in a previous
5990 : * iteration of this (say heap_lock_tuple was forced to
5991 : * restart the locking loop because of a change in xmax), then
5992 : * we hold the lock already on this tuple version and we don't
5993 : * need to do anything; and this is not an error condition
5994 : * either. We just need to skip this tuple and continue
5995 : * locking the next version in the update chain.
5996 : */
5997 29 : if (result == TM_SelfModified)
5998 0 : goto next;
5999 :
6000 29 : if (needwait)
6001 : {
6002 8 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
6003 8 : XactLockTableWait(rawxmax, rel, &mytup.t_self,
6004 : XLTW_LockUpdated);
6005 8 : goto l4;
6006 : }
6007 21 : if (result != TM_Ok)
6008 : {
6009 8 : goto out_locked;
6010 : }
6011 : }
6012 : }
6013 :
6014 : /* compute the new Xmax and infomask values for the tuple ... */
6015 2181 : compute_new_xmax_infomask(xmax, old_infomask, mytup.t_data->t_infomask2,
6016 : xid, mode, false,
6017 : &new_xmax, &new_infomask, &new_infomask2);
6018 :
6019 2181 : if (PageIsAllVisible(BufferGetPage(buf)) &&
6020 0 : visibilitymap_clear(rel, block, vmbuffer,
6021 : VISIBILITYMAP_ALL_FROZEN))
6022 0 : cleared_all_frozen = true;
6023 :
6024 2181 : START_CRIT_SECTION();
6025 :
6026 : /* ... and set them */
6027 2181 : HeapTupleHeaderSetXmax(mytup.t_data, new_xmax);
6028 2181 : mytup.t_data->t_infomask &= ~HEAP_XMAX_BITS;
6029 2181 : mytup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
6030 2181 : mytup.t_data->t_infomask |= new_infomask;
6031 2181 : mytup.t_data->t_infomask2 |= new_infomask2;
6032 :
6033 2181 : MarkBufferDirty(buf);
6034 :
6035 : /* XLOG stuff */
6036 2181 : if (RelationNeedsWAL(rel))
6037 : {
6038 : xl_heap_lock_updated xlrec;
6039 : XLogRecPtr recptr;
6040 2181 : Page page = BufferGetPage(buf);
6041 :
6042 2181 : XLogBeginInsert();
6043 2181 : XLogRegisterBuffer(0, buf, REGBUF_STANDARD);
6044 :
6045 2181 : xlrec.offnum = ItemPointerGetOffsetNumber(&mytup.t_self);
6046 2181 : xlrec.xmax = new_xmax;
6047 2181 : xlrec.infobits_set = compute_infobits(new_infomask, new_infomask2);
6048 2181 : xlrec.flags =
6049 2181 : cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
6050 :
6051 2181 : XLogRegisterData(&xlrec, SizeOfHeapLockUpdated);
6052 :
6053 2181 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_LOCK_UPDATED);
6054 :
6055 2181 : PageSetLSN(page, recptr);
6056 : }
6057 :
6058 2181 : END_CRIT_SECTION();
6059 :
6060 2181 : next:
6061 : /* if we find the end of update chain, we're done. */
6062 4362 : if (mytup.t_data->t_infomask & HEAP_XMAX_INVALID ||
6063 4362 : HeapTupleHeaderIndicatesMovedPartitions(mytup.t_data) ||
6064 2185 : ItemPointerEquals(&mytup.t_self, &mytup.t_data->t_ctid) ||
6065 4 : HeapTupleHeaderIsOnlyLocked(mytup.t_data))
6066 : {
6067 2178 : result = TM_Ok;
6068 2178 : goto out_locked;
6069 : }
6070 :
6071 : /* tail recursion */
6072 3 : priorXmax = HeapTupleHeaderGetUpdateXid(mytup.t_data);
6073 3 : ItemPointerCopy(&(mytup.t_data->t_ctid), &tupid);
6074 3 : UnlockReleaseBuffer(buf);
6075 : }
6076 :
6077 : result = TM_Ok;
6078 :
6079 2212 : out_locked:
6080 2212 : UnlockReleaseBuffer(buf);
6081 :
6082 2212 : out_unlocked:
6083 2212 : if (vmbuffer != InvalidBuffer)
6084 0 : ReleaseBuffer(vmbuffer);
6085 :
6086 2212 : return result;
6087 : }
6088 :
6089 : /*
6090 : * heap_lock_updated_tuple
6091 : * Follow update chain when locking an updated tuple, acquiring locks (row
6092 : * marks) on the updated versions.
6093 : *
6094 : * 'prior_infomask', 'prior_raw_xmax' and 'prior_ctid' are the corresponding
6095 : * fields from the initial tuple. We will lock the tuples starting from the
6096 : * one that 'prior_ctid' points to. Note: This function does not lock the
6097 : * initial tuple itself.
6098 : *
6099 : * This function doesn't check visibility, it just unconditionally marks the
6100 : * tuple(s) as locked. If any tuple in the updated chain is being deleted
6101 : * concurrently (or updated with the key being modified), sleep until the
6102 : * transaction doing it is finished.
6103 : *
6104 : * Note that we don't acquire heavyweight tuple locks on the tuples we walk
6105 : * when we have to wait for other transactions to release them, as opposed to
6106 : * what heap_lock_tuple does. The reason is that having more than one
6107 : * transaction walking the chain is probably uncommon enough that risk of
6108 : * starvation is not likely: one of the preconditions for being here is that
6109 : * the snapshot in use predates the update that created this tuple (because we
6110 : * started at an earlier version of the tuple), but at the same time such a
6111 : * transaction cannot be using repeatable read or serializable isolation
6112 : * levels, because that would lead to a serializability failure.
6113 : */
6114 : static TM_Result
6115 2214 : heap_lock_updated_tuple(Relation rel,
6116 : uint16 prior_infomask,
6117 : TransactionId prior_raw_xmax,
6118 : const ItemPointerData *prior_ctid,
6119 : TransactionId xid, LockTupleMode mode)
6120 : {
6121 2214 : INJECTION_POINT("heap_lock_updated_tuple", NULL);
6122 :
6123 : /*
6124 : * If the tuple has moved into another partition (effectively a delete)
6125 : * stop here.
6126 : */
6127 2214 : if (!ItemPointerIndicatesMovedPartitions(prior_ctid))
6128 : {
6129 : TransactionId prior_xmax;
6130 :
6131 : /*
6132 : * If this is the first possibly-multixact-able operation in the
6133 : * current transaction, set my per-backend OldestMemberMXactId
6134 : * setting. We can be certain that the transaction will never become a
6135 : * member of any older MultiXactIds than that. (We have to do this
6136 : * even if we end up just using our own TransactionId below, since
6137 : * some other backend could incorporate our XID into a MultiXact
6138 : * immediately afterwards.)
6139 : */
6140 2212 : MultiXactIdSetOldestMember();
6141 :
6142 4424 : prior_xmax = (prior_infomask & HEAP_XMAX_IS_MULTI) ?
6143 2212 : MultiXactIdGetUpdateXid(prior_raw_xmax, prior_infomask) : prior_raw_xmax;
6144 2212 : return heap_lock_updated_tuple_rec(rel, prior_xmax, prior_ctid, xid, mode);
6145 : }
6146 :
6147 : /* nothing to lock */
6148 2 : return TM_Ok;
6149 : }
6150 :
6151 : /*
6152 : * heap_finish_speculative - mark speculative insertion as successful
6153 : *
6154 : * To successfully finish a speculative insertion we have to clear speculative
6155 : * token from tuple. To do so the t_ctid field, which will contain a
6156 : * speculative token value, is modified in place to point to the tuple itself,
6157 : * which is characteristic of a newly inserted ordinary tuple.
6158 : *
6159 : * NB: It is not ok to commit without either finishing or aborting a
6160 : * speculative insertion. We could treat speculative tuples of committed
6161 : * transactions implicitly as completed, but then we would have to be prepared
6162 : * to deal with speculative tokens on committed tuples. That wouldn't be
6163 : * difficult - no-one looks at the ctid field of a tuple with invalid xmax -
6164 : * but clearing the token at completion isn't very expensive either.
6165 : * An explicit confirmation WAL record also makes logical decoding simpler.
6166 : */
6167 : void
6168 2118 : heap_finish_speculative(Relation relation, const ItemPointerData *tid)
6169 : {
6170 : Buffer buffer;
6171 : Page page;
6172 : OffsetNumber offnum;
6173 : ItemId lp;
6174 : HeapTupleHeader htup;
6175 :
6176 2118 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
6177 2118 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6178 2118 : page = BufferGetPage(buffer);
6179 :
6180 2118 : offnum = ItemPointerGetOffsetNumber(tid);
6181 2118 : if (offnum < 1 || offnum > PageGetMaxOffsetNumber(page))
6182 0 : elog(ERROR, "offnum out of range");
6183 2118 : lp = PageGetItemId(page, offnum);
6184 2118 : if (!ItemIdIsNormal(lp))
6185 0 : elog(ERROR, "invalid lp");
6186 :
6187 2118 : htup = (HeapTupleHeader) PageGetItem(page, lp);
6188 :
6189 : /* NO EREPORT(ERROR) from here till changes are logged */
6190 2118 : START_CRIT_SECTION();
6191 :
6192 : Assert(HeapTupleHeaderIsSpeculative(htup));
6193 :
6194 2118 : MarkBufferDirty(buffer);
6195 :
6196 : /*
6197 : * Replace the speculative insertion token with a real t_ctid, pointing to
6198 : * itself like it does on regular tuples.
6199 : */
6200 2118 : htup->t_ctid = *tid;
6201 :
6202 : /* XLOG stuff */
6203 2118 : if (RelationNeedsWAL(relation))
6204 : {
6205 : xl_heap_confirm xlrec;
6206 : XLogRecPtr recptr;
6207 :
6208 2102 : xlrec.offnum = ItemPointerGetOffsetNumber(tid);
6209 :
6210 2102 : XLogBeginInsert();
6211 :
6212 : /* We want the same filtering on this as on a plain insert */
6213 2102 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
6214 :
6215 2102 : XLogRegisterData(&xlrec, SizeOfHeapConfirm);
6216 2102 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
6217 :
6218 2102 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_CONFIRM);
6219 :
6220 2102 : PageSetLSN(page, recptr);
6221 : }
6222 :
6223 2118 : END_CRIT_SECTION();
6224 :
6225 2118 : UnlockReleaseBuffer(buffer);
6226 2118 : }
6227 :
6228 : /*
6229 : * heap_abort_speculative - kill a speculatively inserted tuple
6230 : *
6231 : * Marks a tuple that was speculatively inserted in the same command as dead,
6232 : * by setting its xmin as invalid. That makes it immediately appear as dead
6233 : * to all transactions, including our own. In particular, it makes
6234 : * HeapTupleSatisfiesDirty() regard the tuple as dead, so that another backend
6235 : * inserting a duplicate key value won't unnecessarily wait for our whole
6236 : * transaction to finish (it'll just wait for our speculative insertion to
6237 : * finish).
6238 : *
6239 : * Killing the tuple prevents "unprincipled deadlocks", which are deadlocks
6240 : * that arise due to a mutual dependency that is not user visible. By
6241 : * definition, unprincipled deadlocks cannot be prevented by the user
6242 : * reordering lock acquisition in client code, because the implementation level
6243 : * lock acquisitions are not under the user's direct control. If speculative
6244 : * inserters did not take this precaution, then under high concurrency they
6245 : * could deadlock with each other, which would not be acceptable.
6246 : *
6247 : * This is somewhat redundant with heap_delete, but we prefer to have a
6248 : * dedicated routine with stripped down requirements. Note that this is also
6249 : * used to delete the TOAST tuples created during speculative insertion.
6250 : *
6251 : * This routine does not affect logical decoding as it only looks at
6252 : * confirmation records.
6253 : */
6254 : void
6255 16 : heap_abort_speculative(Relation relation, const ItemPointerData *tid)
6256 : {
6257 16 : TransactionId xid = GetCurrentTransactionId();
6258 : ItemId lp;
6259 : HeapTupleData tp;
6260 : Page page;
6261 : BlockNumber block;
6262 : Buffer buffer;
6263 :
6264 : Assert(ItemPointerIsValid(tid));
6265 :
6266 16 : block = ItemPointerGetBlockNumber(tid);
6267 16 : buffer = ReadBuffer(relation, block);
6268 16 : page = BufferGetPage(buffer);
6269 :
6270 16 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6271 :
6272 : /*
6273 : * Page can't be all visible, we just inserted into it, and are still
6274 : * running.
6275 : */
6276 : Assert(!PageIsAllVisible(page));
6277 :
6278 16 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
6279 : Assert(ItemIdIsNormal(lp));
6280 :
6281 16 : tp.t_tableOid = RelationGetRelid(relation);
6282 16 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
6283 16 : tp.t_len = ItemIdGetLength(lp);
6284 16 : tp.t_self = *tid;
6285 :
6286 : /*
6287 : * Sanity check that the tuple really is a speculatively inserted tuple,
6288 : * inserted by us.
6289 : */
6290 16 : if (tp.t_data->t_choice.t_heap.t_xmin != xid)
6291 0 : elog(ERROR, "attempted to kill a tuple inserted by another transaction");
6292 16 : if (!(IsToastRelation(relation) || HeapTupleHeaderIsSpeculative(tp.t_data)))
6293 0 : elog(ERROR, "attempted to kill a non-speculative tuple");
6294 : Assert(!HeapTupleHeaderIsHeapOnly(tp.t_data));
6295 :
6296 : /*
6297 : * No need to check for serializable conflicts here. There is never a
6298 : * need for a combo CID, either. No need to extract replica identity, or
6299 : * do anything special with infomask bits.
6300 : */
6301 :
6302 16 : START_CRIT_SECTION();
6303 :
6304 : /*
6305 : * The tuple will become DEAD immediately. Flag that this page is a
6306 : * candidate for pruning by setting xmin to TransactionXmin. While not
6307 : * immediately prunable, it is the oldest xid we can cheaply determine
6308 : * that's safe against wraparound / being older than the table's
6309 : * relfrozenxid. To defend against the unlikely case of a new relation
6310 : * having a newer relfrozenxid than our TransactionXmin, use relfrozenxid
6311 : * if so (vacuum can't subsequently move relfrozenxid to beyond
6312 : * TransactionXmin, so there's no race here).
6313 : */
6314 : Assert(TransactionIdIsValid(TransactionXmin));
6315 : {
6316 16 : TransactionId relfrozenxid = relation->rd_rel->relfrozenxid;
6317 : TransactionId prune_xid;
6318 :
6319 16 : if (TransactionIdPrecedes(TransactionXmin, relfrozenxid))
6320 0 : prune_xid = relfrozenxid;
6321 : else
6322 16 : prune_xid = TransactionXmin;
6323 16 : PageSetPrunable(page, prune_xid);
6324 : }
6325 :
6326 : /* store transaction information of xact deleting the tuple */
6327 16 : tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
6328 16 : tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
6329 :
6330 : /*
6331 : * Set the tuple header xmin to InvalidTransactionId. This makes the
6332 : * tuple immediately invisible everyone. (In particular, to any
6333 : * transactions waiting on the speculative token, woken up later.)
6334 : */
6335 16 : HeapTupleHeaderSetXmin(tp.t_data, InvalidTransactionId);
6336 :
6337 : /* Clear the speculative insertion token too */
6338 16 : tp.t_data->t_ctid = tp.t_self;
6339 :
6340 16 : MarkBufferDirty(buffer);
6341 :
6342 : /*
6343 : * XLOG stuff
6344 : *
6345 : * The WAL records generated here match heap_delete(). The same recovery
6346 : * routines are used.
6347 : */
6348 16 : if (RelationNeedsWAL(relation))
6349 : {
6350 : xl_heap_delete xlrec;
6351 : XLogRecPtr recptr;
6352 :
6353 12 : xlrec.flags = XLH_DELETE_IS_SUPER;
6354 24 : xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
6355 12 : tp.t_data->t_infomask2);
6356 12 : xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
6357 12 : xlrec.xmax = xid;
6358 :
6359 12 : XLogBeginInsert();
6360 12 : XLogRegisterData(&xlrec, SizeOfHeapDelete);
6361 12 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
6362 :
6363 : /* No replica identity & replication origin logged */
6364 :
6365 12 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
6366 :
6367 12 : PageSetLSN(page, recptr);
6368 : }
6369 :
6370 16 : END_CRIT_SECTION();
6371 :
6372 16 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6373 :
6374 16 : if (HeapTupleHasExternal(&tp))
6375 : {
6376 : Assert(!IsToastRelation(relation));
6377 1 : heap_toast_delete(relation, &tp, true);
6378 : }
6379 :
6380 : /*
6381 : * Never need to mark tuple for invalidation, since catalogs don't support
6382 : * speculative insertion
6383 : */
6384 :
6385 : /* Now we can release the buffer */
6386 16 : ReleaseBuffer(buffer);
6387 :
6388 : /* count deletion, as we counted the insertion too */
6389 16 : pgstat_count_heap_delete(relation);
6390 16 : }
6391 :
6392 : /*
6393 : * heap_inplace_lock - protect inplace update from concurrent heap_update()
6394 : *
6395 : * Evaluate whether the tuple's state is compatible with a no-key update.
6396 : * Current transaction rowmarks are fine, as is KEY SHARE from any
6397 : * transaction. If compatible, return true with the buffer exclusive-locked,
6398 : * and the caller must release that by calling
6399 : * heap_inplace_update_and_unlock(), calling heap_inplace_unlock(), or raising
6400 : * an error. Otherwise, call release_callback(arg), wait for blocking
6401 : * transactions to end, and return false.
6402 : *
6403 : * Since this is intended for system catalogs and SERIALIZABLE doesn't cover
6404 : * DDL, this doesn't guarantee any particular predicate locking.
6405 : *
6406 : * heap_delete() is a rarer source of blocking transactions (xwait). We'll
6407 : * wait for such a transaction just like for the normal heap_update() case.
6408 : * Normal concurrent DROP commands won't cause that, because all inplace
6409 : * updaters take some lock that conflicts with DROP. An explicit SQL "DELETE
6410 : * FROM pg_class" can cause it. By waiting, if the concurrent transaction
6411 : * executed both "DELETE FROM pg_class" and "INSERT INTO pg_class", our caller
6412 : * can find the successor tuple.
6413 : *
6414 : * Readers of inplace-updated fields expect changes to those fields are
6415 : * durable. For example, vac_truncate_clog() reads datfrozenxid from
6416 : * pg_database tuples via catalog snapshots. A future snapshot must not
6417 : * return a lower datfrozenxid for the same database OID (lower in the
6418 : * FullTransactionIdPrecedes() sense). We achieve that since no update of a
6419 : * tuple can start while we hold a lock on its buffer. In cases like
6420 : * BEGIN;GRANT;CREATE INDEX;COMMIT we're inplace-updating a tuple visible only
6421 : * to this transaction. ROLLBACK then is one case where it's okay to lose
6422 : * inplace updates. (Restoring relhasindex=false on ROLLBACK is fine, since
6423 : * any concurrent CREATE INDEX would have blocked, then inplace-updated the
6424 : * committed tuple.)
