Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * heapam.c
4 : * heap access method code
5 : *
6 : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/access/heap/heapam.c
12 : *
13 : *
14 : * INTERFACE ROUTINES
15 : * heap_beginscan - begin relation scan
16 : * heap_rescan - restart a relation scan
17 : * heap_endscan - end relation scan
18 : * heap_getnext - retrieve next tuple in scan
19 : * heap_fetch - retrieve tuple with given tid
20 : * heap_insert - insert tuple into a relation
21 : * heap_multi_insert - insert multiple tuples into a relation
22 : * heap_delete - delete a tuple from a relation
23 : * heap_update - replace a tuple in a relation with another tuple
24 : *
25 : * NOTES
26 : * This file contains the heap_ routines which implement
27 : * the POSTGRES heap access method used for all POSTGRES
28 : * relations.
29 : *
30 : *-------------------------------------------------------------------------
31 : */
32 : #include "postgres.h"
33 :
34 : #include "access/heapam.h"
35 : #include "access/heaptoast.h"
36 : #include "access/hio.h"
37 : #include "access/multixact.h"
38 : #include "access/subtrans.h"
39 : #include "access/syncscan.h"
40 : #include "access/valid.h"
41 : #include "access/visibilitymap.h"
42 : #include "access/xloginsert.h"
43 : #include "catalog/pg_database.h"
44 : #include "catalog/pg_database_d.h"
45 : #include "commands/vacuum.h"
46 : #include "pgstat.h"
47 : #include "port/pg_bitutils.h"
48 : #include "storage/lmgr.h"
49 : #include "storage/predicate.h"
50 : #include "storage/procarray.h"
51 : #include "utils/datum.h"
52 : #include "utils/injection_point.h"
53 : #include "utils/inval.h"
54 : #include "utils/spccache.h"
55 : #include "utils/syscache.h"
56 :
57 :
58 : static HeapTuple heap_prepare_insert(Relation relation, HeapTuple tup,
59 : TransactionId xid, CommandId cid, int options);
60 : static XLogRecPtr log_heap_update(Relation reln, Buffer oldbuf,
61 : Buffer newbuf, HeapTuple oldtup,
62 : HeapTuple newtup, HeapTuple old_key_tuple,
63 : bool all_visible_cleared, bool new_all_visible_cleared);
64 : #ifdef USE_ASSERT_CHECKING
65 : static void check_lock_if_inplace_updateable_rel(Relation relation,
66 : const ItemPointerData *otid,
67 : HeapTuple newtup);
68 : static void check_inplace_rel_lock(HeapTuple oldtup);
69 : #endif
70 : static Bitmapset *HeapDetermineColumnsInfo(Relation relation,
71 : Bitmapset *interesting_cols,
72 : Bitmapset *external_cols,
73 : HeapTuple oldtup, HeapTuple newtup,
74 : bool *has_external);
75 : static bool heap_acquire_tuplock(Relation relation, const ItemPointerData *tid,
76 : LockTupleMode mode, LockWaitPolicy wait_policy,
77 : bool *have_tuple_lock);
78 : static inline BlockNumber heapgettup_advance_block(HeapScanDesc scan,
79 : BlockNumber block,
80 : ScanDirection dir);
81 : static pg_noinline BlockNumber heapgettup_initial_block(HeapScanDesc scan,
82 : ScanDirection dir);
83 : static void compute_new_xmax_infomask(TransactionId xmax, uint16 old_infomask,
84 : uint16 old_infomask2, TransactionId add_to_xmax,
85 : LockTupleMode mode, bool is_update,
86 : TransactionId *result_xmax, uint16 *result_infomask,
87 : uint16 *result_infomask2);
88 : static TM_Result heap_lock_updated_tuple(Relation rel,
89 : uint16 prior_infomask,
90 : TransactionId prior_raw_xmax,
91 : const ItemPointerData *prior_ctid,
92 : TransactionId xid,
93 : LockTupleMode mode);
94 : static void GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
95 : uint16 *new_infomask2);
96 : static TransactionId MultiXactIdGetUpdateXid(TransactionId xmax,
97 : uint16 t_infomask);
98 : static bool DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
99 : LockTupleMode lockmode, bool *current_is_member);
100 : static void MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
101 : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
102 : int *remaining);
103 : static bool ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status,
104 : uint16 infomask, Relation rel, int *remaining,
105 : bool logLockFailure);
106 : static void index_delete_sort(TM_IndexDeleteOp *delstate);
107 : static int bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate);
108 : static XLogRecPtr log_heap_new_cid(Relation relation, HeapTuple tup);
109 : static HeapTuple ExtractReplicaIdentity(Relation relation, HeapTuple tp, bool key_required,
110 : bool *copy);
111 :
112 :
113 : /*
114 : * This table lists the heavyweight lock mode that corresponds to each tuple
115 : * lock mode, as well as one or two corresponding MultiXactStatus values:
116 : * .lockstatus to merely lock tuples, and .updstatus to update them. The
117 : * latter is set to -1 if the corresponding tuple lock mode does not allow
118 : * updating tuples -- see get_mxact_status_for_lock().
119 : *
120 : * These interact with InplaceUpdateTupleLock, an alias for ExclusiveLock.
121 : *
122 : * Don't look at lockstatus/updstatus directly! Use get_mxact_status_for_lock
123 : * instead.
124 : */
125 : static const struct
126 : {
127 : LOCKMODE hwlock;
128 : int lockstatus;
129 : int updstatus;
130 : } tupleLockExtraInfo[] =
131 :
132 : {
133 : [LockTupleKeyShare] = {
134 : .hwlock = AccessShareLock,
135 : .lockstatus = MultiXactStatusForKeyShare,
136 : /* KeyShare does not allow updating tuples */
137 : .updstatus = -1
138 : },
139 : [LockTupleShare] = {
140 : .hwlock = RowShareLock,
141 : .lockstatus = MultiXactStatusForShare,
142 : /* Share does not allow updating tuples */
143 : .updstatus = -1
144 : },
145 : [LockTupleNoKeyExclusive] = {
146 : .hwlock = ExclusiveLock,
147 : .lockstatus = MultiXactStatusForNoKeyUpdate,
148 : .updstatus = MultiXactStatusNoKeyUpdate
149 : },
150 : [LockTupleExclusive] = {
151 : .hwlock = AccessExclusiveLock,
152 : .lockstatus = MultiXactStatusForUpdate,
153 : .updstatus = MultiXactStatusUpdate
154 : }
155 : };
156 :
157 : /* Get the LOCKMODE for a given MultiXactStatus */
158 : #define LOCKMODE_from_mxstatus(status) \
159 : (tupleLockExtraInfo[TUPLOCK_from_mxstatus((status))].hwlock)
160 :
161 : /*
162 : * Acquire heavyweight locks on tuples, using a LockTupleMode strength value.
163 : * This is more readable than having every caller translate it to lock.h's
164 : * LOCKMODE.
165 : */
166 : #define LockTupleTuplock(rel, tup, mode) \
167 : LockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
168 : #define UnlockTupleTuplock(rel, tup, mode) \
169 : UnlockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
170 : #define ConditionalLockTupleTuplock(rel, tup, mode, log) \
171 : ConditionalLockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock, (log))
172 :
173 : #ifdef USE_PREFETCH
174 : /*
175 : * heap_index_delete_tuples and index_delete_prefetch_buffer use this
176 : * structure to coordinate prefetching activity
177 : */
178 : typedef struct
179 : {
180 : BlockNumber cur_hblkno;
181 : int next_item;
182 : int ndeltids;
183 : TM_IndexDelete *deltids;
184 : } IndexDeletePrefetchState;
185 : #endif
186 :
187 : /* heap_index_delete_tuples bottom-up index deletion costing constants */
188 : #define BOTTOMUP_MAX_NBLOCKS 6
189 : #define BOTTOMUP_TOLERANCE_NBLOCKS 3
190 :
191 : /*
192 : * heap_index_delete_tuples uses this when determining which heap blocks it
193 : * must visit to help its bottom-up index deletion caller
194 : */
195 : typedef struct IndexDeleteCounts
196 : {
197 : int16 npromisingtids; /* Number of "promising" TIDs in group */
198 : int16 ntids; /* Number of TIDs in group */
199 : int16 ifirsttid; /* Offset to group's first deltid */
200 : } IndexDeleteCounts;
201 :
202 : /*
203 : * This table maps tuple lock strength values for each particular
204 : * MultiXactStatus value.
205 : */
206 : static const int MultiXactStatusLock[MaxMultiXactStatus + 1] =
207 : {
208 : LockTupleKeyShare, /* ForKeyShare */
209 : LockTupleShare, /* ForShare */
210 : LockTupleNoKeyExclusive, /* ForNoKeyUpdate */
211 : LockTupleExclusive, /* ForUpdate */
212 : LockTupleNoKeyExclusive, /* NoKeyUpdate */
213 : LockTupleExclusive /* Update */
214 : };
215 :
216 : /* Get the LockTupleMode for a given MultiXactStatus */
217 : #define TUPLOCK_from_mxstatus(status) \
218 : (MultiXactStatusLock[(status)])
219 :
220 : /*
221 : * Check that we have a valid snapshot if we might need TOAST access.
222 : */
223 : static inline void
224 21193444 : AssertHasSnapshotForToast(Relation rel)
225 : {
226 : #ifdef USE_ASSERT_CHECKING
227 :
228 : /* bootstrap mode in particular breaks this rule */
229 : if (!IsNormalProcessingMode())
230 : return;
231 :
232 : /* if the relation doesn't have a TOAST table, we are good */
233 : if (!OidIsValid(rel->rd_rel->reltoastrelid))
234 : return;
235 :
236 : Assert(HaveRegisteredOrActiveSnapshot());
237 :
238 : #endif /* USE_ASSERT_CHECKING */
239 21193444 : }
240 :
241 : /* ----------------------------------------------------------------
242 : * heap support routines
243 : * ----------------------------------------------------------------
244 : */
245 :
246 : /*
247 : * Streaming read API callback for parallel sequential scans. Returns the next
248 : * block the caller wants from the read stream or InvalidBlockNumber when done.
249 : */
250 : static BlockNumber
251 203416 : heap_scan_stream_read_next_parallel(ReadStream *stream,
252 : void *callback_private_data,
253 : void *per_buffer_data)
254 : {
255 203416 : HeapScanDesc scan = (HeapScanDesc) callback_private_data;
256 :
257 : Assert(ScanDirectionIsForward(scan->rs_dir));
258 : Assert(scan->rs_base.rs_parallel);
259 :
260 203416 : if (unlikely(!scan->rs_inited))
261 : {
262 : /* parallel scan */
263 3292 : table_block_parallelscan_startblock_init(scan->rs_base.rs_rd,
264 3292 : scan->rs_parallelworkerdata,
265 3292 : (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel,
266 : scan->rs_startblock,
267 : scan->rs_numblocks);
268 :
269 : /* may return InvalidBlockNumber if there are no more blocks */
270 6584 : scan->rs_prefetch_block = table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
271 3292 : scan->rs_parallelworkerdata,
272 3292 : (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel);
273 3292 : scan->rs_inited = true;
274 : }
275 : else
276 : {
277 200124 : scan->rs_prefetch_block = table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
278 200124 : scan->rs_parallelworkerdata, (ParallelBlockTableScanDesc)
279 200124 : scan->rs_base.rs_parallel);
280 : }
281 :
282 203416 : return scan->rs_prefetch_block;
283 : }
284 :
285 : /*
286 : * Streaming read API callback for serial sequential and TID range scans.
287 : * Returns the next block the caller wants from the read stream or
288 : * InvalidBlockNumber when done.
289 : */
290 : static BlockNumber
291 7734508 : heap_scan_stream_read_next_serial(ReadStream *stream,
292 : void *callback_private_data,
293 : void *per_buffer_data)
294 : {
295 7734508 : HeapScanDesc scan = (HeapScanDesc) callback_private_data;
296 :
297 7734508 : if (unlikely(!scan->rs_inited))
298 : {
299 1983004 : scan->rs_prefetch_block = heapgettup_initial_block(scan, scan->rs_dir);
300 1983004 : scan->rs_inited = true;
301 : }
302 : else
303 5751504 : scan->rs_prefetch_block = heapgettup_advance_block(scan,
304 : scan->rs_prefetch_block,
305 : scan->rs_dir);
306 :
307 7734508 : return scan->rs_prefetch_block;
308 : }
309 :
310 : /*
311 : * Read stream API callback for bitmap heap scans.
312 : * Returns the next block the caller wants from the read stream or
313 : * InvalidBlockNumber when done.
314 : */
315 : static BlockNumber
316 434088 : bitmapheap_stream_read_next(ReadStream *pgsr, void *private_data,
317 : void *per_buffer_data)
318 : {
319 434088 : TBMIterateResult *tbmres = per_buffer_data;
320 434088 : BitmapHeapScanDesc bscan = (BitmapHeapScanDesc) private_data;
321 434088 : HeapScanDesc hscan = (HeapScanDesc) bscan;
322 434088 : TableScanDesc sscan = &hscan->rs_base;
323 :
324 : for (;;)
325 : {
326 434088 : CHECK_FOR_INTERRUPTS();
327 :
328 : /* no more entries in the bitmap */
329 434088 : if (!tbm_iterate(&sscan->st.rs_tbmiterator, tbmres))
330 26346 : return InvalidBlockNumber;
331 :
332 : /*
333 : * Ignore any claimed entries past what we think is the end of the
334 : * relation. It may have been extended after the start of our scan (we
335 : * only hold an AccessShareLock, and it could be inserts from this
336 : * backend). We don't take this optimization in SERIALIZABLE
337 : * isolation though, as we need to examine all invisible tuples
338 : * reachable by the index.
339 : */
340 407742 : if (!IsolationIsSerializable() &&
341 407524 : tbmres->blockno >= hscan->rs_nblocks)
342 0 : continue;
343 :
344 407742 : return tbmres->blockno;
345 : }
346 :
347 : /* not reachable */
348 : Assert(false);
349 : }
350 :
351 : /* ----------------
352 : * initscan - scan code common to heap_beginscan and heap_rescan
353 : * ----------------
354 : */
355 : static void
356 2032882 : initscan(HeapScanDesc scan, ScanKey key, bool keep_startblock)
357 : {
358 2032882 : ParallelBlockTableScanDesc bpscan = NULL;
359 : bool allow_strat;
360 : bool allow_sync;
361 :
362 : /*
363 : * Determine the number of blocks we have to scan.
364 : *
365 : * It is sufficient to do this once at scan start, since any tuples added
366 : * while the scan is in progress will be invisible to my snapshot anyway.
367 : * (That is not true when using a non-MVCC snapshot. However, we couldn't
368 : * guarantee to return tuples added after scan start anyway, since they
369 : * might go into pages we already scanned. To guarantee consistent
370 : * results for a non-MVCC snapshot, the caller must hold some higher-level
371 : * lock that ensures the interesting tuple(s) won't change.)
372 : */
373 2032882 : if (scan->rs_base.rs_parallel != NULL)
374 : {
375 4470 : bpscan = (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel;
376 4470 : scan->rs_nblocks = bpscan->phs_nblocks;
377 : }
378 : else
379 2028412 : scan->rs_nblocks = RelationGetNumberOfBlocks(scan->rs_base.rs_rd);
380 :
381 : /*
382 : * If the table is large relative to NBuffers, use a bulk-read access
383 : * strategy and enable synchronized scanning (see syncscan.c). Although
384 : * the thresholds for these features could be different, we make them the
385 : * same so that there are only two behaviors to tune rather than four.
386 : * (However, some callers need to be able to disable one or both of these
387 : * behaviors, independently of the size of the table; also there is a GUC
388 : * variable that can disable synchronized scanning.)
389 : *
390 : * Note that table_block_parallelscan_initialize has a very similar test;
391 : * if you change this, consider changing that one, too.
392 : */
393 2032878 : if (!RelationUsesLocalBuffers(scan->rs_base.rs_rd) &&
394 2018224 : scan->rs_nblocks > NBuffers / 4)
395 : {
396 28118 : allow_strat = (scan->rs_base.rs_flags & SO_ALLOW_STRAT) != 0;
397 28118 : allow_sync = (scan->rs_base.rs_flags & SO_ALLOW_SYNC) != 0;
398 : }
399 : else
400 2004760 : allow_strat = allow_sync = false;
401 :
402 2032878 : if (allow_strat)
403 : {
404 : /* During a rescan, keep the previous strategy object. */
405 25454 : if (scan->rs_strategy == NULL)
406 25088 : scan->rs_strategy = GetAccessStrategy(BAS_BULKREAD);
407 : }
408 : else
409 : {
410 2007424 : if (scan->rs_strategy != NULL)
411 0 : FreeAccessStrategy(scan->rs_strategy);
412 2007424 : scan->rs_strategy = NULL;
413 : }
414 :
415 2032878 : if (scan->rs_base.rs_parallel != NULL)
416 : {
417 : /* For parallel scan, believe whatever ParallelTableScanDesc says. */
418 4470 : if (scan->rs_base.rs_parallel->phs_syncscan)
419 4 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
420 : else
421 4466 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
422 :
423 : /*
424 : * If not rescanning, initialize the startblock. Finding the actual
425 : * start location is done in table_block_parallelscan_startblock_init,
426 : * based on whether an alternative start location has been set with
427 : * heap_setscanlimits, or using the syncscan location, when syncscan
428 : * is enabled.
429 : */
430 4470 : if (!keep_startblock)
431 4242 : scan->rs_startblock = InvalidBlockNumber;
432 : }
433 : else
434 : {
435 2028408 : if (keep_startblock)
436 : {
437 : /*
438 : * When rescanning, we want to keep the previous startblock
439 : * setting, so that rewinding a cursor doesn't generate surprising
440 : * results. Reset the active syncscan setting, though.
441 : */
442 1246526 : if (allow_sync && synchronize_seqscans)
443 100 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
444 : else
445 1246426 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
446 : }
447 781882 : else if (allow_sync && synchronize_seqscans)
448 : {
449 144 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
450 144 : scan->rs_startblock = ss_get_location(scan->rs_base.rs_rd, scan->rs_nblocks);
451 : }
452 : else
453 : {
454 781738 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
455 781738 : scan->rs_startblock = 0;
456 : }
457 : }
458 :
459 2032878 : scan->rs_numblocks = InvalidBlockNumber;
460 2032878 : scan->rs_inited = false;
461 2032878 : scan->rs_ctup.t_data = NULL;
462 2032878 : ItemPointerSetInvalid(&scan->rs_ctup.t_self);
463 2032878 : scan->rs_cbuf = InvalidBuffer;
464 2032878 : scan->rs_cblock = InvalidBlockNumber;
465 2032878 : scan->rs_ntuples = 0;
466 2032878 : scan->rs_cindex = 0;
467 :
468 : /*
469 : * Initialize to ForwardScanDirection because it is most common and
470 : * because heap scans go forward before going backward (e.g. CURSORs).
471 : */
472 2032878 : scan->rs_dir = ForwardScanDirection;
473 2032878 : scan->rs_prefetch_block = InvalidBlockNumber;
474 :
475 : /* page-at-a-time fields are always invalid when not rs_inited */
476 :
477 : /*
478 : * copy the scan key, if appropriate
479 : */
480 2032878 : if (key != NULL && scan->rs_base.rs_nkeys > 0)
481 450836 : memcpy(scan->rs_base.rs_key, key, scan->rs_base.rs_nkeys * sizeof(ScanKeyData));
482 :
483 : /*
484 : * Currently, we only have a stats counter for sequential heap scans (but
485 : * e.g for bitmap scans the underlying bitmap index scans will be counted,
486 : * and for sample scans we update stats for tuple fetches).
487 : */
488 2032878 : if (scan->rs_base.rs_flags & SO_TYPE_SEQSCAN)
489 1985534 : pgstat_count_heap_scan(scan->rs_base.rs_rd);
490 2032878 : }
491 :
492 : /*
493 : * heap_setscanlimits - restrict range of a heapscan
494 : *
495 : * startBlk is the page to start at
496 : * numBlks is number of pages to scan (InvalidBlockNumber means "all")
497 : */
498 : void
499 5716 : heap_setscanlimits(TableScanDesc sscan, BlockNumber startBlk, BlockNumber numBlks)
500 : {
501 5716 : HeapScanDesc scan = (HeapScanDesc) sscan;
502 :
503 : Assert(!scan->rs_inited); /* else too late to change */
504 : /* else rs_startblock is significant */
505 : Assert(!(scan->rs_base.rs_flags & SO_ALLOW_SYNC));
506 :
507 : /* Check startBlk is valid (but allow case of zero blocks...) */
508 : Assert(startBlk == 0 || startBlk < scan->rs_nblocks);
509 :
510 5716 : scan->rs_startblock = startBlk;
511 5716 : scan->rs_numblocks = numBlks;
512 5716 : }
513 :
514 : /*
515 : * Per-tuple loop for heap_prepare_pagescan(). Pulled out so it can be called
516 : * multiple times, with constant arguments for all_visible,
517 : * check_serializable.
518 : */
519 : pg_attribute_always_inline
520 : static int
521 5730760 : page_collect_tuples(HeapScanDesc scan, Snapshot snapshot,
522 : Page page, Buffer buffer,
523 : BlockNumber block, int lines,
524 : bool all_visible, bool check_serializable)
525 : {
526 5730760 : Oid relid = RelationGetRelid(scan->rs_base.rs_rd);
527 5730760 : int ntup = 0;
528 5730760 : int nvis = 0;
529 : BatchMVCCState batchmvcc;
530 :
531 : /* page at a time should have been disabled otherwise */
532 : Assert(IsMVCCSnapshot(snapshot));
533 :
534 : /* first find all tuples on the page */
535 289765424 : for (OffsetNumber lineoff = FirstOffsetNumber; lineoff <= lines; lineoff++)
536 : {
537 284034664 : ItemId lpp = PageGetItemId(page, lineoff);
538 : HeapTuple tup;
539 :
540 284034664 : if (unlikely(!ItemIdIsNormal(lpp)))
541 63813276 : continue;
542 :
543 : /*
544 : * If the page is not all-visible or we need to check serializability,
545 : * maintain enough state to be able to refind the tuple efficiently,
546 : * without again first needing to fetch the item and then via that the
547 : * tuple.
548 : */
549 220221388 : if (!all_visible || check_serializable)
550 : {
551 134369774 : tup = &batchmvcc.tuples[ntup];
552 :
553 134369774 : tup->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
554 134369774 : tup->t_len = ItemIdGetLength(lpp);
555 134369774 : tup->t_tableOid = relid;
556 134369774 : ItemPointerSet(&(tup->t_self), block, lineoff);
557 : }
558 :
559 : /*
560 : * If the page is all visible, these fields otherwise won't be
561 : * populated in loop below.
562 : */
563 220221388 : if (all_visible)
564 : {
565 85851614 : if (check_serializable)
566 : {
567 0 : batchmvcc.visible[ntup] = true;
568 : }
569 85851614 : scan->rs_vistuples[ntup] = lineoff;
570 : }
571 :
572 220221388 : ntup++;
573 : }
574 :
575 : Assert(ntup <= MaxHeapTuplesPerPage);
576 :
577 : /*
578 : * Unless the page is all visible, test visibility for all tuples one go.
579 : * That is considerably more efficient than calling
580 : * HeapTupleSatisfiesMVCC() one-by-one.
581 : */
582 5730760 : if (all_visible)
583 2025226 : nvis = ntup;
584 : else
585 3705534 : nvis = HeapTupleSatisfiesMVCCBatch(snapshot, buffer,
586 : ntup,
587 : &batchmvcc,
588 3705534 : scan->rs_vistuples);
589 :
590 : /*
591 : * So far we don't have batch API for testing serializabilty, so do so
592 : * one-by-one.
593 : */
594 5730760 : if (check_serializable)
595 : {
596 4090 : for (int i = 0; i < ntup; i++)
597 : {
598 2842 : HeapCheckForSerializableConflictOut(batchmvcc.visible[i],
599 : scan->rs_base.rs_rd,
600 : &batchmvcc.tuples[i],
601 : buffer, snapshot);
602 : }
603 : }
604 :
605 5730744 : return nvis;
606 : }
607 :
608 : /*
609 : * heap_prepare_pagescan - Prepare current scan page to be scanned in pagemode
610 : *
611 : * Preparation currently consists of 1. prune the scan's rs_cbuf page, and 2.
612 : * fill the rs_vistuples[] array with the OffsetNumbers of visible tuples.
613 : */
614 : void
615 5730760 : heap_prepare_pagescan(TableScanDesc sscan)
616 : {
617 5730760 : HeapScanDesc scan = (HeapScanDesc) sscan;
618 5730760 : Buffer buffer = scan->rs_cbuf;
619 5730760 : BlockNumber block = scan->rs_cblock;
620 : Snapshot snapshot;
621 : Page page;
622 : int lines;
623 : bool all_visible;
624 : bool check_serializable;
625 :
626 : Assert(BufferGetBlockNumber(buffer) == block);
627 :
628 : /* ensure we're not accidentally being used when not in pagemode */
629 : Assert(scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE);
630 5730760 : snapshot = scan->rs_base.rs_snapshot;
631 :
632 : /*
633 : * Prune and repair fragmentation for the whole page, if possible.
634 : */
635 5730760 : heap_page_prune_opt(scan->rs_base.rs_rd, buffer);
636 :
637 : /*
638 : * We must hold share lock on the buffer content while examining tuple
639 : * visibility. Afterwards, however, the tuples we have found to be
640 : * visible are guaranteed good as long as we hold the buffer pin.
641 : */
642 5730760 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
643 :
644 5730760 : page = BufferGetPage(buffer);
645 5730760 : lines = PageGetMaxOffsetNumber(page);
646 :
647 : /*
648 : * If the all-visible flag indicates that all tuples on the page are
649 : * visible to everyone, we can skip the per-tuple visibility tests.
650 : *
651 : * Note: In hot standby, a tuple that's already visible to all
652 : * transactions on the primary might still be invisible to a read-only
653 : * transaction in the standby. We partly handle this problem by tracking
654 : * the minimum xmin of visible tuples as the cut-off XID while marking a
655 : * page all-visible on the primary and WAL log that along with the
656 : * visibility map SET operation. In hot standby, we wait for (or abort)
657 : * all transactions that can potentially may not see one or more tuples on
658 : * the page. That's how index-only scans work fine in hot standby. A
659 : * crucial difference between index-only scans and heap scans is that the
660 : * index-only scan completely relies on the visibility map where as heap
661 : * scan looks at the page-level PD_ALL_VISIBLE flag. We are not sure if
662 : * the page-level flag can be trusted in the same way, because it might
663 : * get propagated somehow without being explicitly WAL-logged, e.g. via a
664 : * full page write. Until we can prove that beyond doubt, let's check each
665 : * tuple for visibility the hard way.
666 : */
667 5730760 : all_visible = PageIsAllVisible(page) && !snapshot->takenDuringRecovery;
668 : check_serializable =
669 5730760 : CheckForSerializableConflictOutNeeded(scan->rs_base.rs_rd, snapshot);
670 :
671 : /*
672 : * We call page_collect_tuples() with constant arguments, to get the
673 : * compiler to constant fold the constant arguments. Separate calls with
674 : * constant arguments, rather than variables, are needed on several
675 : * compilers to actually perform constant folding.
676 : */
677 5730760 : if (likely(all_visible))
678 : {
679 2025226 : if (likely(!check_serializable))
680 2025226 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
681 : block, lines, true, false);
682 : else
683 0 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
684 : block, lines, true, true);
685 : }
686 : else
687 : {
688 3705534 : if (likely(!check_serializable))
689 3704270 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
690 : block, lines, false, false);
691 : else
692 1264 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
693 : block, lines, false, true);
694 : }
695 :
696 5730744 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
697 5730744 : }
698 :
699 : /*
700 : * heap_fetch_next_buffer - read and pin the next block from MAIN_FORKNUM.
701 : *
702 : * Read the next block of the scan relation from the read stream and save it
703 : * in the scan descriptor. It is already pinned.
704 : */
705 : static inline void
706 7561116 : heap_fetch_next_buffer(HeapScanDesc scan, ScanDirection dir)
707 : {
708 : Assert(scan->rs_read_stream);
709 :
710 : /* release previous scan buffer, if any */
711 7561116 : if (BufferIsValid(scan->rs_cbuf))
712 : {
713 5574818 : ReleaseBuffer(scan->rs_cbuf);
714 5574818 : scan->rs_cbuf = InvalidBuffer;
715 : }
716 :
717 : /*
718 : * Be sure to check for interrupts at least once per page. Checks at
719 : * higher code levels won't be able to stop a seqscan that encounters many
720 : * pages' worth of consecutive dead tuples.
721 : */
722 7561116 : CHECK_FOR_INTERRUPTS();
723 :
724 : /*
725 : * If the scan direction is changing, reset the prefetch block to the
726 : * current block. Otherwise, we will incorrectly prefetch the blocks
727 : * between the prefetch block and the current block again before
728 : * prefetching blocks in the new, correct scan direction.
729 : */
730 7561114 : if (unlikely(scan->rs_dir != dir))
731 : {
732 154 : scan->rs_prefetch_block = scan->rs_cblock;
733 154 : read_stream_reset(scan->rs_read_stream);
734 : }
735 :
736 7561114 : scan->rs_dir = dir;
737 :
738 7561114 : scan->rs_cbuf = read_stream_next_buffer(scan->rs_read_stream, NULL);
739 7561064 : if (BufferIsValid(scan->rs_cbuf))
740 5907476 : scan->rs_cblock = BufferGetBlockNumber(scan->rs_cbuf);
741 7561064 : }
742 :
743 : /*
744 : * heapgettup_initial_block - return the first BlockNumber to scan
745 : *
746 : * Returns InvalidBlockNumber when there are no blocks to scan. This can
747 : * occur with empty tables and in parallel scans when parallel workers get all
748 : * of the pages before we can get a chance to get our first page.
749 : */
750 : static pg_noinline BlockNumber
751 1983004 : heapgettup_initial_block(HeapScanDesc scan, ScanDirection dir)
752 : {
753 : Assert(!scan->rs_inited);
754 : Assert(scan->rs_base.rs_parallel == NULL);
755 :
756 : /* When there are no pages to scan, return InvalidBlockNumber */
757 1983004 : if (scan->rs_nblocks == 0 || scan->rs_numblocks == 0)
758 1009246 : return InvalidBlockNumber;
759 :
760 973758 : if (ScanDirectionIsForward(dir))
761 : {
762 973694 : return scan->rs_startblock;
763 : }
764 : else
765 : {
766 : /*
767 : * Disable reporting to syncscan logic in a backwards scan; it's not
768 : * very likely anyone else is doing the same thing at the same time,
769 : * and much more likely that we'll just bollix things for forward
770 : * scanners.
771 : */
772 64 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
773 :
774 : /*
775 : * Start from last page of the scan. Ensure we take into account
776 : * rs_numblocks if it's been adjusted by heap_setscanlimits().
777 : */
778 64 : if (scan->rs_numblocks != InvalidBlockNumber)
779 6 : return (scan->rs_startblock + scan->rs_numblocks - 1) % scan->rs_nblocks;
780 :
781 58 : if (scan->rs_startblock > 0)
782 0 : return scan->rs_startblock - 1;
783 :
784 58 : return scan->rs_nblocks - 1;
785 : }
786 : }
787 :
788 :
789 : /*
790 : * heapgettup_start_page - helper function for heapgettup()
791 : *
792 : * Return the next page to scan based on the scan->rs_cbuf and set *linesleft
793 : * to the number of tuples on this page. Also set *lineoff to the first
794 : * offset to scan with forward scans getting the first offset and backward
795 : * getting the final offset on the page.
796 : */
797 : static Page
798 185270 : heapgettup_start_page(HeapScanDesc scan, ScanDirection dir, int *linesleft,
799 : OffsetNumber *lineoff)
800 : {
801 : Page page;
802 :
803 : Assert(scan->rs_inited);
804 : Assert(BufferIsValid(scan->rs_cbuf));
805 :
806 : /* Caller is responsible for ensuring buffer is locked if needed */
807 185270 : page = BufferGetPage(scan->rs_cbuf);
808 :
809 185270 : *linesleft = PageGetMaxOffsetNumber(page) - FirstOffsetNumber + 1;
810 :
811 185270 : if (ScanDirectionIsForward(dir))
812 185270 : *lineoff = FirstOffsetNumber;
813 : else
814 0 : *lineoff = (OffsetNumber) (*linesleft);
815 :
816 : /* lineoff now references the physically previous or next tid */
817 185270 : return page;
818 : }
819 :
820 :
821 : /*
822 : * heapgettup_continue_page - helper function for heapgettup()
823 : *
824 : * Return the next page to scan based on the scan->rs_cbuf and set *linesleft
825 : * to the number of tuples left to scan on this page. Also set *lineoff to
826 : * the next offset to scan according to the ScanDirection in 'dir'.
827 : */
828 : static inline Page
829 14965950 : heapgettup_continue_page(HeapScanDesc scan, ScanDirection dir, int *linesleft,
830 : OffsetNumber *lineoff)
831 : {
832 : Page page;
833 :
834 : Assert(scan->rs_inited);
835 : Assert(BufferIsValid(scan->rs_cbuf));
836 :
837 : /* Caller is responsible for ensuring buffer is locked if needed */
838 14965950 : page = BufferGetPage(scan->rs_cbuf);
839 :
840 14965950 : if (ScanDirectionIsForward(dir))
841 : {
842 14965950 : *lineoff = OffsetNumberNext(scan->rs_coffset);
843 14965950 : *linesleft = PageGetMaxOffsetNumber(page) - (*lineoff) + 1;
844 : }
845 : else
846 : {
847 : /*
848 : * The previous returned tuple may have been vacuumed since the
849 : * previous scan when we use a non-MVCC snapshot, so we must
850 : * re-establish the lineoff <= PageGetMaxOffsetNumber(page) invariant
851 : */
852 0 : *lineoff = Min(PageGetMaxOffsetNumber(page), OffsetNumberPrev(scan->rs_coffset));
853 0 : *linesleft = *lineoff;
854 : }
855 :
856 : /* lineoff now references the physically previous or next tid */
857 14965950 : return page;
858 : }
859 :
860 : /*
861 : * heapgettup_advance_block - helper for heap_fetch_next_buffer()
862 : *
863 : * Given the current block number, the scan direction, and various information
864 : * contained in the scan descriptor, calculate the BlockNumber to scan next
865 : * and return it. If there are no further blocks to scan, return
866 : * InvalidBlockNumber to indicate this fact to the caller.
867 : *
868 : * This should not be called to determine the initial block number -- only for
869 : * subsequent blocks.
870 : *
871 : * This also adjusts rs_numblocks when a limit has been imposed by
872 : * heap_setscanlimits().
873 : */
874 : static inline BlockNumber
875 5751504 : heapgettup_advance_block(HeapScanDesc scan, BlockNumber block, ScanDirection dir)
876 : {
877 : Assert(scan->rs_base.rs_parallel == NULL);
878 :
879 5751504 : if (likely(ScanDirectionIsForward(dir)))
880 : {
881 5751386 : block++;
882 :
883 : /* wrap back to the start of the heap */
884 5751386 : if (block >= scan->rs_nblocks)
885 767878 : block = 0;
886 :
887 : /*
888 : * Report our new scan position for synchronization purposes. We don't
889 : * do that when moving backwards, however. That would just mess up any
890 : * other forward-moving scanners.
891 : *
892 : * Note: we do this before checking for end of scan so that the final
893 : * state of the position hint is back at the start of the rel. That's
894 : * not strictly necessary, but otherwise when you run the same query
895 : * multiple times the starting position would shift a little bit
896 : * backwards on every invocation, which is confusing. We don't
897 : * guarantee any specific ordering in general, though.
898 : */
899 5751386 : if (scan->rs_base.rs_flags & SO_ALLOW_SYNC)
900 22530 : ss_report_location(scan->rs_base.rs_rd, block);
901 :
902 : /* we're done if we're back at where we started */
903 5751386 : if (block == scan->rs_startblock)
904 767796 : return InvalidBlockNumber;
905 :
906 : /* check if the limit imposed by heap_setscanlimits() is met */
907 4983590 : if (scan->rs_numblocks != InvalidBlockNumber)
908 : {
909 4968 : if (--scan->rs_numblocks == 0)
910 3092 : return InvalidBlockNumber;
911 : }
912 :
913 4980498 : return block;
914 : }
915 : else
916 : {
917 : /* we're done if the last block is the start position */
918 118 : if (block == scan->rs_startblock)
919 118 : return InvalidBlockNumber;
920 :
921 : /* check if the limit imposed by heap_setscanlimits() is met */
922 0 : if (scan->rs_numblocks != InvalidBlockNumber)
923 : {
924 0 : if (--scan->rs_numblocks == 0)
925 0 : return InvalidBlockNumber;
926 : }
927 :
928 : /* wrap to the end of the heap when the last page was page 0 */
929 0 : if (block == 0)
930 0 : block = scan->rs_nblocks;
931 :
932 0 : block--;
933 :
934 0 : return block;
935 : }
936 : }
937 :
938 : /* ----------------
939 : * heapgettup - fetch next heap tuple
940 : *
941 : * Initialize the scan if not already done; then advance to the next
942 : * tuple as indicated by "dir"; return the next tuple in scan->rs_ctup,
943 : * or set scan->rs_ctup.t_data = NULL if no more tuples.
944 : *
945 : * Note: the reason nkeys/key are passed separately, even though they are
946 : * kept in the scan descriptor, is that the caller may not want us to check
947 : * the scankeys.
948 : *
949 : * Note: when we fall off the end of the scan in either direction, we
950 : * reset rs_inited. This means that a further request with the same
951 : * scan direction will restart the scan, which is a bit odd, but a
952 : * request with the opposite scan direction will start a fresh scan
953 : * in the proper direction. The latter is required behavior for cursors,
954 : * while the former case is generally undefined behavior in Postgres
955 : * so we don't care too much.
956 : * ----------------
957 : */
958 : static void
959 15008008 : heapgettup(HeapScanDesc scan,
960 : ScanDirection dir,
961 : int nkeys,
962 : ScanKey key)
963 : {
964 15008008 : HeapTuple tuple = &(scan->rs_ctup);
965 : Page page;
966 : OffsetNumber lineoff;
967 : int linesleft;
968 :
969 15008008 : if (likely(scan->rs_inited))
970 : {
971 : /* continue from previously returned page/tuple */
972 14965950 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
973 14965950 : page = heapgettup_continue_page(scan, dir, &linesleft, &lineoff);
974 14965950 : goto continue_page;
975 : }
976 :
977 : /*
978 : * advance the scan until we find a qualifying tuple or run out of stuff
979 : * to scan
980 : */
981 : while (true)
982 : {
983 227028 : heap_fetch_next_buffer(scan, dir);
984 :
985 : /* did we run out of blocks to scan? */
986 227028 : if (!BufferIsValid(scan->rs_cbuf))
987 41758 : break;
988 :
989 : Assert(BufferGetBlockNumber(scan->rs_cbuf) == scan->rs_cblock);
990 :
991 185270 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
992 185270 : page = heapgettup_start_page(scan, dir, &linesleft, &lineoff);
993 15151220 : continue_page:
994 :
995 : /*
996 : * Only continue scanning the page while we have lines left.
997 : *
998 : * Note that this protects us from accessing line pointers past
999 : * PageGetMaxOffsetNumber(); both for forward scans when we resume the
1000 : * table scan, and for when we start scanning a new page.
1001 : */
1002 15237214 : for (; linesleft > 0; linesleft--, lineoff += dir)
1003 : {
1004 : bool visible;
1005 15052244 : ItemId lpp = PageGetItemId(page, lineoff);
1006 :
1007 15052244 : if (!ItemIdIsNormal(lpp))
1008 75532 : continue;
1009 :
1010 14976712 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
1011 14976712 : tuple->t_len = ItemIdGetLength(lpp);
1012 14976712 : ItemPointerSet(&(tuple->t_self), scan->rs_cblock, lineoff);
1013 :
1014 14976712 : visible = HeapTupleSatisfiesVisibility(tuple,
1015 : scan->rs_base.rs_snapshot,
1016 : scan->rs_cbuf);
1017 :
1018 14976712 : HeapCheckForSerializableConflictOut(visible, scan->rs_base.rs_rd,
1019 : tuple, scan->rs_cbuf,
1020 : scan->rs_base.rs_snapshot);
1021 :
1022 : /* skip tuples not visible to this snapshot */
1023 14976712 : if (!visible)
1024 10462 : continue;
1025 :
1026 : /* skip any tuples that don't match the scan key */
1027 14966250 : if (key != NULL &&
1028 0 : !HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
1029 : nkeys, key))
1030 0 : continue;
1031 :
1032 14966250 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1033 14966250 : scan->rs_coffset = lineoff;
1034 14966250 : return;
1035 : }
1036 :
1037 : /*
1038 : * if we get here, it means we've exhausted the items on this page and
1039 : * it's time to move to the next.
1040 : */
1041 184970 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1042 : }
1043 :
1044 : /* end of scan */
1045 41758 : if (BufferIsValid(scan->rs_cbuf))
1046 0 : ReleaseBuffer(scan->rs_cbuf);
1047 :
1048 41758 : scan->rs_cbuf = InvalidBuffer;
1049 41758 : scan->rs_cblock = InvalidBlockNumber;
1050 41758 : scan->rs_prefetch_block = InvalidBlockNumber;
1051 41758 : tuple->t_data = NULL;
1052 41758 : scan->rs_inited = false;
1053 : }
1054 :
1055 : /* ----------------
1056 : * heapgettup_pagemode - fetch next heap tuple in page-at-a-time mode
1057 : *
1058 : * Same API as heapgettup, but used in page-at-a-time mode
1059 : *
1060 : * The internal logic is much the same as heapgettup's too, but there are some
1061 : * differences: we do not take the buffer content lock (that only needs to
1062 : * happen inside heap_prepare_pagescan), and we iterate through just the
1063 : * tuples listed in rs_vistuples[] rather than all tuples on the page. Notice
1064 : * that lineindex is 0-based, where the corresponding loop variable lineoff in
1065 : * heapgettup is 1-based.
1066 : * ----------------
1067 : */
1068 : static void
1069 99568590 : heapgettup_pagemode(HeapScanDesc scan,
1070 : ScanDirection dir,
1071 : int nkeys,
1072 : ScanKey key)
1073 : {
1074 99568590 : HeapTuple tuple = &(scan->rs_ctup);
1075 : Page page;
1076 : uint32 lineindex;
1077 : uint32 linesleft;
1078 :
1079 99568590 : if (likely(scan->rs_inited))
1080 : {
1081 : /* continue from previously returned page/tuple */
1082 97624350 : page = BufferGetPage(scan->rs_cbuf);
1083 :
1084 97624350 : lineindex = scan->rs_cindex + dir;
1085 97624350 : if (ScanDirectionIsForward(dir))
1086 97623692 : linesleft = scan->rs_ntuples - lineindex;
1087 : else
1088 658 : linesleft = scan->rs_cindex;
1089 : /* lineindex now references the next or previous visible tid */
1090 :
1091 97624350 : goto continue_page;
1092 : }
1093 :
1094 : /*
1095 : * advance the scan until we find a qualifying tuple or run out of stuff
1096 : * to scan
1097 : */
1098 : while (true)
1099 : {
1100 7334088 : heap_fetch_next_buffer(scan, dir);
1101 :
1102 : /* did we run out of blocks to scan? */
1103 7334036 : if (!BufferIsValid(scan->rs_cbuf))
1104 1611830 : break;
1105 :
1106 : Assert(BufferGetBlockNumber(scan->rs_cbuf) == scan->rs_cblock);
1107 :
1108 : /* prune the page and determine visible tuple offsets */
1109 5722206 : heap_prepare_pagescan((TableScanDesc) scan);
1110 5722190 : page = BufferGetPage(scan->rs_cbuf);
1111 5722190 : linesleft = scan->rs_ntuples;
1112 5722190 : lineindex = ScanDirectionIsForward(dir) ? 0 : linesleft - 1;
1113 :
1114 : /* block is the same for all tuples, set it once outside the loop */
1115 5722190 : ItemPointerSetBlockNumber(&tuple->t_self, scan->rs_cblock);
1116 :
1117 : /* lineindex now references the next or previous visible tid */
1118 103346540 : continue_page:
1119 :
1120 202085508 : for (; linesleft > 0; linesleft--, lineindex += dir)
1121 : {
1122 : ItemId lpp;
1123 : OffsetNumber lineoff;
1124 :
1125 : Assert(lineindex < scan->rs_ntuples);
1126 196695660 : lineoff = scan->rs_vistuples[lineindex];
1127 196695660 : lpp = PageGetItemId(page, lineoff);
1128 : Assert(ItemIdIsNormal(lpp));
1129 :
1130 196695660 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
1131 196695660 : tuple->t_len = ItemIdGetLength(lpp);
1132 196695660 : ItemPointerSetOffsetNumber(&tuple->t_self, lineoff);
1133 :
1134 : /* skip any tuples that don't match the scan key */
1135 196695660 : if (key != NULL &&
1136 99505842 : !HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
1137 : nkeys, key))
1138 98738968 : continue;
1139 :
1140 97956692 : scan->rs_cindex = lineindex;
1141 97956692 : return;
1142 : }
1143 : }
1144 :
1145 : /* end of scan */
1146 1611830 : if (BufferIsValid(scan->rs_cbuf))
1147 0 : ReleaseBuffer(scan->rs_cbuf);
1148 1611830 : scan->rs_cbuf = InvalidBuffer;
1149 1611830 : scan->rs_cblock = InvalidBlockNumber;
1150 1611830 : scan->rs_prefetch_block = InvalidBlockNumber;
1151 1611830 : tuple->t_data = NULL;
1152 1611830 : scan->rs_inited = false;
1153 : }
1154 :
1155 :
1156 : /* ----------------------------------------------------------------
1157 : * heap access method interface
1158 : * ----------------------------------------------------------------
1159 : */
1160 :
1161 :
1162 : TableScanDesc
1163 786128 : heap_beginscan(Relation relation, Snapshot snapshot,
1164 : int nkeys, ScanKey key,
1165 : ParallelTableScanDesc parallel_scan,
1166 : uint32 flags)
1167 : {
1168 : HeapScanDesc scan;
1169 :
1170 : /*
1171 : * increment relation ref count while scanning relation
1172 : *
1173 : * This is just to make really sure the relcache entry won't go away while
1174 : * the scan has a pointer to it. Caller should be holding the rel open
1175 : * anyway, so this is redundant in all normal scenarios...
1176 : */
1177 786128 : RelationIncrementReferenceCount(relation);
1178 :
1179 : /*
1180 : * allocate and initialize scan descriptor
1181 : */
1182 786128 : if (flags & SO_TYPE_BITMAPSCAN)
1183 : {
1184 22088 : BitmapHeapScanDesc bscan = palloc_object(BitmapHeapScanDescData);
1185 :
1186 : /*
1187 : * Bitmap Heap scans do not have any fields that a normal Heap Scan
1188 : * does not have, so no special initializations required here.
1189 : */
1190 22088 : scan = (HeapScanDesc) bscan;
1191 : }
1192 : else
1193 764040 : scan = (HeapScanDesc) palloc_object(HeapScanDescData);
1194 :
1195 786128 : scan->rs_base.rs_rd = relation;
1196 786128 : scan->rs_base.rs_snapshot = snapshot;
1197 786128 : scan->rs_base.rs_nkeys = nkeys;
1198 786128 : scan->rs_base.rs_flags = flags;
1199 786128 : scan->rs_base.rs_parallel = parallel_scan;
1200 786128 : scan->rs_strategy = NULL; /* set in initscan */
1201 786128 : scan->rs_cbuf = InvalidBuffer;
1202 :
1203 : /*
1204 : * Disable page-at-a-time mode if it's not a MVCC-safe snapshot.
1205 : */
1206 786128 : if (!(snapshot && IsMVCCSnapshot(snapshot)))
1207 59858 : scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
1208 :
1209 : /* Check that a historic snapshot is not used for non-catalog tables */
1210 786128 : if (snapshot &&
1211 768328 : IsHistoricMVCCSnapshot(snapshot) &&
1212 1316 : !RelationIsAccessibleInLogicalDecoding(relation))
1213 : {
1214 0 : ereport(ERROR,
1215 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
1216 : errmsg("cannot query non-catalog table \"%s\" during logical decoding",
1217 : RelationGetRelationName(relation))));
1218 : }
1219 :
1220 : /*
1221 : * For seqscan and sample scans in a serializable transaction, acquire a
1222 : * predicate lock on the entire relation. This is required not only to
1223 : * lock all the matching tuples, but also to conflict with new insertions
1224 : * into the table. In an indexscan, we take page locks on the index pages
1225 : * covering the range specified in the scan qual, but in a heap scan there
1226 : * is nothing more fine-grained to lock. A bitmap scan is a different
1227 : * story, there we have already scanned the index and locked the index
1228 : * pages covering the predicate. But in that case we still have to lock
1229 : * any matching heap tuples. For sample scan we could optimize the locking
1230 : * to be at least page-level granularity, but we'd need to add per-tuple
1231 : * locking for that.
1232 : */
1233 786128 : if (scan->rs_base.rs_flags & (SO_TYPE_SEQSCAN | SO_TYPE_SAMPLESCAN))
1234 : {
1235 : /*
1236 : * Ensure a missing snapshot is noticed reliably, even if the
1237 : * isolation mode means predicate locking isn't performed (and
1238 : * therefore the snapshot isn't used here).
1239 : */
1240 : Assert(snapshot);
1241 743474 : PredicateLockRelation(relation, snapshot);
1242 : }
1243 :
1244 : /* we only need to set this up once */
1245 786128 : scan->rs_ctup.t_tableOid = RelationGetRelid(relation);
1246 :
1247 : /*
1248 : * Allocate memory to keep track of page allocation for parallel workers
1249 : * when doing a parallel scan.
1250 : */
1251 786128 : if (parallel_scan != NULL)
1252 4242 : scan->rs_parallelworkerdata = palloc_object(ParallelBlockTableScanWorkerData);
1253 : else
1254 781886 : scan->rs_parallelworkerdata = NULL;
1255 :
1256 : /*
1257 : * we do this here instead of in initscan() because heap_rescan also calls
1258 : * initscan() and we don't want to allocate memory again
1259 : */
1260 786128 : if (nkeys > 0)
1261 450836 : scan->rs_base.rs_key = palloc_array(ScanKeyData, nkeys);
1262 : else
1263 335292 : scan->rs_base.rs_key = NULL;
1264 :
1265 786128 : initscan(scan, key, false);
1266 :
1267 786124 : scan->rs_read_stream = NULL;
1268 :
1269 : /*
1270 : * Set up a read stream for sequential scans and TID range scans. This
1271 : * should be done after initscan() because initscan() allocates the
1272 : * BufferAccessStrategy object passed to the read stream API.
1273 : */
1274 786124 : if (scan->rs_base.rs_flags & SO_TYPE_SEQSCAN ||
1275 42800 : scan->rs_base.rs_flags & SO_TYPE_TIDRANGESCAN)
1276 745304 : {
1277 : ReadStreamBlockNumberCB cb;
1278 :
1279 745304 : if (scan->rs_base.rs_parallel)
1280 4242 : cb = heap_scan_stream_read_next_parallel;
1281 : else
1282 741062 : cb = heap_scan_stream_read_next_serial;
1283 :
1284 : /* ---
1285 : * It is safe to use batchmode as the only locks taken by `cb`
1286 : * are never taken while waiting for IO:
1287 : * - SyncScanLock is used in the non-parallel case
1288 : * - in the parallel case, only spinlocks and atomics are used
1289 : * ---
1290 : */
1291 745304 : scan->rs_read_stream = read_stream_begin_relation(READ_STREAM_SEQUENTIAL |
1292 : READ_STREAM_USE_BATCHING,
1293 : scan->rs_strategy,
1294 : scan->rs_base.rs_rd,
1295 : MAIN_FORKNUM,
1296 : cb,
1297 : scan,
1298 : 0);
1299 : }
1300 40820 : else if (scan->rs_base.rs_flags & SO_TYPE_BITMAPSCAN)
1301 : {
1302 22088 : scan->rs_read_stream = read_stream_begin_relation(READ_STREAM_DEFAULT |
1303 : READ_STREAM_USE_BATCHING,
1304 : scan->rs_strategy,
1305 : scan->rs_base.rs_rd,
1306 : MAIN_FORKNUM,
1307 : bitmapheap_stream_read_next,
1308 : scan,
1309 : sizeof(TBMIterateResult));
1310 : }
1311 :
1312 :
1313 786124 : return (TableScanDesc) scan;
1314 : }
1315 :
1316 : void
1317 1246754 : heap_rescan(TableScanDesc sscan, ScanKey key, bool set_params,
1318 : bool allow_strat, bool allow_sync, bool allow_pagemode)
1319 : {
1320 1246754 : HeapScanDesc scan = (HeapScanDesc) sscan;
1321 :
1322 1246754 : if (set_params)
1323 : {
1324 30 : if (allow_strat)
1325 30 : scan->rs_base.rs_flags |= SO_ALLOW_STRAT;
1326 : else
1327 0 : scan->rs_base.rs_flags &= ~SO_ALLOW_STRAT;
1328 :
1329 30 : if (allow_sync)
1330 12 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
1331 : else
1332 18 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
1333 :
1334 30 : if (allow_pagemode && scan->rs_base.rs_snapshot &&
1335 30 : IsMVCCSnapshot(scan->rs_base.rs_snapshot))
1336 30 : scan->rs_base.rs_flags |= SO_ALLOW_PAGEMODE;
1337 : else
1338 0 : scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
1339 : }
1340 :
1341 : /*
1342 : * unpin scan buffers
1343 : */
1344 1246754 : if (BufferIsValid(scan->rs_cbuf))
1345 : {
1346 3246 : ReleaseBuffer(scan->rs_cbuf);
1347 3246 : scan->rs_cbuf = InvalidBuffer;
1348 : }
1349 :
1350 : /*
1351 : * SO_TYPE_BITMAPSCAN would be cleaned up here, but it does not hold any
1352 : * additional data vs a normal HeapScan
1353 : */
1354 :
1355 : /*
1356 : * The read stream is reset on rescan. This must be done before
1357 : * initscan(), as some state referred to by read_stream_reset() is reset
1358 : * in initscan().