6425 : *
6426 : * In principle, we could avoid waiting by overwriting every tuple in the
6427 : * updated tuple chain. Reader expectations permit updating a tuple only if
6428 : * it's aborted, is the tail of the chain, or we already updated the tuple
6429 : * referenced in its t_ctid. Hence, we would need to overwrite the tuples in
6430 : * order from tail to head. That would imply either (a) mutating all tuples
6431 : * in one critical section or (b) accepting a chance of partial completion.
6432 : * Partial completion of a relfrozenxid update would have the weird
6433 : * consequence that the table's next VACUUM could see the table's relfrozenxid
6434 : * move forward between vacuum_get_cutoffs() and finishing.
6435 : */
6436 : bool
6437 201829 : heap_inplace_lock(Relation relation,
6438 : HeapTuple oldtup_ptr, Buffer buffer,
6439 : void (*release_callback) (void *), void *arg)
6440 : {
6441 201829 : HeapTupleData oldtup = *oldtup_ptr; /* minimize diff vs. heap_update() */
6442 : TM_Result result;
6443 : bool ret;
6444 :
6445 : #ifdef USE_ASSERT_CHECKING
6446 : if (RelationGetRelid(relation) == RelationRelationId)
6447 : check_inplace_rel_lock(oldtup_ptr);
6448 : #endif
6449 :
6450 : Assert(BufferIsValid(buffer));
6451 :
6452 : /*
6453 : * Register shared cache invals if necessary. Other sessions may finish
6454 : * inplace updates of this tuple between this step and LockTuple(). Since
6455 : * inplace updates don't change cache keys, that's harmless.
6456 : *
6457 : * While it's tempting to register invals only after confirming we can
6458 : * return true, the following obstacle precludes reordering steps that
6459 : * way. Registering invals might reach a CatalogCacheInitializeCache()
6460 : * that locks "buffer". That would hang indefinitely if running after our
6461 : * own LockBuffer(). Hence, we must register invals before LockBuffer().
6462 : */
6463 201829 : CacheInvalidateHeapTupleInplace(relation, oldtup_ptr);
6464 :
6465 201829 : LockTuple(relation, &oldtup.t_self, InplaceUpdateTupleLock);
6466 201829 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6467 :
6468 : /*----------
6469 : * Interpret HeapTupleSatisfiesUpdate() like heap_update() does, except:
6470 : *
6471 : * - wait unconditionally
6472 : * - already locked tuple above, since inplace needs that unconditionally
6473 : * - don't recheck header after wait: simpler to defer to next iteration
6474 : * - don't try to continue even if the updater aborts: likewise
6475 : * - no crosscheck
6476 : */
6477 201829 : result = HeapTupleSatisfiesUpdate(&oldtup, GetCurrentCommandId(false),
6478 : buffer);
6479 :
6480 201829 : if (result == TM_Invisible)
6481 : {
6482 : /* no known way this can happen */
6483 0 : ereport(ERROR,
6484 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6485 : errmsg_internal("attempted to overwrite invisible tuple")));
6486 : }
6487 201829 : else if (result == TM_SelfModified)
6488 : {
6489 : /*
6490 : * CREATE INDEX might reach this if an expression is silly enough to
6491 : * call e.g. SELECT ... FROM pg_class FOR SHARE. C code of other SQL
6492 : * statements might get here after a heap_update() of the same row, in
6493 : * the absence of an intervening CommandCounterIncrement().
6494 : */
6495 0 : ereport(ERROR,
6496 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6497 : errmsg("tuple to be updated was already modified by an operation triggered by the current command")));
6498 : }
6499 201829 : else if (result == TM_BeingModified)
6500 : {
6501 : TransactionId xwait;
6502 : uint16 infomask;
6503 :
6504 31 : xwait = HeapTupleHeaderGetRawXmax(oldtup.t_data);
6505 31 : infomask = oldtup.t_data->t_infomask;
6506 :
6507 31 : if (infomask & HEAP_XMAX_IS_MULTI)
6508 : {
6509 5 : LockTupleMode lockmode = LockTupleNoKeyExclusive;
6510 5 : MultiXactStatus mxact_status = MultiXactStatusNoKeyUpdate;
6511 : int remain;
6512 :
6513 5 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
6514 : lockmode, NULL))
6515 : {
6516 2 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6517 2 : release_callback(arg);
6518 2 : ret = false;
6519 2 : MultiXactIdWait((MultiXactId) xwait, mxact_status, infomask,
6520 : relation, &oldtup.t_self, XLTW_Update,
6521 : &remain);
6522 : }
6523 : else
6524 3 : ret = true;
6525 : }
6526 26 : else if (TransactionIdIsCurrentTransactionId(xwait))
6527 1 : ret = true;
6528 25 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask))
6529 1 : ret = true;
6530 : else
6531 : {
6532 24 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6533 24 : release_callback(arg);
6534 24 : ret = false;
6535 24 : XactLockTableWait(xwait, relation, &oldtup.t_self,
6536 : XLTW_Update);
6537 : }
6538 : }
6539 : else
6540 : {
6541 201798 : ret = (result == TM_Ok);
6542 201798 : if (!ret)
6543 : {
6544 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6545 0 : release_callback(arg);
6546 : }
6547 : }
6548 :
6549 : /*
6550 : * GetCatalogSnapshot() relies on invalidation messages to know when to
6551 : * take a new snapshot. COMMIT of xwait is responsible for sending the
6552 : * invalidation. We're not acquiring heavyweight locks sufficient to
6553 : * block if not yet sent, so we must take a new snapshot to ensure a later
6554 : * attempt has a fair chance. While we don't need this if xwait aborted,
6555 : * don't bother optimizing that.
6556 : */
6557 201829 : if (!ret)
6558 : {
6559 26 : UnlockTuple(relation, &oldtup.t_self, InplaceUpdateTupleLock);
6560 26 : ForgetInplace_Inval();
6561 26 : InvalidateCatalogSnapshot();
6562 : }
6563 201829 : return ret;
6564 : }
6565 :
6566 : /*
6567 : * heap_inplace_update_and_unlock - core of systable_inplace_update_finish
6568 : *
6569 : * The tuple cannot change size, and therefore its header fields and null
6570 : * bitmap (if any) don't change either.
6571 : *
6572 : * Since we hold LOCKTAG_TUPLE, no updater has a local copy of this tuple.
6573 : */
6574 : void
6575 83943 : heap_inplace_update_and_unlock(Relation relation,
6576 : HeapTuple oldtup, HeapTuple tuple,
6577 : Buffer buffer)
6578 : {
6579 83943 : HeapTupleHeader htup = oldtup->t_data;
6580 : uint32 oldlen;
6581 : uint32 newlen;
6582 : char *dst;
6583 : char *src;
6584 83943 : int nmsgs = 0;
6585 83943 : SharedInvalidationMessage *invalMessages = NULL;
6586 83943 : bool RelcacheInitFileInval = false;
6587 :
6588 : Assert(ItemPointerEquals(&oldtup->t_self, &tuple->t_self));
6589 83943 : oldlen = oldtup->t_len - htup->t_hoff;
6590 83943 : newlen = tuple->t_len - tuple->t_data->t_hoff;
6591 83943 : if (oldlen != newlen || htup->t_hoff != tuple->t_data->t_hoff)
6592 0 : elog(ERROR, "wrong tuple length");
6593 :
6594 83943 : dst = (char *) htup + htup->t_hoff;
6595 83943 : src = (char *) tuple->t_data + tuple->t_data->t_hoff;
6596 :
6597 : /* Like RecordTransactionCommit(), log only if needed */
6598 83943 : if (XLogStandbyInfoActive())
6599 58673 : nmsgs = inplaceGetInvalidationMessages(&invalMessages,
6600 : &RelcacheInitFileInval);
6601 :
6602 : /*
6603 : * Unlink relcache init files as needed. If unlinking, acquire
6604 : * RelCacheInitLock until after associated invalidations. By doing this
6605 : * in advance, if we checkpoint and then crash between inplace
6606 : * XLogInsert() and inval, we don't rely on StartupXLOG() ->
6607 : * RelationCacheInitFileRemove(). That uses elevel==LOG, so replay would
6608 : * neglect to PANIC on EIO.
6609 : */
6610 83943 : PreInplace_Inval();
6611 :
6612 : /*----------
6613 : * NO EREPORT(ERROR) from here till changes are complete
6614 : *
6615 : * Our buffer lock won't stop a reader having already pinned and checked
6616 : * visibility for this tuple. Hence, we write WAL first, then mutate the
6617 : * buffer. Like in MarkBufferDirtyHint() or RecordTransactionCommit(),
6618 : * checkpoint delay makes that acceptable. With the usual order of
6619 : * changes, a crash after memcpy() and before XLogInsert() could allow
6620 : * datfrozenxid to overtake relfrozenxid:
6621 : *
6622 : * ["D" is a VACUUM (ONLY_DATABASE_STATS)]
6623 : * ["R" is a VACUUM tbl]
6624 : * D: vac_update_datfrozenxid() -> systable_beginscan(pg_class)
6625 : * D: systable_getnext() returns pg_class tuple of tbl
6626 : * R: memcpy() into pg_class tuple of tbl
6627 : * D: raise pg_database.datfrozenxid, XLogInsert(), finish
6628 : * [crash]
6629 : * [recovery restores datfrozenxid w/o relfrozenxid]
6630 : *
6631 : * Mimic MarkBufferDirtyHint() subroutine XLogSaveBufferForHint().
6632 : * Specifically, use DELAY_CHKPT_START, and copy the buffer to the stack.
6633 : * The stack copy facilitates a FPI of the post-mutation block before we
6634 : * accept other sessions seeing it. DELAY_CHKPT_START allows us to
6635 : * XLogInsert() before MarkBufferDirty(). Since XLogSaveBufferForHint()
6636 : * can operate under BUFFER_LOCK_SHARED, it can't avoid DELAY_CHKPT_START.
6637 : * This function, however, likely could avoid it with the following order
6638 : * of operations: MarkBufferDirty(), XLogInsert(), memcpy(). Opt to use
6639 : * DELAY_CHKPT_START here, too, as a way to have fewer distinct code
6640 : * patterns to analyze. Inplace update isn't so frequent that it should
6641 : * pursue the small optimization of skipping DELAY_CHKPT_START.
6642 : */
6643 : Assert((MyProc->delayChkptFlags & DELAY_CHKPT_START) == 0);
6644 83943 : START_CRIT_SECTION();
6645 83943 : MyProc->delayChkptFlags |= DELAY_CHKPT_START;
6646 :
6647 : /* XLOG stuff */
6648 83943 : if (RelationNeedsWAL(relation))
6649 : {
6650 : xl_heap_inplace xlrec;
6651 : PGAlignedBlock copied_buffer;
6652 83939 : char *origdata = (char *) BufferGetBlock(buffer);
6653 83939 : Page page = BufferGetPage(buffer);
6654 83939 : uint16 lower = ((PageHeader) page)->pd_lower;
6655 83939 : uint16 upper = ((PageHeader) page)->pd_upper;
6656 : uintptr_t dst_offset_in_block;
6657 : RelFileLocator rlocator;
6658 : ForkNumber forkno;
6659 : BlockNumber blkno;
6660 : XLogRecPtr recptr;
6661 :
6662 83939 : xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
6663 83939 : xlrec.dbId = MyDatabaseId;
6664 83939 : xlrec.tsId = MyDatabaseTableSpace;
6665 83939 : xlrec.relcacheInitFileInval = RelcacheInitFileInval;
6666 83939 : xlrec.nmsgs = nmsgs;
6667 :
6668 83939 : XLogBeginInsert();
6669 83939 : XLogRegisterData(&xlrec, MinSizeOfHeapInplace);
6670 83939 : if (nmsgs != 0)
6671 43730 : XLogRegisterData(invalMessages,
6672 : nmsgs * sizeof(SharedInvalidationMessage));
6673 :
6674 : /* register block matching what buffer will look like after changes */
6675 83939 : memcpy(copied_buffer.data, origdata, lower);
6676 83939 : memcpy(copied_buffer.data + upper, origdata + upper, BLCKSZ - upper);
6677 83939 : dst_offset_in_block = dst - origdata;
6678 83939 : memcpy(copied_buffer.data + dst_offset_in_block, src, newlen);
6679 83939 : BufferGetTag(buffer, &rlocator, &forkno, &blkno);
6680 : Assert(forkno == MAIN_FORKNUM);
6681 83939 : XLogRegisterBlock(0, &rlocator, forkno, blkno, copied_buffer.data,
6682 : REGBUF_STANDARD);
6683 83939 : XLogRegisterBufData(0, src, newlen);
6684 :
6685 : /* inplace updates aren't decoded atm, don't log the origin */
6686 :
6687 83939 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_INPLACE);
6688 :
6689 83939 : PageSetLSN(page, recptr);
6690 : }
6691 :
6692 83943 : memcpy(dst, src, newlen);
6693 :
6694 83943 : MarkBufferDirty(buffer);
6695 :
6696 83943 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6697 :
6698 : /*
6699 : * Send invalidations to shared queue. SearchSysCacheLocked1() assumes we
6700 : * do this before UnlockTuple().
6701 : */
6702 83943 : AtInplace_Inval();
6703 :
6704 83943 : MyProc->delayChkptFlags &= ~DELAY_CHKPT_START;
6705 83943 : END_CRIT_SECTION();
6706 83943 : UnlockTuple(relation, &tuple->t_self, InplaceUpdateTupleLock);
6707 :
6708 83943 : AcceptInvalidationMessages(); /* local processing of just-sent inval */
6709 :
6710 : /*
6711 : * Queue a transactional inval, for logical decoding and for third-party
6712 : * code that might have been relying on it since long before inplace
6713 : * update adopted immediate invalidation. See README.tuplock section
6714 : * "Reading inplace-updated columns" for logical decoding details.
6715 : */
6716 83943 : if (!IsBootstrapProcessingMode())
6717 69000 : CacheInvalidateHeapTuple(relation, tuple, NULL);
6718 83943 : }
6719 :
6720 : /*
6721 : * heap_inplace_unlock - reverse of heap_inplace_lock
6722 : */
6723 : void
6724 117860 : heap_inplace_unlock(Relation relation,
6725 : HeapTuple oldtup, Buffer buffer)
6726 : {
6727 117860 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6728 117860 : UnlockTuple(relation, &oldtup->t_self, InplaceUpdateTupleLock);
6729 117860 : ForgetInplace_Inval();
6730 117860 : }
6731 :
6732 : #define FRM_NOOP 0x0001
6733 : #define FRM_INVALIDATE_XMAX 0x0002
6734 : #define FRM_RETURN_IS_XID 0x0004
6735 : #define FRM_RETURN_IS_MULTI 0x0008
6736 : #define FRM_MARK_COMMITTED 0x0010
6737 :
6738 : /*
6739 : * FreezeMultiXactId
6740 : * Determine what to do during freezing when a tuple is marked by a
6741 : * MultiXactId.
6742 : *
6743 : * "flags" is an output value; it's used to tell caller what to do on return.
6744 : * "pagefrz" is an input/output value, used to manage page level freezing.
6745 : *
6746 : * Possible values that we can set in "flags":
6747 : * FRM_NOOP
6748 : * don't do anything -- keep existing Xmax
6749 : * FRM_INVALIDATE_XMAX
6750 : * mark Xmax as InvalidTransactionId and set XMAX_INVALID flag.
6751 : * FRM_RETURN_IS_XID
6752 : * The Xid return value is a single update Xid to set as xmax.
6753 : * FRM_MARK_COMMITTED
6754 : * Xmax can be marked as HEAP_XMAX_COMMITTED
6755 : * FRM_RETURN_IS_MULTI
6756 : * The return value is a new MultiXactId to set as new Xmax.
6757 : * (caller must obtain proper infomask bits using GetMultiXactIdHintBits)
6758 : *
6759 : * Caller delegates control of page freezing to us. In practice we always
6760 : * force freezing of caller's page unless FRM_NOOP processing is indicated.
6761 : * We help caller ensure that XIDs < FreezeLimit and MXIDs < MultiXactCutoff
6762 : * can never be left behind. We freely choose when and how to process each
6763 : * Multi, without ever violating the cutoff postconditions for freezing.
6764 : *
6765 : * It's useful to remove Multis on a proactive timeline (relative to freezing
6766 : * XIDs) to keep MultiXact member SLRU buffer misses to a minimum. It can also
6767 : * be cheaper in the short run, for us, since we too can avoid SLRU buffer
6768 : * misses through eager processing.
6769 : *
6770 : * NB: Creates a _new_ MultiXactId when FRM_RETURN_IS_MULTI is set, though only
6771 : * when FreezeLimit and/or MultiXactCutoff cutoffs leave us with no choice.
6772 : * This can usually be put off, which is usually enough to avoid it altogether.
6773 : * Allocating new multis during VACUUM should be avoided on general principle;
6774 : * only VACUUM can advance relminmxid, so allocating new Multis here comes with
6775 : * its own special risks.
6776 : *
6777 : * NB: Caller must maintain "no freeze" NewRelfrozenXid/NewRelminMxid trackers
6778 : * using heap_tuple_should_freeze when we haven't forced page-level freezing.
6779 : *
6780 : * NB: Caller should avoid needlessly calling heap_tuple_should_freeze when we
6781 : * have already forced page-level freezing, since that might incur the same
6782 : * SLRU buffer misses that we specifically intended to avoid by freezing.
6783 : */
6784 : static TransactionId
6785 6 : FreezeMultiXactId(MultiXactId multi, uint16 t_infomask,
6786 : const struct VacuumCutoffs *cutoffs, uint16 *flags,
6787 : HeapPageFreeze *pagefrz)
6788 : {
6789 : TransactionId newxmax;
6790 : MultiXactMember *members;
6791 : int nmembers;
6792 : bool need_replace;
6793 : int nnewmembers;
6794 : MultiXactMember *newmembers;
6795 : bool has_lockers;
6796 : TransactionId update_xid;
6797 : bool update_committed;
6798 : TransactionId FreezePageRelfrozenXid;
6799 :
6800 6 : *flags = 0;
6801 :
6802 : /* We should only be called in Multis */
6803 : Assert(t_infomask & HEAP_XMAX_IS_MULTI);
6804 :
6805 12 : if (!MultiXactIdIsValid(multi) ||
6806 6 : HEAP_LOCKED_UPGRADED(t_infomask))
6807 : {
6808 0 : *flags |= FRM_INVALIDATE_XMAX;
6809 0 : pagefrz->freeze_required = true;
6810 0 : return InvalidTransactionId;
6811 : }
6812 6 : else if (MultiXactIdPrecedes(multi, cutoffs->relminmxid))
6813 0 : ereport(ERROR,
6814 : (errcode(ERRCODE_DATA_CORRUPTED),
6815 : errmsg_internal("found multixact %u from before relminmxid %u",
6816 : multi, cutoffs->relminmxid)));
6817 6 : else if (MultiXactIdPrecedes(multi, cutoffs->OldestMxact))
6818 : {
6819 : TransactionId update_xact;
6820 :
6821 : /*
6822 : * This old multi cannot possibly have members still running, but
6823 : * verify just in case. If it was a locker only, it can be removed
6824 : * without any further consideration; but if it contained an update,
6825 : * we might need to preserve it.