1359 : */
1360 1246754 : if (scan->rs_read_stream)
1361 1246718 : read_stream_reset(scan->rs_read_stream);
1362 :
1363 : /*
1364 : * reinitialize scan descriptor
1365 : */
1366 1246754 : initscan(scan, key, true);
1367 1246754 : }
1368 :
1369 : void
1370 781390 : heap_endscan(TableScanDesc sscan)
1371 : {
1372 781390 : HeapScanDesc scan = (HeapScanDesc) sscan;
1373 :
1374 : /* Note: no locking manipulations needed */
1375 :
1376 : /*
1377 : * unpin scan buffers
1378 : */
1379 781390 : if (BufferIsValid(scan->rs_cbuf))
1380 326150 : ReleaseBuffer(scan->rs_cbuf);
1381 :
1382 : /*
1383 : * Must free the read stream before freeing the BufferAccessStrategy.
1384 : */
1385 781390 : if (scan->rs_read_stream)
1386 762764 : read_stream_end(scan->rs_read_stream);
1387 :
1388 : /*
1389 : * decrement relation reference count and free scan descriptor storage
1390 : */
1391 781390 : RelationDecrementReferenceCount(scan->rs_base.rs_rd);
1392 :
1393 781390 : if (scan->rs_base.rs_key)
1394 450782 : pfree(scan->rs_base.rs_key);
1395 :
1396 781390 : if (scan->rs_strategy != NULL)
1397 25070 : FreeAccessStrategy(scan->rs_strategy);
1398 :
1399 781390 : if (scan->rs_parallelworkerdata != NULL)
1400 4242 : pfree(scan->rs_parallelworkerdata);
1401 :
1402 781390 : if (scan->rs_base.rs_flags & SO_TEMP_SNAPSHOT)
1403 77814 : UnregisterSnapshot(scan->rs_base.rs_snapshot);
1404 :
1405 781390 : pfree(scan);
1406 781390 : }
1407 :
1408 : HeapTuple
1409 19366694 : heap_getnext(TableScanDesc sscan, ScanDirection direction)
1410 : {
1411 19366694 : HeapScanDesc scan = (HeapScanDesc) sscan;
1412 :
1413 : /*
1414 : * This is still widely used directly, without going through table AM, so
1415 : * add a safety check. It's possible we should, at a later point,
1416 : * downgrade this to an assert. The reason for checking the AM routine,
1417 : * rather than the AM oid, is that this allows to write regression tests
1418 : * that create another AM reusing the heap handler.
1419 : */
1420 19366694 : if (unlikely(sscan->rs_rd->rd_tableam != GetHeapamTableAmRoutine()))
1421 0 : ereport(ERROR,
1422 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1423 : errmsg_internal("only heap AM is supported")));
1424 :
1425 : /* Note: no locking manipulations needed */
1426 :
1427 19366694 : if (scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE)
1428 5335432 : heapgettup_pagemode(scan, direction,
1429 5335432 : scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1430 : else
1431 14031262 : heapgettup(scan, direction,
1432 14031262 : scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1433 :
1434 19366692 : if (scan->rs_ctup.t_data == NULL)
1435 131452 : return NULL;
1436 :
1437 : /*
1438 : * if we get here it means we have a new current scan tuple, so point to
1439 : * the proper return buffer and return the tuple.
1440 : */
1441 :
1442 19235240 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1443 :
1444 19235240 : return &scan->rs_ctup;
1445 : }
1446 :
1447 : bool
1448 95198446 : heap_getnextslot(TableScanDesc sscan, ScanDirection direction, TupleTableSlot *slot)
1449 : {
1450 95198446 : HeapScanDesc scan = (HeapScanDesc) sscan;
1451 :
1452 : /* Note: no locking manipulations needed */
1453 :
1454 95198446 : if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1455 94221700 : heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1456 : else
1457 976746 : heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1458 :
1459 95198396 : if (scan->rs_ctup.t_data == NULL)
1460 : {
1461 1521928 : ExecClearTuple(slot);
1462 1521928 : return false;
1463 : }
1464 :
1465 : /*
1466 : * if we get here it means we have a new current scan tuple, so point to
1467 : * the proper return buffer and return the tuple.
1468 : */
1469 :
1470 93676468 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1471 :
1472 93676468 : ExecStoreBufferHeapTuple(&scan->rs_ctup, slot,
1473 : scan->rs_cbuf);
1474 93676468 : return true;
1475 : }
1476 :
1477 : void
1478 2070 : heap_set_tidrange(TableScanDesc sscan, ItemPointer mintid,
1479 : ItemPointer maxtid)
1480 : {
1481 2070 : HeapScanDesc scan = (HeapScanDesc) sscan;
1482 : BlockNumber startBlk;
1483 : BlockNumber numBlks;
1484 : ItemPointerData highestItem;
1485 : ItemPointerData lowestItem;
1486 :
1487 : /*
1488 : * For relations without any pages, we can simply leave the TID range
1489 : * unset. There will be no tuples to scan, therefore no tuples outside
1490 : * the given TID range.
1491 : */
1492 2070 : if (scan->rs_nblocks == 0)
1493 48 : return;
1494 :
1495 : /*
1496 : * Set up some ItemPointers which point to the first and last possible
1497 : * tuples in the heap.
1498 : */
1499 2058 : ItemPointerSet(&highestItem, scan->rs_nblocks - 1, MaxOffsetNumber);
1500 2058 : ItemPointerSet(&lowestItem, 0, FirstOffsetNumber);
1501 :
1502 : /*
1503 : * If the given maximum TID is below the highest possible TID in the
1504 : * relation, then restrict the range to that, otherwise we scan to the end
1505 : * of the relation.
1506 : */
1507 2058 : if (ItemPointerCompare(maxtid, &highestItem) < 0)
1508 260 : ItemPointerCopy(maxtid, &highestItem);
1509 :
1510 : /*
1511 : * If the given minimum TID is above the lowest possible TID in the
1512 : * relation, then restrict the range to only scan for TIDs above that.
1513 : */
1514 2058 : if (ItemPointerCompare(mintid, &lowestItem) > 0)
1515 1822 : ItemPointerCopy(mintid, &lowestItem);
1516 :
1517 : /*
1518 : * Check for an empty range and protect from would be negative results
1519 : * from the numBlks calculation below.
1520 : */
1521 2058 : if (ItemPointerCompare(&highestItem, &lowestItem) < 0)
1522 : {
1523 : /* Set an empty range of blocks to scan */
1524 36 : heap_setscanlimits(sscan, 0, 0);
1525 36 : return;
1526 : }
1527 :
1528 : /*
1529 : * Calculate the first block and the number of blocks we must scan. We
1530 : * could be more aggressive here and perform some more validation to try
1531 : * and further narrow the scope of blocks to scan by checking if the
1532 : * lowestItem has an offset above MaxOffsetNumber. In this case, we could
1533 : * advance startBlk by one. Likewise, if highestItem has an offset of 0
1534 : * we could scan one fewer blocks. However, such an optimization does not
1535 : * seem worth troubling over, currently.
1536 : */
1537 2022 : startBlk = ItemPointerGetBlockNumberNoCheck(&lowestItem);
1538 :
1539 2022 : numBlks = ItemPointerGetBlockNumberNoCheck(&highestItem) -
1540 2022 : ItemPointerGetBlockNumberNoCheck(&lowestItem) + 1;
1541 :
1542 : /* Set the start block and number of blocks to scan */
1543 2022 : heap_setscanlimits(sscan, startBlk, numBlks);
1544 :
1545 : /* Finally, set the TID range in sscan */
1546 2022 : ItemPointerCopy(&lowestItem, &sscan->st.tidrange.rs_mintid);
1547 2022 : ItemPointerCopy(&highestItem, &sscan->st.tidrange.rs_maxtid);
1548 : }
1549 :
1550 : bool
1551 11272 : heap_getnextslot_tidrange(TableScanDesc sscan, ScanDirection direction,
1552 : TupleTableSlot *slot)
1553 : {
1554 11272 : HeapScanDesc scan = (HeapScanDesc) sscan;
1555 11272 : ItemPointer mintid = &sscan->st.tidrange.rs_mintid;
1556 11272 : ItemPointer maxtid = &sscan->st.tidrange.rs_maxtid;
1557 :
1558 : /* Note: no locking manipulations needed */
1559 : for (;;)
1560 : {
1561 11458 : if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1562 11458 : heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1563 : else
1564 0 : heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1565 :
1566 11442 : if (scan->rs_ctup.t_data == NULL)
1567 : {
1568 208 : ExecClearTuple(slot);
1569 208 : return false;
1570 : }
1571 :
1572 : /*
1573 : * heap_set_tidrange will have used heap_setscanlimits to limit the
1574 : * range of pages we scan to only ones that can contain the TID range
1575 : * we're scanning for. Here we must filter out any tuples from these
1576 : * pages that are outside of that range.
1577 : */
1578 11234 : if (ItemPointerCompare(&scan->rs_ctup.t_self, mintid) < 0)
1579 : {
1580 186 : ExecClearTuple(slot);
1581 :
1582 : /*
1583 : * When scanning backwards, the TIDs will be in descending order.
1584 : * Future tuples in this direction will be lower still, so we can
1585 : * just return false to indicate there will be no more tuples.
1586 : */
1587 186 : if (ScanDirectionIsBackward(direction))
1588 0 : return false;
1589 :
1590 186 : continue;
1591 : }
1592 :
1593 : /*
1594 : * Likewise for the final page, we must filter out TIDs greater than
1595 : * maxtid.
1596 : */
1597 11048 : if (ItemPointerCompare(&scan->rs_ctup.t_self, maxtid) > 0)
1598 : {
1599 112 : ExecClearTuple(slot);
1600 :
1601 : /*
1602 : * When scanning forward, the TIDs will be in ascending order.
1603 : * Future tuples in this direction will be higher still, so we can
1604 : * just return false to indicate there will be no more tuples.
1605 : */
1606 112 : if (ScanDirectionIsForward(direction))
1607 112 : return false;
1608 0 : continue;
1609 : }
1610 :
1611 10936 : break;
1612 : }
1613 :
1614 : /*
1615 : * if we get here it means we have a new current scan tuple, so point to
1616 : * the proper return buffer and return the tuple.
1617 : */
1618 10936 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1619 :
1620 10936 : ExecStoreBufferHeapTuple(&scan->rs_ctup, slot, scan->rs_cbuf);
1621 10936 : return true;
1622 : }
1623 :
1624 : /*
1625 : * heap_fetch - retrieve tuple with given tid
1626 : *
1627 : * On entry, tuple->t_self is the TID to fetch. We pin the buffer holding
1628 : * the tuple, fill in the remaining fields of *tuple, and check the tuple
1629 : * against the specified snapshot.
1630 : *
1631 : * If successful (tuple found and passes snapshot time qual), then *userbuf
1632 : * is set to the buffer holding the tuple and true is returned. The caller
1633 : * must unpin the buffer when done with the tuple.
1634 : *
1635 : * If the tuple is not found (ie, item number references a deleted slot),
1636 : * then tuple->t_data is set to NULL, *userbuf is set to InvalidBuffer,
1637 : * and false is returned.
1638 : *
1639 : * If the tuple is found but fails the time qual check, then the behavior
1640 : * depends on the keep_buf parameter. If keep_buf is false, the results
1641 : * are the same as for the tuple-not-found case. If keep_buf is true,
1642 : * then tuple->t_data and *userbuf are returned as for the success case,
1643 : * and again the caller must unpin the buffer; but false is returned.
1644 : *
1645 : * heap_fetch does not follow HOT chains: only the exact TID requested will
1646 : * be fetched.
1647 : *
1648 : * It is somewhat inconsistent that we ereport() on invalid block number but
1649 : * return false on invalid item number. There are a couple of reasons though.
1650 : * One is that the caller can relatively easily check the block number for
1651 : * validity, but cannot check the item number without reading the page
1652 : * himself. Another is that when we are following a t_ctid link, we can be
1653 : * reasonably confident that the page number is valid (since VACUUM shouldn't
1654 : * truncate off the destination page without having killed the referencing
1655 : * tuple first), but the item number might well not be good.
1656 : */
1657 : bool
1658 362588 : heap_fetch(Relation relation,
1659 : Snapshot snapshot,
1660 : HeapTuple tuple,
1661 : Buffer *userbuf,
1662 : bool keep_buf)
1663 : {
1664 362588 : ItemPointer tid = &(tuple->t_self);
1665 : ItemId lp;
1666 : Buffer buffer;
1667 : Page page;
1668 : OffsetNumber offnum;
1669 : bool valid;
1670 :
1671 : /*
1672 : * Fetch and pin the appropriate page of the relation.
1673 : */
1674 362588 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
1675 :
1676 : /*
1677 : * Need share lock on buffer to examine tuple commit status.
1678 : */
1679 362572 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
1680 362572 : page = BufferGetPage(buffer);
1681 :
1682 : /*
1683 : * We'd better check for out-of-range offnum in case of VACUUM since the
1684 : * TID was obtained.
1685 : */
1686 362572 : offnum = ItemPointerGetOffsetNumber(tid);
1687 362572 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1688 : {
1689 6 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1690 6 : ReleaseBuffer(buffer);
1691 6 : *userbuf = InvalidBuffer;
1692 6 : tuple->t_data = NULL;
1693 6 : return false;
1694 : }
1695 :
1696 : /*
1697 : * get the item line pointer corresponding to the requested tid
1698 : */
1699 362566 : lp = PageGetItemId(page, offnum);
1700 :
1701 : /*
1702 : * Must check for deleted tuple.
1703 : */
1704 362566 : if (!ItemIdIsNormal(lp))
1705 : {
1706 700 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1707 700 : ReleaseBuffer(buffer);
1708 700 : *userbuf = InvalidBuffer;
1709 700 : tuple->t_data = NULL;
1710 700 : return false;
1711 : }
1712 :
1713 : /*
1714 : * fill in *tuple fields
1715 : */
1716 361866 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
1717 361866 : tuple->t_len = ItemIdGetLength(lp);
1718 361866 : tuple->t_tableOid = RelationGetRelid(relation);
1719 :
1720 : /*
1721 : * check tuple visibility, then release lock
1722 : */
1723 361866 : valid = HeapTupleSatisfiesVisibility(tuple, snapshot, buffer);
1724 :
1725 361866 : if (valid)
1726 361760 : PredicateLockTID(relation, &(tuple->t_self), snapshot,
1727 361760 : HeapTupleHeaderGetXmin(tuple->t_data));
1728 :
1729 361866 : HeapCheckForSerializableConflictOut(valid, relation, tuple, buffer, snapshot);
1730 :
1731 361866 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1732 :
1733 361866 : if (valid)
1734 : {
1735 : /*
1736 : * All checks passed, so return the tuple as valid. Caller is now
1737 : * responsible for releasing the buffer.
1738 : */
1739 361760 : *userbuf = buffer;
1740 :
1741 361760 : return true;
1742 : }
1743 :
1744 : /* Tuple failed time qual, but maybe caller wants to see it anyway. */
1745 106 : if (keep_buf)
1746 68 : *userbuf = buffer;
1747 : else
1748 : {
1749 38 : ReleaseBuffer(buffer);
1750 38 : *userbuf = InvalidBuffer;
1751 38 : tuple->t_data = NULL;
1752 : }
1753 :
1754 106 : return false;
1755 : }
1756 :
1757 : /*
1758 : * heap_hot_search_buffer - search HOT chain for tuple satisfying snapshot
1759 : *
1760 : * On entry, *tid is the TID of a tuple (either a simple tuple, or the root
1761 : * of a HOT chain), and buffer is the buffer holding this tuple. We search
1762 : * for the first chain member satisfying the given snapshot. If one is
1763 : * found, we update *tid to reference that tuple's offset number, and
1764 : * return true. If no match, return false without modifying *tid.
1765 : *
1766 : * heapTuple is a caller-supplied buffer. When a match is found, we return
1767 : * the tuple here, in addition to updating *tid. If no match is found, the
1768 : * contents of this buffer on return are undefined.
1769 : *
1770 : * If all_dead is not NULL, we check non-visible tuples to see if they are
1771 : * globally dead; *all_dead is set true if all members of the HOT chain
1772 : * are vacuumable, false if not.
1773 : *
1774 : * Unlike heap_fetch, the caller must already have pin and (at least) share
1775 : * lock on the buffer; it is still pinned/locked at exit.
1776 : */
1777 : bool
1778 45266994 : heap_hot_search_buffer(ItemPointer tid, Relation relation, Buffer buffer,
1779 : Snapshot snapshot, HeapTuple heapTuple,
1780 : bool *all_dead, bool first_call)
1781 : {
1782 45266994 : Page page = BufferGetPage(buffer);
1783 45266994 : TransactionId prev_xmax = InvalidTransactionId;
1784 : BlockNumber blkno;
1785 : OffsetNumber offnum;
1786 : bool at_chain_start;
1787 : bool valid;
1788 : bool skip;
1789 45266994 : GlobalVisState *vistest = NULL;
1790 :
1791 : /* If this is not the first call, previous call returned a (live!) tuple */
1792 45266994 : if (all_dead)
1793 38589262 : *all_dead = first_call;
1794 :
1795 45266994 : blkno = ItemPointerGetBlockNumber(tid);
1796 45266994 : offnum = ItemPointerGetOffsetNumber(tid);
1797 45266994 : at_chain_start = first_call;
1798 45266994 : skip = !first_call;
1799 :
1800 : /* XXX: we should assert that a snapshot is pushed or registered */
1801 : Assert(TransactionIdIsValid(RecentXmin));
1802 : Assert(BufferGetBlockNumber(buffer) == blkno);
1803 :
1804 : /* Scan through possible multiple members of HOT-chain */
1805 : for (;;)
1806 3078190 : {
1807 : ItemId lp;
1808 :
1809 : /* check for bogus TID */
1810 48345184 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1811 : break;
1812 :
1813 48345184 : lp = PageGetItemId(page, offnum);
1814 :
1815 : /* check for unused, dead, or redirected items */
1816 48345184 : if (!ItemIdIsNormal(lp))
1817 : {
1818 : /* We should only see a redirect at start of chain */
1819 1700066 : if (ItemIdIsRedirected(lp) && at_chain_start)
1820 : {
1821 : /* Follow the redirect */
1822 957428 : offnum = ItemIdGetRedirect(lp);
1823 957428 : at_chain_start = false;
1824 957428 : continue;
1825 : }
1826 : /* else must be end of chain */
1827 742638 : break;
1828 : }
1829 :
1830 : /*
1831 : * Update heapTuple to point to the element of the HOT chain we're
1832 : * currently investigating. Having t_self set correctly is important
1833 : * because the SSI checks and the *Satisfies routine for historical
1834 : * MVCC snapshots need the correct tid to decide about the visibility.
1835 : */
1836 46645118 : heapTuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
1837 46645118 : heapTuple->t_len = ItemIdGetLength(lp);
1838 46645118 : heapTuple->t_tableOid = RelationGetRelid(relation);
1839 46645118 : ItemPointerSet(&heapTuple->t_self, blkno, offnum);
1840 :
1841 : /*
1842 : * Shouldn't see a HEAP_ONLY tuple at chain start.
1843 : */
1844 46645118 : if (at_chain_start && HeapTupleIsHeapOnly(heapTuple))
1845 0 : break;
1846 :
1847 : /*
1848 : * The xmin should match the previous xmax value, else chain is
1849 : * broken.
1850 : */
1851 48765880 : if (TransactionIdIsValid(prev_xmax) &&
1852 2120762 : !TransactionIdEquals(prev_xmax,
1853 : HeapTupleHeaderGetXmin(heapTuple->t_data)))
1854 0 : break;
1855 :
1856 : /*
1857 : * When first_call is true (and thus, skip is initially false) we'll
1858 : * return the first tuple we find. But on later passes, heapTuple
1859 : * will initially be pointing to the tuple we returned last time.
1860 : * Returning it again would be incorrect (and would loop forever), so
1861 : * we skip it and return the next match we find.
1862 : */
1863 46645118 : if (!skip)
1864 : {
1865 : /* If it's visible per the snapshot, we must return it */
1866 46471158 : valid = HeapTupleSatisfiesVisibility(heapTuple, snapshot, buffer);
1867 46471158 : HeapCheckForSerializableConflictOut(valid, relation, heapTuple,
1868 : buffer, snapshot);
1869 :
1870 46471148 : if (valid)
1871 : {
1872 32040538 : ItemPointerSetOffsetNumber(tid, offnum);
1873 32040538 : PredicateLockTID(relation, &heapTuple->t_self, snapshot,
1874 32040538 : HeapTupleHeaderGetXmin(heapTuple->t_data));
1875 32040538 : if (all_dead)
1876 25948712 : *all_dead = false;
1877 32040538 : return true;
1878 : }
1879 : }
1880 14604570 : skip = false;
1881 :
1882 : /*
1883 : * If we can't see it, maybe no one else can either. At caller
1884 : * request, check whether all chain members are dead to all
1885 : * transactions.
1886 : *
1887 : * Note: if you change the criterion here for what is "dead", fix the
1888 : * planner's get_actual_variable_range() function to match.
1889 : */
1890 14604570 : if (all_dead && *all_dead)
1891 : {
1892 12971386 : if (!vistest)
1893 12729522 : vistest = GlobalVisTestFor(relation);
1894 :
1895 12971386 : if (!HeapTupleIsSurelyDead(heapTuple, vistest))
1896 12261482 : *all_dead = false;
1897 : }
1898 :
1899 : /*
1900 : * Check to see if HOT chain continues past this tuple; if so fetch
1901 : * the next offnum and loop around.
1902 : */
1903 14604570 : if (HeapTupleIsHotUpdated(heapTuple))
1904 : {
1905 : Assert(ItemPointerGetBlockNumber(&heapTuple->t_data->t_ctid) ==
1906 : blkno);
1907 2120762 : offnum = ItemPointerGetOffsetNumber(&heapTuple->t_data->t_ctid);
1908 2120762 : at_chain_start = false;
1909 2120762 : prev_xmax = HeapTupleHeaderGetUpdateXid(heapTuple->t_data);
1910 : }
1911 : else
1912 12483808 : break; /* end of chain */
1913 : }
1914 :
1915 13226446 : return false;
1916 : }
1917 :
1918 : /*
1919 : * heap_get_latest_tid - get the latest tid of a specified tuple
1920 : *
1921 : * Actually, this gets the latest version that is visible according to the
1922 : * scan's snapshot. Create a scan using SnapshotDirty to get the very latest,
1923 : * possibly uncommitted version.
1924 : *
1925 : * *tid is both an input and an output parameter: it is updated to
1926 : * show the latest version of the row. Note that it will not be changed
1927 : * if no version of the row passes the snapshot test.
1928 : */
1929 : void
1930 300 : heap_get_latest_tid(TableScanDesc sscan,
1931 : ItemPointer tid)
1932 : {
1933 300 : Relation relation = sscan->rs_rd;
1934 300 : Snapshot snapshot = sscan->rs_snapshot;
1935 : ItemPointerData ctid;
1936 : TransactionId priorXmax;
1937 :
1938 : /*
1939 : * table_tuple_get_latest_tid() verified that the passed in tid is valid.
1940 : * Assume that t_ctid links are valid however - there shouldn't be invalid
1941 : * ones in the table.
1942 : */
1943 : Assert(ItemPointerIsValid(tid));
1944 :
1945 : /*
1946 : * Loop to chase down t_ctid links. At top of loop, ctid is the tuple we
1947 : * need to examine, and *tid is the TID we will return if ctid turns out
1948 : * to be bogus.
1949 : *
1950 : * Note that we will loop until we reach the end of the t_ctid chain.
1951 : * Depending on the snapshot passed, there might be at most one visible
1952 : * version of the row, but we don't try to optimize for that.
1953 : */
1954 300 : ctid = *tid;
1955 300 : priorXmax = InvalidTransactionId; /* cannot check first XMIN */
1956 : for (;;)
1957 90 : {
1958 : Buffer buffer;
1959 : Page page;
1960 : OffsetNumber offnum;
1961 : ItemId lp;
1962 : HeapTupleData tp;
1963 : bool valid;
1964 :
1965 : /*
1966 : * Read, pin, and lock the page.
1967 : */
1968 390 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(&ctid));
1969 390 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
1970 390 : page = BufferGetPage(buffer);
1971 :
1972 : /*
1973 : * Check for bogus item number. This is not treated as an error
1974 : * condition because it can happen while following a t_ctid link. We
1975 : * just assume that the prior tid is OK and return it unchanged.
1976 : */
1977 390 : offnum = ItemPointerGetOffsetNumber(&ctid);
1978 390 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1979 : {
1980 0 : UnlockReleaseBuffer(buffer);
1981 0 : break;
1982 : }
1983 390 : lp = PageGetItemId(page, offnum);
1984 390 : if (!ItemIdIsNormal(lp))
1985 : {
1986 0 : UnlockReleaseBuffer(buffer);
1987 0 : break;
1988 : }
1989 :
1990 : /* OK to access the tuple */
1991 390 : tp.t_self = ctid;
1992 390 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
1993 390 : tp.t_len = ItemIdGetLength(lp);
1994 390 : tp.t_tableOid = RelationGetRelid(relation);
1995 :
1996 : /*
1997 : * After following a t_ctid link, we might arrive at an unrelated
1998 : * tuple. Check for XMIN match.
1999 : */
2000 480 : if (TransactionIdIsValid(priorXmax) &&
2001 90 : !TransactionIdEquals(priorXmax, HeapTupleHeaderGetXmin(tp.t_data)))
2002 : {
2003 0 : UnlockReleaseBuffer(buffer);
2004 0 : break;
2005 : }
2006 :
2007 : /*
2008 : * Check tuple visibility; if visible, set it as the new result
2009 : * candidate.
2010 : */
2011 390 : valid = HeapTupleSatisfiesVisibility(&tp, snapshot, buffer);
2012 390 : HeapCheckForSerializableConflictOut(valid, relation, &tp, buffer, snapshot);
2013 390 : if (valid)
2014 276 : *tid = ctid;
2015 :
2016 : /*
2017 : * If there's a valid t_ctid link, follow it, else we're done.
2018 : */
2019 552 : if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
2020 276 : HeapTupleHeaderIsOnlyLocked(tp.t_data) ||
2021 228 : HeapTupleHeaderIndicatesMovedPartitions(tp.t_data) ||
2022 114 : ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
2023 : {
2024 300 : UnlockReleaseBuffer(buffer);
2025 300 : break;
2026 : }
2027 :
2028 90 : ctid = tp.t_data->t_ctid;
2029 90 : priorXmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
2030 90 : UnlockReleaseBuffer(buffer);
2031 : } /* end of loop */
2032 300 : }
2033 :
2034 :
2035 : /*
2036 : * UpdateXmaxHintBits - update tuple hint bits after xmax transaction ends
2037 : *
2038 : * This is called after we have waited for the XMAX transaction to terminate.
2039 : * If the transaction aborted, we guarantee the XMAX_INVALID hint bit will
2040 : * be set on exit. If the transaction committed, we set the XMAX_COMMITTED
2041 : * hint bit if possible --- but beware that that may not yet be possible,
2042 : * if the transaction committed asynchronously.
2043 : *
2044 : * Note that if the transaction was a locker only, we set HEAP_XMAX_INVALID
2045 : * even if it commits.
2046 : *
2047 : * Hence callers should look only at XMAX_INVALID.
2048 : *
2049 : * Note this is not allowed for tuples whose xmax is a multixact.