6826 : */
6827 4 : if (MultiXactIdIsRunning(multi,
6828 4 : HEAP_XMAX_IS_LOCKED_ONLY(t_infomask)))
6829 0 : ereport(ERROR,
6830 : (errcode(ERRCODE_DATA_CORRUPTED),
6831 : errmsg_internal("multixact %u from before multi freeze cutoff %u found to be still running",
6832 : multi, cutoffs->OldestMxact)));
6833 :
6834 4 : if (HEAP_XMAX_IS_LOCKED_ONLY(t_infomask))
6835 : {
6836 4 : *flags |= FRM_INVALIDATE_XMAX;
6837 4 : pagefrz->freeze_required = true;
6838 4 : return InvalidTransactionId;
6839 : }
6840 :
6841 : /* replace multi with single XID for its updater? */
6842 0 : update_xact = MultiXactIdGetUpdateXid(multi, t_infomask);
6843 0 : if (TransactionIdPrecedes(update_xact, cutoffs->relfrozenxid))
6844 0 : ereport(ERROR,
6845 : (errcode(ERRCODE_DATA_CORRUPTED),
6846 : errmsg_internal("multixact %u contains update XID %u from before relfrozenxid %u",
6847 : multi, update_xact,
6848 : cutoffs->relfrozenxid)));
6849 0 : else if (TransactionIdPrecedes(update_xact, cutoffs->OldestXmin))
6850 : {
6851 : /*
6852 : * Updater XID has to have aborted (otherwise the tuple would have
6853 : * been pruned away instead, since updater XID is < OldestXmin).
6854 : * Just remove xmax.
6855 : */
6856 0 : if (TransactionIdDidCommit(update_xact))
6857 0 : ereport(ERROR,
6858 : (errcode(ERRCODE_DATA_CORRUPTED),
6859 : errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
6860 : multi, update_xact,
6861 : cutoffs->OldestXmin)));
6862 0 : *flags |= FRM_INVALIDATE_XMAX;
6863 0 : pagefrz->freeze_required = true;
6864 0 : return InvalidTransactionId;
6865 : }
6866 :
6867 : /* Have to keep updater XID as new xmax */
6868 0 : *flags |= FRM_RETURN_IS_XID;
6869 0 : pagefrz->freeze_required = true;
6870 0 : return update_xact;
6871 : }
6872 :
6873 : /*
6874 : * Some member(s) of this Multi may be below FreezeLimit xid cutoff, so we
6875 : * need to walk the whole members array to figure out what to do, if
6876 : * anything.
6877 : */
6878 : nmembers =
6879 2 : GetMultiXactIdMembers(multi, &members, false,
6880 2 : HEAP_XMAX_IS_LOCKED_ONLY(t_infomask));
6881 2 : if (nmembers <= 0)
6882 : {
6883 : /* Nothing worth keeping */
6884 0 : *flags |= FRM_INVALIDATE_XMAX;
6885 0 : pagefrz->freeze_required = true;
6886 0 : return InvalidTransactionId;
6887 : }
6888 :
6889 : /*
6890 : * The FRM_NOOP case is the only case where we might need to ratchet back
6891 : * FreezePageRelfrozenXid or FreezePageRelminMxid. It is also the only
6892 : * case where our caller might ratchet back its NoFreezePageRelfrozenXid
6893 : * or NoFreezePageRelminMxid "no freeze" trackers to deal with a multi.
6894 : * FRM_NOOP handling should result in the NewRelfrozenXid/NewRelminMxid
6895 : * trackers managed by VACUUM being ratcheting back by xmax to the degree
6896 : * required to make it safe to leave xmax undisturbed, independent of
6897 : * whether or not page freezing is triggered somewhere else.
6898 : *
6899 : * Our policy is to force freezing in every case other than FRM_NOOP,
6900 : * which obviates the need to maintain either set of trackers, anywhere.
6901 : * Every other case will reliably execute a freeze plan for xmax that
6902 : * either replaces xmax with an XID/MXID >= OldestXmin/OldestMxact, or
6903 : * sets xmax to an InvalidTransactionId XID, rendering xmax fully frozen.
6904 : * (VACUUM's NewRelfrozenXid/NewRelminMxid trackers are initialized with
6905 : * OldestXmin/OldestMxact, so later values never need to be tracked here.)
6906 : */
6907 2 : need_replace = false;
6908 2 : FreezePageRelfrozenXid = pagefrz->FreezePageRelfrozenXid;
6909 4 : for (int i = 0; i < nmembers; i++)
6910 : {
6911 3 : TransactionId xid = members[i].xid;
6912 :
6913 : Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
6914 :
6915 3 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
6916 : {
6917 : /* Can't violate the FreezeLimit postcondition */
6918 1 : need_replace = true;
6919 1 : break;
6920 : }
6921 2 : if (TransactionIdPrecedes(xid, FreezePageRelfrozenXid))
6922 0 : FreezePageRelfrozenXid = xid;
6923 : }
6924 :
6925 : /* Can't violate the MultiXactCutoff postcondition, either */
6926 2 : if (!need_replace)
6927 1 : need_replace = MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff);
6928 :
6929 2 : if (!need_replace)
6930 : {
6931 : /*
6932 : * vacuumlazy.c might ratchet back NewRelminMxid, NewRelfrozenXid, or
6933 : * both together to make it safe to retain this particular multi after
6934 : * freezing its page
6935 : */
6936 1 : *flags |= FRM_NOOP;
6937 1 : pagefrz->FreezePageRelfrozenXid = FreezePageRelfrozenXid;
6938 1 : if (MultiXactIdPrecedes(multi, pagefrz->FreezePageRelminMxid))
6939 0 : pagefrz->FreezePageRelminMxid = multi;
6940 1 : pfree(members);
6941 1 : return multi;
6942 : }
6943 :
6944 : /*
6945 : * Do a more thorough second pass over the multi to figure out which
6946 : * member XIDs actually need to be kept. Checking the precise status of
6947 : * individual members might even show that we don't need to keep anything.
6948 : * That is quite possible even though the Multi must be >= OldestMxact,
6949 : * since our second pass only keeps member XIDs when it's truly necessary;
6950 : * even member XIDs >= OldestXmin often won't be kept by second pass.
6951 : */
6952 1 : nnewmembers = 0;
6953 1 : newmembers = palloc_array(MultiXactMember, nmembers);
6954 1 : has_lockers = false;
6955 1 : update_xid = InvalidTransactionId;
6956 1 : update_committed = false;
6957 :
6958 : /*
6959 : * Determine whether to keep each member xid, or to ignore it instead
6960 : */
6961 3 : for (int i = 0; i < nmembers; i++)
6962 : {
6963 2 : TransactionId xid = members[i].xid;
6964 2 : MultiXactStatus mstatus = members[i].status;
6965 :
6966 : Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
6967 :
6968 2 : if (!ISUPDATE_from_mxstatus(mstatus))
6969 : {
6970 : /*
6971 : * Locker XID (not updater XID). We only keep lockers that are
6972 : * still running.
6973 : */
6974 4 : if (TransactionIdIsCurrentTransactionId(xid) ||
6975 2 : TransactionIdIsInProgress(xid))
6976 : {
6977 1 : if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
6978 0 : ereport(ERROR,
6979 : (errcode(ERRCODE_DATA_CORRUPTED),
6980 : errmsg_internal("multixact %u contains running locker XID %u from before removable cutoff %u",
6981 : multi, xid,
6982 : cutoffs->OldestXmin)));
6983 1 : newmembers[nnewmembers++] = members[i];
6984 1 : has_lockers = true;
6985 : }
6986 :
6987 2 : continue;
6988 : }
6989 :
6990 : /*
6991 : * Updater XID (not locker XID). Should we keep it?
6992 : *
6993 : * Since the tuple wasn't totally removed when vacuum pruned, the
6994 : * update Xid cannot possibly be older than OldestXmin cutoff unless
6995 : * the updater XID aborted. If the updater transaction is known
6996 : * aborted or crashed then it's okay to ignore it, otherwise not.
6997 : *
6998 : * In any case the Multi should never contain two updaters, whatever
6999 : * their individual commit status. Check for that first, in passing.
7000 : */
7001 0 : if (TransactionIdIsValid(update_xid))
7002 0 : ereport(ERROR,
7003 : (errcode(ERRCODE_DATA_CORRUPTED),
7004 : errmsg_internal("multixact %u has two or more updating members",
7005 : multi),
7006 : errdetail_internal("First updater XID=%u second updater XID=%u.",
7007 : update_xid, xid)));
7008 :
7009 : /*
7010 : * As with all tuple visibility routines, it's critical to test
7011 : * TransactionIdIsInProgress before TransactionIdDidCommit, because of
7012 : * race conditions explained in detail in heapam_visibility.c.
7013 : */
7014 0 : if (TransactionIdIsCurrentTransactionId(xid) ||
7015 0 : TransactionIdIsInProgress(xid))
7016 0 : update_xid = xid;
7017 0 : else if (TransactionIdDidCommit(xid))
7018 : {
7019 : /*
7020 : * The transaction committed, so we can tell caller to set
7021 : * HEAP_XMAX_COMMITTED. (We can only do this because we know the
7022 : * transaction is not running.)
7023 : */
7024 0 : update_committed = true;
7025 0 : update_xid = xid;
7026 : }
7027 : else
7028 : {
7029 : /*
7030 : * Not in progress, not committed -- must be aborted or crashed;
7031 : * we can ignore it.
7032 : */
7033 0 : continue;
7034 : }
7035 :
7036 : /*
7037 : * We determined that updater must be kept -- add it to pending new
7038 : * members list
7039 : */
7040 0 : if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
7041 0 : ereport(ERROR,
7042 : (errcode(ERRCODE_DATA_CORRUPTED),
7043 : errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
7044 : multi, xid, cutoffs->OldestXmin)));
7045 0 : newmembers[nnewmembers++] = members[i];
7046 : }
7047 :
7048 1 : pfree(members);
7049 :
7050 : /*
7051 : * Determine what to do with caller's multi based on information gathered
7052 : * during our second pass
7053 : */
7054 1 : if (nnewmembers == 0)
7055 : {
7056 : /* Nothing worth keeping */
7057 0 : *flags |= FRM_INVALIDATE_XMAX;
7058 0 : newxmax = InvalidTransactionId;
7059 : }
7060 1 : else if (TransactionIdIsValid(update_xid) && !has_lockers)
7061 : {
7062 : /*
7063 : * If there's a single member and it's an update, pass it back alone
7064 : * without creating a new Multi. (XXX we could do this when there's a
7065 : * single remaining locker, too, but that would complicate the API too
7066 : * much; moreover, the case with the single updater is more
7067 : * interesting, because those are longer-lived.)
7068 : */
7069 : Assert(nnewmembers == 1);
7070 0 : *flags |= FRM_RETURN_IS_XID;
7071 0 : if (update_committed)
7072 0 : *flags |= FRM_MARK_COMMITTED;
7073 0 : newxmax = update_xid;
7074 : }
7075 : else
7076 : {
7077 : /*
7078 : * Create a new multixact with the surviving members of the previous
7079 : * one, to set as new Xmax in the tuple
7080 : */
7081 1 : newxmax = MultiXactIdCreateFromMembers(nnewmembers, newmembers);
7082 1 : *flags |= FRM_RETURN_IS_MULTI;
7083 : }
7084 :
7085 1 : pfree(newmembers);
7086 :
7087 1 : pagefrz->freeze_required = true;
7088 1 : return newxmax;
7089 : }
7090 :
7091 : /*
7092 : * heap_prepare_freeze_tuple
7093 : *
7094 : * Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
7095 : * are older than the OldestXmin and/or OldestMxact freeze cutoffs. If so,
7096 : * setup enough state (in the *frz output argument) to enable caller to
7097 : * process this tuple as part of freezing its page, and return true. Return
7098 : * false if nothing can be changed about the tuple right now.
7099 : *
7100 : * Also sets *totally_frozen to true if the tuple will be totally frozen once
7101 : * caller executes returned freeze plan (or if the tuple was already totally
7102 : * frozen by an earlier VACUUM). This indicates that there are no remaining
7103 : * XIDs or MultiXactIds that will need to be processed by a future VACUUM.
7104 : *
7105 : * VACUUM caller must assemble HeapTupleFreeze freeze plan entries for every
7106 : * tuple that we returned true for, and then execute freezing. Caller must
7107 : * initialize pagefrz fields for page as a whole before first call here for
7108 : * each heap page.
7109 : *
7110 : * VACUUM caller decides on whether or not to freeze the page as a whole.
7111 : * We'll often prepare freeze plans for a page that caller just discards.
7112 : * However, VACUUM doesn't always get to make a choice; it must freeze when
7113 : * pagefrz.freeze_required is set, to ensure that any XIDs < FreezeLimit (and
7114 : * MXIDs < MultiXactCutoff) can never be left behind. We help to make sure
7115 : * that VACUUM always follows that rule.
7116 : *
7117 : * We sometimes force freezing of xmax MultiXactId values long before it is
7118 : * strictly necessary to do so just to ensure the FreezeLimit postcondition.
7119 : * It's worth processing MultiXactIds proactively when it is cheap to do so,
7120 : * and it's convenient to make that happen by piggy-backing it on the "force
7121 : * freezing" mechanism. Conversely, we sometimes delay freezing MultiXactIds
7122 : * because it is expensive right now (though only when it's still possible to
7123 : * do so without violating the FreezeLimit/MultiXactCutoff postcondition).
7124 : *
7125 : * It is assumed that the caller has checked the tuple with
7126 : * HeapTupleSatisfiesVacuum() and determined that it is not HEAPTUPLE_DEAD
7127 : * (else we should be removing the tuple, not freezing it).
7128 : *
7129 : * NB: This function has side effects: it might allocate a new MultiXactId.
7130 : * It will be set as tuple's new xmax when our *frz output is processed within
7131 : * heap_execute_freeze_tuple later on. If the tuple is in a shared buffer
7132 : * then caller had better have an exclusive lock on it already.
7133 : */
7134 : bool
7135 23037084 : heap_prepare_freeze_tuple(HeapTupleHeader tuple,
7136 : const struct VacuumCutoffs *cutoffs,
7137 : HeapPageFreeze *pagefrz,
7138 : HeapTupleFreeze *frz, bool *totally_frozen)
7139 : {
7140 23037084 : bool xmin_already_frozen = false,
7141 23037084 : xmax_already_frozen = false;
7142 23037084 : bool freeze_xmin = false,
7143 23037084 : replace_xvac = false,
7144 23037084 : replace_xmax = false,
7145 23037084 : freeze_xmax = false;
7146 : TransactionId xid;
7147 :
7148 23037084 : frz->xmax = HeapTupleHeaderGetRawXmax(tuple);
7149 23037084 : frz->t_infomask2 = tuple->t_infomask2;
7150 23037084 : frz->t_infomask = tuple->t_infomask;
7151 23037084 : frz->frzflags = 0;
7152 23037084 : frz->checkflags = 0;
7153 :
7154 : /*
7155 : * Process xmin, while keeping track of whether it's already frozen, or
7156 : * will become frozen iff our freeze plan is executed by caller (could be
7157 : * neither).
7158 : */
7159 23037084 : xid = HeapTupleHeaderGetXmin(tuple);
7160 23037084 : if (!TransactionIdIsNormal(xid))
7161 19909024 : xmin_already_frozen = true;
7162 : else
7163 : {
7164 3128060 : if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
7165 0 : ereport(ERROR,
7166 : (errcode(ERRCODE_DATA_CORRUPTED),
7167 : errmsg_internal("found xmin %u from before relfrozenxid %u",
7168 : xid, cutoffs->relfrozenxid)));
7169 :
7170 : /* Will set freeze_xmin flags in freeze plan below */
7171 3128060 : freeze_xmin = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
7172 :
7173 : /* Verify that xmin committed if and when freeze plan is executed */
7174 3128060 : if (freeze_xmin)
7175 2482137 : frz->checkflags |= HEAP_FREEZE_CHECK_XMIN_COMMITTED;
7176 : }
7177 :
7178 : /*
7179 : * Old-style VACUUM FULL is gone, but we have to process xvac for as long
7180 : * as we support having MOVED_OFF/MOVED_IN tuples in the database
7181 : */
7182 23037084 : xid = HeapTupleHeaderGetXvac(tuple);
7183 23037084 : if (TransactionIdIsNormal(xid))
7184 : {
7185 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7186 : Assert(TransactionIdPrecedes(xid, cutoffs->OldestXmin));
7187 :
7188 : /*
7189 : * For Xvac, we always freeze proactively. This allows totally_frozen
7190 : * tracking to ignore xvac.
7191 : */
7192 0 : replace_xvac = pagefrz->freeze_required = true;
7193 :
7194 : /* Will set replace_xvac flags in freeze plan below */
7195 : }
7196 :
7197 : /* Now process xmax */
7198 23037084 : xid = frz->xmax;
7199 23037084 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7200 : {
7201 : /* Raw xmax is a MultiXactId */
7202 : TransactionId newxmax;
7203 : uint16 flags;
7204 :
7205 : /*
7206 : * We will either remove xmax completely (in the "freeze_xmax" path),
7207 : * process xmax by replacing it (in the "replace_xmax" path), or
7208 : * perform no-op xmax processing. The only constraint is that the
7209 : * FreezeLimit/MultiXactCutoff postcondition must never be violated.
7210 : */
7211 6 : newxmax = FreezeMultiXactId(xid, tuple->t_infomask, cutoffs,
7212 : &flags, pagefrz);
7213 :
7214 6 : if (flags & FRM_NOOP)
7215 : {
7216 : /*
7217 : * xmax is a MultiXactId, and nothing about it changes for now.
7218 : * This is the only case where 'freeze_required' won't have been
7219 : * set for us by FreezeMultiXactId, as well as the only case where
7220 : * neither freeze_xmax nor replace_xmax are set (given a multi).
7221 : *
7222 : * This is a no-op, but the call to FreezeMultiXactId might have
7223 : * ratcheted back NewRelfrozenXid and/or NewRelminMxid trackers
7224 : * for us (the "freeze page" variants, specifically). That'll
7225 : * make it safe for our caller to freeze the page later on, while
7226 : * leaving this particular xmax undisturbed.
7227 : *
7228 : * FreezeMultiXactId is _not_ responsible for the "no freeze"
7229 : * NewRelfrozenXid/NewRelminMxid trackers, though -- that's our
7230 : * job. A call to heap_tuple_should_freeze for this same tuple
7231 : * will take place below if 'freeze_required' isn't set already.
7232 : * (This repeats work from FreezeMultiXactId, but allows "no
7233 : * freeze" tracker maintenance to happen in only one place.)
7234 : */
7235 : Assert(!MultiXactIdPrecedes(newxmax, cutoffs->MultiXactCutoff));
7236 : Assert(MultiXactIdIsValid(newxmax) && xid == newxmax);
7237 : }
7238 5 : else if (flags & FRM_RETURN_IS_XID)
7239 : {
7240 : /*
7241 : * xmax will become an updater Xid (original MultiXact's updater
7242 : * member Xid will be carried forward as a simple Xid in Xmax).
7243 : */
7244 : Assert(!TransactionIdPrecedes(newxmax, cutoffs->OldestXmin));
7245 :
7246 : /*
7247 : * NB -- some of these transformations are only valid because we
7248 : * know the return Xid is a tuple updater (i.e. not merely a
7249 : * locker.) Also note that the only reason we don't explicitly
7250 : * worry about HEAP_KEYS_UPDATED is because it lives in
7251 : * t_infomask2 rather than t_infomask.
7252 : */
7253 0 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7254 0 : frz->xmax = newxmax;
7255 0 : if (flags & FRM_MARK_COMMITTED)
7256 0 : frz->t_infomask |= HEAP_XMAX_COMMITTED;
7257 0 : replace_xmax = true;
7258 : }
7259 5 : else if (flags & FRM_RETURN_IS_MULTI)
7260 : {
7261 : uint16 newbits;
7262 : uint16 newbits2;
7263 :
7264 : /*
7265 : * xmax is an old MultiXactId that we have to replace with a new
7266 : * MultiXactId, to carry forward two or more original member XIDs.
7267 : */
7268 : Assert(!MultiXactIdPrecedes(newxmax, cutoffs->OldestMxact));
7269 :
7270 : /*
7271 : * We can't use GetMultiXactIdHintBits directly on the new multi
7272 : * here; that routine initializes the masks to all zeroes, which
7273 : * would lose other bits we need. Doing it this way ensures all
7274 : * unrelated bits remain untouched.
7275 : */
7276 1 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7277 1 : frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
7278 1 : GetMultiXactIdHintBits(newxmax, &newbits, &newbits2);
7279 1 : frz->t_infomask |= newbits;
7280 1 : frz->t_infomask2 |= newbits2;
7281 1 : frz->xmax = newxmax;
7282 1 : replace_xmax = true;
7283 : }
7284 : else
7285 : {
7286 : /*
7287 : * Freeze plan for tuple "freezes xmax" in the strictest sense:
7288 : * it'll leave nothing in xmax (neither an Xid nor a MultiXactId).
7289 : */
7290 : Assert(flags & FRM_INVALIDATE_XMAX);
7291 : Assert(!TransactionIdIsValid(newxmax));
7292 :
7293 : /* Will set freeze_xmax flags in freeze plan below */
7294 4 : freeze_xmax = true;
7295 : }
7296 :
7297 : /* MultiXactId processing forces freezing (barring FRM_NOOP case) */
7298 : Assert(pagefrz->freeze_required || (!freeze_xmax && !replace_xmax));
7299 : }
7300 23037078 : else if (TransactionIdIsNormal(xid))
7301 : {
7302 : /* Raw xmax is normal XID */
7303 6692685 : if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
7304 0 : ereport(ERROR,
7305 : (errcode(ERRCODE_DATA_CORRUPTED),
7306 : errmsg_internal("found xmax %u from before relfrozenxid %u",
7307 : xid, cutoffs->relfrozenxid)));
7308 :
7309 : /* Will set freeze_xmax flags in freeze plan below */
7310 6692685 : freeze_xmax = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
7311 :
7312 : /*
7313 : * Verify that xmax aborted if and when freeze plan is executed,
7314 : * provided it's from an update. (A lock-only xmax can be removed
7315 : * independent of this, since the lock is released at xact end.)
7316 : */
7317 6692685 : if (freeze_xmax && !HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask))
7318 1379 : frz->checkflags |= HEAP_FREEZE_CHECK_XMAX_ABORTED;
7319 : }
7320 16344393 : else if (!TransactionIdIsValid(xid))
7321 : {
7322 : /* Raw xmax is InvalidTransactionId XID */
7323 : Assert((tuple->t_infomask & HEAP_XMAX_IS_MULTI) == 0);
7324 16344393 : xmax_already_frozen = true;
7325 : }
7326 : else
7327 0 : ereport(ERROR,
7328 : (errcode(ERRCODE_DATA_CORRUPTED),
7329 : errmsg_internal("found raw xmax %u (infomask 0x%04x) not invalid and not multi",
7330 : xid, tuple->t_infomask)));
7331 :
7332 23037084 : if (freeze_xmin)
7333 : {
7334 : Assert(!xmin_already_frozen);
7335 :
7336 2482137 : frz->t_infomask |= HEAP_XMIN_FROZEN;
7337 : }
7338 23037084 : if (replace_xvac)
7339 : {
7340 : /*
7341 : * If a MOVED_OFF tuple is not dead, the xvac transaction must have
7342 : * failed; whereas a non-dead MOVED_IN tuple must mean the xvac
7343 : * transaction succeeded.
7344 : */
7345 : Assert(pagefrz->freeze_required);
7346 0 : if (tuple->t_infomask & HEAP_MOVED_OFF)
7347 0 : frz->frzflags |= XLH_INVALID_XVAC;
7348 : else
7349 0 : frz->frzflags |= XLH_FREEZE_XVAC;
7350 : }
7351 : if (replace_xmax)
7352 : {
7353 : Assert(!xmax_already_frozen && !freeze_xmax);
7354 : Assert(pagefrz->freeze_required);
7355 :
7356 : /* Already set replace_xmax flags in freeze plan earlier */
7357 : }
7358 23037084 : if (freeze_xmax)
7359 : {
7360 : Assert(!xmax_already_frozen && !replace_xmax);
7361 :
7362 2355 : frz->xmax = InvalidTransactionId;
7363 :
7364 : /*
7365 : * The tuple might be marked either XMAX_INVALID or XMAX_COMMITTED +
7366 : * LOCKED. Normalize to INVALID just to be sure no one gets confused.
7367 : * Also get rid of the HEAP_KEYS_UPDATED bit.
7368 : */
7369 2355 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7370 2355 : frz->t_infomask |= HEAP_XMAX_INVALID;
7371 2355 : frz->t_infomask2 &= ~HEAP_HOT_UPDATED;
7372 2355 : frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
7373 : }
7374 :
7375 : /*
7376 : * Determine if this tuple is already totally frozen, or will become
7377 : * totally frozen (provided caller executes freeze plans for the page)
7378 : */
7379 45425890 : *totally_frozen = ((freeze_xmin || xmin_already_frozen) &&
7380 22388806 : (freeze_xmax || xmax_already_frozen));
7381 :
7382 23037084 : if (!pagefrz->freeze_required && !(xmin_already_frozen &&
7383 : xmax_already_frozen))
7384 : {
7385 : /*
7386 : * So far no previous tuple from the page made freezing mandatory.
7387 : * Does this tuple force caller to freeze the entire page?
7388 : */
7389 8263077 : pagefrz->freeze_required =
7390 8263077 : heap_tuple_should_freeze(tuple, cutoffs,
7391 : &pagefrz->NoFreezePageRelfrozenXid,
7392 : &pagefrz->NoFreezePageRelminMxid);
7393 : }
7394 :
7395 : /* Tell caller if this tuple has a usable freeze plan set in *frz */
7396 23037084 : return freeze_xmin || replace_xvac || replace_xmax || freeze_xmax;
7397 : }
7398 :
7399 : /*
7400 : * Perform xmin/xmax XID status sanity checks before actually executing freeze
7401 : * plans.
7402 : *
7403 : * heap_prepare_freeze_tuple doesn't perform these checks directly because
7404 : * pg_xact lookups are relatively expensive. They shouldn't be repeated by
7405 : * successive VACUUMs that each decide against freezing the same page.
7406 : */
7407 : void
7408 22270 : heap_pre_freeze_checks(Buffer buffer,
7409 : HeapTupleFreeze *tuples, int ntuples)
7410 : {
7411 22270 : Page page = BufferGetPage(buffer);
7412 :
7413 971660 : for (int i = 0; i < ntuples; i++)
7414 : {
7415 949390 : HeapTupleFreeze *frz = tuples + i;
7416 949390 : ItemId itemid = PageGetItemId(page, frz->offset);
7417 : HeapTupleHeader htup;
7418 :
7419 949390 : htup = (HeapTupleHeader) PageGetItem(page, itemid);
7420 :
7421 : /* Deliberately avoid relying on tuple hint bits here */
7422 949390 : if (frz->checkflags & HEAP_FREEZE_CHECK_XMIN_COMMITTED)
7423 : {
7424 949389 : TransactionId xmin = HeapTupleHeaderGetRawXmin(htup);
7425 :
7426 : Assert(!HeapTupleHeaderXminFrozen(htup));
7427 949389 : if (unlikely(!TransactionIdDidCommit(xmin)))
7428 0 : ereport(ERROR,
7429 : (errcode(ERRCODE_DATA_CORRUPTED),
7430 : errmsg_internal("uncommitted xmin %u needs to be frozen",
7431 : xmin)));
7432 : }
7433 :
7434 : /*
7435 : * TransactionIdDidAbort won't work reliably in the presence of XIDs
7436 : * left behind by transactions that were in progress during a crash,
7437 : * so we can only check that xmax didn't commit
7438 : */
7439 949390 : if (frz->checkflags & HEAP_FREEZE_CHECK_XMAX_ABORTED)
7440 : {
7441 430 : TransactionId xmax = HeapTupleHeaderGetRawXmax(htup);
7442 :
7443 : Assert(TransactionIdIsNormal(xmax));
7444 430 : if (unlikely(TransactionIdDidCommit(xmax)))
7445 0 : ereport(ERROR,
7446 : (errcode(ERRCODE_DATA_CORRUPTED),
7447 : errmsg_internal("cannot freeze committed xmax %u",
7448 : xmax)));
7449 : }
7450 : }
7451 22270 : }
7452 :
7453 : /*
7454 : * Helper which executes freezing of one or more heap tuples on a page on
7455 : * behalf of caller. Caller passes an array of tuple plans from
7456 : * heap_prepare_freeze_tuple. Caller must set 'offset' in each plan for us.
7457 : * Must be called in a critical section that also marks the buffer dirty and,
7458 : * if needed, emits WAL.
7459 : */
7460 : void
7461 22270 : heap_freeze_prepared_tuples(Buffer buffer, HeapTupleFreeze *tuples, int ntuples)
7462 : {
7463 22270 : Page page = BufferGetPage(buffer);
7464 :
7465 971660 : for (int i = 0; i < ntuples; i++)
7466 : {
7467 949390 : HeapTupleFreeze *frz = tuples + i;
7468 949390 : ItemId itemid = PageGetItemId(page, frz->offset);
7469 : HeapTupleHeader htup;
7470 :
7471 949390 : htup = (HeapTupleHeader) PageGetItem(page, itemid);
7472 949390 : heap_execute_freeze_tuple(htup, frz);
7473 : }
7474 22270 : }
7475 :
7476 : /*
7477 : * heap_freeze_tuple
7478 : * Freeze tuple in place, without WAL logging.
7479 : *
7480 : * Useful for callers like CLUSTER that perform their own WAL logging.
7481 : */
7482 : bool
7483 362454 : heap_freeze_tuple(HeapTupleHeader tuple,
7484 : TransactionId relfrozenxid, TransactionId relminmxid,
7485 : TransactionId FreezeLimit, TransactionId MultiXactCutoff)
7486 : {
7487 : HeapTupleFreeze frz;
7488 : bool do_freeze;
7489 : bool totally_frozen;
7490 : struct VacuumCutoffs cutoffs;
7491 : HeapPageFreeze pagefrz;
7492 :
7493 362454 : cutoffs.relfrozenxid = relfrozenxid;
7494 362454 : cutoffs.relminmxid = relminmxid;
7495 362454 : cutoffs.OldestXmin = FreezeLimit;
7496 362454 : cutoffs.OldestMxact = MultiXactCutoff;
7497 362454 : cutoffs.FreezeLimit = FreezeLimit;
7498 362454 : cutoffs.MultiXactCutoff = MultiXactCutoff;
7499 :
7500 362454 : pagefrz.freeze_required = true;
7501 362454 : pagefrz.FreezePageRelfrozenXid = FreezeLimit;
7502 362454 : pagefrz.FreezePageRelminMxid = MultiXactCutoff;
7503 362454 : pagefrz.NoFreezePageRelfrozenXid = FreezeLimit;
7504 362454 : pagefrz.NoFreezePageRelminMxid = MultiXactCutoff;
7505 :
7506 362454 : do_freeze = heap_prepare_freeze_tuple(tuple, &cutoffs,
7507 : &pagefrz, &frz, &totally_frozen);
7508 :
7509 : /*
7510 : * Note that because this is not a WAL-logged operation, we don't need to
7511 : * fill in the offset in the freeze record.
7512 : */
7513 :
7514 362454 : if (do_freeze)
7515 251986 : heap_execute_freeze_tuple(tuple, &frz);
7516 362454 : return do_freeze;
7517 : }
7518 :
7519 : /*
7520 : * For a given MultiXactId, return the hint bits that should be set in the
7521 : * tuple's infomask.
7522 : *
7523 : * Normally this should be called for a multixact that was just created, and
7524 : * so is on our local cache, so the GetMembers call is fast.
7525 : */
7526 : static void
7527 76801 : GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
7528 : uint16 *new_infomask2)
7529 : {
7530 : int nmembers;
7531 : MultiXactMember *members;
7532 : int i;
7533 76801 : uint16 bits = HEAP_XMAX_IS_MULTI;
7534 76801 : uint16 bits2 = 0;
7535 76801 : bool has_update = false;
7536 76801 : LockTupleMode strongest = LockTupleKeyShare;
7537 :
7538 : /*
7539 : * We only use this in multis we just created, so they cannot be values
7540 : * pre-pg_upgrade.
7541 : */
7542 76801 : nmembers = GetMultiXactIdMembers(multi, &members, false, false);
7543 :
7544 1472618 : for (i = 0; i < nmembers; i++)
7545 : {
7546 : LockTupleMode mode;
7547 :
7548 : /*
7549 : * Remember the strongest lock mode held by any member of the
7550 : * multixact.
7551 : */
7552 1395817 : mode = TUPLOCK_from_mxstatus(members[i].status);
7553 1395817 : if (mode > strongest)
7554 2888 : strongest = mode;
7555 :
7556 : /* See what other bits we need */
7557 1395817 : switch (members[i].status)
7558 : {
7559 1393399 : case MultiXactStatusForKeyShare:
7560 : case MultiXactStatusForShare:
7561 : case MultiXactStatusForNoKeyUpdate:
7562 1393399 : break;
7563 :
7564 52 : case MultiXactStatusForUpdate:
7565 52 : bits2 |= HEAP_KEYS_UPDATED;
7566 52 : break;
7567 :
7568 2356 : case MultiXactStatusNoKeyUpdate:
7569 2356 : has_update = true;
7570 2356 : break;
7571 :
7572 10 : case MultiXactStatusUpdate:
7573 10 : bits2 |= HEAP_KEYS_UPDATED;
7574 10 : has_update = true;
7575 10 : break;
7576 : }
7577 : }
7578 :
7579 76801 : if (strongest == LockTupleExclusive ||
7580 : strongest == LockTupleNoKeyExclusive)
7581 2446 : bits |= HEAP_XMAX_EXCL_LOCK;
7582 74355 : else if (strongest == LockTupleShare)
7583 439 : bits |= HEAP_XMAX_SHR_LOCK;
7584 73916 : else if (strongest == LockTupleKeyShare)
7585 73916 : bits |= HEAP_XMAX_KEYSHR_LOCK;
7586 :
7587 76801 : if (!has_update)
7588 74435 : bits |= HEAP_XMAX_LOCK_ONLY;
7589 :
7590 76801 : if (nmembers > 0)
7591 76801 : pfree(members);
7592 :
7593 76801 : *new_infomask = bits;
7594 76801 : *new_infomask2 = bits2;
7595 76801 : }
7596 :
7597 : /*
7598 : * MultiXactIdGetUpdateXid
7599 : *
7600 : * Given a multixact Xmax and corresponding infomask, which does not have the
7601 : * HEAP_XMAX_LOCK_ONLY bit set, obtain and return the Xid of the updating
7602 : * transaction.
7603 : *
7604 : * Caller is expected to check the status of the updating transaction, if
7605 : * necessary.
7606 : */
7607 : static TransactionId
7608 162019 : MultiXactIdGetUpdateXid(TransactionId xmax, uint16 t_infomask)
7609 : {
7610 162019 : TransactionId update_xact = InvalidTransactionId;
7611 : MultiXactMember *members;
7612 : int nmembers;
7613 :
7614 : Assert(!(t_infomask & HEAP_XMAX_LOCK_ONLY));
7615 : Assert(t_infomask & HEAP_XMAX_IS_MULTI);
7616 :
7617 : /*
7618 : * Since we know the LOCK_ONLY bit is not set, this cannot be a multi from
7619 : * pre-pg_upgrade.
7620 : */
7621 162019 : nmembers = GetMultiXactIdMembers(xmax, &members, false, false);
7622 :
7623 162019 : if (nmembers > 0)
7624 : {
7625 : int i;
7626 :
7627 245780 : for (i = 0; i < nmembers; i++)
7628 : {
7629 : /* Ignore lockers */
7630 245780 : if (!ISUPDATE_from_mxstatus(members[i].status))
7631 83761 : continue;
7632 :
7633 : /* there can be at most one updater */
7634 : Assert(update_xact == InvalidTransactionId);
7635 162019 : update_xact = members[i].xid;
7636 : #ifndef USE_ASSERT_CHECKING
7637 :
7638 : /*
7639 : * in an assert-enabled build, walk the whole array to ensure
7640 : * there's no other updater.
7641 : */
7642 162019 : break;
7643 : #endif
7644 : }
7645 :
7646 162019 : pfree(members);
7647 : }
7648 :
7649 162019 : return update_xact;
7650 : }
7651 :
7652 : /*
7653 : * HeapTupleGetUpdateXid
7654 : * As above, but use a HeapTupleHeader
7655 : *
7656 : * See also HeapTupleHeaderGetUpdateXid, which can be used without previously
7657 : * checking the hint bits.
7658 : */
7659 : TransactionId
7660 159881 : HeapTupleGetUpdateXid(const HeapTupleHeaderData *tup)
7661 : {
7662 159881 : return MultiXactIdGetUpdateXid(HeapTupleHeaderGetRawXmax(tup),
7663 159881 : tup->t_infomask);
7664 : }
7665 :
7666 : /*
7667 : * Does the given multixact conflict with the current transaction grabbing a
7668 : * tuple lock of the given strength?
7669 : *
7670 : * The passed infomask pairs up with the given multixact in the tuple header.
7671 : *
7672 : * If current_is_member is not NULL, it is set to 'true' if the current
7673 : * transaction is a member of the given multixact.