2050 : */
2051 : static void
2052 440 : UpdateXmaxHintBits(HeapTupleHeader tuple, Buffer buffer, TransactionId xid)
2053 : {
2054 : Assert(TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple), xid));
2055 : Assert(!(tuple->t_infomask & HEAP_XMAX_IS_MULTI));
2056 :
2057 440 : if (!(tuple->t_infomask & (HEAP_XMAX_COMMITTED | HEAP_XMAX_INVALID)))
2058 : {
2059 786 : if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask) &&
2060 348 : TransactionIdDidCommit(xid))
2061 294 : HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_COMMITTED,
2062 : xid);
2063 : else
2064 144 : HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_INVALID,
2065 : InvalidTransactionId);
2066 : }
2067 440 : }
2068 :
2069 :
2070 : /*
2071 : * GetBulkInsertState - prepare status object for a bulk insert
2072 : */
2073 : BulkInsertState
2074 5468 : GetBulkInsertState(void)
2075 : {
2076 : BulkInsertState bistate;
2077 :
2078 5468 : bistate = (BulkInsertState) palloc_object(BulkInsertStateData);
2079 5468 : bistate->strategy = GetAccessStrategy(BAS_BULKWRITE);
2080 5468 : bistate->current_buf = InvalidBuffer;
2081 5468 : bistate->next_free = InvalidBlockNumber;
2082 5468 : bistate->last_free = InvalidBlockNumber;
2083 5468 : bistate->already_extended_by = 0;
2084 5468 : return bistate;
2085 : }
2086 :
2087 : /*
2088 : * FreeBulkInsertState - clean up after finishing a bulk insert
2089 : */
2090 : void
2091 5178 : FreeBulkInsertState(BulkInsertState bistate)
2092 : {
2093 5178 : if (bistate->current_buf != InvalidBuffer)
2094 4020 : ReleaseBuffer(bistate->current_buf);
2095 5178 : FreeAccessStrategy(bistate->strategy);
2096 5178 : pfree(bistate);
2097 5178 : }
2098 :
2099 : /*
2100 : * ReleaseBulkInsertStatePin - release a buffer currently held in bistate
2101 : */
2102 : void
2103 161516 : ReleaseBulkInsertStatePin(BulkInsertState bistate)
2104 : {
2105 161516 : if (bistate->current_buf != InvalidBuffer)
2106 60042 : ReleaseBuffer(bistate->current_buf);
2107 161516 : bistate->current_buf = InvalidBuffer;
2108 :
2109 : /*
2110 : * Despite the name, we also reset bulk relation extension state.
2111 : * Otherwise we can end up erroring out due to looking for free space in
2112 : * ->next_free of one partition, even though ->next_free was set when
2113 : * extending another partition. It could obviously also be bad for
2114 : * efficiency to look at existing blocks at offsets from another
2115 : * partition, even if we don't error out.
2116 : */
2117 161516 : bistate->next_free = InvalidBlockNumber;
2118 161516 : bistate->last_free = InvalidBlockNumber;
2119 161516 : }
2120 :
2121 :
2122 : /*
2123 : * heap_insert - insert tuple into a heap
2124 : *
2125 : * The new tuple is stamped with current transaction ID and the specified
2126 : * command ID.
2127 : *
2128 : * See table_tuple_insert for comments about most of the input flags, except
2129 : * that this routine directly takes a tuple rather than a slot.
2130 : *
2131 : * There's corresponding HEAP_INSERT_ options to all the TABLE_INSERT_
2132 : * options, and there additionally is HEAP_INSERT_SPECULATIVE which is used to
2133 : * implement table_tuple_insert_speculative().
2134 : *
2135 : * On return the header fields of *tup are updated to match the stored tuple;
2136 : * in particular tup->t_self receives the actual TID where the tuple was
2137 : * stored. But note that any toasting of fields within the tuple data is NOT
2138 : * reflected into *tup.
2139 : */
2140 : void
2141 16769074 : heap_insert(Relation relation, HeapTuple tup, CommandId cid,
2142 : int options, BulkInsertState bistate)
2143 : {
2144 16769074 : TransactionId xid = GetCurrentTransactionId();
2145 : HeapTuple heaptup;
2146 : Buffer buffer;
2147 16769058 : Buffer vmbuffer = InvalidBuffer;
2148 16769058 : bool all_visible_cleared = false;
2149 :
2150 : /* Cheap, simplistic check that the tuple matches the rel's rowtype. */
2151 : Assert(HeapTupleHeaderGetNatts(tup->t_data) <=
2152 : RelationGetNumberOfAttributes(relation));
2153 :
2154 16769058 : AssertHasSnapshotForToast(relation);
2155 :
2156 : /*
2157 : * Fill in tuple header fields and toast the tuple if necessary.
2158 : *
2159 : * Note: below this point, heaptup is the data we actually intend to store
2160 : * into the relation; tup is the caller's original untoasted data.
2161 : */
2162 16769058 : heaptup = heap_prepare_insert(relation, tup, xid, cid, options);
2163 :
2164 : /*
2165 : * Find buffer to insert this tuple into. If the page is all visible,
2166 : * this will also pin the requisite visibility map page.
2167 : */
2168 16769058 : buffer = RelationGetBufferForTuple(relation, heaptup->t_len,
2169 : InvalidBuffer, options, bistate,
2170 : &vmbuffer, NULL,
2171 : 0);
2172 :
2173 : /*
2174 : * We're about to do the actual insert -- but check for conflict first, to
2175 : * avoid possibly having to roll back work we've just done.
2176 : *
2177 : * This is safe without a recheck as long as there is no possibility of
2178 : * another process scanning the page between this check and the insert
2179 : * being visible to the scan (i.e., an exclusive buffer content lock is
2180 : * continuously held from this point until the tuple insert is visible).
2181 : *
2182 : * For a heap insert, we only need to check for table-level SSI locks. Our
2183 : * new tuple can't possibly conflict with existing tuple locks, and heap
2184 : * page locks are only consolidated versions of tuple locks; they do not
2185 : * lock "gaps" as index page locks do. So we don't need to specify a
2186 : * buffer when making the call, which makes for a faster check.
2187 : */
2188 16769058 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2189 :
2190 : /* NO EREPORT(ERROR) from here till changes are logged */
2191 16769034 : START_CRIT_SECTION();
2192 :
2193 16769034 : RelationPutHeapTuple(relation, buffer, heaptup,
2194 16769034 : (options & HEAP_INSERT_SPECULATIVE) != 0);
2195 :
2196 16769034 : if (PageIsAllVisible(BufferGetPage(buffer)))
2197 : {
2198 14872 : all_visible_cleared = true;
2199 14872 : PageClearAllVisible(BufferGetPage(buffer));
2200 14872 : visibilitymap_clear(relation,
2201 14872 : ItemPointerGetBlockNumber(&(heaptup->t_self)),
2202 : vmbuffer, VISIBILITYMAP_VALID_BITS);
2203 : }
2204 :
2205 : /*
2206 : * XXX Should we set PageSetPrunable on this page ?
2207 : *
2208 : * The inserting transaction may eventually abort thus making this tuple
2209 : * DEAD and hence available for pruning. Though we don't want to optimize
2210 : * for aborts, if no other tuple in this page is UPDATEd/DELETEd, the
2211 : * aborted tuple will never be pruned until next vacuum is triggered.
2212 : *
2213 : * If you do add PageSetPrunable here, add it in heap_xlog_insert too.
2214 : */
2215 :
2216 16769034 : MarkBufferDirty(buffer);
2217 :
2218 : /* XLOG stuff */
2219 16769034 : if (RelationNeedsWAL(relation))
2220 : {
2221 : xl_heap_insert xlrec;
2222 : xl_heap_header xlhdr;
2223 : XLogRecPtr recptr;
2224 14989640 : Page page = BufferGetPage(buffer);
2225 14989640 : uint8 info = XLOG_HEAP_INSERT;
2226 14989640 : int bufflags = 0;
2227 :
2228 : /*
2229 : * If this is a catalog, we need to transmit combo CIDs to properly
2230 : * decode, so log that as well.
2231 : */
2232 14989640 : if (RelationIsAccessibleInLogicalDecoding(relation))
2233 6868 : log_heap_new_cid(relation, heaptup);
2234 :
2235 : /*
2236 : * If this is the single and first tuple on page, we can reinit the
2237 : * page instead of restoring the whole thing. Set flag, and hide
2238 : * buffer references from XLogInsert.
2239 : */
2240 15182912 : if (ItemPointerGetOffsetNumber(&(heaptup->t_self)) == FirstOffsetNumber &&
2241 193272 : PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
2242 : {
2243 191526 : info |= XLOG_HEAP_INIT_PAGE;
2244 191526 : bufflags |= REGBUF_WILL_INIT;
2245 : }
2246 :
2247 14989640 : xlrec.offnum = ItemPointerGetOffsetNumber(&heaptup->t_self);
2248 14989640 : xlrec.flags = 0;
2249 14989640 : if (all_visible_cleared)
2250 14866 : xlrec.flags |= XLH_INSERT_ALL_VISIBLE_CLEARED;
2251 14989640 : if (options & HEAP_INSERT_SPECULATIVE)
2252 4164 : xlrec.flags |= XLH_INSERT_IS_SPECULATIVE;
2253 : Assert(ItemPointerGetBlockNumber(&heaptup->t_self) == BufferGetBlockNumber(buffer));
2254 :
2255 : /*
2256 : * For logical decoding, we need the tuple even if we're doing a full
2257 : * page write, so make sure it's included even if we take a full-page
2258 : * image. (XXX We could alternatively store a pointer into the FPW).
2259 : */
2260 14989640 : if (RelationIsLogicallyLogged(relation) &&
2261 501026 : !(options & HEAP_INSERT_NO_LOGICAL))
2262 : {
2263 500972 : xlrec.flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
2264 500972 : bufflags |= REGBUF_KEEP_DATA;
2265 :
2266 500972 : if (IsToastRelation(relation))
2267 3572 : xlrec.flags |= XLH_INSERT_ON_TOAST_RELATION;
2268 : }
2269 :
2270 14989640 : XLogBeginInsert();
2271 14989640 : XLogRegisterData(&xlrec, SizeOfHeapInsert);
2272 :
2273 14989640 : xlhdr.t_infomask2 = heaptup->t_data->t_infomask2;
2274 14989640 : xlhdr.t_infomask = heaptup->t_data->t_infomask;
2275 14989640 : xlhdr.t_hoff = heaptup->t_data->t_hoff;
2276 :
2277 : /*
2278 : * note we mark xlhdr as belonging to buffer; if XLogInsert decides to
2279 : * write the whole page to the xlog, we don't need to store
2280 : * xl_heap_header in the xlog.
2281 : */
2282 14989640 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD | bufflags);
2283 14989640 : XLogRegisterBufData(0, &xlhdr, SizeOfHeapHeader);
2284 : /* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
2285 14989640 : XLogRegisterBufData(0,
2286 14989640 : (char *) heaptup->t_data + SizeofHeapTupleHeader,
2287 14989640 : heaptup->t_len - SizeofHeapTupleHeader);
2288 :
2289 : /* filtering by origin on a row level is much more efficient */
2290 14989640 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
2291 :
2292 14989640 : recptr = XLogInsert(RM_HEAP_ID, info);
2293 :
2294 14989640 : PageSetLSN(page, recptr);
2295 : }
2296 :
2297 16769034 : END_CRIT_SECTION();
2298 :
2299 16769034 : UnlockReleaseBuffer(buffer);
2300 16769034 : if (vmbuffer != InvalidBuffer)
2301 15430 : ReleaseBuffer(vmbuffer);
2302 :
2303 : /*
2304 : * If tuple is cacheable, mark it for invalidation from the caches in case
2305 : * we abort. Note it is OK to do this after releasing the buffer, because
2306 : * the heaptup data structure is all in local memory, not in the shared
2307 : * buffer.
2308 : */
2309 16769034 : CacheInvalidateHeapTuple(relation, heaptup, NULL);
2310 :
2311 : /* Note: speculative insertions are counted too, even if aborted later */
2312 16769034 : pgstat_count_heap_insert(relation, 1);
2313 :
2314 : /*
2315 : * If heaptup is a private copy, release it. Don't forget to copy t_self
2316 : * back to the caller's image, too.
2317 : */
2318 16769034 : if (heaptup != tup)
2319 : {
2320 36884 : tup->t_self = heaptup->t_self;
2321 36884 : heap_freetuple(heaptup);
2322 : }
2323 16769034 : }
2324 :
2325 : /*
2326 : * Subroutine for heap_insert(). Prepares a tuple for insertion. This sets the
2327 : * tuple header fields and toasts the tuple if necessary. Returns a toasted
2328 : * version of the tuple if it was toasted, or the original tuple if not. Note
2329 : * that in any case, the header fields are also set in the original tuple.
2330 : */
2331 : static HeapTuple
2332 19781884 : heap_prepare_insert(Relation relation, HeapTuple tup, TransactionId xid,
2333 : CommandId cid, int options)
2334 : {
2335 : /*
2336 : * To allow parallel inserts, we need to ensure that they are safe to be
2337 : * performed in workers. We have the infrastructure to allow parallel
2338 : * inserts in general except for the cases where inserts generate a new
2339 : * CommandId (eg. inserts into a table having a foreign key column).
2340 : */
2341 19781884 : if (IsParallelWorker())
2342 0 : ereport(ERROR,
2343 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
2344 : errmsg("cannot insert tuples in a parallel worker")));
2345 :
2346 19781884 : tup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
2347 19781884 : tup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
2348 19781884 : tup->t_data->t_infomask |= HEAP_XMAX_INVALID;
2349 19781884 : HeapTupleHeaderSetXmin(tup->t_data, xid);
2350 19781884 : if (options & HEAP_INSERT_FROZEN)
2351 204176 : HeapTupleHeaderSetXminFrozen(tup->t_data);
2352 :
2353 19781884 : HeapTupleHeaderSetCmin(tup->t_data, cid);
2354 19781884 : HeapTupleHeaderSetXmax(tup->t_data, 0); /* for cleanliness */
2355 19781884 : tup->t_tableOid = RelationGetRelid(relation);
2356 :
2357 : /*
2358 : * If the new tuple is too big for storage or contains already toasted
2359 : * out-of-line attributes from some other relation, invoke the toaster.
2360 : */
2361 19781884 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
2362 62848 : relation->rd_rel->relkind != RELKIND_MATVIEW)
2363 : {
2364 : /* toast table entries should never be recursively toasted */
2365 : Assert(!HeapTupleHasExternal(tup));
2366 62752 : return tup;
2367 : }
2368 19719132 : else if (HeapTupleHasExternal(tup) || tup->t_len > TOAST_TUPLE_THRESHOLD)
2369 36988 : return heap_toast_insert_or_update(relation, tup, NULL, options);
2370 : else
2371 19682144 : return tup;
2372 : }
2373 :
2374 : /*
2375 : * Helper for heap_multi_insert() that computes the number of entire pages
2376 : * that inserting the remaining heaptuples requires. Used to determine how
2377 : * much the relation needs to be extended by.
2378 : */
2379 : static int
2380 761652 : heap_multi_insert_pages(HeapTuple *heaptuples, int done, int ntuples, Size saveFreeSpace)
2381 : {
2382 761652 : size_t page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
2383 761652 : int npages = 1;
2384 :
2385 4964594 : for (int i = done; i < ntuples; i++)
2386 : {
2387 4202942 : size_t tup_sz = sizeof(ItemIdData) + MAXALIGN(heaptuples[i]->t_len);
2388 :
2389 4202942 : if (page_avail < tup_sz)
2390 : {
2391 31200 : npages++;
2392 31200 : page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
2393 : }
2394 4202942 : page_avail -= tup_sz;
2395 : }
2396 :
2397 761652 : return npages;
2398 : }
2399 :
2400 : /*
2401 : * heap_multi_insert - insert multiple tuples into a heap
2402 : *
2403 : * This is like heap_insert(), but inserts multiple tuples in one operation.
2404 : * That's faster than calling heap_insert() in a loop, because when multiple
2405 : * tuples can be inserted on a single page, we can write just a single WAL
2406 : * record covering all of them, and only need to lock/unlock the page once.
2407 : *
2408 : * Note: this leaks memory into the current memory context. You can create a
2409 : * temporary context before calling this, if that's a problem.
2410 : */
2411 : void
2412 748104 : heap_multi_insert(Relation relation, TupleTableSlot **slots, int ntuples,
2413 : CommandId cid, int options, BulkInsertState bistate)
2414 : {
2415 748104 : TransactionId xid = GetCurrentTransactionId();
2416 : HeapTuple *heaptuples;
2417 : int i;
2418 : int ndone;
2419 : PGAlignedBlock scratch;
2420 : Page page;
2421 748104 : Buffer vmbuffer = InvalidBuffer;
2422 : bool needwal;
2423 : Size saveFreeSpace;
2424 748104 : bool need_tuple_data = RelationIsLogicallyLogged(relation);
2425 748104 : bool need_cids = RelationIsAccessibleInLogicalDecoding(relation);
2426 748104 : bool starting_with_empty_page = false;
2427 748104 : int npages = 0;
2428 748104 : int npages_used = 0;
2429 :
2430 : /* currently not needed (thus unsupported) for heap_multi_insert() */
2431 : Assert(!(options & HEAP_INSERT_NO_LOGICAL));
2432 :
2433 748104 : AssertHasSnapshotForToast(relation);
2434 :
2435 748104 : needwal = RelationNeedsWAL(relation);
2436 748104 : saveFreeSpace = RelationGetTargetPageFreeSpace(relation,
2437 : HEAP_DEFAULT_FILLFACTOR);
2438 :
2439 : /* Toast and set header data in all the slots */
2440 748104 : heaptuples = palloc(ntuples * sizeof(HeapTuple));
2441 3760930 : for (i = 0; i < ntuples; i++)
2442 : {
2443 : HeapTuple tuple;
2444 :
2445 3012826 : tuple = ExecFetchSlotHeapTuple(slots[i], true, NULL);
2446 3012826 : slots[i]->tts_tableOid = RelationGetRelid(relation);
2447 3012826 : tuple->t_tableOid = slots[i]->tts_tableOid;
2448 3012826 : heaptuples[i] = heap_prepare_insert(relation, tuple, xid, cid,
2449 : options);
2450 : }
2451 :
2452 : /*
2453 : * We're about to do the actual inserts -- but check for conflict first,
2454 : * to minimize the possibility of having to roll back work we've just
2455 : * done.
2456 : *
2457 : * A check here does not definitively prevent a serialization anomaly;
2458 : * that check MUST be done at least past the point of acquiring an
2459 : * exclusive buffer content lock on every buffer that will be affected,
2460 : * and MAY be done after all inserts are reflected in the buffers and
2461 : * those locks are released; otherwise there is a race condition. Since
2462 : * multiple buffers can be locked and unlocked in the loop below, and it
2463 : * would not be feasible to identify and lock all of those buffers before
2464 : * the loop, we must do a final check at the end.
2465 : *
2466 : * The check here could be omitted with no loss of correctness; it is
2467 : * present strictly as an optimization.
2468 : *
2469 : * For heap inserts, we only need to check for table-level SSI locks. Our
2470 : * new tuples can't possibly conflict with existing tuple locks, and heap
2471 : * page locks are only consolidated versions of tuple locks; they do not
2472 : * lock "gaps" as index page locks do. So we don't need to specify a
2473 : * buffer when making the call, which makes for a faster check.
2474 : */
2475 748104 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2476 :
2477 748104 : ndone = 0;
2478 1526166 : while (ndone < ntuples)
2479 : {
2480 : Buffer buffer;
2481 778062 : bool all_visible_cleared = false;
2482 778062 : bool all_frozen_set = false;
2483 : int nthispage;
2484 :
2485 778062 : CHECK_FOR_INTERRUPTS();
2486 :
2487 : /*
2488 : * Compute number of pages needed to fit the to-be-inserted tuples in
2489 : * the worst case. This will be used to determine how much to extend
2490 : * the relation by in RelationGetBufferForTuple(), if needed. If we
2491 : * filled a prior page from scratch, we can just update our last
2492 : * computation, but if we started with a partially filled page,
2493 : * recompute from scratch, the number of potentially required pages
2494 : * can vary due to tuples needing to fit onto the page, page headers
2495 : * etc.
2496 : */
2497 778062 : if (ndone == 0 || !starting_with_empty_page)
2498 : {
2499 761652 : npages = heap_multi_insert_pages(heaptuples, ndone, ntuples,
2500 : saveFreeSpace);
2501 761652 : npages_used = 0;
2502 : }
2503 : else
2504 16410 : npages_used++;
2505 :
2506 : /*
2507 : * Find buffer where at least the next tuple will fit. If the page is
2508 : * all-visible, this will also pin the requisite visibility map page.
2509 : *
2510 : * Also pin visibility map page if COPY FREEZE inserts tuples into an
2511 : * empty page. See all_frozen_set below.
2512 : */
2513 778062 : buffer = RelationGetBufferForTuple(relation, heaptuples[ndone]->t_len,
2514 : InvalidBuffer, options, bistate,
2515 : &vmbuffer, NULL,
2516 : npages - npages_used);
2517 778062 : page = BufferGetPage(buffer);
2518 :
2519 778062 : starting_with_empty_page = PageGetMaxOffsetNumber(page) == 0;
2520 :
2521 778062 : if (starting_with_empty_page && (options & HEAP_INSERT_FROZEN))
2522 : {
2523 3322 : all_frozen_set = true;
2524 : /* Lock the vmbuffer before entering the critical section */
2525 3322 : LockBuffer(vmbuffer, BUFFER_LOCK_EXCLUSIVE);
2526 : }
2527 :
2528 : /* NO EREPORT(ERROR) from here till changes are logged */
2529 778062 : START_CRIT_SECTION();
2530 :
2531 : /*
2532 : * RelationGetBufferForTuple has ensured that the first tuple fits.
2533 : * Put that on the page, and then as many other tuples as fit.
2534 : */
2535 778062 : RelationPutHeapTuple(relation, buffer, heaptuples[ndone], false);
2536 :
2537 : /*
2538 : * For logical decoding we need combo CIDs to properly decode the
2539 : * catalog.
2540 : */
2541 778062 : if (needwal && need_cids)
2542 10124 : log_heap_new_cid(relation, heaptuples[ndone]);
2543 :
2544 3012826 : for (nthispage = 1; ndone + nthispage < ntuples; nthispage++)
2545 : {
2546 2264722 : HeapTuple heaptup = heaptuples[ndone + nthispage];
2547 :
2548 2264722 : if (PageGetHeapFreeSpace(page) < MAXALIGN(heaptup->t_len) + saveFreeSpace)
2549 29958 : break;
2550 :
2551 2234764 : RelationPutHeapTuple(relation, buffer, heaptup, false);
2552 :
2553 : /*
2554 : * For logical decoding we need combo CIDs to properly decode the
2555 : * catalog.
2556 : */
2557 2234764 : if (needwal && need_cids)
2558 9478 : log_heap_new_cid(relation, heaptup);
2559 : }
2560 :
2561 : /*
2562 : * If the page is all visible, need to clear that, unless we're only
2563 : * going to add further frozen rows to it.
2564 : *
2565 : * If we're only adding already frozen rows to a previously empty
2566 : * page, mark it as all-frozen and update the visibility map. We're
2567 : * already holding a pin on the vmbuffer.
2568 : */
2569 778062 : if (PageIsAllVisible(page) && !(options & HEAP_INSERT_FROZEN))
2570 : {
2571 5876 : all_visible_cleared = true;
2572 5876 : PageClearAllVisible(page);
2573 5876 : visibilitymap_clear(relation,
2574 : BufferGetBlockNumber(buffer),
2575 : vmbuffer, VISIBILITYMAP_VALID_BITS);
2576 : }
2577 772186 : else if (all_frozen_set)
2578 : {
2579 3322 : PageSetAllVisible(page);
2580 3322 : visibilitymap_set_vmbits(BufferGetBlockNumber(buffer),
2581 : vmbuffer,
2582 : VISIBILITYMAP_ALL_VISIBLE |
2583 : VISIBILITYMAP_ALL_FROZEN,
2584 : relation->rd_locator);
2585 : }
2586 :
2587 : /*
2588 : * XXX Should we set PageSetPrunable on this page ? See heap_insert()
2589 : */
2590 :
2591 778062 : MarkBufferDirty(buffer);
2592 :
2593 : /* XLOG stuff */
2594 778062 : if (needwal)
2595 : {
2596 : XLogRecPtr recptr;
2597 : xl_heap_multi_insert *xlrec;
2598 770410 : uint8 info = XLOG_HEAP2_MULTI_INSERT;
2599 : char *tupledata;
2600 : int totaldatalen;
2601 770410 : char *scratchptr = scratch.data;
2602 : bool init;
2603 770410 : int bufflags = 0;
2604 :
2605 : /*
2606 : * If the page was previously empty, we can reinit the page
2607 : * instead of restoring the whole thing.
2608 : */
2609 770410 : init = starting_with_empty_page;
2610 :
2611 : /* allocate xl_heap_multi_insert struct from the scratch area */
2612 770410 : xlrec = (xl_heap_multi_insert *) scratchptr;
2613 770410 : scratchptr += SizeOfHeapMultiInsert;
2614 :
2615 : /*
2616 : * Allocate offsets array. Unless we're reinitializing the page,
2617 : * in that case the tuples are stored in order starting at
2618 : * FirstOffsetNumber and we don't need to store the offsets
2619 : * explicitly.
2620 : */
2621 770410 : if (!init)
2622 743454 : scratchptr += nthispage * sizeof(OffsetNumber);
2623 :
2624 : /* the rest of the scratch space is used for tuple data */
2625 770410 : tupledata = scratchptr;
2626 :
2627 : /* check that the mutually exclusive flags are not both set */
2628 : Assert(!(all_visible_cleared && all_frozen_set));
2629 :
2630 770410 : xlrec->flags = 0;
2631 770410 : if (all_visible_cleared)
2632 5876 : xlrec->flags = XLH_INSERT_ALL_VISIBLE_CLEARED;
2633 :
2634 : /*
2635 : * We don't have to worry about including a conflict xid in the
2636 : * WAL record, as HEAP_INSERT_FROZEN intentionally violates
2637 : * visibility rules.
2638 : */
2639 770410 : if (all_frozen_set)
2640 34 : xlrec->flags = XLH_INSERT_ALL_FROZEN_SET;
2641 :
2642 770410 : xlrec->ntuples = nthispage;
2643 :
2644 : /*
2645 : * Write out an xl_multi_insert_tuple and the tuple data itself
2646 : * for each tuple.
2647 : */
2648 3372420 : for (i = 0; i < nthispage; i++)
2649 : {
2650 2602010 : HeapTuple heaptup = heaptuples[ndone + i];
2651 : xl_multi_insert_tuple *tuphdr;
2652 : int datalen;
2653 :
2654 2602010 : if (!init)
2655 1547178 : xlrec->offsets[i] = ItemPointerGetOffsetNumber(&heaptup->t_self);
2656 : /* xl_multi_insert_tuple needs two-byte alignment. */
2657 2602010 : tuphdr = (xl_multi_insert_tuple *) SHORTALIGN(scratchptr);
2658 2602010 : scratchptr = ((char *) tuphdr) + SizeOfMultiInsertTuple;
2659 :
2660 2602010 : tuphdr->t_infomask2 = heaptup->t_data->t_infomask2;
2661 2602010 : tuphdr->t_infomask = heaptup->t_data->t_infomask;
2662 2602010 : tuphdr->t_hoff = heaptup->t_data->t_hoff;
2663 :
2664 : /* write bitmap [+ padding] [+ oid] + data */
2665 2602010 : datalen = heaptup->t_len - SizeofHeapTupleHeader;
2666 2602010 : memcpy(scratchptr,
2667 2602010 : (char *) heaptup->t_data + SizeofHeapTupleHeader,
2668 : datalen);
2669 2602010 : tuphdr->datalen = datalen;
2670 2602010 : scratchptr += datalen;
2671 : }
2672 770410 : totaldatalen = scratchptr - tupledata;
2673 : Assert((scratchptr - scratch.data) < BLCKSZ);
2674 :
2675 770410 : if (need_tuple_data)
2676 144 : xlrec->flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
2677 :
2678 : /*
2679 : * Signal that this is the last xl_heap_multi_insert record
2680 : * emitted by this call to heap_multi_insert(). Needed for logical
2681 : * decoding so it knows when to cleanup temporary data.