7674 : */
7675 : static bool
7676 218 : DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
7677 : LockTupleMode lockmode, bool *current_is_member)
7678 : {
7679 : int nmembers;
7680 : MultiXactMember *members;
7681 218 : bool result = false;
7682 218 : LOCKMODE wanted = tupleLockExtraInfo[lockmode].hwlock;
7683 :
7684 218 : if (HEAP_LOCKED_UPGRADED(infomask))
7685 0 : return false;
7686 :
7687 218 : nmembers = GetMultiXactIdMembers(multi, &members, false,
7688 218 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
7689 218 : if (nmembers >= 0)
7690 : {
7691 : int i;
7692 :
7693 2682 : for (i = 0; i < nmembers; i++)
7694 : {
7695 : TransactionId memxid;
7696 : LOCKMODE memlockmode;
7697 :
7698 2471 : if (result && (current_is_member == NULL || *current_is_member))
7699 : break;
7700 :
7701 2464 : memlockmode = LOCKMODE_from_mxstatus(members[i].status);
7702 :
7703 : /* ignore members from current xact (but track their presence) */
7704 2464 : memxid = members[i].xid;
7705 2464 : if (TransactionIdIsCurrentTransactionId(memxid))
7706 : {
7707 92 : if (current_is_member != NULL)
7708 78 : *current_is_member = true;
7709 92 : continue;
7710 : }
7711 2372 : else if (result)
7712 8 : continue;
7713 :
7714 : /* ignore members that don't conflict with the lock we want */
7715 2364 : if (!DoLockModesConflict(memlockmode, wanted))
7716 2325 : continue;
7717 :
7718 39 : if (ISUPDATE_from_mxstatus(members[i].status))
7719 : {
7720 : /* ignore aborted updaters */
7721 17 : if (TransactionIdDidAbort(memxid))
7722 1 : continue;
7723 : }
7724 : else
7725 : {
7726 : /* ignore lockers-only that are no longer in progress */
7727 22 : if (!TransactionIdIsInProgress(memxid))
7728 7 : continue;
7729 : }
7730 :
7731 : /*
7732 : * Whatever remains are either live lockers that conflict with our
7733 : * wanted lock, and updaters that are not aborted. Those conflict
7734 : * with what we want. Set up to return true, but keep going to
7735 : * look for the current transaction among the multixact members,
7736 : * if needed.
7737 : */
7738 31 : result = true;
7739 : }
7740 218 : pfree(members);
7741 : }
7742 :
7743 218 : return result;
7744 : }
7745 :
7746 : /*
7747 : * Do_MultiXactIdWait
7748 : * Actual implementation for the two functions below.
7749 : *
7750 : * 'multi', 'status' and 'infomask' indicate what to sleep on (the status is
7751 : * needed to ensure we only sleep on conflicting members, and the infomask is
7752 : * used to optimize multixact access in case it's a lock-only multi); 'nowait'
7753 : * indicates whether to use conditional lock acquisition, to allow callers to
7754 : * fail if lock is unavailable. 'rel', 'ctid' and 'oper' are used to set up
7755 : * context information for error messages. 'remaining', if not NULL, receives
7756 : * the number of members that are still running, including any (non-aborted)
7757 : * subtransactions of our own transaction. 'logLockFailure' indicates whether
7758 : * to log details when a lock acquisition fails with 'nowait' enabled.
7759 : *
7760 : * We do this by sleeping on each member using XactLockTableWait. Any
7761 : * members that belong to the current backend are *not* waited for, however;
7762 : * this would not merely be useless but would lead to Assert failure inside
7763 : * XactLockTableWait. By the time this returns, it is certain that all
7764 : * transactions *of other backends* that were members of the MultiXactId
7765 : * that conflict with the requested status are dead (and no new ones can have
7766 : * been added, since it is not legal to add members to an existing
7767 : * MultiXactId).
7768 : *
7769 : * But by the time we finish sleeping, someone else may have changed the Xmax
7770 : * of the containing tuple, so the caller needs to iterate on us somehow.
7771 : *
7772 : * Note that in case we return false, the number of remaining members is
7773 : * not to be trusted.
7774 : */
7775 : static bool
7776 58 : Do_MultiXactIdWait(MultiXactId multi, MultiXactStatus status,
7777 : uint16 infomask, bool nowait,
7778 : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
7779 : int *remaining, bool logLockFailure)
7780 : {
7781 58 : bool result = true;
7782 : MultiXactMember *members;
7783 : int nmembers;
7784 58 : int remain = 0;
7785 :
7786 : /* for pre-pg_upgrade tuples, no need to sleep at all */
7787 58 : nmembers = HEAP_LOCKED_UPGRADED(infomask) ? -1 :
7788 58 : GetMultiXactIdMembers(multi, &members, false,
7789 58 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
7790 :
7791 58 : if (nmembers >= 0)
7792 : {
7793 : int i;
7794 :
7795 187 : for (i = 0; i < nmembers; i++)
7796 : {
7797 133 : TransactionId memxid = members[i].xid;
7798 133 : MultiXactStatus memstatus = members[i].status;
7799 :
7800 133 : if (TransactionIdIsCurrentTransactionId(memxid))
7801 : {
7802 24 : remain++;
7803 24 : continue;
7804 : }
7805 :
7806 109 : if (!DoLockModesConflict(LOCKMODE_from_mxstatus(memstatus),
7807 109 : LOCKMODE_from_mxstatus(status)))
7808 : {
7809 22 : if (remaining && TransactionIdIsInProgress(memxid))
7810 8 : remain++;
7811 22 : continue;
7812 : }
7813 :
7814 : /*
7815 : * This member conflicts with our multi, so we have to sleep (or
7816 : * return failure, if asked to avoid waiting.)
7817 : *
7818 : * Note that we don't set up an error context callback ourselves,
7819 : * but instead we pass the info down to XactLockTableWait. This
7820 : * might seem a bit wasteful because the context is set up and
7821 : * tore down for each member of the multixact, but in reality it
7822 : * should be barely noticeable, and it avoids duplicate code.
7823 : */
7824 87 : if (nowait)
7825 : {
7826 4 : result = ConditionalXactLockTableWait(memxid, logLockFailure);
7827 4 : if (!result)
7828 4 : break;
7829 : }
7830 : else
7831 83 : XactLockTableWait(memxid, rel, ctid, oper);
7832 : }
7833 :
7834 58 : pfree(members);
7835 : }
7836 :
7837 58 : if (remaining)
7838 10 : *remaining = remain;
7839 :
7840 58 : return result;
7841 : }
7842 :
7843 : /*
7844 : * MultiXactIdWait
7845 : * Sleep on a MultiXactId.
7846 : *
7847 : * By the time we finish sleeping, someone else may have changed the Xmax
7848 : * of the containing tuple, so the caller needs to iterate on us somehow.
7849 : *
7850 : * We return (in *remaining, if not NULL) the number of members that are still
7851 : * running, including any (non-aborted) subtransactions of our own transaction.
7852 : */
7853 : static void
7854 54 : MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
7855 : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
7856 : int *remaining)
7857 : {
7858 54 : (void) Do_MultiXactIdWait(multi, status, infomask, false,
7859 : rel, ctid, oper, remaining, false);
7860 54 : }
7861 :
7862 : /*
7863 : * ConditionalMultiXactIdWait
7864 : * As above, but only lock if we can get the lock without blocking.
7865 : *
7866 : * By the time we finish sleeping, someone else may have changed the Xmax
7867 : * of the containing tuple, so the caller needs to iterate on us somehow.
7868 : *
7869 : * If the multixact is now all gone, return true. Returns false if some
7870 : * transactions might still be running.
7871 : *
7872 : * We return (in *remaining, if not NULL) the number of members that are still
7873 : * running, including any (non-aborted) subtransactions of our own transaction.
7874 : */
7875 : static bool
7876 4 : ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status,
7877 : uint16 infomask, Relation rel, int *remaining,
7878 : bool logLockFailure)
7879 : {
7880 4 : return Do_MultiXactIdWait(multi, status, infomask, true,
7881 : rel, NULL, XLTW_None, remaining, logLockFailure);
7882 : }
7883 :
7884 : /*
7885 : * heap_tuple_needs_eventual_freeze
7886 : *
7887 : * Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
7888 : * will eventually require freezing (if tuple isn't removed by pruning first).
7889 : */
7890 : bool
7891 129134 : heap_tuple_needs_eventual_freeze(HeapTupleHeader tuple)
7892 : {
7893 : TransactionId xid;
7894 :
7895 : /*
7896 : * If xmin is a normal transaction ID, this tuple is definitely not
7897 : * frozen.
7898 : */
7899 129134 : xid = HeapTupleHeaderGetXmin(tuple);
7900 129134 : if (TransactionIdIsNormal(xid))
7901 3136 : return true;
7902 :
7903 : /*
7904 : * If xmax is a valid xact or multixact, this tuple is also not frozen.
7905 : */
7906 125998 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7907 : {
7908 : MultiXactId multi;
7909 :
7910 0 : multi = HeapTupleHeaderGetRawXmax(tuple);
7911 0 : if (MultiXactIdIsValid(multi))
7912 0 : return true;
7913 : }
7914 : else
7915 : {
7916 125998 : xid = HeapTupleHeaderGetRawXmax(tuple);
7917 125998 : if (TransactionIdIsNormal(xid))
7918 7 : return true;
7919 : }
7920 :
7921 125991 : if (tuple->t_infomask & HEAP_MOVED)
7922 : {
7923 0 : xid = HeapTupleHeaderGetXvac(tuple);
7924 0 : if (TransactionIdIsNormal(xid))
7925 0 : return true;
7926 : }
7927 :
7928 125991 : return false;
7929 : }
7930 :
7931 : /*
7932 : * heap_tuple_should_freeze
7933 : *
7934 : * Return value indicates if heap_prepare_freeze_tuple sibling function would
7935 : * (or should) force freezing of the heap page that contains caller's tuple.
7936 : * Tuple header XIDs/MXIDs < FreezeLimit/MultiXactCutoff trigger freezing.
7937 : * This includes (xmin, xmax, xvac) fields, as well as MultiXact member XIDs.
7938 : *
7939 : * The *NoFreezePageRelfrozenXid and *NoFreezePageRelminMxid input/output
7940 : * arguments help VACUUM track the oldest extant XID/MXID remaining in rel.
7941 : * Our working assumption is that caller won't decide to freeze this tuple.
7942 : * It's up to caller to only ratchet back its own top-level trackers after the
7943 : * point that it fully commits to not freezing the tuple/page in question.
7944 : */
7945 : bool
7946 8264770 : heap_tuple_should_freeze(HeapTupleHeader tuple,
7947 : const struct VacuumCutoffs *cutoffs,
7948 : TransactionId *NoFreezePageRelfrozenXid,
7949 : MultiXactId *NoFreezePageRelminMxid)
7950 : {
7951 : TransactionId xid;
7952 : MultiXactId multi;
7953 8264770 : bool freeze = false;
7954 :
7955 : /* First deal with xmin */
7956 8264770 : xid = HeapTupleHeaderGetXmin(tuple);
7957 8264770 : if (TransactionIdIsNormal(xid))
7958 : {
7959 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7960 1935324 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7961 22613 : *NoFreezePageRelfrozenXid = xid;
7962 1935324 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7963 21157 : freeze = true;
7964 : }
7965 :
7966 : /* Now deal with xmax */
7967 8264770 : xid = InvalidTransactionId;
7968 8264770 : multi = InvalidMultiXactId;
7969 8264770 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7970 2 : multi = HeapTupleHeaderGetRawXmax(tuple);
7971 : else
7972 8264768 : xid = HeapTupleHeaderGetRawXmax(tuple);
7973 :
7974 8264770 : if (TransactionIdIsNormal(xid))
7975 : {
7976 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7977 : /* xmax is a non-permanent XID */
7978 6622738 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7979 4 : *NoFreezePageRelfrozenXid = xid;
7980 6622738 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7981 5 : freeze = true;
7982 : }
7983 1642032 : else if (!MultiXactIdIsValid(multi))
7984 : {
7985 : /* xmax is a permanent XID or invalid MultiXactId/XID */
7986 : }
7987 2 : else if (HEAP_LOCKED_UPGRADED(tuple->t_infomask))
7988 : {
7989 : /* xmax is a pg_upgrade'd MultiXact, which can't have updater XID */
7990 0 : if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
7991 0 : *NoFreezePageRelminMxid = multi;
7992 : /* heap_prepare_freeze_tuple always freezes pg_upgrade'd xmax */
7993 0 : freeze = true;
7994 : }
7995 : else
7996 : {
7997 : /* xmax is a MultiXactId that may have an updater XID */
7998 : MultiXactMember *members;
7999 : int nmembers;
8000 :
8001 : Assert(MultiXactIdPrecedesOrEquals(cutoffs->relminmxid, multi));
8002 2 : if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
8003 2 : *NoFreezePageRelminMxid = multi;
8004 2 : if (MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff))
8005 2 : freeze = true;
8006 :
8007 : /* need to check whether any member of the mxact is old */
8008 2 : nmembers = GetMultiXactIdMembers(multi, &members, false,
8009 2 : HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask));
8010 :
8011 5 : for (int i = 0; i < nmembers; i++)
8012 : {
8013 3 : xid = members[i].xid;
8014 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
8015 3 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
8016 0 : *NoFreezePageRelfrozenXid = xid;
8017 3 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
8018 0 : freeze = true;
8019 : }
8020 2 : if (nmembers > 0)
8021 1 : pfree(members);
8022 : }
8023 :
8024 8264770 : if (tuple->t_infomask & HEAP_MOVED)
8025 : {
8026 0 : xid = HeapTupleHeaderGetXvac(tuple);
8027 0 : if (TransactionIdIsNormal(xid))
8028 : {
8029 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
8030 0 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
8031 0 : *NoFreezePageRelfrozenXid = xid;
8032 : /* heap_prepare_freeze_tuple forces xvac freezing */
8033 0 : freeze = true;
8034 : }
8035 : }
8036 :
8037 8264770 : return freeze;
8038 : }
8039 :
8040 : /*
8041 : * Maintain snapshotConflictHorizon for caller by ratcheting forward its value
8042 : * using any committed XIDs contained in 'tuple', an obsolescent heap tuple
8043 : * that caller is in the process of physically removing, e.g. via HOT pruning
8044 : * or index deletion.
8045 : *
8046 : * Caller must initialize its value to InvalidTransactionId, which is
8047 : * generally interpreted as "definitely no need for a recovery conflict".
8048 : * Final value must reflect all heap tuples that caller will physically remove
8049 : * (or remove TID references to) via its ongoing pruning/deletion operation.
8050 : * ResolveRecoveryConflictWithSnapshot() is passed the final value (taken from
8051 : * caller's WAL record) by REDO routine when it replays caller's operation.
8052 : */
8053 : void
8054 1572617 : HeapTupleHeaderAdvanceConflictHorizon(HeapTupleHeader tuple,
8055 : TransactionId *snapshotConflictHorizon)
8056 : {
8057 1572617 : TransactionId xmin = HeapTupleHeaderGetXmin(tuple);
8058 1572617 : TransactionId xmax = HeapTupleHeaderGetUpdateXid(tuple);
8059 1572617 : TransactionId xvac = HeapTupleHeaderGetXvac(tuple);
8060 :
8061 1572617 : if (tuple->t_infomask & HEAP_MOVED)
8062 : {
8063 0 : if (TransactionIdPrecedes(*snapshotConflictHorizon, xvac))
8064 0 : *snapshotConflictHorizon = xvac;
8065 : }
8066 :
8067 : /*
8068 : * Ignore tuples inserted by an aborted transaction or if the tuple was
8069 : * updated/deleted by the inserting transaction.
8070 : *
8071 : * Look for a committed hint bit, or if no xmin bit is set, check clog.
8072 : */
8073 1572617 : if (HeapTupleHeaderXminCommitted(tuple) ||
8074 106822 : (!HeapTupleHeaderXminInvalid(tuple) && TransactionIdDidCommit(xmin)))
8075 : {
8076 2847445 : if (xmax != xmin &&
8077 1353945 : TransactionIdFollows(xmax, *snapshotConflictHorizon))
8078 100440 : *snapshotConflictHorizon = xmax;
8079 : }
8080 1572617 : }
8081 :
8082 : #ifdef USE_PREFETCH
8083 : /*
8084 : * Helper function for heap_index_delete_tuples. Issues prefetch requests for
8085 : * prefetch_count buffers. The prefetch_state keeps track of all the buffers
8086 : * we can prefetch, and which have already been prefetched; each call to this
8087 : * function picks up where the previous call left off.
8088 : *
8089 : * Note: we expect the deltids array to be sorted in an order that groups TIDs
8090 : * by heap block, with all TIDs for each block appearing together in exactly
8091 : * one group.
8092 : */
8093 : static void
8094 19838 : index_delete_prefetch_buffer(Relation rel,
8095 : IndexDeletePrefetchState *prefetch_state,
8096 : int prefetch_count)
8097 : {
8098 19838 : BlockNumber cur_hblkno = prefetch_state->cur_hblkno;
8099 19838 : int count = 0;
8100 : int i;
8101 19838 : int ndeltids = prefetch_state->ndeltids;
8102 19838 : TM_IndexDelete *deltids = prefetch_state->deltids;
8103 :
8104 19838 : for (i = prefetch_state->next_item;
8105 685439 : i < ndeltids && count < prefetch_count;
8106 665601 : i++)
8107 : {
8108 665601 : ItemPointer htid = &deltids[i].tid;
8109 :
8110 1325335 : if (cur_hblkno == InvalidBlockNumber ||
8111 659734 : ItemPointerGetBlockNumber(htid) != cur_hblkno)
8112 : {
8113 18102 : cur_hblkno = ItemPointerGetBlockNumber(htid);
8114 18102 : PrefetchBuffer(rel, MAIN_FORKNUM, cur_hblkno);
8115 18102 : count++;
8116 : }
8117 : }
8118 :
8119 : /*
8120 : * Save the prefetch position so that next time we can continue from that
8121 : * position.
8122 : */
8123 19838 : prefetch_state->next_item = i;
8124 19838 : prefetch_state->cur_hblkno = cur_hblkno;
8125 19838 : }
8126 : #endif
8127 :
8128 : /*
8129 : * Helper function for heap_index_delete_tuples. Checks for index corruption
8130 : * involving an invalid TID in index AM caller's index page.
8131 : *
8132 : * This is an ideal place for these checks. The index AM must hold a buffer
8133 : * lock on the index page containing the TIDs we examine here, so we don't
8134 : * have to worry about concurrent VACUUMs at all. We can be sure that the
8135 : * index is corrupt when htid points directly to an LP_UNUSED item or
8136 : * heap-only tuple, which is not the case during standard index scans.