2682 : */
2683 770410 : if (ndone + nthispage == ntuples)
2684 747270 : xlrec->flags |= XLH_INSERT_LAST_IN_MULTI;
2685 :
2686 770410 : if (init)
2687 : {
2688 26956 : info |= XLOG_HEAP_INIT_PAGE;
2689 26956 : bufflags |= REGBUF_WILL_INIT;
2690 : }
2691 :
2692 : /*
2693 : * If we're doing logical decoding, include the new tuple data
2694 : * even if we take a full-page image of the page.
2695 : */
2696 770410 : if (need_tuple_data)
2697 144 : bufflags |= REGBUF_KEEP_DATA;
2698 :
2699 770410 : XLogBeginInsert();
2700 770410 : XLogRegisterData(xlrec, tupledata - scratch.data);
2701 770410 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD | bufflags);
2702 770410 : if (all_frozen_set)
2703 34 : XLogRegisterBuffer(1, vmbuffer, 0);
2704 :
2705 770410 : XLogRegisterBufData(0, tupledata, totaldatalen);
2706 :
2707 : /* filtering by origin on a row level is much more efficient */
2708 770410 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
2709 :
2710 770410 : recptr = XLogInsert(RM_HEAP2_ID, info);
2711 :
2712 770410 : PageSetLSN(page, recptr);
2713 770410 : if (all_frozen_set)
2714 : {
2715 : Assert(BufferIsDirty(vmbuffer));
2716 34 : PageSetLSN(BufferGetPage(vmbuffer), recptr);
2717 : }
2718 : }
2719 :
2720 778062 : END_CRIT_SECTION();
2721 :
2722 778062 : if (all_frozen_set)
2723 3322 : LockBuffer(vmbuffer, BUFFER_LOCK_UNLOCK);
2724 :
2725 778062 : UnlockReleaseBuffer(buffer);
2726 778062 : ndone += nthispage;
2727 :
2728 : /*
2729 : * NB: Only release vmbuffer after inserting all tuples - it's fairly
2730 : * likely that we'll insert into subsequent heap pages that are likely
2731 : * to use the same vm page.
2732 : */
2733 : }
2734 :
2735 : /* We're done with inserting all tuples, so release the last vmbuffer. */
2736 748104 : if (vmbuffer != InvalidBuffer)
2737 6078 : ReleaseBuffer(vmbuffer);
2738 :
2739 : /*
2740 : * We're done with the actual inserts. Check for conflicts again, to
2741 : * ensure that all rw-conflicts in to these inserts are detected. Without
2742 : * this final check, a sequential scan of the heap may have locked the
2743 : * table after the "before" check, missing one opportunity to detect the
2744 : * conflict, and then scanned the table before the new tuples were there,
2745 : * missing the other chance to detect the conflict.
2746 : *
2747 : * For heap inserts, we only need to check for table-level SSI locks. Our
2748 : * new tuples can't possibly conflict with existing tuple locks, and heap
2749 : * page locks are only consolidated versions of tuple locks; they do not
2750 : * lock "gaps" as index page locks do. So we don't need to specify a
2751 : * buffer when making the call.
2752 : */
2753 748104 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2754 :
2755 : /*
2756 : * If tuples are cacheable, mark them for invalidation from the caches in
2757 : * case we abort. Note it is OK to do this after releasing the buffer,
2758 : * because the heaptuples data structure is all in local memory, not in
2759 : * the shared buffer.
2760 : */
2761 748104 : if (IsCatalogRelation(relation))
2762 : {
2763 2553810 : for (i = 0; i < ntuples; i++)
2764 1808154 : CacheInvalidateHeapTuple(relation, heaptuples[i], NULL);
2765 : }
2766 :
2767 : /* copy t_self fields back to the caller's slots */
2768 3760930 : for (i = 0; i < ntuples; i++)
2769 3012826 : slots[i]->tts_tid = heaptuples[i]->t_self;
2770 :
2771 748104 : pgstat_count_heap_insert(relation, ntuples);
2772 748104 : }
2773 :
2774 : /*
2775 : * simple_heap_insert - insert a tuple
2776 : *
2777 : * Currently, this routine differs from heap_insert only in supplying
2778 : * a default command ID and not allowing access to the speedup options.
2779 : *
2780 : * This should be used rather than using heap_insert directly in most places
2781 : * where we are modifying system catalogs.
2782 : */
2783 : void
2784 1856018 : simple_heap_insert(Relation relation, HeapTuple tup)
2785 : {
2786 1856018 : heap_insert(relation, tup, GetCurrentCommandId(true), 0, NULL);
2787 1856018 : }
2788 :
2789 : /*
2790 : * Given infomask/infomask2, compute the bits that must be saved in the
2791 : * "infobits" field of xl_heap_delete, xl_heap_update, xl_heap_lock,
2792 : * xl_heap_lock_updated WAL records.
2793 : *
2794 : * See fix_infomask_from_infobits.
2795 : */
2796 : static uint8
2797 4128262 : compute_infobits(uint16 infomask, uint16 infomask2)
2798 : {
2799 : return
2800 4128262 : ((infomask & HEAP_XMAX_IS_MULTI) != 0 ? XLHL_XMAX_IS_MULTI : 0) |
2801 4128262 : ((infomask & HEAP_XMAX_LOCK_ONLY) != 0 ? XLHL_XMAX_LOCK_ONLY : 0) |
2802 4128262 : ((infomask & HEAP_XMAX_EXCL_LOCK) != 0 ? XLHL_XMAX_EXCL_LOCK : 0) |
2803 : /* note we ignore HEAP_XMAX_SHR_LOCK here */
2804 8256524 : ((infomask & HEAP_XMAX_KEYSHR_LOCK) != 0 ? XLHL_XMAX_KEYSHR_LOCK : 0) |
2805 : ((infomask2 & HEAP_KEYS_UPDATED) != 0 ?
2806 4128262 : XLHL_KEYS_UPDATED : 0);
2807 : }
2808 :
2809 : /*
2810 : * Given two versions of the same t_infomask for a tuple, compare them and
2811 : * return whether the relevant status for a tuple Xmax has changed. This is
2812 : * used after a buffer lock has been released and reacquired: we want to ensure
2813 : * that the tuple state continues to be the same it was when we previously
2814 : * examined it.
2815 : *
2816 : * Note the Xmax field itself must be compared separately.
2817 : */
2818 : static inline bool
2819 10758 : xmax_infomask_changed(uint16 new_infomask, uint16 old_infomask)
2820 : {
2821 10758 : const uint16 interesting =
2822 : HEAP_XMAX_IS_MULTI | HEAP_XMAX_LOCK_ONLY | HEAP_LOCK_MASK;
2823 :
2824 10758 : if ((new_infomask & interesting) != (old_infomask & interesting))
2825 32 : return true;
2826 :
2827 10726 : return false;
2828 : }
2829 :
2830 : /*
2831 : * heap_delete - delete a tuple
2832 : *
2833 : * See table_tuple_delete() for an explanation of the parameters, except that
2834 : * this routine directly takes a tuple rather than a slot.
2835 : *
2836 : * In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
2837 : * t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
2838 : * only for TM_SelfModified, since we cannot obtain cmax from a combo CID
2839 : * generated by another transaction).
2840 : */
2841 : TM_Result
2842 3052000 : heap_delete(Relation relation, const ItemPointerData *tid,
2843 : CommandId cid, Snapshot crosscheck, bool wait,
2844 : TM_FailureData *tmfd, bool changingPart)
2845 : {
2846 : TM_Result result;
2847 3052000 : TransactionId xid = GetCurrentTransactionId();
2848 : ItemId lp;
2849 : HeapTupleData tp;
2850 : Page page;
2851 : BlockNumber block;
2852 : Buffer buffer;
2853 3052000 : Buffer vmbuffer = InvalidBuffer;
2854 : TransactionId new_xmax;
2855 : uint16 new_infomask,
2856 : new_infomask2;
2857 3052000 : bool have_tuple_lock = false;
2858 : bool iscombo;
2859 3052000 : bool all_visible_cleared = false;
2860 3052000 : HeapTuple old_key_tuple = NULL; /* replica identity of the tuple */
2861 3052000 : bool old_key_copied = false;
2862 :
2863 : Assert(ItemPointerIsValid(tid));
2864 :
2865 3052000 : AssertHasSnapshotForToast(relation);
2866 :
2867 : /*
2868 : * Forbid this during a parallel operation, lest it allocate a combo CID.
2869 : * Other workers might need that combo CID for visibility checks, and we
2870 : * have no provision for broadcasting it to them.
2871 : */
2872 3052000 : if (IsInParallelMode())
2873 0 : ereport(ERROR,
2874 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
2875 : errmsg("cannot delete tuples during a parallel operation")));
2876 :
2877 3052000 : block = ItemPointerGetBlockNumber(tid);
2878 3052000 : buffer = ReadBuffer(relation, block);
2879 3052000 : page = BufferGetPage(buffer);
2880 :
2881 : /*
2882 : * Before locking the buffer, pin the visibility map page if it appears to
2883 : * be necessary. Since we haven't got the lock yet, someone else might be
2884 : * in the middle of changing this, so we'll need to recheck after we have
2885 : * the lock.
2886 : */
2887 3052000 : if (PageIsAllVisible(page))
2888 476 : visibilitymap_pin(relation, block, &vmbuffer);
2889 :
2890 3052000 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2891 :
2892 3052000 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
2893 : Assert(ItemIdIsNormal(lp));
2894 :
2895 3052000 : tp.t_tableOid = RelationGetRelid(relation);
2896 3052000 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
2897 3052000 : tp.t_len = ItemIdGetLength(lp);
2898 3052000 : tp.t_self = *tid;
2899 :
2900 2 : l1:
2901 :
2902 : /*
2903 : * If we didn't pin the visibility map page and the page has become all
2904 : * visible while we were busy locking the buffer, we'll have to unlock and
2905 : * re-lock, to avoid holding the buffer lock across an I/O. That's a bit
2906 : * unfortunate, but hopefully shouldn't happen often.
2907 : */
2908 3052002 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
2909 : {
2910 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2911 0 : visibilitymap_pin(relation, block, &vmbuffer);
2912 0 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2913 : }
2914 :
2915 3052002 : result = HeapTupleSatisfiesUpdate(&tp, cid, buffer);
2916 :
2917 3052002 : if (result == TM_Invisible)
2918 : {
2919 0 : UnlockReleaseBuffer(buffer);
2920 0 : ereport(ERROR,
2921 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2922 : errmsg("attempted to delete invisible tuple")));
2923 : }
2924 3052002 : else if (result == TM_BeingModified && wait)
2925 : {
2926 : TransactionId xwait;
2927 : uint16 infomask;
2928 :
2929 : /* must copy state data before unlocking buffer */
2930 81124 : xwait = HeapTupleHeaderGetRawXmax(tp.t_data);
2931 81124 : infomask = tp.t_data->t_infomask;
2932 :
2933 : /*
2934 : * Sleep until concurrent transaction ends -- except when there's a
2935 : * single locker and it's our own transaction. Note we don't care
2936 : * which lock mode the locker has, because we need the strongest one.
2937 : *
2938 : * Before sleeping, we need to acquire tuple lock to establish our
2939 : * priority for the tuple (see heap_lock_tuple). LockTuple will
2940 : * release us when we are next-in-line for the tuple.
2941 : *
2942 : * If we are forced to "start over" below, we keep the tuple lock;
2943 : * this arranges that we stay at the head of the line while rechecking
2944 : * tuple state.
2945 : */
2946 81124 : if (infomask & HEAP_XMAX_IS_MULTI)
2947 : {
2948 16 : bool current_is_member = false;
2949 :
2950 16 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
2951 : LockTupleExclusive, ¤t_is_member))
2952 : {
2953 16 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2954 :
2955 : /*
2956 : * Acquire the lock, if necessary (but skip it when we're
2957 : * requesting a lock and already have one; avoids deadlock).
2958 : */
2959 16 : if (!current_is_member)
2960 12 : heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
2961 : LockWaitBlock, &have_tuple_lock);
2962 :
2963 : /* wait for multixact */
2964 16 : MultiXactIdWait((MultiXactId) xwait, MultiXactStatusUpdate, infomask,
2965 : relation, &(tp.t_self), XLTW_Delete,
2966 : NULL);
2967 16 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2968 :
2969 : /*
2970 : * If xwait had just locked the tuple then some other xact
2971 : * could update this tuple before we get to this point. Check
2972 : * for xmax change, and start over if so.
2973 : *
2974 : * We also must start over if we didn't pin the VM page, and
2975 : * the page has become all visible.
2976 : */
2977 32 : if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
2978 32 : xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
2979 16 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
2980 : xwait))
2981 0 : goto l1;
2982 : }
2983 :
2984 : /*
2985 : * You might think the multixact is necessarily done here, but not
2986 : * so: it could have surviving members, namely our own xact or
2987 : * other subxacts of this backend. It is legal for us to delete
2988 : * the tuple in either case, however (the latter case is
2989 : * essentially a situation of upgrading our former shared lock to
2990 : * exclusive). We don't bother changing the on-disk hint bits
2991 : * since we are about to overwrite the xmax altogether.
2992 : */
2993 : }
2994 81108 : else if (!TransactionIdIsCurrentTransactionId(xwait))
2995 : {
2996 : /*
2997 : * Wait for regular transaction to end; but first, acquire tuple
2998 : * lock.
2999 : */
3000 104 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3001 104 : heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
3002 : LockWaitBlock, &have_tuple_lock);
3003 104 : XactLockTableWait(xwait, relation, &(tp.t_self), XLTW_Delete);
3004 96 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3005 :
3006 : /*
3007 : * xwait is done, but if xwait had just locked the tuple then some
3008 : * other xact could update this tuple before we get to this point.
3009 : * Check for xmax change, and start over if so.
3010 : *
3011 : * We also must start over if we didn't pin the VM page, and the
3012 : * page has become all visible.
3013 : */
3014 192 : if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
3015 190 : xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
3016 94 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
3017 : xwait))
3018 2 : goto l1;
3019 :
3020 : /* Otherwise check if it committed or aborted */
3021 94 : UpdateXmaxHintBits(tp.t_data, buffer, xwait);
3022 : }
3023 :
3024 : /*
3025 : * We may overwrite if previous xmax aborted, or if it committed but
3026 : * only locked the tuple without updating it.
3027 : */
3028 162188 : if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
3029 81136 : HEAP_XMAX_IS_LOCKED_ONLY(tp.t_data->t_infomask) ||
3030 62 : HeapTupleHeaderIsOnlyLocked(tp.t_data))
3031 81060 : result = TM_Ok;
3032 54 : else if (!ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
3033 46 : result = TM_Updated;
3034 : else
3035 8 : result = TM_Deleted;
3036 : }
3037 :
3038 : /* sanity check the result HeapTupleSatisfiesUpdate() and the logic above */
3039 : if (result != TM_Ok)
3040 : {
3041 : Assert(result == TM_SelfModified ||
3042 : result == TM_Updated ||
3043 : result == TM_Deleted ||
3044 : result == TM_BeingModified);
3045 : Assert(!(tp.t_data->t_infomask & HEAP_XMAX_INVALID));
3046 : Assert(result != TM_Updated ||
3047 : !ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid));
3048 : }
3049 :
3050 3051992 : if (crosscheck != InvalidSnapshot && result == TM_Ok)
3051 : {
3052 : /* Perform additional check for transaction-snapshot mode RI updates */
3053 2 : if (!HeapTupleSatisfiesVisibility(&tp, crosscheck, buffer))
3054 2 : result = TM_Updated;
3055 : }
3056 :
3057 3051992 : if (result != TM_Ok)
3058 : {
3059 124 : tmfd->ctid = tp.t_data->t_ctid;
3060 124 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
3061 124 : if (result == TM_SelfModified)
3062 42 : tmfd->cmax = HeapTupleHeaderGetCmax(tp.t_data);
3063 : else
3064 82 : tmfd->cmax = InvalidCommandId;
3065 124 : UnlockReleaseBuffer(buffer);
3066 124 : if (have_tuple_lock)
3067 54 : UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
3068 124 : if (vmbuffer != InvalidBuffer)
3069 0 : ReleaseBuffer(vmbuffer);
3070 124 : return result;
3071 : }
3072 :
3073 : /*
3074 : * We're about to do the actual delete -- check for conflict first, to
3075 : * avoid possibly having to roll back work we've just done.
3076 : *
3077 : * This is safe without a recheck as long as there is no possibility of
3078 : * another process scanning the page between this check and the delete
3079 : * being visible to the scan (i.e., an exclusive buffer content lock is
3080 : * continuously held from this point until the tuple delete is visible).
3081 : */
3082 3051868 : CheckForSerializableConflictIn(relation, tid, BufferGetBlockNumber(buffer));
3083 :
3084 : /* replace cid with a combo CID if necessary */
3085 3051840 : HeapTupleHeaderAdjustCmax(tp.t_data, &cid, &iscombo);
3086 :
3087 : /*
3088 : * Compute replica identity tuple before entering the critical section so
3089 : * we don't PANIC upon a memory allocation failure.
3090 : */
3091 3051840 : old_key_tuple = ExtractReplicaIdentity(relation, &tp, true, &old_key_copied);
3092 :
3093 : /*
3094 : * If this is the first possibly-multixact-able operation in the current
3095 : * transaction, set my per-backend OldestMemberMXactId setting. We can be
3096 : * certain that the transaction will never become a member of any older
3097 : * MultiXactIds than that. (We have to do this even if we end up just
3098 : * using our own TransactionId below, since some other backend could
3099 : * incorporate our XID into a MultiXact immediately afterwards.)
3100 : */
3101 3051840 : MultiXactIdSetOldestMember();
3102 :
3103 3051840 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(tp.t_data),
3104 3051840 : tp.t_data->t_infomask, tp.t_data->t_infomask2,
3105 : xid, LockTupleExclusive, true,
3106 : &new_xmax, &new_infomask, &new_infomask2);
3107 :
3108 3051840 : START_CRIT_SECTION();
3109 :
3110 : /*
3111 : * If this transaction commits, the tuple will become DEAD sooner or
3112 : * later. Set flag that this page is a candidate for pruning once our xid
3113 : * falls below the OldestXmin horizon. If the transaction finally aborts,
3114 : * the subsequent page pruning will be a no-op and the hint will be
3115 : * cleared.
3116 : */
3117 3051840 : PageSetPrunable(page, xid);
3118 :
3119 3051840 : if (PageIsAllVisible(page))
3120 : {
3121 476 : all_visible_cleared = true;
3122 476 : PageClearAllVisible(page);
3123 476 : visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
3124 : vmbuffer, VISIBILITYMAP_VALID_BITS);
3125 : }
3126 :
3127 : /* store transaction information of xact deleting the tuple */
3128 3051840 : tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
3129 3051840 : tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
3130 3051840 : tp.t_data->t_infomask |= new_infomask;
3131 3051840 : tp.t_data->t_infomask2 |= new_infomask2;
3132 3051840 : HeapTupleHeaderClearHotUpdated(tp.t_data);
3133 3051840 : HeapTupleHeaderSetXmax(tp.t_data, new_xmax);
3134 3051840 : HeapTupleHeaderSetCmax(tp.t_data, cid, iscombo);
3135 : /* Make sure there is no forward chain link in t_ctid */
3136 3051840 : tp.t_data->t_ctid = tp.t_self;
3137 :
3138 : /* Signal that this is actually a move into another partition */
3139 3051840 : if (changingPart)
3140 986 : HeapTupleHeaderSetMovedPartitions(tp.t_data);
3141 :
3142 3051840 : MarkBufferDirty(buffer);
3143 :
3144 : /*
3145 : * XLOG stuff
3146 : *
3147 : * NB: heap_abort_speculative() uses the same xlog record and replay
3148 : * routines.
3149 : */
3150 3051840 : if (RelationNeedsWAL(relation))
3151 : {
3152 : xl_heap_delete xlrec;
3153 : xl_heap_header xlhdr;
3154 : XLogRecPtr recptr;
3155 :
3156 : /*
3157 : * For logical decode we need combo CIDs to properly decode the
3158 : * catalog
3159 : */
3160 2926652 : if (RelationIsAccessibleInLogicalDecoding(relation))
3161 12584 : log_heap_new_cid(relation, &tp);
3162 :
3163 2926652 : xlrec.flags = 0;
3164 2926652 : if (all_visible_cleared)
3165 476 : xlrec.flags |= XLH_DELETE_ALL_VISIBLE_CLEARED;
3166 2926652 : if (changingPart)
3167 986 : xlrec.flags |= XLH_DELETE_IS_PARTITION_MOVE;
3168 5853304 : xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
3169 2926652 : tp.t_data->t_infomask2);
3170 2926652 : xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
3171 2926652 : xlrec.xmax = new_xmax;
3172 :
3173 2926652 : if (old_key_tuple != NULL)
3174 : {
3175 94036 : if (relation->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
3176 264 : xlrec.flags |= XLH_DELETE_CONTAINS_OLD_TUPLE;
3177 : else
3178 93772 : xlrec.flags |= XLH_DELETE_CONTAINS_OLD_KEY;
3179 : }
3180 :
3181 2926652 : XLogBeginInsert();
3182 2926652 : XLogRegisterData(&xlrec, SizeOfHeapDelete);
3183 :
3184 2926652 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
3185 :
3186 : /*
3187 : * Log replica identity of the deleted tuple if there is one
3188 : */
3189 2926652 : if (old_key_tuple != NULL)
3190 : {
3191 94036 : xlhdr.t_infomask2 = old_key_tuple->t_data->t_infomask2;
3192 94036 : xlhdr.t_infomask = old_key_tuple->t_data->t_infomask;
3193 94036 : xlhdr.t_hoff = old_key_tuple->t_data->t_hoff;
3194 :
3195 94036 : XLogRegisterData(&xlhdr, SizeOfHeapHeader);
3196 94036 : XLogRegisterData((char *) old_key_tuple->t_data
3197 : + SizeofHeapTupleHeader,
3198 94036 : old_key_tuple->t_len
3199 : - SizeofHeapTupleHeader);
3200 : }
3201 :
3202 : /* filtering by origin on a row level is much more efficient */
3203 2926652 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
3204 :
3205 2926652 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
3206 :
3207 2926652 : PageSetLSN(page, recptr);
3208 : }
3209 :
3210 3051840 : END_CRIT_SECTION();
3211 :
3212 3051840 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3213 :
3214 3051840 : if (vmbuffer != InvalidBuffer)
3215 476 : ReleaseBuffer(vmbuffer);
3216 :
3217 : /*
3218 : * If the tuple has toasted out-of-line attributes, we need to delete
3219 : * those items too. We have to do this before releasing the buffer
3220 : * because we need to look at the contents of the tuple, but it's OK to
3221 : * release the content lock on the buffer first.
3222 : */
3223 3051840 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
3224 5152 : relation->rd_rel->relkind != RELKIND_MATVIEW)
3225 : {
3226 : /* toast table entries should never be recursively toasted */
3227 : Assert(!HeapTupleHasExternal(&tp));
3228 : }
3229 3046708 : else if (HeapTupleHasExternal(&tp))
3230 588 : heap_toast_delete(relation, &tp, false);
3231 :
3232 : /*
3233 : * Mark tuple for invalidation from system caches at next command
3234 : * boundary. We have to do this before releasing the buffer because we
3235 : * need to look at the contents of the tuple.
3236 : */
3237 3051840 : CacheInvalidateHeapTuple(relation, &tp, NULL);
3238 :
3239 : /* Now we can release the buffer */
3240 3051840 : ReleaseBuffer(buffer);
3241 :
3242 : /*
3243 : * Release the lmgr tuple lock, if we had it.
3244 : */
3245 3051840 : if (have_tuple_lock)
3246 52 : UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
3247 :
3248 3051840 : pgstat_count_heap_delete(relation);
3249 :
3250 3051840 : if (old_key_tuple != NULL && old_key_copied)
3251 93774 : heap_freetuple(old_key_tuple);
3252 :
3253 3051840 : return TM_Ok;
3254 : }
3255 :
3256 : /*
3257 : * simple_heap_delete - delete a tuple
3258 : *
3259 : * This routine may be used to delete a tuple when concurrent updates of
3260 : * the target tuple are not expected (for example, because we have a lock
3261 : * on the relation associated with the tuple). Any failure is reported
3262 : * via ereport().
3263 : */
3264 : void
3265 1320046 : simple_heap_delete(Relation relation, const ItemPointerData *tid)
3266 : {
3267 : TM_Result result;
3268 : TM_FailureData tmfd;
3269 :
3270 1320046 : result = heap_delete(relation, tid,
3271 : GetCurrentCommandId(true), InvalidSnapshot,
3272 : true /* wait for commit */ ,
3273 : &tmfd, false /* changingPart */ );
3274 1320046 : switch (result)
3275 : {
3276 0 : case TM_SelfModified:
3277 : /* Tuple was already updated in current command? */
3278 0 : elog(ERROR, "tuple already updated by self");
3279 : break;
3280 :
3281 1320046 : case TM_Ok:
3282 : /* done successfully */
3283 1320046 : break;
3284 :
3285 0 : case TM_Updated:
3286 0 : elog(ERROR, "tuple concurrently updated");
3287 : break;
3288 :
3289 0 : case TM_Deleted:
3290 0 : elog(ERROR, "tuple concurrently deleted");
3291 : break;
3292 :
3293 0 : default:
3294 0 : elog(ERROR, "unrecognized heap_delete status: %u", result);
3295 : break;
3296 : }
3297 1320046 : }
3298 :
3299 : /*
3300 : * heap_update - replace a tuple
3301 : *
3302 : * See table_tuple_update() for an explanation of the parameters, except that
3303 : * this routine directly takes a tuple rather than a slot.
3304 : *
3305 : * In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
3306 : * t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
3307 : * only for TM_SelfModified, since we cannot obtain cmax from a combo CID
3308 : * generated by another transaction).
3309 : */
3310 : TM_Result
3311 624282 : heap_update(Relation relation, const ItemPointerData *otid, HeapTuple newtup,
3312 : CommandId cid, Snapshot crosscheck, bool wait,
3313 : TM_FailureData *tmfd, LockTupleMode *lockmode,
3314 : TU_UpdateIndexes *update_indexes)
3315 : {
3316 : TM_Result result;
3317 624282 : TransactionId xid = GetCurrentTransactionId();
3318 : Bitmapset *hot_attrs;
3319 : Bitmapset *sum_attrs;
3320 : Bitmapset *key_attrs;
3321 : Bitmapset *id_attrs;
3322 : Bitmapset *interesting_attrs;
3323 : Bitmapset *modified_attrs;
3324 : ItemId lp;
3325 : HeapTupleData oldtup;
3326 : HeapTuple heaptup;
3327 624282 : HeapTuple old_key_tuple = NULL;
3328 624282 : bool old_key_copied = false;
3329 : Page page;
3330 : BlockNumber block;
3331 : MultiXactStatus mxact_status;
3332 : Buffer buffer,
3333 : newbuf,
3334 624282 : vmbuffer = InvalidBuffer,
3335 624282 : vmbuffer_new = InvalidBuffer;
3336 : bool need_toast;
3337 : Size newtupsize,
3338 : pagefree;
3339 624282 : bool have_tuple_lock = false;
3340 : bool iscombo;
3341 624282 : bool use_hot_update = false;
3342 624282 : bool summarized_update = false;
3343 : bool key_intact;
3344 624282 : bool all_visible_cleared = false;
3345 624282 : bool all_visible_cleared_new = false;
3346 : bool checked_lockers;
3347 : bool locker_remains;
3348 624282 : bool id_has_external = false;
3349 : TransactionId xmax_new_tuple,
3350 : xmax_old_tuple;
3351 : uint16 infomask_old_tuple,
3352 : infomask2_old_tuple,
3353 : infomask_new_tuple,
3354 : infomask2_new_tuple;
3355 :
3356 : Assert(ItemPointerIsValid(otid));
3357 :
3358 : /* Cheap, simplistic check that the tuple matches the rel's rowtype. */
3359 : Assert(HeapTupleHeaderGetNatts(newtup->t_data) <=
3360 : RelationGetNumberOfAttributes(relation));
3361 :
3362 624282 : AssertHasSnapshotForToast(relation);
3363 :
3364 : /*
3365 : * Forbid this during a parallel operation, lest it allocate a combo CID.
3366 : * Other workers might need that combo CID for visibility checks, and we
3367 : * have no provision for broadcasting it to them.
3368 : */
3369 624282 : if (IsInParallelMode())
3370 0 : ereport(ERROR,
3371 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
3372 : errmsg("cannot update tuples during a parallel operation")));
3373 :
3374 : #ifdef USE_ASSERT_CHECKING
3375 : check_lock_if_inplace_updateable_rel(relation, otid, newtup);
3376 : #endif
3377 :
3378 : /*
3379 : * Fetch the list of attributes to be checked for various operations.
3380 : *
3381 : * For HOT considerations, this is wasted effort if we fail to update or
3382 : * have to put the new tuple on a different page. But we must compute the
3383 : * list before obtaining buffer lock --- in the worst case, if we are
3384 : * doing an update on one of the relevant system catalogs, we could
3385 : * deadlock if we try to fetch the list later. In any case, the relcache
3386 : * caches the data so this is usually pretty cheap.
3387 : *
3388 : * We also need columns used by the replica identity and columns that are
3389 : * considered the "key" of rows in the table.
3390 : *
3391 : * Note that we get copies of each bitmap, so we need not worry about
3392 : * relcache flush happening midway through.
3393 : */
3394 624282 : hot_attrs = RelationGetIndexAttrBitmap(relation,
3395 : INDEX_ATTR_BITMAP_HOT_BLOCKING);
3396 624282 : sum_attrs = RelationGetIndexAttrBitmap(relation,
3397 : INDEX_ATTR_BITMAP_SUMMARIZED);
3398 624282 : key_attrs = RelationGetIndexAttrBitmap(relation, INDEX_ATTR_BITMAP_KEY);
3399 624282 : id_attrs = RelationGetIndexAttrBitmap(relation,
3400 : INDEX_ATTR_BITMAP_IDENTITY_KEY);
3401 624282 : interesting_attrs = NULL;
3402 624282 : interesting_attrs = bms_add_members(interesting_attrs, hot_attrs);
3403 624282 : interesting_attrs = bms_add_members(interesting_attrs, sum_attrs);
3404 624282 : interesting_attrs = bms_add_members(interesting_attrs, key_attrs);
3405 624282 : interesting_attrs = bms_add_members(interesting_attrs, id_attrs);
3406 :
3407 624282 : block = ItemPointerGetBlockNumber(otid);
3408 624282 : INJECTION_POINT("heap_update-before-pin", NULL);
3409 624282 : buffer = ReadBuffer(relation, block);
3410 624282 : page = BufferGetPage(buffer);
3411 :
3412 : /*
3413 : * Before locking the buffer, pin the visibility map page if it appears to
3414 : * be necessary. Since we haven't got the lock yet, someone else might be
3415 : * in the middle of changing this, so we'll need to recheck after we have
3416 : * the lock.
3417 : */
3418 624282 : if (PageIsAllVisible(page))
3419 3058 : visibilitymap_pin(relation, block, &vmbuffer);
3420 :
3421 624282 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3422 :
3423 624282 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(otid));
3424 :
3425 : /*
3426 : * Usually, a buffer pin and/or snapshot blocks pruning of otid, ensuring
3427 : * we see LP_NORMAL here. When the otid origin is a syscache, we may have
3428 : * neither a pin nor a snapshot. Hence, we may see other LP_ states, each
3429 : * of which indicates concurrent pruning.
3430 : *
3431 : * Failing with TM_Updated would be most accurate. However, unlike other
3432 : * TM_Updated scenarios, we don't know the successor ctid in LP_UNUSED and
3433 : * LP_DEAD cases. While the distinction between TM_Updated and TM_Deleted
3434 : * does matter to SQL statements UPDATE and MERGE, those SQL statements
3435 : * hold a snapshot that ensures LP_NORMAL. Hence, the choice between
3436 : * TM_Updated and TM_Deleted affects only the wording of error messages.
3437 : * Settle on TM_Deleted, for two reasons. First, it avoids complicating
3438 : * the specification of when tmfd->ctid is valid. Second, it creates
3439 : * error log evidence that we took this branch.
3440 : *
3441 : * Since it's possible to see LP_UNUSED at otid, it's also possible to see
3442 : * LP_NORMAL for a tuple that replaced LP_UNUSED. If it's a tuple for an
3443 : * unrelated row, we'll fail with "duplicate key value violates unique".
3444 : * XXX if otid is the live, newer version of the newtup row, we'll discard
3445 : * changes originating in versions of this catalog row after the version
3446 : * the caller got from syscache. See syscache-update-pruned.spec.
3447 : */
3448 624282 : if (!ItemIdIsNormal(lp))
3449 : {
3450 : Assert(RelationSupportsSysCache(RelationGetRelid(relation)));
3451 :
3452 2 : UnlockReleaseBuffer(buffer);
3453 : Assert(!have_tuple_lock);
3454 2 : if (vmbuffer != InvalidBuffer)
3455 2 : ReleaseBuffer(vmbuffer);
3456 2 : tmfd->ctid = *otid;
3457 2 : tmfd->xmax = InvalidTransactionId;
3458 2 : tmfd->cmax = InvalidCommandId;
3459 2 : *update_indexes = TU_None;
3460 :
3461 2 : bms_free(hot_attrs);
3462 2 : bms_free(sum_attrs);
3463 2 : bms_free(key_attrs);
3464 2 : bms_free(id_attrs);
3465 : /* modified_attrs not yet initialized */
3466 2 : bms_free(interesting_attrs);
3467 2 : return TM_Deleted;
3468 : }
3469 :
3470 : /*
3471 : * Fill in enough data in oldtup for HeapDetermineColumnsInfo to work
3472 : * properly.
3473 : */
3474 624280 : oldtup.t_tableOid = RelationGetRelid(relation);
3475 624280 : oldtup.t_data = (HeapTupleHeader) PageGetItem(page, lp);
3476 624280 : oldtup.t_len = ItemIdGetLength(lp);
3477 624280 : oldtup.t_self = *otid;
3478 :
3479 : /* the new tuple is ready, except for this: */
3480 624280 : newtup->t_tableOid = RelationGetRelid(relation);
3481 :
3482 : /*
3483 : * Determine columns modified by the update. Additionally, identify
3484 : * whether any of the unmodified replica identity key attributes in the
3485 : * old tuple is externally stored or not. This is required because for
3486 : * such attributes the flattened value won't be WAL logged as part of the
3487 : * new tuple so we must include it as part of the old_key_tuple. See
3488 : * ExtractReplicaIdentity.
3489 : */
3490 624280 : modified_attrs = HeapDetermineColumnsInfo(relation, interesting_attrs,
3491 : id_attrs, &oldtup,
3492 : newtup, &id_has_external);
3493 :
3494 : /*
3495 : * If we're not updating any "key" column, we can grab a weaker lock type.
3496 : * This allows for more concurrency when we are running simultaneously
3497 : * with foreign key checks.
3498 : *
3499 : * Note that if a column gets detoasted while executing the update, but
3500 : * the value ends up being the same, this test will fail and we will use
3501 : * the stronger lock. This is acceptable; the important case to optimize
3502 : * is updates that don't manipulate key columns, not those that
3503 : * serendipitously arrive at the same key values.
3504 : */
3505 624280 : if (!bms_overlap(modified_attrs, key_attrs))
3506 : {
3507 615656 : *lockmode = LockTupleNoKeyExclusive;
3508 615656 : mxact_status = MultiXactStatusNoKeyUpdate;
3509 615656 : key_intact = true;
3510 :
3511 : /*
3512 : * If this is the first possibly-multixact-able operation in the
3513 : * current transaction, set my per-backend OldestMemberMXactId
3514 : * setting. We can be certain that the transaction will never become a
3515 : * member of any older MultiXactIds than that. (We have to do this
3516 : * even if we end up just using our own TransactionId below, since
3517 : * some other backend could incorporate our XID into a MultiXact
3518 : * immediately afterwards.)
3519 : */
3520 615656 : MultiXactIdSetOldestMember();
3521 : }
3522 : else
3523 : {
3524 8624 : *lockmode = LockTupleExclusive;
3525 8624 : mxact_status = MultiXactStatusUpdate;
3526 8624 : key_intact = false;
3527 : }
3528 :
3529 : /*
3530 : * Note: beyond this point, use oldtup not otid to refer to old tuple.
3531 : * otid may very well point at newtup->t_self, which we will overwrite
3532 : * with the new tuple's location, so there's great risk of confusion if we
3533 : * use otid anymore.
3534 : */
3535 :
3536 624280 : l2:
3537 624282 : checked_lockers = false;
3538 624282 : locker_remains = false;
3539 624282 : result = HeapTupleSatisfiesUpdate(&oldtup, cid, buffer);
3540 :
3541 : /* see below about the "no wait" case */
3542 : Assert(result != TM_BeingModified || wait);
3543 :
3544 624282 : if (result == TM_Invisible)
3545 : {
3546 0 : UnlockReleaseBuffer(buffer);
3547 0 : ereport(ERROR,
3548 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3549 : errmsg("attempted to update invisible tuple")));
3550 : }
3551 624282 : else if (result == TM_BeingModified && wait)
3552 : {
3553 : TransactionId xwait;
3554 : uint16 infomask;
3555 72214 : bool can_continue = false;
3556 :
3557 : /*
3558 : * XXX note that we don't consider the "no wait" case here. This
3559 : * isn't a problem currently because no caller uses that case, but it
3560 : * should be fixed if such a caller is introduced. It wasn't a
3561 : * problem previously because this code would always wait, but now
3562 : * that some tuple locks do not conflict with one of the lock modes we
3563 : * use, it is possible that this case is interesting to handle
3564 : * specially.
3565 : *
3566 : * This may cause failures with third-party code that calls
3567 : * heap_update directly.
3568 : */
3569 :
3570 : /* must copy state data before unlocking buffer */
3571 72214 : xwait = HeapTupleHeaderGetRawXmax(oldtup.t_data);
3572 72214 : infomask = oldtup.t_data->t_infomask;
3573 :
3574 : /*
3575 : * Now we have to do something about the existing locker. If it's a
3576 : * multi, sleep on it; we might be awakened before it is completely
3577 : * gone (or even not sleep at all in some cases); we need to preserve
3578 : * it as locker, unless it is gone completely.
3579 : *
3580 : * If it's not a multi, we need to check for sleeping conditions
3581 : * before actually going to sleep. If the update doesn't conflict
3582 : * with the locks, we just continue without sleeping (but making sure
3583 : * it is preserved).
3584 : *
3585 : * Before sleeping, we need to acquire tuple lock to establish our
3586 : * priority for the tuple (see heap_lock_tuple). LockTuple will
3587 : * release us when we are next-in-line for the tuple. Note we must
3588 : * not acquire the tuple lock until we're sure we're going to sleep;
3589 : * otherwise we're open for race conditions with other transactions
3590 : * holding the tuple lock which sleep on us.
3591 : *
3592 : * If we are forced to "start over" below, we keep the tuple lock;
3593 : * this arranges that we stay at the head of the line while rechecking
3594 : * tuple state.
3595 : */
3596 72214 : if (infomask & HEAP_XMAX_IS_MULTI)
3597 : {
3598 : TransactionId update_xact;
3599 : int remain;
3600 358 : bool current_is_member = false;
3601 :
3602 358 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
3603 : *lockmode, ¤t_is_member))
3604 : {
3605 16 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3606 :
3607 : /*
3608 : * Acquire the lock, if necessary (but skip it when we're
3609 : * requesting a lock and already have one; avoids deadlock).
3610 : */
3611 16 : if (!current_is_member)
3612 0 : heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3613 : LockWaitBlock, &have_tuple_lock);
3614 :
3615 : /* wait for multixact */
3616 16 : MultiXactIdWait((MultiXactId) xwait, mxact_status, infomask,
3617 : relation, &oldtup.t_self, XLTW_Update,
3618 : &remain);
3619 16 : checked_lockers = true;
3620 16 : locker_remains = remain != 0;
3621 16 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3622 :
3623 : /*
3624 : * If xwait had just locked the tuple then some other xact
3625 : * could update this tuple before we get to this point. Check
3626 : * for xmax change, and start over if so.
3627 : */
3628 16 : if (xmax_infomask_changed(oldtup.t_data->t_infomask,
3629 16 : infomask) ||
3630 16 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3631 : xwait))
3632 0 : goto l2;
3633 : }
3634 :
3635 : /*
3636 : * Note that the multixact may not be done by now. It could have
3637 : * surviving members; our own xact or other subxacts of this
3638 : * backend, and also any other concurrent transaction that locked
3639 : * the tuple with LockTupleKeyShare if we only got
3640 : * LockTupleNoKeyExclusive. If this is the case, we have to be
3641 : * careful to mark the updated tuple with the surviving members in
3642 : * Xmax.
3643 : *
3644 : * Note that there could have been another update in the
3645 : * MultiXact. In that case, we need to check whether it committed
3646 : * or aborted. If it aborted we are safe to update it again;
3647 : * otherwise there is an update conflict, and we have to return
3648 : * TableTuple{Deleted, Updated} below.
3649 : *
3650 : * In the LockTupleExclusive case, we still need to preserve the
3651 : * surviving members: those would include the tuple locks we had
3652 : * before this one, which are important to keep in case this
3653 : * subxact aborts.
3654 : */
3655 358 : if (!HEAP_XMAX_IS_LOCKED_ONLY(oldtup.t_data->t_infomask))
3656 16 : update_xact = HeapTupleGetUpdateXid(oldtup.t_data);
3657 : else
3658 342 : update_xact = InvalidTransactionId;
3659 :
3660 : /*
3661 : * There was no UPDATE in the MultiXact; or it aborted. No
3662 : * TransactionIdIsInProgress() call needed here, since we called
3663 : * MultiXactIdWait() above.
3664 : */
3665 374 : if (!TransactionIdIsValid(update_xact) ||
3666 16 : TransactionIdDidAbort(update_xact))
3667 344 : can_continue = true;
3668 : }
3669 71856 : else if (TransactionIdIsCurrentTransactionId(xwait))
3670 : {
3671 : /*
3672 : * The only locker is ourselves; we can avoid grabbing the tuple
3673 : * lock here, but must preserve our locking information.
3674 : */
3675 71642 : checked_lockers = true;
3676 71642 : locker_remains = true;
3677 71642 : can_continue = true;
3678 : }
3679 214 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) && key_intact)
3680 : {
3681 : /*
3682 : * If it's just a key-share locker, and we're not changing the key
3683 : * columns, we don't need to wait for it to end; but we need to
3684 : * preserve it as locker.
3685 : */
3686 58 : checked_lockers = true;
3687 58 : locker_remains = true;
3688 58 : can_continue = true;
3689 : }
3690 : else
3691 : {
3692 : /*
3693 : * Wait for regular transaction to end; but first, acquire tuple
3694 : * lock.
3695 : */
3696 156 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3697 156 : heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3698 : LockWaitBlock, &have_tuple_lock);
3699 156 : XactLockTableWait(xwait, relation, &oldtup.t_self,
3700 : XLTW_Update);
3701 156 : checked_lockers = true;
3702 156 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3703 :
3704 : /*
3705 : * xwait is done, but if xwait had just locked the tuple then some
3706 : * other xact could update this tuple before we get to this point.
3707 : * Check for xmax change, and start over if so.
3708 : */
3709 310 : if (xmax_infomask_changed(oldtup.t_data->t_infomask, infomask) ||
3710 154 : !TransactionIdEquals(xwait,
3711 : HeapTupleHeaderGetRawXmax(oldtup.t_data)))
3712 2 : goto l2;
3713 :
3714 : /* Otherwise check if it committed or aborted */
3715 154 : UpdateXmaxHintBits(oldtup.t_data, buffer, xwait);
3716 154 : if (oldtup.t_data->t_infomask & HEAP_XMAX_INVALID)
3717 44 : can_continue = true;
3718 : }
3719 :
3720 72212 : if (can_continue)
3721 72088 : result = TM_Ok;
3722 124 : else if (!ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid))
3723 114 : result = TM_Updated;
3724 : else
3725 10 : result = TM_Deleted;
3726 : }
3727 :
3728 : /* Sanity check the result HeapTupleSatisfiesUpdate() and the logic above */
3729 : if (result != TM_Ok)
3730 : {
3731 : Assert(result == TM_SelfModified ||
3732 : result == TM_Updated ||
3733 : result == TM_Deleted ||
3734 : result == TM_BeingModified);
3735 : Assert(!(oldtup.t_data->t_infomask & HEAP_XMAX_INVALID));
3736 : Assert(result != TM_Updated ||
3737 : !ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid));
3738 : }
3739 :
3740 624280 : if (crosscheck != InvalidSnapshot && result == TM_Ok)
3741 : {
3742 : /* Perform additional check for transaction-snapshot mode RI updates */
3743 2 : if (!HeapTupleSatisfiesVisibility(&oldtup, crosscheck, buffer))
3744 2 : result = TM_Updated;
3745 : }
3746 :
3747 624280 : if (result != TM_Ok)
3748 : {
3749 320 : tmfd->ctid = oldtup.t_data->t_ctid;
3750 320 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(oldtup.t_data);
3751 320 : if (result == TM_SelfModified)
3752 104 : tmfd->cmax = HeapTupleHeaderGetCmax(oldtup.t_data);
3753 : else
3754 216 : tmfd->cmax = InvalidCommandId;
3755 320 : UnlockReleaseBuffer(buffer);
3756 320 : if (have_tuple_lock)
3757 110 : UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
3758 320 : if (vmbuffer != InvalidBuffer)
3759 0 : ReleaseBuffer(vmbuffer);
3760 320 : *update_indexes = TU_None;
3761 :
3762 320 : bms_free(hot_attrs);
3763 320 : bms_free(sum_attrs);
3764 320 : bms_free(key_attrs);
3765 320 : bms_free(id_attrs);
3766 320 : bms_free(modified_attrs);
3767 320 : bms_free(interesting_attrs);
3768 320 : return result;
3769 : }
3770 :
3771 : /*
3772 : * If we didn't pin the visibility map page and the page has become all
3773 : * visible while we were busy locking the buffer, or during some
3774 : * subsequent window during which we had it unlocked, we'll have to unlock
3775 : * and re-lock, to avoid holding the buffer lock across an I/O. That's a
3776 : * bit unfortunate, especially since we'll now have to recheck whether the
3777 : * tuple has been locked or updated under us, but hopefully it won't
3778 : * happen very often.
3779 : */
3780 623960 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
3781 : {
3782 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3783 0 : visibilitymap_pin(relation, block, &vmbuffer);
3784 0 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3785 0 : goto l2;
3786 : }
3787 :
3788 : /* Fill in transaction status data */
3789 :
3790 : /*
3791 : * If the tuple we're updating is locked, we need to preserve the locking
3792 : * info in the old tuple's Xmax. Prepare a new Xmax value for this.
3793 : */
3794 623960 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3795 623960 : oldtup.t_data->t_infomask,
3796 623960 : oldtup.t_data->t_infomask2,
3797 : xid, *lockmode, true,
3798 : &xmax_old_tuple, &infomask_old_tuple,
3799 : &infomask2_old_tuple);
3800 :
3801 : /*
3802 : * And also prepare an Xmax value for the new copy of the tuple. If there
3803 : * was no xmax previously, or there was one but all lockers are now gone,
3804 : * then use InvalidTransactionId; otherwise, get the xmax from the old
3805 : * tuple. (In rare cases that might also be InvalidTransactionId and yet
3806 : * not have the HEAP_XMAX_INVALID bit set; that's fine.)
3807 : */
3808 696004 : if ((oldtup.t_data->t_infomask & HEAP_XMAX_INVALID) ||
3809 144088 : HEAP_LOCKED_UPGRADED(oldtup.t_data->t_infomask) ||
3810 71702 : (checked_lockers && !locker_remains))
3811 551916 : xmax_new_tuple = InvalidTransactionId;
3812 : else
3813 72044 : xmax_new_tuple = HeapTupleHeaderGetRawXmax(oldtup.t_data);
3814 :
3815 623960 : if (!TransactionIdIsValid(xmax_new_tuple))
3816 : {
3817 551916 : infomask_new_tuple = HEAP_XMAX_INVALID;
3818 551916 : infomask2_new_tuple = 0;
3819 : }
3820 : else
3821 : {
3822 : /*
3823 : * If we found a valid Xmax for the new tuple, then the infomask bits
3824 : * to use on the new tuple depend on what was there on the old one.
3825 : * Note that since we're doing an update, the only possibility is that
3826 : * the lockers had FOR KEY SHARE lock.
3827 : */
3828 72044 : if (oldtup.t_data->t_infomask & HEAP_XMAX_IS_MULTI)
3829 : {
3830 344 : GetMultiXactIdHintBits(xmax_new_tuple, &infomask_new_tuple,
3831 : &infomask2_new_tuple);
3832 : }
3833 : else
3834 : {
3835 71700 : infomask_new_tuple = HEAP_XMAX_KEYSHR_LOCK | HEAP_XMAX_LOCK_ONLY;
3836 71700 : infomask2_new_tuple = 0;
3837 : }
3838 : }
3839 :
3840 : /*
3841 : * Prepare the new tuple with the appropriate initial values of Xmin and
3842 : * Xmax, as well as initial infomask bits as computed above.
3843 : */
3844 623960 : newtup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
3845 623960 : newtup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
3846 623960 : HeapTupleHeaderSetXmin(newtup->t_data, xid);
3847 623960 : HeapTupleHeaderSetCmin(newtup->t_data, cid);
3848 623960 : newtup->t_data->t_infomask |= HEAP_UPDATED | infomask_new_tuple;
3849 623960 : newtup->t_data->t_infomask2 |= infomask2_new_tuple;
3850 623960 : HeapTupleHeaderSetXmax(newtup->t_data, xmax_new_tuple);
3851 :
3852 : /*
3853 : * Replace cid with a combo CID if necessary. Note that we already put
3854 : * the plain cid into the new tuple.
3855 : */
3856 623960 : HeapTupleHeaderAdjustCmax(oldtup.t_data, &cid, &iscombo);
3857 :
3858 : /*
3859 : * If the toaster needs to be activated, OR if the new tuple will not fit
3860 : * on the same page as the old, then we need to release the content lock
3861 : * (but not the pin!) on the old tuple's buffer while we are off doing
3862 : * TOAST and/or table-file-extension work. We must mark the old tuple to
3863 : * show that it's locked, else other processes may try to update it
3864 : * themselves.
3865 : *
3866 : * We need to invoke the toaster if there are already any out-of-line
3867 : * toasted values present, or if the new tuple is over-threshold.
3868 : */
3869 623960 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
3870 0 : relation->rd_rel->relkind != RELKIND_MATVIEW)
3871 : {
3872 : /* toast table entries should never be recursively toasted */
3873 : Assert(!HeapTupleHasExternal(&oldtup));
3874 : Assert(!HeapTupleHasExternal(newtup));
3875 0 : need_toast = false;
3876 : }
3877 : else
3878 1871128 : need_toast = (HeapTupleHasExternal(&oldtup) ||
3879 1247168 : HeapTupleHasExternal(newtup) ||
3880 623160 : newtup->t_len > TOAST_TUPLE_THRESHOLD);
3881 :
3882 623960 : pagefree = PageGetHeapFreeSpace(page);
3883 :
3884 623960 : newtupsize = MAXALIGN(newtup->t_len);
3885 :
3886 623960 : if (need_toast || newtupsize > pagefree)
3887 303772 : {
3888 : TransactionId xmax_lock_old_tuple;
3889 : uint16 infomask_lock_old_tuple,
3890 : infomask2_lock_old_tuple;
3891 303772 : bool cleared_all_frozen = false;
3892 :
3893 : /*
3894 : * To prevent concurrent sessions from updating the tuple, we have to
3895 : * temporarily mark it locked, while we release the page-level lock.
3896 : *
3897 : * To satisfy the rule that any xid potentially appearing in a buffer
3898 : * written out to disk, we unfortunately have to WAL log this
3899 : * temporary modification. We can reuse xl_heap_lock for this
3900 : * purpose. If we crash/error before following through with the
3901 : * actual update, xmax will be of an aborted transaction, allowing
3902 : * other sessions to proceed.
3903 : */
3904 :
3905 : /*
3906 : * Compute xmax / infomask appropriate for locking the tuple. This has
3907 : * to be done separately from the combo that's going to be used for
3908 : * updating, because the potentially created multixact would otherwise
3909 : * be wrong.
3910 : */
3911 303772 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3912 303772 : oldtup.t_data->t_infomask,
3913 303772 : oldtup.t_data->t_infomask2,
3914 : xid, *lockmode, false,
3915 : &xmax_lock_old_tuple, &infomask_lock_old_tuple,
3916 : &infomask2_lock_old_tuple);
3917 :
3918 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(infomask_lock_old_tuple));
3919 :
3920 303772 : START_CRIT_SECTION();
3921 :
3922 : /* Clear obsolete visibility flags ... */
3923 303772 : oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
3924 303772 : oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
3925 303772 : HeapTupleClearHotUpdated(&oldtup);
3926 : /* ... and store info about transaction updating this tuple */
3927 : Assert(TransactionIdIsValid(xmax_lock_old_tuple));
3928 303772 : HeapTupleHeaderSetXmax(oldtup.t_data, xmax_lock_old_tuple);
3929 303772 : oldtup.t_data->t_infomask |= infomask_lock_old_tuple;
3930 303772 : oldtup.t_data->t_infomask2 |= infomask2_lock_old_tuple;
3931 303772 : HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
3932 :
3933 : /* temporarily make it look not-updated, but locked */
3934 303772 : oldtup.t_data->t_ctid = oldtup.t_self;
3935 :
3936 : /*
3937 : * Clear all-frozen bit on visibility map if needed. We could
3938 : * immediately reset ALL_VISIBLE, but given that the WAL logging
3939 : * overhead would be unchanged, that doesn't seem necessarily
3940 : * worthwhile.