8137 : */
8138 : static inline void
8139 551090 : index_delete_check_htid(TM_IndexDeleteOp *delstate,
8140 : Page page, OffsetNumber maxoff,
8141 : const ItemPointerData *htid, TM_IndexStatus *istatus)
8142 : {
8143 551090 : OffsetNumber indexpagehoffnum = ItemPointerGetOffsetNumber(htid);
8144 : ItemId iid;
8145 :
8146 : Assert(OffsetNumberIsValid(istatus->idxoffnum));
8147 :
8148 551090 : if (unlikely(indexpagehoffnum > maxoff))
8149 0 : ereport(ERROR,
8150 : (errcode(ERRCODE_INDEX_CORRUPTED),
8151 : 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\"",
8152 : ItemPointerGetBlockNumber(htid),
8153 : indexpagehoffnum,
8154 : istatus->idxoffnum, delstate->iblknum,
8155 : RelationGetRelationName(delstate->irel))));
8156 :
8157 551090 : iid = PageGetItemId(page, indexpagehoffnum);
8158 551090 : if (unlikely(!ItemIdIsUsed(iid)))
8159 0 : ereport(ERROR,
8160 : (errcode(ERRCODE_INDEX_CORRUPTED),
8161 : errmsg_internal("heap tid from index tuple (%u,%u) points to unused heap page item at offset %u of block %u in index \"%s\"",
8162 : ItemPointerGetBlockNumber(htid),
8163 : indexpagehoffnum,
8164 : istatus->idxoffnum, delstate->iblknum,
8165 : RelationGetRelationName(delstate->irel))));
8166 :
8167 551090 : if (ItemIdHasStorage(iid))
8168 : {
8169 : HeapTupleHeader htup;
8170 :
8171 : Assert(ItemIdIsNormal(iid));
8172 325690 : htup = (HeapTupleHeader) PageGetItem(page, iid);
8173 :
8174 325690 : if (unlikely(HeapTupleHeaderIsHeapOnly(htup)))
8175 0 : ereport(ERROR,
8176 : (errcode(ERRCODE_INDEX_CORRUPTED),
8177 : errmsg_internal("heap tid from index tuple (%u,%u) points to heap-only tuple at offset %u of block %u in index \"%s\"",
8178 : ItemPointerGetBlockNumber(htid),
8179 : indexpagehoffnum,
8180 : istatus->idxoffnum, delstate->iblknum,
8181 : RelationGetRelationName(delstate->irel))));
8182 : }
8183 551090 : }
8184 :
8185 : /*
8186 : * heapam implementation of tableam's index_delete_tuples interface.
8187 : *
8188 : * This helper function is called by index AMs during index tuple deletion.
8189 : * See tableam header comments for an explanation of the interface implemented
8190 : * here and a general theory of operation. Note that each call here is either
8191 : * a simple index deletion call, or a bottom-up index deletion call.
8192 : *
8193 : * It's possible for this to generate a fair amount of I/O, since we may be
8194 : * deleting hundreds of tuples from a single index block. To amortize that
8195 : * cost to some degree, this uses prefetching and combines repeat accesses to
8196 : * the same heap block.
8197 : */
8198 : TransactionId
8199 5867 : heap_index_delete_tuples(Relation rel, TM_IndexDeleteOp *delstate)
8200 : {
8201 : /* Initial assumption is that earlier pruning took care of conflict */
8202 5867 : TransactionId snapshotConflictHorizon = InvalidTransactionId;
8203 5867 : BlockNumber blkno = InvalidBlockNumber;
8204 5867 : Buffer buf = InvalidBuffer;
8205 5867 : Page page = NULL;
8206 5867 : OffsetNumber maxoff = InvalidOffsetNumber;
8207 : TransactionId priorXmax;
8208 : #ifdef USE_PREFETCH
8209 : IndexDeletePrefetchState prefetch_state;
8210 : int prefetch_distance;
8211 : #endif
8212 : SnapshotData SnapshotNonVacuumable;
8213 5867 : int finalndeltids = 0,
8214 5867 : nblocksaccessed = 0;
8215 :
8216 : /* State that's only used in bottom-up index deletion case */
8217 5867 : int nblocksfavorable = 0;
8218 5867 : int curtargetfreespace = delstate->bottomupfreespace,
8219 5867 : lastfreespace = 0,
8220 5867 : actualfreespace = 0;
8221 5867 : bool bottomup_final_block = false;
8222 :
8223 5867 : InitNonVacuumableSnapshot(SnapshotNonVacuumable, GlobalVisTestFor(rel));
8224 :
8225 : /* Sort caller's deltids array by TID for further processing */
8226 5867 : index_delete_sort(delstate);
8227 :
8228 : /*
8229 : * Bottom-up case: resort deltids array in an order attuned to where the
8230 : * greatest number of promising TIDs are to be found, and determine how
8231 : * many blocks from the start of sorted array should be considered
8232 : * favorable. This will also shrink the deltids array in order to
8233 : * eliminate completely unfavorable blocks up front.
8234 : */
8235 5867 : if (delstate->bottomup)
8236 1965 : nblocksfavorable = bottomup_sort_and_shrink(delstate);
8237 :
8238 : #ifdef USE_PREFETCH
8239 : /* Initialize prefetch state. */
8240 5867 : prefetch_state.cur_hblkno = InvalidBlockNumber;
8241 5867 : prefetch_state.next_item = 0;
8242 5867 : prefetch_state.ndeltids = delstate->ndeltids;
8243 5867 : prefetch_state.deltids = delstate->deltids;
8244 :
8245 : /*
8246 : * Determine the prefetch distance that we will attempt to maintain.
8247 : *
8248 : * Since the caller holds a buffer lock somewhere in rel, we'd better make
8249 : * sure that isn't a catalog relation before we call code that does
8250 : * syscache lookups, to avoid risk of deadlock.
8251 : */
8252 5867 : if (IsCatalogRelation(rel))
8253 4276 : prefetch_distance = maintenance_io_concurrency;
8254 : else
8255 : prefetch_distance =
8256 1591 : get_tablespace_maintenance_io_concurrency(rel->rd_rel->reltablespace);
8257 :
8258 : /* Cap initial prefetch distance for bottom-up deletion caller */
8259 5867 : if (delstate->bottomup)
8260 : {
8261 : Assert(nblocksfavorable >= 1);
8262 : Assert(nblocksfavorable <= BOTTOMUP_MAX_NBLOCKS);
8263 1965 : prefetch_distance = Min(prefetch_distance, nblocksfavorable);
8264 : }
8265 :
8266 : /* Start prefetching. */
8267 5867 : index_delete_prefetch_buffer(rel, &prefetch_state, prefetch_distance);
8268 : #endif
8269 :
8270 : /* Iterate over deltids, determine which to delete, check their horizon */
8271 : Assert(delstate->ndeltids > 0);
8272 556957 : for (int i = 0; i < delstate->ndeltids; i++)
8273 : {
8274 553055 : TM_IndexDelete *ideltid = &delstate->deltids[i];
8275 553055 : TM_IndexStatus *istatus = delstate->status + ideltid->id;
8276 553055 : ItemPointer htid = &ideltid->tid;
8277 : OffsetNumber offnum;
8278 :
8279 : /*
8280 : * Read buffer, and perform required extra steps each time a new block
8281 : * is encountered. Avoid refetching if it's the same block as the one
8282 : * from the last htid.
8283 : */
8284 1100243 : if (blkno == InvalidBlockNumber ||
8285 547188 : ItemPointerGetBlockNumber(htid) != blkno)
8286 : {
8287 : /*
8288 : * Consider giving up early for bottom-up index deletion caller
8289 : * first. (Only prefetch next-next block afterwards, when it
8290 : * becomes clear that we're at least going to access the next
8291 : * block in line.)
8292 : *
8293 : * Sometimes the first block frees so much space for bottom-up
8294 : * caller that the deletion process can end without accessing any
8295 : * more blocks. It is usually necessary to access 2 or 3 blocks
8296 : * per bottom-up deletion operation, though.
8297 : */
8298 15936 : if (delstate->bottomup)
8299 : {
8300 : /*
8301 : * We often allow caller to delete a few additional items
8302 : * whose entries we reached after the point that space target
8303 : * from caller was satisfied. The cost of accessing the page
8304 : * was already paid at that point, so it made sense to finish
8305 : * it off. When that happened, we finalize everything here
8306 : * (by finishing off the whole bottom-up deletion operation
8307 : * without needlessly paying the cost of accessing any more
8308 : * blocks).
8309 : */
8310 4423 : if (bottomup_final_block)
8311 152 : break;
8312 :
8313 : /*
8314 : * Give up when we didn't enable our caller to free any
8315 : * additional space as a result of processing the page that we
8316 : * just finished up with. This rule is the main way in which
8317 : * we keep the cost of bottom-up deletion under control.
8318 : */
8319 4271 : if (nblocksaccessed >= 1 && actualfreespace == lastfreespace)
8320 1813 : break;
8321 2458 : lastfreespace = actualfreespace; /* for next time */
8322 :
8323 : /*
8324 : * Deletion operation (which is bottom-up) will definitely
8325 : * access the next block in line. Prepare for that now.
8326 : *
8327 : * Decay target free space so that we don't hang on for too
8328 : * long with a marginal case. (Space target is only truly
8329 : * helpful when it allows us to recognize that we don't need
8330 : * to access more than 1 or 2 blocks to satisfy caller due to
8331 : * agreeable workload characteristics.)
8332 : *
8333 : * We are a bit more patient when we encounter contiguous
8334 : * blocks, though: these are treated as favorable blocks. The
8335 : * decay process is only applied when the next block in line
8336 : * is not a favorable/contiguous block. This is not an
8337 : * exception to the general rule; we still insist on finding
8338 : * at least one deletable item per block accessed. See
8339 : * bottomup_nblocksfavorable() for full details of the theory
8340 : * behind favorable blocks and heap block locality in general.
8341 : *
8342 : * Note: The first block in line is always treated as a
8343 : * favorable block, so the earliest possible point that the
8344 : * decay can be applied is just before we access the second
8345 : * block in line. The Assert() verifies this for us.
8346 : */
8347 : Assert(nblocksaccessed > 0 || nblocksfavorable > 0);
8348 2458 : if (nblocksfavorable > 0)
8349 2209 : nblocksfavorable--;
8350 : else
8351 249 : curtargetfreespace /= 2;
8352 : }
8353 :
8354 : /* release old buffer */
8355 13971 : if (BufferIsValid(buf))
8356 8104 : UnlockReleaseBuffer(buf);
8357 :
8358 13971 : blkno = ItemPointerGetBlockNumber(htid);
8359 13971 : buf = ReadBuffer(rel, blkno);
8360 13971 : nblocksaccessed++;
8361 : Assert(!delstate->bottomup ||
8362 : nblocksaccessed <= BOTTOMUP_MAX_NBLOCKS);
8363 :
8364 : #ifdef USE_PREFETCH
8365 :
8366 : /*
8367 : * To maintain the prefetch distance, prefetch one more page for
8368 : * each page we read.
8369 : */
8370 13971 : index_delete_prefetch_buffer(rel, &prefetch_state, 1);
8371 : #endif
8372 :
8373 13971 : LockBuffer(buf, BUFFER_LOCK_SHARE);
8374 :
8375 13971 : page = BufferGetPage(buf);
8376 13971 : maxoff = PageGetMaxOffsetNumber(page);
8377 : }
8378 :
8379 : /*
8380 : * In passing, detect index corruption involving an index page with a
8381 : * TID that points to a location in the heap that couldn't possibly be
8382 : * correct. We only do this with actual TIDs from caller's index page
8383 : * (not items reached by traversing through a HOT chain).
8384 : */
8385 551090 : index_delete_check_htid(delstate, page, maxoff, htid, istatus);
8386 :
8387 551090 : if (istatus->knowndeletable)
8388 : Assert(!delstate->bottomup && !istatus->promising);
8389 : else
8390 : {
8391 415152 : ItemPointerData tmp = *htid;
8392 : HeapTupleData heapTuple;
8393 :
8394 : /* Are any tuples from this HOT chain non-vacuumable? */
8395 415152 : if (heap_hot_search_buffer(&tmp, rel, buf, &SnapshotNonVacuumable,
8396 : &heapTuple, NULL, true))
8397 246881 : continue; /* can't delete entry */
8398 :
8399 : /* Caller will delete, since whole HOT chain is vacuumable */
8400 168271 : istatus->knowndeletable = true;
8401 :
8402 : /* Maintain index free space info for bottom-up deletion case */
8403 168271 : if (delstate->bottomup)
8404 : {
8405 : Assert(istatus->freespace > 0);
8406 9193 : actualfreespace += istatus->freespace;
8407 9193 : if (actualfreespace >= curtargetfreespace)
8408 2157 : bottomup_final_block = true;
8409 : }
8410 : }
8411 :
8412 : /*
8413 : * Maintain snapshotConflictHorizon value for deletion operation as a
8414 : * whole by advancing current value using heap tuple headers. This is
8415 : * loosely based on the logic for pruning a HOT chain.
8416 : */
8417 304209 : offnum = ItemPointerGetOffsetNumber(htid);
8418 304209 : priorXmax = InvalidTransactionId; /* cannot check first XMIN */
8419 : for (;;)
8420 21435 : {
8421 : ItemId lp;
8422 : HeapTupleHeader htup;
8423 :
8424 : /* Sanity check (pure paranoia) */
8425 325644 : if (offnum < FirstOffsetNumber)
8426 0 : break;
8427 :
8428 : /*
8429 : * An offset past the end of page's line pointer array is possible
8430 : * when the array was truncated
8431 : */
8432 325644 : if (offnum > maxoff)
8433 0 : break;
8434 :
8435 325644 : lp = PageGetItemId(page, offnum);
8436 325644 : if (ItemIdIsRedirected(lp))
8437 : {
8438 9385 : offnum = ItemIdGetRedirect(lp);
8439 9385 : continue;
8440 : }
8441 :
8442 : /*
8443 : * We'll often encounter LP_DEAD line pointers (especially with an
8444 : * entry marked knowndeletable by our caller up front). No heap
8445 : * tuple headers get examined for an htid that leads us to an
8446 : * LP_DEAD item. This is okay because the earlier pruning
8447 : * operation that made the line pointer LP_DEAD in the first place
8448 : * must have considered the original tuple header as part of
8449 : * generating its own snapshotConflictHorizon value.
8450 : *
8451 : * Relying on XLOG_HEAP2_PRUNE_VACUUM_SCAN records like this is
8452 : * the same strategy that index vacuuming uses in all cases. Index
8453 : * VACUUM WAL records don't even have a snapshotConflictHorizon
8454 : * field of their own for this reason.
8455 : */
8456 316259 : if (!ItemIdIsNormal(lp))
8457 200379 : break;
8458 :
8459 115880 : htup = (HeapTupleHeader) PageGetItem(page, lp);
8460 :
8461 : /*
8462 : * Check the tuple XMIN against prior XMAX, if any
8463 : */
8464 127930 : if (TransactionIdIsValid(priorXmax) &&
8465 12050 : !TransactionIdEquals(HeapTupleHeaderGetXmin(htup), priorXmax))
8466 0 : break;
8467 :
8468 115880 : HeapTupleHeaderAdvanceConflictHorizon(htup,
8469 : &snapshotConflictHorizon);
8470 :
8471 : /*
8472 : * If the tuple is not HOT-updated, then we are at the end of this
8473 : * HOT-chain. No need to visit later tuples from the same update
8474 : * chain (they get their own index entries) -- just move on to
8475 : * next htid from index AM caller.
8476 : */
8477 115880 : if (!HeapTupleHeaderIsHotUpdated(htup))
8478 103830 : break;
8479 :
8480 : /* Advance to next HOT chain member */
8481 : Assert(ItemPointerGetBlockNumber(&htup->t_ctid) == blkno);
8482 12050 : offnum = ItemPointerGetOffsetNumber(&htup->t_ctid);
8483 12050 : priorXmax = HeapTupleHeaderGetUpdateXid(htup);
8484 : }
8485 :
8486 : /* Enable further/final shrinking of deltids for caller */
8487 304209 : finalndeltids = i + 1;
8488 : }
8489 :
8490 5867 : UnlockReleaseBuffer(buf);
8491 :
8492 : /*
8493 : * Shrink deltids array to exclude non-deletable entries at the end. This
8494 : * is not just a minor optimization. Final deltids array size might be
8495 : * zero for a bottom-up caller. Index AM is explicitly allowed to rely on
8496 : * ndeltids being zero in all cases with zero total deletable entries.
8497 : */
8498 : Assert(finalndeltids > 0 || delstate->bottomup);
8499 5867 : delstate->ndeltids = finalndeltids;
8500 :
8501 5867 : return snapshotConflictHorizon;
8502 : }
8503 :
8504 : /*
8505 : * Specialized inlineable comparison function for index_delete_sort()
8506 : */
8507 : static inline int
8508 12893443 : index_delete_sort_cmp(TM_IndexDelete *deltid1, TM_IndexDelete *deltid2)
8509 : {
8510 12893443 : ItemPointer tid1 = &deltid1->tid;
8511 12893443 : ItemPointer tid2 = &deltid2->tid;
8512 :
8513 : {
8514 12893443 : BlockNumber blk1 = ItemPointerGetBlockNumber(tid1);
8515 12893443 : BlockNumber blk2 = ItemPointerGetBlockNumber(tid2);
8516 :
8517 12893443 : if (blk1 != blk2)
8518 5251798 : return (blk1 < blk2) ? -1 : 1;
8519 : }
8520 : {
8521 7641645 : OffsetNumber pos1 = ItemPointerGetOffsetNumber(tid1);
8522 7641645 : OffsetNumber pos2 = ItemPointerGetOffsetNumber(tid2);
8523 :
8524 7641645 : if (pos1 != pos2)
8525 7641645 : return (pos1 < pos2) ? -1 : 1;
8526 : }
8527 :
8528 : Assert(false);
8529 :
8530 0 : return 0;
8531 : }
8532 :
8533 : /*
8534 : * Sort deltids array from delstate by TID. This prepares it for further
8535 : * processing by heap_index_delete_tuples().
8536 : *
8537 : * This operation becomes a noticeable consumer of CPU cycles with some
8538 : * workloads, so we go to the trouble of specialization/micro optimization.
8539 : * We use shellsort for this because it's easy to specialize, compiles to
8540 : * relatively few instructions, and is adaptive to presorted inputs/subsets
8541 : * (which are typical here).
8542 : */
8543 : static void
8544 5867 : index_delete_sort(TM_IndexDeleteOp *delstate)
8545 : {
8546 5867 : TM_IndexDelete *deltids = delstate->deltids;
8547 5867 : int ndeltids = delstate->ndeltids;
8548 :
8549 : /*
8550 : * Shellsort gap sequence (taken from Sedgewick-Incerpi paper).
8551 : *
8552 : * This implementation is fast with array sizes up to ~4500. This covers
8553 : * all supported BLCKSZ values.
8554 : */
8555 5867 : const int gaps[9] = {1968, 861, 336, 112, 48, 21, 7, 3, 1};
8556 :
8557 : /* Think carefully before changing anything here -- keep swaps cheap */
8558 : StaticAssertDecl(sizeof(TM_IndexDelete) <= 8,
8559 : "element size exceeds 8 bytes");
8560 :
8561 58670 : for (int g = 0; g < lengthof(gaps); g++)
8562 : {
8563 7681906 : for (int hi = gaps[g], i = hi; i < ndeltids; i++)
8564 : {
8565 7629103 : TM_IndexDelete d = deltids[i];
8566 7629103 : int j = i;
8567 :
8568 13272407 : while (j >= hi && index_delete_sort_cmp(&deltids[j - hi], &d) >= 0)
8569 : {
8570 5643304 : deltids[j] = deltids[j - hi];
8571 5643304 : j -= hi;
8572 : }
8573 7629103 : deltids[j] = d;
8574 : }
8575 : }
8576 5867 : }
8577 :
8578 : /*
8579 : * Returns how many blocks should be considered favorable/contiguous for a
8580 : * bottom-up index deletion pass. This is a number of heap blocks that starts
8581 : * from and includes the first block in line.
8582 : *
8583 : * There is always at least one favorable block during bottom-up index
8584 : * deletion. In the worst case (i.e. with totally random heap blocks) the
8585 : * first block in line (the only favorable block) can be thought of as a
8586 : * degenerate array of contiguous blocks that consists of a single block.