3941 : */
3942 305620 : if (PageIsAllVisible(page) &&
3943 1848 : visibilitymap_clear(relation, block, vmbuffer,
3944 : VISIBILITYMAP_ALL_FROZEN))
3945 1550 : cleared_all_frozen = true;
3946 :
3947 303772 : MarkBufferDirty(buffer);
3948 :
3949 303772 : if (RelationNeedsWAL(relation))
3950 : {
3951 : xl_heap_lock xlrec;
3952 : XLogRecPtr recptr;
3953 :
3954 283518 : XLogBeginInsert();
3955 283518 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
3956 :
3957 283518 : xlrec.offnum = ItemPointerGetOffsetNumber(&oldtup.t_self);
3958 283518 : xlrec.xmax = xmax_lock_old_tuple;
3959 567036 : xlrec.infobits_set = compute_infobits(oldtup.t_data->t_infomask,
3960 283518 : oldtup.t_data->t_infomask2);
3961 283518 : xlrec.flags =
3962 283518 : cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
3963 283518 : XLogRegisterData(&xlrec, SizeOfHeapLock);
3964 283518 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
3965 283518 : PageSetLSN(page, recptr);
3966 : }
3967 :
3968 303772 : END_CRIT_SECTION();
3969 :
3970 303772 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3971 :
3972 : /*
3973 : * Let the toaster do its thing, if needed.
3974 : *
3975 : * Note: below this point, heaptup is the data we actually intend to
3976 : * store into the relation; newtup is the caller's original untoasted
3977 : * data.
3978 : */
3979 303772 : if (need_toast)
3980 : {
3981 : /* Note we always use WAL and FSM during updates */
3982 3346 : heaptup = heap_toast_insert_or_update(relation, newtup, &oldtup, 0);
3983 3346 : newtupsize = MAXALIGN(heaptup->t_len);
3984 : }
3985 : else
3986 300426 : heaptup = newtup;
3987 :
3988 : /*
3989 : * Now, do we need a new page for the tuple, or not? This is a bit
3990 : * tricky since someone else could have added tuples to the page while
3991 : * we weren't looking. We have to recheck the available space after
3992 : * reacquiring the buffer lock. But don't bother to do that if the
3993 : * former amount of free space is still not enough; it's unlikely
3994 : * there's more free now than before.
3995 : *
3996 : * What's more, if we need to get a new page, we will need to acquire
3997 : * buffer locks on both old and new pages. To avoid deadlock against
3998 : * some other backend trying to get the same two locks in the other
3999 : * order, we must be consistent about the order we get the locks in.
4000 : * We use the rule "lock the lower-numbered page of the relation
4001 : * first". To implement this, we must do RelationGetBufferForTuple
4002 : * while not holding the lock on the old page, and we must rely on it
4003 : * to get the locks on both pages in the correct order.
4004 : *
4005 : * Another consideration is that we need visibility map page pin(s) if
4006 : * we will have to clear the all-visible flag on either page. If we
4007 : * call RelationGetBufferForTuple, we rely on it to acquire any such
4008 : * pins; but if we don't, we have to handle that here. Hence we need
4009 : * a loop.
4010 : */
4011 : for (;;)
4012 : {
4013 303772 : if (newtupsize > pagefree)
4014 : {
4015 : /* It doesn't fit, must use RelationGetBufferForTuple. */
4016 302636 : newbuf = RelationGetBufferForTuple(relation, heaptup->t_len,
4017 : buffer, 0, NULL,
4018 : &vmbuffer_new, &vmbuffer,
4019 : 0);
4020 : /* We're all done. */
4021 302636 : break;
4022 : }
4023 : /* Acquire VM page pin if needed and we don't have it. */
4024 1136 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
4025 0 : visibilitymap_pin(relation, block, &vmbuffer);
4026 : /* Re-acquire the lock on the old tuple's page. */
4027 1136 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
4028 : /* Re-check using the up-to-date free space */
4029 1136 : pagefree = PageGetHeapFreeSpace(page);
4030 1136 : if (newtupsize > pagefree ||
4031 1136 : (vmbuffer == InvalidBuffer && PageIsAllVisible(page)))
4032 : {
4033 : /*
4034 : * Rats, it doesn't fit anymore, or somebody just now set the
4035 : * all-visible flag. We must now unlock and loop to avoid
4036 : * deadlock. Fortunately, this path should seldom be taken.
4037 : */
4038 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
4039 : }
4040 : else
4041 : {
4042 : /* We're all done. */
4043 1136 : newbuf = buffer;
4044 1136 : break;
4045 : }
4046 : }
4047 : }
4048 : else
4049 : {
4050 : /* No TOAST work needed, and it'll fit on same page */
4051 320188 : newbuf = buffer;
4052 320188 : heaptup = newtup;
4053 : }
4054 :
4055 : /*
4056 : * We're about to do the actual update -- check for conflict first, to
4057 : * avoid possibly having to roll back work we've just done.
4058 : *
4059 : * This is safe without a recheck as long as there is no possibility of
4060 : * another process scanning the pages between this check and the update
4061 : * being visible to the scan (i.e., exclusive buffer content lock(s) are
4062 : * continuously held from this point until the tuple update is visible).
4063 : *
4064 : * For the new tuple the only check needed is at the relation level, but
4065 : * since both tuples are in the same relation and the check for oldtup
4066 : * will include checking the relation level, there is no benefit to a
4067 : * separate check for the new tuple.
4068 : */
4069 623960 : CheckForSerializableConflictIn(relation, &oldtup.t_self,
4070 : BufferGetBlockNumber(buffer));
4071 :
4072 : /*
4073 : * At this point newbuf and buffer are both pinned and locked, and newbuf
4074 : * has enough space for the new tuple. If they are the same buffer, only
4075 : * one pin is held.
4076 : */
4077 :
4078 623936 : if (newbuf == buffer)
4079 : {
4080 : /*
4081 : * Since the new tuple is going into the same page, we might be able
4082 : * to do a HOT update. Check if any of the index columns have been
4083 : * changed.
4084 : */
4085 321300 : if (!bms_overlap(modified_attrs, hot_attrs))
4086 : {
4087 296648 : use_hot_update = true;
4088 :
4089 : /*
4090 : * If none of the columns that are used in hot-blocking indexes
4091 : * were updated, we can apply HOT, but we do still need to check
4092 : * if we need to update the summarizing indexes, and update those
4093 : * indexes if the columns were updated, or we may fail to detect
4094 : * e.g. value bound changes in BRIN minmax indexes.
4095 : */
4096 296648 : if (bms_overlap(modified_attrs, sum_attrs))
4097 3282 : summarized_update = true;
4098 : }
4099 : }
4100 : else
4101 : {
4102 : /* Set a hint that the old page could use prune/defrag */
4103 302636 : PageSetFull(page);
4104 : }
4105 :
4106 : /*
4107 : * Compute replica identity tuple before entering the critical section so
4108 : * we don't PANIC upon a memory allocation failure.
4109 : * ExtractReplicaIdentity() will return NULL if nothing needs to be
4110 : * logged. Pass old key required as true only if the replica identity key
4111 : * columns are modified or it has external data.
4112 : */
4113 623936 : old_key_tuple = ExtractReplicaIdentity(relation, &oldtup,
4114 623936 : bms_overlap(modified_attrs, id_attrs) ||
4115 : id_has_external,
4116 : &old_key_copied);
4117 :
4118 : /* NO EREPORT(ERROR) from here till changes are logged */
4119 623936 : START_CRIT_SECTION();
4120 :
4121 : /*
4122 : * If this transaction commits, the old tuple will become DEAD sooner or
4123 : * later. Set flag that this page is a candidate for pruning once our xid
4124 : * falls below the OldestXmin horizon. If the transaction finally aborts,
4125 : * the subsequent page pruning will be a no-op and the hint will be
4126 : * cleared.
4127 : *
4128 : * XXX Should we set hint on newbuf as well? If the transaction aborts,
4129 : * there would be a prunable tuple in the newbuf; but for now we choose
4130 : * not to optimize for aborts. Note that heap_xlog_update must be kept in
4131 : * sync if this decision changes.
4132 : */
4133 623936 : PageSetPrunable(page, xid);
4134 :
4135 623936 : if (use_hot_update)
4136 : {
4137 : /* Mark the old tuple as HOT-updated */
4138 296648 : HeapTupleSetHotUpdated(&oldtup);
4139 : /* And mark the new tuple as heap-only */
4140 296648 : HeapTupleSetHeapOnly(heaptup);
4141 : /* Mark the caller's copy too, in case different from heaptup */
4142 296648 : HeapTupleSetHeapOnly(newtup);
4143 : }
4144 : else
4145 : {
4146 : /* Make sure tuples are correctly marked as not-HOT */
4147 327288 : HeapTupleClearHotUpdated(&oldtup);
4148 327288 : HeapTupleClearHeapOnly(heaptup);
4149 327288 : HeapTupleClearHeapOnly(newtup);
4150 : }
4151 :
4152 623936 : RelationPutHeapTuple(relation, newbuf, heaptup, false); /* insert new tuple */
4153 :
4154 :
4155 : /* Clear obsolete visibility flags, possibly set by ourselves above... */
4156 623936 : oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
4157 623936 : oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
4158 : /* ... and store info about transaction updating this tuple */
4159 : Assert(TransactionIdIsValid(xmax_old_tuple));
4160 623936 : HeapTupleHeaderSetXmax(oldtup.t_data, xmax_old_tuple);
4161 623936 : oldtup.t_data->t_infomask |= infomask_old_tuple;
4162 623936 : oldtup.t_data->t_infomask2 |= infomask2_old_tuple;
4163 623936 : HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
4164 :
4165 : /* record address of new tuple in t_ctid of old one */
4166 623936 : oldtup.t_data->t_ctid = heaptup->t_self;
4167 :
4168 : /* clear PD_ALL_VISIBLE flags, reset all visibilitymap bits */
4169 623936 : if (PageIsAllVisible(BufferGetPage(buffer)))
4170 : {
4171 3056 : all_visible_cleared = true;
4172 3056 : PageClearAllVisible(BufferGetPage(buffer));
4173 3056 : visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
4174 : vmbuffer, VISIBILITYMAP_VALID_BITS);
4175 : }
4176 623936 : if (newbuf != buffer && PageIsAllVisible(BufferGetPage(newbuf)))
4177 : {
4178 1616 : all_visible_cleared_new = true;
4179 1616 : PageClearAllVisible(BufferGetPage(newbuf));
4180 1616 : visibilitymap_clear(relation, BufferGetBlockNumber(newbuf),
4181 : vmbuffer_new, VISIBILITYMAP_VALID_BITS);
4182 : }
4183 :
4184 623936 : if (newbuf != buffer)
4185 302636 : MarkBufferDirty(newbuf);
4186 623936 : MarkBufferDirty(buffer);
4187 :
4188 : /* XLOG stuff */
4189 623936 : if (RelationNeedsWAL(relation))
4190 : {
4191 : XLogRecPtr recptr;
4192 :
4193 : /*
4194 : * For logical decoding we need combo CIDs to properly decode the
4195 : * catalog.
4196 : */
4197 601206 : if (RelationIsAccessibleInLogicalDecoding(relation))
4198 : {
4199 5134 : log_heap_new_cid(relation, &oldtup);
4200 5134 : log_heap_new_cid(relation, heaptup);
4201 : }
4202 :
4203 601206 : recptr = log_heap_update(relation, buffer,
4204 : newbuf, &oldtup, heaptup,
4205 : old_key_tuple,
4206 : all_visible_cleared,
4207 : all_visible_cleared_new);
4208 601206 : if (newbuf != buffer)
4209 : {
4210 282394 : PageSetLSN(BufferGetPage(newbuf), recptr);
4211 : }
4212 601206 : PageSetLSN(BufferGetPage(buffer), recptr);
4213 : }
4214 :
4215 623936 : END_CRIT_SECTION();
4216 :
4217 623936 : if (newbuf != buffer)
4218 302636 : LockBuffer(newbuf, BUFFER_LOCK_UNLOCK);
4219 623936 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
4220 :
4221 : /*
4222 : * Mark old tuple for invalidation from system caches at next command
4223 : * boundary, and mark the new tuple for invalidation in case we abort. We
4224 : * have to do this before releasing the buffer because oldtup is in the
4225 : * buffer. (heaptup is all in local memory, but it's necessary to process
4226 : * both tuple versions in one call to inval.c so we can avoid redundant
4227 : * sinval messages.)
4228 : */
4229 623936 : CacheInvalidateHeapTuple(relation, &oldtup, heaptup);
4230 :
4231 : /* Now we can release the buffer(s) */
4232 623936 : if (newbuf != buffer)
4233 302636 : ReleaseBuffer(newbuf);
4234 623936 : ReleaseBuffer(buffer);
4235 623936 : if (BufferIsValid(vmbuffer_new))
4236 1616 : ReleaseBuffer(vmbuffer_new);
4237 623936 : if (BufferIsValid(vmbuffer))
4238 3056 : ReleaseBuffer(vmbuffer);
4239 :
4240 : /*
4241 : * Release the lmgr tuple lock, if we had it.
4242 : */
4243 623936 : if (have_tuple_lock)
4244 44 : UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
4245 :
4246 623936 : pgstat_count_heap_update(relation, use_hot_update, newbuf != buffer);
4247 :
4248 : /*
4249 : * If heaptup is a private copy, release it. Don't forget to copy t_self
4250 : * back to the caller's image, too.
4251 : */
4252 623936 : if (heaptup != newtup)
4253 : {
4254 3250 : newtup->t_self = heaptup->t_self;
4255 3250 : heap_freetuple(heaptup);
4256 : }
4257 :
4258 : /*
4259 : * If it is a HOT update, the update may still need to update summarized
4260 : * indexes, lest we fail to update those summaries and get incorrect
4261 : * results (for example, minmax bounds of the block may change with this
4262 : * update).
4263 : */
4264 623936 : if (use_hot_update)
4265 : {
4266 296648 : if (summarized_update)
4267 3282 : *update_indexes = TU_Summarizing;
4268 : else
4269 293366 : *update_indexes = TU_None;
4270 : }
4271 : else
4272 327288 : *update_indexes = TU_All;
4273 :
4274 623936 : if (old_key_tuple != NULL && old_key_copied)
4275 168 : heap_freetuple(old_key_tuple);
4276 :
4277 623936 : bms_free(hot_attrs);
4278 623936 : bms_free(sum_attrs);
4279 623936 : bms_free(key_attrs);
4280 623936 : bms_free(id_attrs);
4281 623936 : bms_free(modified_attrs);
4282 623936 : bms_free(interesting_attrs);
4283 :
4284 623936 : return TM_Ok;
4285 : }
4286 :
4287 : #ifdef USE_ASSERT_CHECKING
4288 : /*
4289 : * Confirm adequate lock held during heap_update(), per rules from
4290 : * README.tuplock section "Locking to write inplace-updated tables".
4291 : */
4292 : static void
4293 : check_lock_if_inplace_updateable_rel(Relation relation,
4294 : const ItemPointerData *otid,
4295 : HeapTuple newtup)
4296 : {
4297 : /* LOCKTAG_TUPLE acceptable for any catalog */
4298 : switch (RelationGetRelid(relation))
4299 : {
4300 : case RelationRelationId:
4301 : case DatabaseRelationId:
4302 : {
4303 : LOCKTAG tuptag;
4304 :
4305 : SET_LOCKTAG_TUPLE(tuptag,
4306 : relation->rd_lockInfo.lockRelId.dbId,
4307 : relation->rd_lockInfo.lockRelId.relId,
4308 : ItemPointerGetBlockNumber(otid),
4309 : ItemPointerGetOffsetNumber(otid));
4310 : if (LockHeldByMe(&tuptag, InplaceUpdateTupleLock, false))
4311 : return;
4312 : }
4313 : break;
4314 : default:
4315 : Assert(!IsInplaceUpdateRelation(relation));
4316 : return;
4317 : }
4318 :
4319 : switch (RelationGetRelid(relation))
4320 : {
4321 : case RelationRelationId:
4322 : {
4323 : /* LOCKTAG_TUPLE or LOCKTAG_RELATION ok */
4324 : Form_pg_class classForm = (Form_pg_class) GETSTRUCT(newtup);
4325 : Oid relid = classForm->oid;
4326 : Oid dbid;
4327 : LOCKTAG tag;
4328 :
4329 : if (IsSharedRelation(relid))
4330 : dbid = InvalidOid;
4331 : else
4332 : dbid = MyDatabaseId;
4333 :
4334 : if (classForm->relkind == RELKIND_INDEX)
4335 : {
4336 : Relation irel = index_open(relid, AccessShareLock);
4337 :
4338 : SET_LOCKTAG_RELATION(tag, dbid, irel->rd_index->indrelid);
4339 : index_close(irel, AccessShareLock);
4340 : }
4341 : else
4342 : SET_LOCKTAG_RELATION(tag, dbid, relid);
4343 :
4344 : if (!LockHeldByMe(&tag, ShareUpdateExclusiveLock, false) &&
4345 : !LockHeldByMe(&tag, ShareRowExclusiveLock, true))
4346 : elog(WARNING,
4347 : "missing lock for relation \"%s\" (OID %u, relkind %c) @ TID (%u,%u)",
4348 : NameStr(classForm->relname),
4349 : relid,
4350 : classForm->relkind,
4351 : ItemPointerGetBlockNumber(otid),
4352 : ItemPointerGetOffsetNumber(otid));
4353 : }
4354 : break;
4355 : case DatabaseRelationId:
4356 : {
4357 : /* LOCKTAG_TUPLE required */
4358 : Form_pg_database dbForm = (Form_pg_database) GETSTRUCT(newtup);
4359 :
4360 : elog(WARNING,
4361 : "missing lock on database \"%s\" (OID %u) @ TID (%u,%u)",
4362 : NameStr(dbForm->datname),
4363 : dbForm->oid,
4364 : ItemPointerGetBlockNumber(otid),
4365 : ItemPointerGetOffsetNumber(otid));
4366 : }
4367 : break;
4368 : }
4369 : }
4370 :
4371 : /*
4372 : * Confirm adequate relation lock held, per rules from README.tuplock section
4373 : * "Locking to write inplace-updated tables".
4374 : */
4375 : static void
4376 : check_inplace_rel_lock(HeapTuple oldtup)
4377 : {
4378 : Form_pg_class classForm = (Form_pg_class) GETSTRUCT(oldtup);
4379 : Oid relid = classForm->oid;
4380 : Oid dbid;
4381 : LOCKTAG tag;
4382 :
4383 : if (IsSharedRelation(relid))
4384 : dbid = InvalidOid;
4385 : else
4386 : dbid = MyDatabaseId;
4387 :
4388 : if (classForm->relkind == RELKIND_INDEX)
4389 : {
4390 : Relation irel = index_open(relid, AccessShareLock);
4391 :
4392 : SET_LOCKTAG_RELATION(tag, dbid, irel->rd_index->indrelid);
4393 : index_close(irel, AccessShareLock);
4394 : }
4395 : else
4396 : SET_LOCKTAG_RELATION(tag, dbid, relid);
4397 :
4398 : if (!LockHeldByMe(&tag, ShareUpdateExclusiveLock, true))
4399 : elog(WARNING,
4400 : "missing lock for relation \"%s\" (OID %u, relkind %c) @ TID (%u,%u)",
4401 : NameStr(classForm->relname),
4402 : relid,
4403 : classForm->relkind,
4404 : ItemPointerGetBlockNumber(&oldtup->t_self),
4405 : ItemPointerGetOffsetNumber(&oldtup->t_self));
4406 : }
4407 : #endif
4408 :
4409 : /*
4410 : * Check if the specified attribute's values are the same. Subroutine for
4411 : * HeapDetermineColumnsInfo.
4412 : */
4413 : static bool
4414 1531728 : heap_attr_equals(TupleDesc tupdesc, int attrnum, Datum value1, Datum value2,
4415 : bool isnull1, bool isnull2)
4416 : {
4417 : /*
4418 : * If one value is NULL and other is not, then they are certainly not
4419 : * equal
4420 : */
4421 1531728 : if (isnull1 != isnull2)
4422 90 : return false;
4423 :
4424 : /*
4425 : * If both are NULL, they can be considered equal.
4426 : */
4427 1531638 : if (isnull1)
4428 9982 : return true;
4429 :
4430 : /*
4431 : * We do simple binary comparison of the two datums. This may be overly
4432 : * strict because there can be multiple binary representations for the
4433 : * same logical value. But we should be OK as long as there are no false
4434 : * positives. Using a type-specific equality operator is messy because
4435 : * there could be multiple notions of equality in different operator
4436 : * classes; furthermore, we cannot safely invoke user-defined functions
4437 : * while holding exclusive buffer lock.
4438 : */
4439 1521656 : if (attrnum <= 0)
4440 : {
4441 : /* The only allowed system columns are OIDs, so do this */
4442 0 : return (DatumGetObjectId(value1) == DatumGetObjectId(value2));
4443 : }
4444 : else
4445 : {
4446 : CompactAttribute *att;
4447 :
4448 : Assert(attrnum <= tupdesc->natts);
4449 1521656 : att = TupleDescCompactAttr(tupdesc, attrnum - 1);
4450 1521656 : return datumIsEqual(value1, value2, att->attbyval, att->attlen);
4451 : }
4452 : }
4453 :
4454 : /*
4455 : * Check which columns are being updated.
4456 : *
4457 : * Given an updated tuple, determine (and return into the output bitmapset),
4458 : * from those listed as interesting, the set of columns that changed.
4459 : *
4460 : * has_external indicates if any of the unmodified attributes (from those
4461 : * listed as interesting) of the old tuple is a member of external_cols and is
4462 : * stored externally.
4463 : */
4464 : static Bitmapset *
4465 624280 : HeapDetermineColumnsInfo(Relation relation,
4466 : Bitmapset *interesting_cols,
4467 : Bitmapset *external_cols,
4468 : HeapTuple oldtup, HeapTuple newtup,
4469 : bool *has_external)
4470 : {
4471 : int attidx;
4472 624280 : Bitmapset *modified = NULL;
4473 624280 : TupleDesc tupdesc = RelationGetDescr(relation);
4474 :
4475 624280 : attidx = -1;
4476 2156008 : while ((attidx = bms_next_member(interesting_cols, attidx)) >= 0)
4477 : {
4478 : /* attidx is zero-based, attrnum is the normal attribute number */
4479 1531728 : AttrNumber attrnum = attidx + FirstLowInvalidHeapAttributeNumber;
4480 : Datum value1,
4481 : value2;
4482 : bool isnull1,
4483 : isnull2;
4484 :
4485 : /*
4486 : * If it's a whole-tuple reference, say "not equal". It's not really
4487 : * worth supporting this case, since it could only succeed after a
4488 : * no-op update, which is hardly a case worth optimizing for.
4489 : */
4490 1531728 : if (attrnum == 0)
4491 : {
4492 0 : modified = bms_add_member(modified, attidx);
4493 1468084 : continue;
4494 : }
4495 :
4496 : /*
4497 : * Likewise, automatically say "not equal" for any system attribute
4498 : * other than tableOID; we cannot expect these to be consistent in a
4499 : * HOT chain, or even to be set correctly yet in the new tuple.
4500 : */
4501 1531728 : if (attrnum < 0)
4502 : {
4503 0 : if (attrnum != TableOidAttributeNumber)
4504 : {
4505 0 : modified = bms_add_member(modified, attidx);
4506 0 : continue;
4507 : }
4508 : }
4509 :
4510 : /*
4511 : * Extract the corresponding values. XXX this is pretty inefficient
4512 : * if there are many indexed columns. Should we do a single
4513 : * heap_deform_tuple call on each tuple, instead? But that doesn't
4514 : * work for system columns ...
4515 : */
4516 1531728 : value1 = heap_getattr(oldtup, attrnum, tupdesc, &isnull1);
4517 1531728 : value2 = heap_getattr(newtup, attrnum, tupdesc, &isnull2);
4518 :
4519 1531728 : if (!heap_attr_equals(tupdesc, attrnum, value1,
4520 : value2, isnull1, isnull2))
4521 : {
4522 54486 : modified = bms_add_member(modified, attidx);
4523 54486 : continue;
4524 : }
4525 :
4526 : /*
4527 : * No need to check attributes that can't be stored externally. Note
4528 : * that system attributes can't be stored externally.
4529 : */
4530 1477242 : if (attrnum < 0 || isnull1 ||
4531 1467260 : TupleDescCompactAttr(tupdesc, attrnum - 1)->attlen != -1)
4532 1413598 : continue;
4533 :
4534 : /*
4535 : * Check if the old tuple's attribute is stored externally and is a
4536 : * member of external_cols.
4537 : */
4538 63654 : if (VARATT_IS_EXTERNAL((struct varlena *) DatumGetPointer(value1)) &&
4539 10 : bms_is_member(attidx, external_cols))
4540 4 : *has_external = true;
4541 : }
4542 :
4543 624280 : return modified;
4544 : }
4545 :
4546 : /*
4547 : * simple_heap_update - replace a tuple
4548 : *
4549 : * This routine may be used to update a tuple when concurrent updates of
4550 : * the target tuple are not expected (for example, because we have a lock
4551 : * on the relation associated with the tuple). Any failure is reported
4552 : * via ereport().
4553 : */
4554 : void
4555 234810 : simple_heap_update(Relation relation, const ItemPointerData *otid, HeapTuple tup,
4556 : TU_UpdateIndexes *update_indexes)
4557 : {
4558 : TM_Result result;
4559 : TM_FailureData tmfd;
4560 : LockTupleMode lockmode;
4561 :
4562 234810 : result = heap_update(relation, otid, tup,
4563 : GetCurrentCommandId(true), InvalidSnapshot,
4564 : true /* wait for commit */ ,
4565 : &tmfd, &lockmode, update_indexes);
4566 234810 : switch (result)
4567 : {
4568 0 : case TM_SelfModified:
4569 : /* Tuple was already updated in current command? */
4570 0 : elog(ERROR, "tuple already updated by self");
4571 : break;
4572 :
4573 234808 : case TM_Ok:
4574 : /* done successfully */
4575 234808 : break;
4576 :
4577 0 : case TM_Updated:
4578 0 : elog(ERROR, "tuple concurrently updated");
4579 : break;
4580 :
4581 2 : case TM_Deleted:
4582 2 : elog(ERROR, "tuple concurrently deleted");
4583 : break;
4584 :
4585 0 : default:
4586 0 : elog(ERROR, "unrecognized heap_update status: %u", result);
4587 : break;
4588 : }
4589 234808 : }
4590 :
4591 :
4592 : /*
4593 : * Return the MultiXactStatus corresponding to the given tuple lock mode.
4594 : */
4595 : static MultiXactStatus
4596 230884 : get_mxact_status_for_lock(LockTupleMode mode, bool is_update)
4597 : {
4598 : int retval;
4599 :
4600 230884 : if (is_update)
4601 430 : retval = tupleLockExtraInfo[mode].updstatus;
4602 : else
4603 230454 : retval = tupleLockExtraInfo[mode].lockstatus;
4604 :
4605 230884 : if (retval == -1)
4606 0 : elog(ERROR, "invalid lock tuple mode %d/%s", mode,
4607 : is_update ? "true" : "false");
4608 :
4609 230884 : return (MultiXactStatus) retval;
4610 : }
4611 :
4612 : /*
4613 : * heap_lock_tuple - lock a tuple in shared or exclusive mode
4614 : *
4615 : * Note that this acquires a buffer pin, which the caller must release.