8587 : * heap_index_delete_tuples() will expect this.
8588 : *
8589 : * Caller passes blockgroups, a description of the final order that deltids
8590 : * will be sorted in for heap_index_delete_tuples() bottom-up index deletion
8591 : * processing. Note that deltids need not actually be sorted just yet (caller
8592 : * only passes deltids to us so that we can interpret blockgroups).
8593 : *
8594 : * You might guess that the existence of contiguous blocks cannot matter much,
8595 : * since in general the main factor that determines which blocks we visit is
8596 : * the number of promising TIDs, which is a fixed hint from the index AM.
8597 : * We're not really targeting the general case, though -- the actual goal is
8598 : * to adapt our behavior to a wide variety of naturally occurring conditions.
8599 : * The effects of most of the heuristics we apply are only noticeable in the
8600 : * aggregate, over time and across many _related_ bottom-up index deletion
8601 : * passes.
8602 : *
8603 : * Deeming certain blocks favorable allows heapam to recognize and adapt to
8604 : * workloads where heap blocks visited during bottom-up index deletion can be
8605 : * accessed contiguously, in the sense that each newly visited block is the
8606 : * neighbor of the block that bottom-up deletion just finished processing (or
8607 : * close enough to it). It will likely be cheaper to access more favorable
8608 : * blocks sooner rather than later (e.g. in this pass, not across a series of
8609 : * related bottom-up passes). Either way it is probably only a matter of time
8610 : * (or a matter of further correlated version churn) before all blocks that
8611 : * appear together as a single large batch of favorable blocks get accessed by
8612 : * _some_ bottom-up pass. Large batches of favorable blocks tend to either
8613 : * appear almost constantly or not even once (it all depends on per-index
8614 : * workload characteristics).
8615 : *
8616 : * Note that the blockgroups sort order applies a power-of-two bucketing
8617 : * scheme that creates opportunities for contiguous groups of blocks to get
8618 : * batched together, at least with workloads that are naturally amenable to
8619 : * being driven by heap block locality. This doesn't just enhance the spatial
8620 : * locality of bottom-up heap block processing in the obvious way. It also
8621 : * enables temporal locality of access, since sorting by heap block number
8622 : * naturally tends to make the bottom-up processing order deterministic.
8623 : *
8624 : * Consider the following example to get a sense of how temporal locality
8625 : * might matter: There is a heap relation with several indexes, each of which
8626 : * is low to medium cardinality. It is subject to constant non-HOT updates.
8627 : * The updates are skewed (in one part of the primary key, perhaps). None of
8628 : * the indexes are logically modified by the UPDATE statements (if they were
8629 : * then bottom-up index deletion would not be triggered in the first place).
8630 : * Naturally, each new round of index tuples (for each heap tuple that gets a
8631 : * heap_update() call) will have the same heap TID in each and every index.
8632 : * Since these indexes are low cardinality and never get logically modified,
8633 : * heapam processing during bottom-up deletion passes will access heap blocks
8634 : * in approximately sequential order. Temporal locality of access occurs due
8635 : * to bottom-up deletion passes behaving very similarly across each of the
8636 : * indexes at any given moment. This keeps the number of buffer misses needed
8637 : * to visit heap blocks to a minimum.
8638 : */
8639 : static int
8640 1965 : bottomup_nblocksfavorable(IndexDeleteCounts *blockgroups, int nblockgroups,
8641 : TM_IndexDelete *deltids)
8642 : {
8643 1965 : int64 lastblock = -1;
8644 1965 : int nblocksfavorable = 0;
8645 :
8646 : Assert(nblockgroups >= 1);
8647 : Assert(nblockgroups <= BOTTOMUP_MAX_NBLOCKS);
8648 :
8649 : /*
8650 : * We tolerate heap blocks that will be accessed only slightly out of
8651 : * physical order. Small blips occur when a pair of almost-contiguous
8652 : * blocks happen to fall into different buckets (perhaps due only to a
8653 : * small difference in npromisingtids that the bucketing scheme didn't
8654 : * quite manage to ignore). We effectively ignore these blips by applying
8655 : * a small tolerance. The precise tolerance we use is a little arbitrary,
8656 : * but it works well enough in practice.
8657 : */
8658 6471 : for (int b = 0; b < nblockgroups; b++)
8659 : {
8660 6151 : IndexDeleteCounts *group = blockgroups + b;
8661 6151 : TM_IndexDelete *firstdtid = deltids + group->ifirsttid;
8662 6151 : BlockNumber block = ItemPointerGetBlockNumber(&firstdtid->tid);
8663 :
8664 6151 : if (lastblock != -1 &&
8665 4186 : ((int64) block < lastblock - BOTTOMUP_TOLERANCE_NBLOCKS ||
8666 3718 : (int64) block > lastblock + BOTTOMUP_TOLERANCE_NBLOCKS))
8667 : break;
8668 :
8669 4506 : nblocksfavorable++;
8670 4506 : lastblock = block;
8671 : }
8672 :
8673 : /* Always indicate that there is at least 1 favorable block */
8674 : Assert(nblocksfavorable >= 1);
8675 :
8676 1965 : return nblocksfavorable;
8677 : }
8678 :
8679 : /*
8680 : * qsort comparison function for bottomup_sort_and_shrink()
8681 : */
8682 : static int
8683 192552 : bottomup_sort_and_shrink_cmp(const void *arg1, const void *arg2)
8684 : {
8685 192552 : const IndexDeleteCounts *group1 = (const IndexDeleteCounts *) arg1;
8686 192552 : const IndexDeleteCounts *group2 = (const IndexDeleteCounts *) arg2;
8687 :
8688 : /*
8689 : * Most significant field is npromisingtids (which we invert the order of
8690 : * so as to sort in desc order).
8691 : *
8692 : * Caller should have already normalized npromisingtids fields into
8693 : * power-of-two values (buckets).
8694 : */
8695 192552 : if (group1->npromisingtids > group2->npromisingtids)
8696 9174 : return -1;
8697 183378 : if (group1->npromisingtids < group2->npromisingtids)
8698 10417 : return 1;
8699 :
8700 : /*
8701 : * Tiebreak: desc ntids sort order.
8702 : *
8703 : * We cannot expect power-of-two values for ntids fields. We should
8704 : * behave as if they were already rounded up for us instead.
8705 : */
8706 172961 : if (group1->ntids != group2->ntids)
8707 : {
8708 123513 : uint32 ntids1 = pg_nextpower2_32((uint32) group1->ntids);
8709 123513 : uint32 ntids2 = pg_nextpower2_32((uint32) group2->ntids);
8710 :
8711 123513 : if (ntids1 > ntids2)
8712 17963 : return -1;
8713 105550 : if (ntids1 < ntids2)
8714 22425 : return 1;
8715 : }
8716 :
8717 : /*
8718 : * Tiebreak: asc offset-into-deltids-for-block (offset to first TID for
8719 : * block in deltids array) order.
8720 : *
8721 : * This is equivalent to sorting in ascending heap block number order
8722 : * (among otherwise equal subsets of the array). This approach allows us
8723 : * to avoid accessing the out-of-line TID. (We rely on the assumption
8724 : * that the deltids array was sorted in ascending heap TID order when
8725 : * these offsets to the first TID from each heap block group were formed.)
8726 : */
8727 132573 : if (group1->ifirsttid > group2->ifirsttid)
8728 64281 : return 1;
8729 68292 : if (group1->ifirsttid < group2->ifirsttid)
8730 68292 : return -1;
8731 :
8732 0 : pg_unreachable();
8733 :
8734 : return 0;
8735 : }
8736 :
8737 : /*
8738 : * heap_index_delete_tuples() helper function for bottom-up deletion callers.
8739 : *
8740 : * Sorts deltids array in the order needed for useful processing by bottom-up
8741 : * deletion. The array should already be sorted in TID order when we're
8742 : * called. The sort process groups heap TIDs from deltids into heap block
8743 : * groupings. Earlier/more-promising groups/blocks are usually those that are
8744 : * known to have the most "promising" TIDs.
8745 : *
8746 : * Sets new size of deltids array (ndeltids) in state. deltids will only have
8747 : * TIDs from the BOTTOMUP_MAX_NBLOCKS most promising heap blocks when we
8748 : * return. This often means that deltids will be shrunk to a small fraction
8749 : * of its original size (we eliminate many heap blocks from consideration for
8750 : * caller up front).
8751 : *
8752 : * Returns the number of "favorable" blocks. See bottomup_nblocksfavorable()
8753 : * for a definition and full details.
8754 : */
8755 : static int
8756 1965 : bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate)
8757 : {
8758 : IndexDeleteCounts *blockgroups;
8759 : TM_IndexDelete *reordereddeltids;
8760 1965 : BlockNumber curblock = InvalidBlockNumber;
8761 1965 : int nblockgroups = 0;
8762 1965 : int ncopied = 0;
8763 1965 : int nblocksfavorable = 0;
8764 :
8765 : Assert(delstate->bottomup);
8766 : Assert(delstate->ndeltids > 0);
8767 :
8768 : /* Calculate per-heap-block count of TIDs */
8769 1965 : blockgroups = palloc_array(IndexDeleteCounts, delstate->ndeltids);
8770 951352 : for (int i = 0; i < delstate->ndeltids; i++)
8771 : {
8772 949387 : TM_IndexDelete *ideltid = &delstate->deltids[i];
8773 949387 : TM_IndexStatus *istatus = delstate->status + ideltid->id;
8774 949387 : ItemPointer htid = &ideltid->tid;
8775 949387 : bool promising = istatus->promising;
8776 :
8777 949387 : if (curblock != ItemPointerGetBlockNumber(htid))
8778 : {
8779 : /* New block group */
8780 37693 : nblockgroups++;
8781 :
8782 : Assert(curblock < ItemPointerGetBlockNumber(htid) ||
8783 : !BlockNumberIsValid(curblock));
8784 :
8785 37693 : curblock = ItemPointerGetBlockNumber(htid);
8786 37693 : blockgroups[nblockgroups - 1].ifirsttid = i;
8787 37693 : blockgroups[nblockgroups - 1].ntids = 1;
8788 37693 : blockgroups[nblockgroups - 1].npromisingtids = 0;
8789 : }
8790 : else
8791 : {
8792 911694 : blockgroups[nblockgroups - 1].ntids++;
8793 : }
8794 :
8795 949387 : if (promising)
8796 121665 : blockgroups[nblockgroups - 1].npromisingtids++;
8797 : }
8798 :
8799 : /*
8800 : * We're about ready to sort block groups to determine the optimal order
8801 : * for visiting heap blocks. But before we do, round the number of
8802 : * promising tuples for each block group up to the next power-of-two,
8803 : * unless it is very low (less than 4), in which case we round up to 4.
8804 : * npromisingtids is far too noisy to trust when choosing between a pair
8805 : * of block groups that both have very low values.
8806 : *
8807 : * This scheme divides heap blocks/block groups into buckets. Each bucket
8808 : * contains blocks that have _approximately_ the same number of promising
8809 : * TIDs as each other. The goal is to ignore relatively small differences
8810 : * in the total number of promising entries, so that the whole process can
8811 : * give a little weight to heapam factors (like heap block locality)
8812 : * instead. This isn't a trade-off, really -- we have nothing to lose. It
8813 : * would be foolish to interpret small differences in npromisingtids
8814 : * values as anything more than noise.
8815 : *
8816 : * We tiebreak on nhtids when sorting block group subsets that have the
8817 : * same npromisingtids, but this has the same issues as npromisingtids,
8818 : * and so nhtids is subject to the same power-of-two bucketing scheme. The
8819 : * only reason that we don't fix nhtids in the same way here too is that
8820 : * we'll need accurate nhtids values after the sort. We handle nhtids
8821 : * bucketization dynamically instead (in the sort comparator).
8822 : *
8823 : * See bottomup_nblocksfavorable() for a full explanation of when and how
8824 : * heap locality/favorable blocks can significantly influence when and how
8825 : * heap blocks are accessed.
8826 : */
8827 39658 : for (int b = 0; b < nblockgroups; b++)
8828 : {
8829 37693 : IndexDeleteCounts *group = blockgroups + b;
8830 :
8831 : /* Better off falling back on nhtids with low npromisingtids */
8832 37693 : if (group->npromisingtids <= 4)
8833 32190 : group->npromisingtids = 4;
8834 : else
8835 5503 : group->npromisingtids =
8836 5503 : pg_nextpower2_32((uint32) group->npromisingtids);
8837 : }
8838 :
8839 : /* Sort groups and rearrange caller's deltids array */
8840 1965 : qsort(blockgroups, nblockgroups, sizeof(IndexDeleteCounts),
8841 : bottomup_sort_and_shrink_cmp);
8842 1965 : reordereddeltids = palloc(delstate->ndeltids * sizeof(TM_IndexDelete));
8843 :
8844 1965 : nblockgroups = Min(BOTTOMUP_MAX_NBLOCKS, nblockgroups);
8845 : /* Determine number of favorable blocks at the start of final deltids */
8846 1965 : nblocksfavorable = bottomup_nblocksfavorable(blockgroups, nblockgroups,
8847 : delstate->deltids);
8848 :
8849 13146 : for (int b = 0; b < nblockgroups; b++)
8850 : {
8851 11181 : IndexDeleteCounts *group = blockgroups + b;
8852 11181 : TM_IndexDelete *firstdtid = delstate->deltids + group->ifirsttid;
8853 :
8854 11181 : memcpy(reordereddeltids + ncopied, firstdtid,
8855 11181 : sizeof(TM_IndexDelete) * group->ntids);
8856 11181 : ncopied += group->ntids;
8857 : }
8858 :
8859 : /* Copy final grouped and sorted TIDs back into start of caller's array */
8860 1965 : memcpy(delstate->deltids, reordereddeltids,
8861 : sizeof(TM_IndexDelete) * ncopied);
8862 1965 : delstate->ndeltids = ncopied;
8863 :
8864 1965 : pfree(reordereddeltids);
8865 1965 : pfree(blockgroups);
8866 :
8867 1965 : return nblocksfavorable;
8868 : }
8869 :
8870 : /*
8871 : * Perform XLogInsert for a heap-visible operation. 'block' is the block
8872 : * being marked all-visible, and vm_buffer is the buffer containing the
8873 : * corresponding visibility map block. Both should have already been modified
8874 : * and dirtied.
8875 : *
8876 : * snapshotConflictHorizon comes from the largest xmin on the page being
8877 : * marked all-visible. REDO routine uses it to generate recovery conflicts.
8878 : *
8879 : * If checksums or wal_log_hints are enabled, we may also generate a full-page
8880 : * image of heap_buffer. Otherwise, we optimize away the FPI (by specifying
8881 : * REGBUF_NO_IMAGE for the heap buffer), in which case the caller should *not*
8882 : * update the heap page's LSN.
8883 : */
8884 : XLogRecPtr
8885 35535 : log_heap_visible(Relation rel, Buffer heap_buffer, Buffer vm_buffer,
8886 : TransactionId snapshotConflictHorizon, uint8 vmflags)
8887 : {
8888 : xl_heap_visible xlrec;
8889 : XLogRecPtr recptr;
8890 : uint8 flags;
8891 :
8892 : Assert(BufferIsValid(heap_buffer));
8893 : Assert(BufferIsValid(vm_buffer));
8894 :
8895 35535 : xlrec.snapshotConflictHorizon = snapshotConflictHorizon;
8896 35535 : xlrec.flags = vmflags;
8897 35535 : if (RelationIsAccessibleInLogicalDecoding(rel))
8898 63 : xlrec.flags |= VISIBILITYMAP_XLOG_CATALOG_REL;
8899 35535 : XLogBeginInsert();
8900 35535 : XLogRegisterData(&xlrec, SizeOfHeapVisible);
8901 :
8902 35535 : XLogRegisterBuffer(0, vm_buffer, 0);
8903 :
8904 35535 : flags = REGBUF_STANDARD;
8905 35535 : if (!XLogHintBitIsNeeded())
8906 3135 : flags |= REGBUF_NO_IMAGE;
8907 35535 : XLogRegisterBuffer(1, heap_buffer, flags);
8908 :
8909 35535 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_VISIBLE);
8910 :
8911 35535 : return recptr;
8912 : }
8913 :
8914 : /*
8915 : * Perform XLogInsert for a heap-update operation. Caller must already
8916 : * have modified the buffer(s) and marked them dirty.
8917 : */
8918 : static XLogRecPtr
8919 299371 : log_heap_update(Relation reln, Buffer oldbuf,
8920 : Buffer newbuf, HeapTuple oldtup, HeapTuple newtup,
8921 : HeapTuple old_key_tuple,
8922 : bool all_visible_cleared, bool new_all_visible_cleared)
8923 : {
8924 : xl_heap_update xlrec;
8925 : xl_heap_header xlhdr;
8926 : xl_heap_header xlhdr_idx;
8927 : uint8 info;
8928 : uint16 prefix_suffix[2];
8929 299371 : uint16 prefixlen = 0,
8930 299371 : suffixlen = 0;
8931 : XLogRecPtr recptr;
8932 299371 : Page page = BufferGetPage(newbuf);
8933 299371 : bool need_tuple_data = RelationIsLogicallyLogged(reln);
8934 : bool init;
8935 : int bufflags;
8936 :
8937 : /* Caller should not call me on a non-WAL-logged relation */
8938 : Assert(RelationNeedsWAL(reln));
8939 :
8940 299371 : XLogBeginInsert();
8941 :
8942 299371 : if (HeapTupleIsHeapOnly(newtup))
8943 146939 : info = XLOG_HEAP_HOT_UPDATE;
8944 : else
8945 152432 : info = XLOG_HEAP_UPDATE;
8946 :
8947 : /*
8948 : * If the old and new tuple are on the same page, we only need to log the
8949 : * parts of the new tuple that were changed. That saves on the amount of
8950 : * WAL we need to write. Currently, we just count any unchanged bytes in
8951 : * the beginning and end of the tuple. That's quick to check, and
8952 : * perfectly covers the common case that only one field is updated.
8953 : *
8954 : * We could do this even if the old and new tuple are on different pages,
8955 : * but only if we don't make a full-page image of the old page, which is
8956 : * difficult to know in advance. Also, if the old tuple is corrupt for
8957 : * some reason, it would allow the corruption to propagate the new page,
8958 : * so it seems best to avoid. Under the general assumption that most
8959 : * updates tend to create the new tuple version on the same page, there
8960 : * isn't much to be gained by doing this across pages anyway.
8961 : *
8962 : * Skip this if we're taking a full-page image of the new page, as we
8963 : * don't include the new tuple in the WAL record in that case. Also
8964 : * disable if effective_wal_level='logical', as logical decoding needs to
8965 : * be able to read the new tuple in whole from the WAL record alone.
8966 : */
8967 299371 : if (oldbuf == newbuf && !need_tuple_data &&
8968 147335 : !XLogCheckBufferNeedsBackup(newbuf))
8969 : {
8970 146716 : char *oldp = (char *) oldtup->t_data + oldtup->t_data->t_hoff;
8971 146716 : char *newp = (char *) newtup->t_data + newtup->t_data->t_hoff;
8972 146716 : int oldlen = oldtup->t_len - oldtup->t_data->t_hoff;
8973 146716 : int newlen = newtup->t_len - newtup->t_data->t_hoff;
8974 :
8975 : /* Check for common prefix between old and new tuple */
8976 12331801 : for (prefixlen = 0; prefixlen < Min(oldlen, newlen); prefixlen++)
8977 : {
8978 12305637 : if (newp[prefixlen] != oldp[prefixlen])
8979 120552 : break;
8980 : }
8981 :
8982 : /*
8983 : * Storing the length of the prefix takes 2 bytes, so we need to save
8984 : * at least 3 bytes or there's no point.
8985 : */
8986 146716 : if (prefixlen < 3)
8987 22105 : prefixlen = 0;
8988 :
8989 : /* Same for suffix */
8990 4828440 : for (suffixlen = 0; suffixlen < Min(oldlen, newlen) - prefixlen; suffixlen++)
8991 : {
8992 4802017 : if (newp[newlen - suffixlen - 1] != oldp[oldlen - suffixlen - 1])
8993 120293 : break;
8994 : }
8995 146716 : if (suffixlen < 3)
8996 36955 : suffixlen = 0;
8997 : }
8998 :
8999 : /* Prepare main WAL data chain */
9000 299371 : xlrec.flags = 0;
9001 299371 : if (all_visible_cleared)
9002 1602 : xlrec.flags |= XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED;
9003 299371 : if (new_all_visible_cleared)
9004 926 : xlrec.flags |= XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED;
9005 299371 : if (prefixlen > 0)
9006 124611 : xlrec.flags |= XLH_UPDATE_PREFIX_FROM_OLD;
9007 299371 : if (suffixlen > 0)
9008 109761 : xlrec.flags |= XLH_UPDATE_SUFFIX_FROM_OLD;
9009 299371 : if (need_tuple_data)
9010 : {
9011 47022 : xlrec.flags |= XLH_UPDATE_CONTAINS_NEW_TUPLE;
9012 47022 : if (old_key_tuple)
9013 : {
9014 146 : if (reln->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
9015 65 : xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_TUPLE;
9016 : else
9017 81 : xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_KEY;
9018 : }
9019 : }
9020 :
9021 : /* If new tuple is the single and first tuple on page... */
9022 302945 : if (ItemPointerGetOffsetNumber(&(newtup->t_self)) == FirstOffsetNumber &&
9023 3574 : PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
9024 : {
9025 3275 : info |= XLOG_HEAP_INIT_PAGE;
9026 3275 : init = true;
9027 : }
9028 : else
9029 296096 : init = false;
9030 :
9031 : /* Prepare WAL data for the old page */
9032 299371 : xlrec.old_offnum = ItemPointerGetOffsetNumber(&oldtup->t_self);
9033 299371 : xlrec.old_xmax = HeapTupleHeaderGetRawXmax(oldtup->t_data);
9034 598742 : xlrec.old_infobits_set = compute_infobits(oldtup->t_data->t_infomask,
9035 299371 : oldtup->t_data->t_infomask2);
9036 :
9037 : /* Prepare WAL data for the new page */
9038 299371 : xlrec.new_offnum = ItemPointerGetOffsetNumber(&newtup->t_self);
9039 299371 : xlrec.new_xmax = HeapTupleHeaderGetRawXmax(newtup->t_data);
9040 :
9041 299371 : bufflags = REGBUF_STANDARD;
9042 299371 : if (init)
9043 3275 : bufflags |= REGBUF_WILL_INIT;
9044 299371 : if (need_tuple_data)
9045 47022 : bufflags |= REGBUF_KEEP_DATA;
9046 :
9047 299371 : XLogRegisterBuffer(0, newbuf, bufflags);
9048 299371 : if (oldbuf != newbuf)
9049 140094 : XLogRegisterBuffer(1, oldbuf, REGBUF_STANDARD);
9050 :
9051 299371 : XLogRegisterData(&xlrec, SizeOfHeapUpdate);
9052 :
9053 : /*
9054 : * Prepare WAL data for the new tuple.
9055 : */
9056 299371 : if (prefixlen > 0 || suffixlen > 0)
9057 : {
9058 146248 : if (prefixlen > 0 && suffixlen > 0)
9059 : {
9060 88124 : prefix_suffix[0] = prefixlen;
9061 88124 : prefix_suffix[1] = suffixlen;
9062 88124 : XLogRegisterBufData(0, &prefix_suffix, sizeof(uint16) * 2);
9063 : }
9064 58124 : else if (prefixlen > 0)
9065 : {
9066 36487 : XLogRegisterBufData(0, &prefixlen, sizeof(uint16));
9067 : }
9068 : else
9069 : {
9070 21637 : XLogRegisterBufData(0, &suffixlen, sizeof(uint16));
9071 : }
9072 : }
9073 :
9074 299371 : xlhdr.t_infomask2 = newtup->t_data->t_infomask2;
9075 299371 : xlhdr.t_infomask = newtup->t_data->t_infomask;
9076 299371 : xlhdr.t_hoff = newtup->t_data->t_hoff;
9077 : Assert(SizeofHeapTupleHeader + prefixlen + suffixlen <= newtup->t_len);
9078 :
9079 : /*
9080 : * PG73FORMAT: write bitmap [+ padding] [+ oid] + data
9081 : *
9082 : * The 'data' doesn't include the common prefix or suffix.
9083 : */
9084 299371 : XLogRegisterBufData(0, &xlhdr, SizeOfHeapHeader);
9085 299371 : if (prefixlen == 0)
9086 : {
9087 174760 : XLogRegisterBufData(0,
9088 174760 : (char *) newtup->t_data + SizeofHeapTupleHeader,
9089 174760 : newtup->t_len - SizeofHeapTupleHeader - suffixlen);
9090 : }
9091 : else
9092 : {
9093 : /*
9094 : * Have to write the null bitmap and data after the common prefix as
9095 : * two separate rdata entries.
9096 : */
9097 : /* bitmap [+ padding] [+ oid] */
9098 124611 : if (newtup->t_data->t_hoff - SizeofHeapTupleHeader > 0)
9099 : {
9100 124611 : XLogRegisterBufData(0,
9101 124611 : (char *) newtup->t_data + SizeofHeapTupleHeader,
9102 124611 : newtup->t_data->t_hoff - SizeofHeapTupleHeader);
9103 : }
9104 :
9105 : /* data after common prefix */
9106 124611 : XLogRegisterBufData(0,
9107 124611 : (char *) newtup->t_data + newtup->t_data->t_hoff + prefixlen,
9108 124611 : newtup->t_len - newtup->t_data->t_hoff - prefixlen - suffixlen);
9109 : }
9110 :
9111 : /* We need to log a tuple identity */
9112 299371 : if (need_tuple_data && old_key_tuple)
9113 : {
9114 : /* don't really need this, but its more comfy to decode */
9115 146 : xlhdr_idx.t_infomask2 = old_key_tuple->t_data->t_infomask2;
9116 146 : xlhdr_idx.t_infomask = old_key_tuple->t_data->t_infomask;
9117 146 : xlhdr_idx.t_hoff = old_key_tuple->t_data->t_hoff;
9118 :
9119 146 : XLogRegisterData(&xlhdr_idx, SizeOfHeapHeader);
9120 :
9121 : /* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
9122 146 : XLogRegisterData((char *) old_key_tuple->t_data + SizeofHeapTupleHeader,
9123 146 : old_key_tuple->t_len - SizeofHeapTupleHeader);
9124 : }
9125 :
9126 : /* filtering by origin on a row level is much more efficient */
9127 299371 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
9128 :
9129 299371 : recptr = XLogInsert(RM_HEAP_ID, info);
9130 :
9131 299371 : return recptr;
9132 : }
9133 :
9134 : /*
9135 : * Perform XLogInsert of an XLOG_HEAP2_NEW_CID record
9136 : *
9137 : * This is only used when effective_wal_level is logical, and only for
9138 : * catalog tuples.
9139 : */
9140 : static XLogRecPtr
9141 24661 : log_heap_new_cid(Relation relation, HeapTuple tup)
9142 : {
9143 : xl_heap_new_cid xlrec;
9144 :
9145 : XLogRecPtr recptr;
9146 24661 : HeapTupleHeader hdr = tup->t_data;
9147 :
9148 : Assert(ItemPointerIsValid(&tup->t_self));
9149 : Assert(tup->t_tableOid != InvalidOid);
9150 :
9151 24661 : xlrec.top_xid = GetTopTransactionId();
9152 24661 : xlrec.target_locator = relation->rd_locator;
9153 24661 : xlrec.target_tid = tup->t_self;
9154 :
9155 : /*
9156 : * If the tuple got inserted & deleted in the same TX we definitely have a
9157 : * combo CID, set cmin and cmax.
9158 : */
9159 24661 : if (hdr->t_infomask & HEAP_COMBOCID)
9160 : {
9161 : Assert(!(hdr->t_infomask & HEAP_XMAX_INVALID));
9162 : Assert(!HeapTupleHeaderXminInvalid(hdr));
9163 2024 : xlrec.cmin = HeapTupleHeaderGetCmin(hdr);
9164 2024 : xlrec.cmax = HeapTupleHeaderGetCmax(hdr);
9165 2024 : xlrec.combocid = HeapTupleHeaderGetRawCommandId(hdr);
9166 : }
9167 : /* No combo CID, so only cmin or cmax can be set by this TX */
9168 : else
9169 : {
9170 : /*
9171 : * Tuple inserted.
9172 : *
9173 : * We need to check for LOCK ONLY because multixacts might be
9174 : * transferred to the new tuple in case of FOR KEY SHARE updates in
9175 : * which case there will be an xmax, although the tuple just got
9176 : * inserted.
9177 : */
9178 29473 : if (hdr->t_infomask & HEAP_XMAX_INVALID ||
9179 6836 : HEAP_XMAX_IS_LOCKED_ONLY(hdr->t_infomask))
9180 : {
9181 15802 : xlrec.cmin = HeapTupleHeaderGetRawCommandId(hdr);
9182 15802 : xlrec.cmax = InvalidCommandId;
9183 : }
9184 : /* Tuple from a different tx updated or deleted. */
9185 : else
9186 : {
9187 6835 : xlrec.cmin = InvalidCommandId;
9188 6835 : xlrec.cmax = HeapTupleHeaderGetRawCommandId(hdr);
9189 : }
9190 22637 : xlrec.combocid = InvalidCommandId;
9191 : }
9192 :
9193 : /*
9194 : * Note that we don't need to register the buffer here, because this
9195 : * operation does not modify the page. The insert/update/delete that
9196 : * called us certainly did, but that's WAL-logged separately.
9197 : */
9198 24661 : XLogBeginInsert();
9199 24661 : XLogRegisterData(&xlrec, SizeOfHeapNewCid);
9200 :
9201 : /* will be looked at irrespective of origin */
9202 :
9203 24661 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_NEW_CID);
9204 :
9205 24661 : return recptr;
9206 : }
9207 :
9208 : /*
9209 : * Build a heap tuple representing the configured REPLICA IDENTITY to represent
9210 : * the old tuple in an UPDATE or DELETE.
9211 : *
9212 : * Returns NULL if there's no need to log an identity or if there's no suitable
9213 : * key defined.
9214 : *
9215 : * Pass key_required true if any replica identity columns changed value, or if
9216 : * any of them have any external data. Delete must always pass true.
9217 : *
9218 : * *copy is set to true if the returned tuple is a modified copy rather than
9219 : * the same tuple that was passed in.
9220 : */
9221 : static HeapTuple
9222 1838098 : ExtractReplicaIdentity(Relation relation, HeapTuple tp, bool key_required,
9223 : bool *copy)
9224 : {
9225 1838098 : TupleDesc desc = RelationGetDescr(relation);
9226 1838098 : char replident = relation->rd_rel->relreplident;
9227 : Bitmapset *idattrs;
9228 : HeapTuple key_tuple;
9229 : bool nulls[MaxHeapAttributeNumber];
9230 : Datum values[MaxHeapAttributeNumber];
9231 :
9232 1838098 : *copy = false;
9233 :
9234 1838098 : if (!RelationIsLogicallyLogged(relation))
9235 1737806 : return NULL;
9236 :
9237 100292 : if (replident == REPLICA_IDENTITY_NOTHING)
9238 231 : return NULL;
9239 :
9240 100061 : if (replident == REPLICA_IDENTITY_FULL)
9241 : {
9242 : /*
9243 : * When logging the entire old tuple, it very well could contain
9244 : * toasted columns. If so, force them to be inlined.
9245 : */
9246 197 : if (HeapTupleHasExternal(tp))
9247 : {
9248 4 : *copy = true;
9249 4 : tp = toast_flatten_tuple(tp, desc);
9250 : }
9251 197 : return tp;
9252 : }
9253 :
9254 : /* if the key isn't required and we're only logging the key, we're done */
9255 99864 : if (!key_required)
9256 46876 : return NULL;
9257 :
9258 : /* find out the replica identity columns */
9259 52988 : idattrs = RelationGetIndexAttrBitmap(relation,
9260 : INDEX_ATTR_BITMAP_IDENTITY_KEY);
9261 :
9262 : /*
9263 : * If there's no defined replica identity columns, treat as !key_required.
9264 : * (This case should not be reachable from heap_update, since that should
9265 : * calculate key_required accurately. But heap_delete just passes
9266 : * constant true for key_required, so we can hit this case in deletes.)
9267 : */
9268 52988 : if (bms_is_empty(idattrs))
9269 6021 : return NULL;
9270 :
9271 : /*
9272 : * Construct a new tuple containing only the replica identity columns,
9273 : * with nulls elsewhere. While we're at it, assert that the replica
9274 : * identity columns aren't null.
9275 : */
9276 46967 : heap_deform_tuple(tp, desc, values, nulls);
9277 :
9278 150895 : for (int i = 0; i < desc->natts; i++)
9279 : {
9280 103928 : if (bms_is_member(i + 1 - FirstLowInvalidHeapAttributeNumber,
9281 : idattrs))
9282 : Assert(!nulls[i]);
9283 : else
9284 56949 : nulls[i] = true;
9285 : }
9286 :
9287 46967 : key_tuple = heap_form_tuple(desc, values, nulls);
9288 46967 : *copy = true;
9289 :
9290 46967 : bms_free(idattrs);
9291 :
9292 : /*
9293 : * If the tuple, which by here only contains indexed columns, still has
9294 : * toasted columns, force them to be inlined. This is somewhat unlikely
9295 : * since there's limits on the size of indexed columns, so we don't
9296 : * duplicate toast_flatten_tuple()s functionality in the above loop over
9297 : * the indexed columns, even if it would be more efficient.
9298 : */
9299 46967 : if (HeapTupleHasExternal(key_tuple))
9300 : {
9301 4 : HeapTuple oldtup = key_tuple;
9302 :
9303 4 : key_tuple = toast_flatten_tuple(oldtup, desc);
9304 4 : heap_freetuple(oldtup);
9305 : }
9306 :
9307 46967 : return key_tuple;
9308 : }
9309 :
9310 : /*
9311 : * HeapCheckForSerializableConflictOut
9312 : * We are reading a tuple. If it's not visible, there may be a
9313 : * rw-conflict out with the inserter. Otherwise, if it is visible to us
9314 : * but has been deleted, there may be a rw-conflict out with the deleter.
9315 : *
9316 : * We will determine the top level xid of the writing transaction with which
9317 : * we may be in conflict, and ask CheckForSerializableConflictOut() to check
9318 : * for overlap with our own transaction.
9319 : *
9320 : * This function should be called just about anywhere in heapam.c where a
9321 : * tuple has been read. The caller must hold at least a shared lock on the
9322 : * buffer, because this function might set hint bits on the tuple. There is
9323 : * currently no known reason to call this function from an index AM.
9324 : */
9325 : void
9326 32450189 : HeapCheckForSerializableConflictOut(bool visible, Relation relation,
9327 : HeapTuple tuple, Buffer buffer,
9328 : Snapshot snapshot)
9329 : {
9330 : TransactionId xid;
9331 : HTSV_Result htsvResult;
9332 :
9333 32450189 : if (!CheckForSerializableConflictOutNeeded(relation, snapshot))
9334 32424815 : return;
9335 :
9336 : /*
9337 : * Check to see whether the tuple has been written to by a concurrent
9338 : * transaction, either to create it not visible to us, or to delete it
9339 : * while it is visible to us. The "visible" bool indicates whether the
9340 : * tuple is visible to us, while HeapTupleSatisfiesVacuum checks what else
9341 : * is going on with it.
9342 : *
9343 : * In the event of a concurrently inserted tuple that also happens to have
9344 : * been concurrently updated (by a separate transaction), the xmin of the
9345 : * tuple will be used -- not the updater's xid.
9346 : */
9347 25374 : htsvResult = HeapTupleSatisfiesVacuum(tuple, TransactionXmin, buffer);
9348 25374 : switch (htsvResult)
9349 : {
9350 24561 : case HEAPTUPLE_LIVE:
9351 24561 : if (visible)
9352 24548 : return;
9353 13 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9354 13 : break;
9355 361 : case HEAPTUPLE_RECENTLY_DEAD:
9356 : case HEAPTUPLE_DELETE_IN_PROGRESS:
9357 361 : if (visible)
9358 285 : xid = HeapTupleHeaderGetUpdateXid(tuple->t_data);
9359 : else
9360 76 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9361 :
9362 361 : if (TransactionIdPrecedes(xid, TransactionXmin))
9363 : {
9364 : /* This is like the HEAPTUPLE_DEAD case */
9365 : Assert(!visible);
9366 67 : return;
9367 : }
9368 294 : break;
9369 328 : case HEAPTUPLE_INSERT_IN_PROGRESS:
9370 328 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9371 328 : break;
9372 124 : case HEAPTUPLE_DEAD:
9373 : Assert(!visible);
9374 124 : return;
9375 0 : default:
9376 :
9377 : /*
9378 : * The only way to get to this default clause is if a new value is
9379 : * added to the enum type without adding it to this switch
9380 : * statement. That's a bug, so elog.
9381 : */
9382 0 : elog(ERROR, "unrecognized return value from HeapTupleSatisfiesVacuum: %u", htsvResult);
9383 :
9384 : /*
9385 : * In spite of having all enum values covered and calling elog on
9386 : * this default, some compilers think this is a code path which
9387 : * allows xid to be used below without initialization. Silence
9388 : * that warning.
9389 : */
9390 : xid = InvalidTransactionId;
9391 : }
9392 :
9393 : Assert(TransactionIdIsValid(xid));
9394 : Assert(TransactionIdFollowsOrEquals(xid, TransactionXmin));
9395 :
9396 : /*
9397 : * Find top level xid. Bail out if xid is too early to be a conflict, or
9398 : * if it's our own xid.
9399 : */
9400 635 : if (TransactionIdEquals(xid, GetTopTransactionIdIfAny()))
9401 64 : return;
9402 571 : xid = SubTransGetTopmostTransaction(xid);
9403 571 : if (TransactionIdPrecedes(xid, TransactionXmin))
9404 0 : return;
9405 :
9406 571 : CheckForSerializableConflictOut(relation, xid, snapshot);
9407 : }
|