4616 : *
4617 : * Input parameters:
4618 : * relation: relation containing tuple (caller must hold suitable lock)
4619 : * cid: current command ID (used for visibility test, and stored into
4620 : * tuple's cmax if lock is successful)
4621 : * mode: indicates if shared or exclusive tuple lock is desired
4622 : * wait_policy: what to do if tuple lock is not available
4623 : * follow_updates: if true, follow the update chain to also lock descendant
4624 : * tuples.
4625 : *
4626 : * Output parameters:
4627 : * *tuple: all fields filled in
4628 : * *buffer: set to buffer holding tuple (pinned but not locked at exit)
4629 : * *tmfd: filled in failure cases (see below)
4630 : *
4631 : * Function results are the same as the ones for table_tuple_lock().
4632 : *
4633 : * In the failure cases other than TM_Invisible, the routine fills
4634 : * *tmfd with the tuple's t_ctid, t_xmax (resolving a possible MultiXact,
4635 : * if necessary), and t_cmax (the last only for TM_SelfModified,
4636 : * since we cannot obtain cmax from a combo CID generated by another
4637 : * transaction).
4638 : * See comments for struct TM_FailureData for additional info.
4639 : *
4640 : * See README.tuplock for a thorough explanation of this mechanism.
4641 : */
4642 : TM_Result
4643 316814 : heap_lock_tuple(Relation relation, HeapTuple tuple,
4644 : CommandId cid, LockTupleMode mode, LockWaitPolicy wait_policy,
4645 : bool follow_updates,
4646 : Buffer *buffer, TM_FailureData *tmfd)
4647 : {
4648 : TM_Result result;
4649 316814 : ItemPointer tid = &(tuple->t_self);
4650 : ItemId lp;
4651 : Page page;
4652 316814 : Buffer vmbuffer = InvalidBuffer;
4653 : BlockNumber block;
4654 : TransactionId xid,
4655 : xmax;
4656 : uint16 old_infomask,
4657 : new_infomask,
4658 : new_infomask2;
4659 316814 : bool first_time = true;
4660 316814 : bool skip_tuple_lock = false;
4661 316814 : bool have_tuple_lock = false;
4662 316814 : bool cleared_all_frozen = false;
4663 :
4664 316814 : *buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
4665 316814 : block = ItemPointerGetBlockNumber(tid);
4666 :
4667 : /*
4668 : * Before locking the buffer, pin the visibility map page if it appears to
4669 : * be necessary. Since we haven't got the lock yet, someone else might be
4670 : * in the middle of changing this, so we'll need to recheck after we have
4671 : * the lock.
4672 : */
4673 316814 : if (PageIsAllVisible(BufferGetPage(*buffer)))
4674 3332 : visibilitymap_pin(relation, block, &vmbuffer);
4675 :
4676 316814 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4677 :
4678 316814 : page = BufferGetPage(*buffer);
4679 316814 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
4680 : Assert(ItemIdIsNormal(lp));
4681 :
4682 316814 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
4683 316814 : tuple->t_len = ItemIdGetLength(lp);
4684 316814 : tuple->t_tableOid = RelationGetRelid(relation);
4685 :
4686 32 : l3:
4687 316846 : result = HeapTupleSatisfiesUpdate(tuple, cid, *buffer);
4688 :
4689 316846 : if (result == TM_Invisible)
4690 : {
4691 : /*
4692 : * This is possible, but only when locking a tuple for ON CONFLICT
4693 : * UPDATE. We return this value here rather than throwing an error in
4694 : * order to give that case the opportunity to throw a more specific
4695 : * error.
4696 : */
4697 24 : result = TM_Invisible;
4698 24 : goto out_locked;
4699 : }
4700 316822 : else if (result == TM_BeingModified ||
4701 154426 : result == TM_Updated ||
4702 : result == TM_Deleted)
4703 : {
4704 : TransactionId xwait;
4705 : uint16 infomask;
4706 : uint16 infomask2;
4707 : bool require_sleep;
4708 : ItemPointerData t_ctid;
4709 :
4710 : /* must copy state data before unlocking buffer */
4711 162398 : xwait = HeapTupleHeaderGetRawXmax(tuple->t_data);
4712 162398 : infomask = tuple->t_data->t_infomask;
4713 162398 : infomask2 = tuple->t_data->t_infomask2;
4714 162398 : ItemPointerCopy(&tuple->t_data->t_ctid, &t_ctid);
4715 :
4716 162398 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
4717 :
4718 : /*
4719 : * If any subtransaction of the current top transaction already holds
4720 : * a lock as strong as or stronger than what we're requesting, we
4721 : * effectively hold the desired lock already. We *must* succeed
4722 : * without trying to take the tuple lock, else we will deadlock
4723 : * against anyone wanting to acquire a stronger lock.
4724 : *
4725 : * Note we only do this the first time we loop on the HTSU result;
4726 : * there is no point in testing in subsequent passes, because
4727 : * evidently our own transaction cannot have acquired a new lock after
4728 : * the first time we checked.
4729 : */
4730 162398 : if (first_time)
4731 : {
4732 162376 : first_time = false;
4733 :
4734 162376 : if (infomask & HEAP_XMAX_IS_MULTI)
4735 : {
4736 : int i;
4737 : int nmembers;
4738 : MultiXactMember *members;
4739 :
4740 : /*
4741 : * We don't need to allow old multixacts here; if that had
4742 : * been the case, HeapTupleSatisfiesUpdate would have returned
4743 : * MayBeUpdated and we wouldn't be here.
4744 : */
4745 : nmembers =
4746 146596 : GetMultiXactIdMembers(xwait, &members, false,
4747 146596 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
4748 :
4749 2845334 : for (i = 0; i < nmembers; i++)
4750 : {
4751 : /* only consider members of our own transaction */
4752 2698766 : if (!TransactionIdIsCurrentTransactionId(members[i].xid))
4753 2698668 : continue;
4754 :
4755 98 : if (TUPLOCK_from_mxstatus(members[i].status) >= mode)
4756 : {
4757 28 : pfree(members);
4758 28 : result = TM_Ok;
4759 28 : goto out_unlocked;
4760 : }
4761 : else
4762 : {
4763 : /*
4764 : * Disable acquisition of the heavyweight tuple lock.
4765 : * Otherwise, when promoting a weaker lock, we might
4766 : * deadlock with another locker that has acquired the
4767 : * heavyweight tuple lock and is waiting for our
4768 : * transaction to finish.
4769 : *
4770 : * Note that in this case we still need to wait for
4771 : * the multixact if required, to avoid acquiring
4772 : * conflicting locks.
4773 : */
4774 70 : skip_tuple_lock = true;
4775 : }
4776 : }
4777 :
4778 146568 : if (members)
4779 146568 : pfree(members);
4780 : }
4781 15780 : else if (TransactionIdIsCurrentTransactionId(xwait))
4782 : {
4783 13142 : switch (mode)
4784 : {
4785 348 : case LockTupleKeyShare:
4786 : Assert(HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) ||
4787 : HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4788 : HEAP_XMAX_IS_EXCL_LOCKED(infomask));
4789 348 : result = TM_Ok;
4790 348 : goto out_unlocked;
4791 40 : case LockTupleShare:
4792 52 : if (HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4793 12 : HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4794 : {
4795 28 : result = TM_Ok;
4796 28 : goto out_unlocked;
4797 : }
4798 12 : break;
4799 144 : case LockTupleNoKeyExclusive:
4800 144 : if (HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4801 : {
4802 120 : result = TM_Ok;
4803 120 : goto out_unlocked;
4804 : }
4805 24 : break;
4806 12610 : case LockTupleExclusive:
4807 12610 : if (HEAP_XMAX_IS_EXCL_LOCKED(infomask) &&
4808 2530 : infomask2 & HEAP_KEYS_UPDATED)
4809 : {
4810 2488 : result = TM_Ok;
4811 2488 : goto out_unlocked;
4812 : }
4813 10122 : break;
4814 : }
4815 : }
4816 : }
4817 :
4818 : /*
4819 : * Initially assume that we will have to wait for the locking
4820 : * transaction(s) to finish. We check various cases below in which
4821 : * this can be turned off.
4822 : */
4823 159386 : require_sleep = true;
4824 159386 : if (mode == LockTupleKeyShare)
4825 : {
4826 : /*
4827 : * If we're requesting KeyShare, and there's no update present, we
4828 : * don't need to wait. Even if there is an update, we can still
4829 : * continue if the key hasn't been modified.
4830 : *
4831 : * However, if there are updates, we need to walk the update chain
4832 : * to mark future versions of the row as locked, too. That way,
4833 : * if somebody deletes that future version, we're protected
4834 : * against the key going away. This locking of future versions
4835 : * could block momentarily, if a concurrent transaction is
4836 : * deleting a key; or it could return a value to the effect that
4837 : * the transaction deleting the key has already committed. So we
4838 : * do this before re-locking the buffer; otherwise this would be
4839 : * prone to deadlocks.
4840 : *
4841 : * Note that the TID we're locking was grabbed before we unlocked
4842 : * the buffer. For it to change while we're not looking, the
4843 : * other properties we're testing for below after re-locking the
4844 : * buffer would also change, in which case we would restart this
4845 : * loop above.
4846 : */
4847 147726 : if (!(infomask2 & HEAP_KEYS_UPDATED))
4848 : {
4849 : bool updated;
4850 :
4851 147640 : updated = !HEAP_XMAX_IS_LOCKED_ONLY(infomask);
4852 :
4853 : /*
4854 : * If there are updates, follow the update chain; bail out if
4855 : * that cannot be done.
4856 : */
4857 147640 : if (follow_updates && updated &&
4858 4338 : !ItemPointerEquals(&tuple->t_self, &t_ctid))
4859 : {
4860 : TM_Result res;
4861 :
4862 4338 : res = heap_lock_updated_tuple(relation,
4863 : infomask, xwait, &t_ctid,
4864 : GetCurrentTransactionId(),
4865 : mode);
4866 4338 : if (res != TM_Ok)
4867 : {
4868 12 : result = res;
4869 : /* recovery code expects to have buffer lock held */
4870 12 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4871 392 : goto failed;
4872 : }
4873 : }
4874 :
4875 147628 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4876 :
4877 : /*
4878 : * Make sure it's still an appropriate lock, else start over.
4879 : * Also, if it wasn't updated before we released the lock, but
4880 : * is updated now, we start over too; the reason is that we
4881 : * now need to follow the update chain to lock the new
4882 : * versions.
4883 : */
4884 147628 : if (!HeapTupleHeaderIsOnlyLocked(tuple->t_data) &&
4885 4302 : ((tuple->t_data->t_infomask2 & HEAP_KEYS_UPDATED) ||
4886 4302 : !updated))
4887 32 : goto l3;
4888 :
4889 : /* Things look okay, so we can skip sleeping */
4890 147628 : require_sleep = false;
4891 :
4892 : /*
4893 : * Note we allow Xmax to change here; other updaters/lockers
4894 : * could have modified it before we grabbed the buffer lock.
4895 : * However, this is not a problem, because with the recheck we
4896 : * just did we ensure that they still don't conflict with the
4897 : * lock we want.
4898 : */
4899 : }
4900 : }
4901 11660 : else if (mode == LockTupleShare)
4902 : {
4903 : /*
4904 : * If we're requesting Share, we can similarly avoid sleeping if
4905 : * there's no update and no exclusive lock present.
4906 : */
4907 886 : if (HEAP_XMAX_IS_LOCKED_ONLY(infomask) &&
4908 886 : !HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4909 : {
4910 874 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4911 :
4912 : /*
4913 : * Make sure it's still an appropriate lock, else start over.
4914 : * See above about allowing xmax to change.
4915 : */
4916 1748 : if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
4917 874 : HEAP_XMAX_IS_EXCL_LOCKED(tuple->t_data->t_infomask))
4918 0 : goto l3;
4919 874 : require_sleep = false;
4920 : }
4921 : }
4922 10774 : else if (mode == LockTupleNoKeyExclusive)
4923 : {
4924 : /*
4925 : * If we're requesting NoKeyExclusive, we might also be able to
4926 : * avoid sleeping; just ensure that there no conflicting lock
4927 : * already acquired.
4928 : */
4929 340 : if (infomask & HEAP_XMAX_IS_MULTI)
4930 : {
4931 52 : if (!DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
4932 : mode, NULL))
4933 : {
4934 : /*
4935 : * No conflict, but if the xmax changed under us in the
4936 : * meantime, start over.
4937 : */
4938 26 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4939 52 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4940 26 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4941 : xwait))
4942 0 : goto l3;
4943 :
4944 : /* otherwise, we're good */
4945 26 : require_sleep = false;
4946 : }
4947 : }
4948 288 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask))
4949 : {
4950 36 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4951 :
4952 : /* if the xmax changed in the meantime, start over */
4953 72 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4954 36 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4955 : xwait))
4956 0 : goto l3;
4957 : /* otherwise, we're good */
4958 36 : require_sleep = false;
4959 : }
4960 : }
4961 :
4962 : /*
4963 : * As a check independent from those above, we can also avoid sleeping
4964 : * if the current transaction is the sole locker of the tuple. Note
4965 : * that the strength of the lock already held is irrelevant; this is
4966 : * not about recording the lock in Xmax (which will be done regardless
4967 : * of this optimization, below). Also, note that the cases where we
4968 : * hold a lock stronger than we are requesting are already handled
4969 : * above by not doing anything.
4970 : *
4971 : * Note we only deal with the non-multixact case here; MultiXactIdWait
4972 : * is well equipped to deal with this situation on its own.
4973 : */
4974 170104 : if (require_sleep && !(infomask & HEAP_XMAX_IS_MULTI) &&
4975 10730 : TransactionIdIsCurrentTransactionId(xwait))
4976 : {
4977 : /* ... but if the xmax changed in the meantime, start over */
4978 10122 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4979 20244 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4980 10122 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4981 : xwait))
4982 0 : goto l3;
4983 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask));
4984 10122 : require_sleep = false;
4985 : }
4986 :
4987 : /*
4988 : * Time to sleep on the other transaction/multixact, if necessary.
4989 : *
4990 : * If the other transaction is an update/delete that's already
4991 : * committed, then sleeping cannot possibly do any good: if we're
4992 : * required to sleep, get out to raise an error instead.
4993 : *
4994 : * By here, we either have already acquired the buffer exclusive lock,
4995 : * or we must wait for the locking transaction or multixact; so below
4996 : * we ensure that we grab buffer lock after the sleep.
4997 : */
4998 159374 : if (require_sleep && (result == TM_Updated || result == TM_Deleted))
4999 : {
5000 304 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5001 304 : goto failed;
5002 : }
5003 159070 : else if (require_sleep)
5004 : {
5005 : /*
5006 : * Acquire tuple lock to establish our priority for the tuple, or
5007 : * die trying. LockTuple will release us when we are next-in-line
5008 : * for the tuple. We must do this even if we are share-locking,
5009 : * but not if we already have a weaker lock on the tuple.
5010 : *
5011 : * If we are forced to "start over" below, we keep the tuple lock;
5012 : * this arranges that we stay at the head of the line while
5013 : * rechecking tuple state.
5014 : */
5015 384 : if (!skip_tuple_lock &&
5016 352 : !heap_acquire_tuplock(relation, tid, mode, wait_policy,
5017 : &have_tuple_lock))
5018 : {
5019 : /*
5020 : * This can only happen if wait_policy is Skip and the lock
5021 : * couldn't be obtained.
5022 : */
5023 2 : result = TM_WouldBlock;
5024 : /* recovery code expects to have buffer lock held */
5025 2 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5026 2 : goto failed;
5027 : }
5028 :
5029 380 : if (infomask & HEAP_XMAX_IS_MULTI)
5030 : {
5031 80 : MultiXactStatus status = get_mxact_status_for_lock(mode, false);
5032 :
5033 : /* We only ever lock tuples, never update them */
5034 80 : if (status >= MultiXactStatusNoKeyUpdate)
5035 0 : elog(ERROR, "invalid lock mode in heap_lock_tuple");
5036 :
5037 : /* wait for multixact to end, or die trying */
5038 80 : switch (wait_policy)
5039 : {
5040 72 : case LockWaitBlock:
5041 72 : MultiXactIdWait((MultiXactId) xwait, status, infomask,
5042 72 : relation, &tuple->t_self, XLTW_Lock, NULL);
5043 72 : break;
5044 4 : case LockWaitSkip:
5045 4 : if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
5046 : status, infomask, relation,
5047 : NULL, false))
5048 : {
5049 4 : result = TM_WouldBlock;
5050 : /* recovery code expects to have buffer lock held */
5051 4 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5052 4 : goto failed;
5053 : }
5054 0 : break;
5055 4 : case LockWaitError:
5056 4 : if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
5057 : status, infomask, relation,
5058 : NULL, log_lock_failures))
5059 4 : ereport(ERROR,
5060 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5061 : errmsg("could not obtain lock on row in relation \"%s\"",
5062 : RelationGetRelationName(relation))));
5063 :
5064 0 : break;
5065 : }
5066 :
5067 : /*
5068 : * Of course, the multixact might not be done here: if we're
5069 : * requesting a light lock mode, other transactions with light
5070 : * locks could still be alive, as well as locks owned by our
5071 : * own xact or other subxacts of this backend. We need to
5072 : * preserve the surviving MultiXact members. Note that it
5073 : * isn't absolutely necessary in the latter case, but doing so
5074 : * is simpler.
5075 : */
5076 : }
5077 : else
5078 : {
5079 : /* wait for regular transaction to end, or die trying */
5080 300 : switch (wait_policy)
5081 : {
5082 222 : case LockWaitBlock:
5083 222 : XactLockTableWait(xwait, relation, &tuple->t_self,
5084 : XLTW_Lock);
5085 222 : break;
5086 66 : case LockWaitSkip:
5087 66 : if (!ConditionalXactLockTableWait(xwait, false))
5088 : {
5089 66 : result = TM_WouldBlock;
5090 : /* recovery code expects to have buffer lock held */
5091 66 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5092 66 : goto failed;
5093 : }
5094 0 : break;
5095 12 : case LockWaitError:
5096 12 : if (!ConditionalXactLockTableWait(xwait, log_lock_failures))
5097 12 : ereport(ERROR,
5098 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5099 : errmsg("could not obtain lock on row in relation \"%s\"",
5100 : RelationGetRelationName(relation))));
5101 0 : break;
5102 : }
5103 : }
5104 :
5105 : /* if there are updates, follow the update chain */
5106 294 : if (follow_updates && !HEAP_XMAX_IS_LOCKED_ONLY(infomask) &&
5107 118 : !ItemPointerEquals(&tuple->t_self, &t_ctid))
5108 : {
5109 : TM_Result res;
5110 :
5111 90 : res = heap_lock_updated_tuple(relation,
5112 : infomask, xwait, &t_ctid,
5113 : GetCurrentTransactionId(),
5114 : mode);
5115 90 : if (res != TM_Ok)
5116 : {
5117 4 : result = res;
5118 : /* recovery code expects to have buffer lock held */
5119 4 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5120 4 : goto failed;
5121 : }
5122 : }
5123 :
5124 290 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5125 :
5126 : /*
5127 : * xwait is done, but if xwait had just locked the tuple then some
5128 : * other xact could update this tuple before we get to this point.
5129 : * Check for xmax change, and start over if so.
5130 : */
5131 552 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
5132 262 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
5133 : xwait))
5134 32 : goto l3;
5135 :
5136 258 : if (!(infomask & HEAP_XMAX_IS_MULTI))
5137 : {
5138 : /*
5139 : * Otherwise check if it committed or aborted. Note we cannot
5140 : * be here if the tuple was only locked by somebody who didn't
5141 : * conflict with us; that would have been handled above. So
5142 : * that transaction must necessarily be gone by now. But
5143 : * don't check for this in the multixact case, because some
5144 : * locker transactions might still be running.
5145 : */
5146 192 : UpdateXmaxHintBits(tuple->t_data, *buffer, xwait);
5147 : }
5148 : }
5149 :
5150 : /* By here, we're certain that we hold buffer exclusive lock again */
5151 :
5152 : /*
5153 : * We may lock if previous xmax aborted, or if it committed but only
5154 : * locked the tuple without updating it; or if we didn't have to wait
5155 : * at all for whatever reason.
5156 : */
5157 158944 : if (!require_sleep ||
5158 454 : (tuple->t_data->t_infomask & HEAP_XMAX_INVALID) ||
5159 360 : HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
5160 164 : HeapTupleHeaderIsOnlyLocked(tuple->t_data))
5161 158792 : result = TM_Ok;
5162 152 : else if (!ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid))
5163 114 : result = TM_Updated;
5164 : else
5165 38 : result = TM_Deleted;
5166 : }
5167 :
5168 154424 : failed:
5169 313760 : if (result != TM_Ok)
5170 : {
5171 : Assert(result == TM_SelfModified || result == TM_Updated ||
5172 : result == TM_Deleted || result == TM_WouldBlock);
5173 :
5174 : /*
5175 : * When locking a tuple under LockWaitSkip semantics and we fail with
5176 : * TM_WouldBlock above, it's possible for concurrent transactions to
5177 : * release the lock and set HEAP_XMAX_INVALID in the meantime. So
5178 : * this assert is slightly different from the equivalent one in
5179 : * heap_delete and heap_update.
5180 : */
5181 : Assert((result == TM_WouldBlock) ||
5182 : !(tuple->t_data->t_infomask & HEAP_XMAX_INVALID));
5183 : Assert(result != TM_Updated ||
5184 : !ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid));
5185 556 : tmfd->ctid = tuple->t_data->t_ctid;
5186 556 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(tuple->t_data);
5187 556 : if (result == TM_SelfModified)
5188 12 : tmfd->cmax = HeapTupleHeaderGetCmax(tuple->t_data);
5189 : else
5190 544 : tmfd->cmax = InvalidCommandId;
5191 556 : goto out_locked;
5192 : }
5193 :
5194 : /*
5195 : * If we didn't pin the visibility map page and the page has become all
5196 : * visible while we were busy locking the buffer, or during some
5197 : * subsequent window during which we had it unlocked, we'll have to unlock
5198 : * and re-lock, to avoid holding the buffer lock across I/O. That's a bit
5199 : * unfortunate, especially since we'll now have to recheck whether the
5200 : * tuple has been locked or updated under us, but hopefully it won't
5201 : * happen very often.
5202 : */
5203 313204 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
5204 : {
5205 0 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
5206 0 : visibilitymap_pin(relation, block, &vmbuffer);
5207 0 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5208 0 : goto l3;
5209 : }
5210 :
5211 313204 : xmax = HeapTupleHeaderGetRawXmax(tuple->t_data);
5212 313204 : old_infomask = tuple->t_data->t_infomask;
5213 :
5214 : /*
5215 : * If this is the first possibly-multixact-able operation in the current
5216 : * transaction, set my per-backend OldestMemberMXactId setting. We can be
5217 : * certain that the transaction will never become a member of any older
5218 : * MultiXactIds than that. (We have to do this even if we end up just
5219 : * using our own TransactionId below, since some other backend could
5220 : * incorporate our XID into a MultiXact immediately afterwards.)
5221 : */
5222 313204 : MultiXactIdSetOldestMember();
5223 :
5224 : /*
5225 : * Compute the new xmax and infomask to store into the tuple. Note we do
5226 : * not modify the tuple just yet, because that would leave it in the wrong
5227 : * state if multixact.c elogs.
5228 : */
5229 313204 : compute_new_xmax_infomask(xmax, old_infomask, tuple->t_data->t_infomask2,
5230 : GetCurrentTransactionId(), mode, false,
5231 : &xid, &new_infomask, &new_infomask2);
5232 :
5233 313204 : START_CRIT_SECTION();
5234 :
5235 : /*
5236 : * Store transaction information of xact locking the tuple.
5237 : *
5238 : * Note: Cmax is meaningless in this context, so don't set it; this avoids
5239 : * possibly generating a useless combo CID. Moreover, if we're locking a
5240 : * previously updated tuple, it's important to preserve the Cmax.
5241 : *
5242 : * Also reset the HOT UPDATE bit, but only if there's no update; otherwise
5243 : * we would break the HOT chain.
5244 : */
5245 313204 : tuple->t_data->t_infomask &= ~HEAP_XMAX_BITS;
5246 313204 : tuple->t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
5247 313204 : tuple->t_data->t_infomask |= new_infomask;
5248 313204 : tuple->t_data->t_infomask2 |= new_infomask2;
5249 313204 : if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
5250 308910 : HeapTupleHeaderClearHotUpdated(tuple->t_data);
5251 313204 : HeapTupleHeaderSetXmax(tuple->t_data, xid);
5252 :
5253 : /*
5254 : * Make sure there is no forward chain link in t_ctid. Note that in the
5255 : * cases where the tuple has been updated, we must not overwrite t_ctid,
5256 : * because it was set by the updater. Moreover, if the tuple has been
5257 : * updated, we need to follow the update chain to lock the new versions of
5258 : * the tuple as well.
5259 : */
5260 313204 : if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
5261 308910 : tuple->t_data->t_ctid = *tid;
5262 :
5263 : /* Clear only the all-frozen bit on visibility map if needed */
5264 316536 : if (PageIsAllVisible(page) &&
5265 3332 : visibilitymap_clear(relation, block, vmbuffer,
5266 : VISIBILITYMAP_ALL_FROZEN))
5267 28 : cleared_all_frozen = true;
5268 :
5269 :
5270 313204 : MarkBufferDirty(*buffer);
5271 :
5272 : /*
5273 : * XLOG stuff. You might think that we don't need an XLOG record because
5274 : * there is no state change worth restoring after a crash. You would be
5275 : * wrong however: we have just written either a TransactionId or a
5276 : * MultiXactId that may never have been seen on disk before, and we need
5277 : * to make sure that there are XLOG entries covering those ID numbers.
5278 : * Else the same IDs might be re-used after a crash, which would be
5279 : * disastrous if this page made it to disk before the crash. Essentially
5280 : * we have to enforce the WAL log-before-data rule even in this case.
5281 : * (Also, in a PITR log-shipping or 2PC environment, we have to have XLOG
5282 : * entries for everything anyway.)
5283 : */
5284 313204 : if (RelationNeedsWAL(relation))
5285 : {
5286 : xl_heap_lock xlrec;
5287 : XLogRecPtr recptr;
5288 :
5289 312500 : XLogBeginInsert();
5290 312500 : XLogRegisterBuffer(0, *buffer, REGBUF_STANDARD);
5291 :
5292 312500 : xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
5293 312500 : xlrec.xmax = xid;
5294 625000 : xlrec.infobits_set = compute_infobits(new_infomask,
5295 312500 : tuple->t_data->t_infomask2);
5296 312500 : xlrec.flags = cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
5297 312500 : XLogRegisterData(&xlrec, SizeOfHeapLock);
5298 :
5299 : /* we don't decode row locks atm, so no need to log the origin */
5300 :
5301 312500 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
5302 :
5303 312500 : PageSetLSN(page, recptr);
5304 : }
5305 :
5306 313204 : END_CRIT_SECTION();
5307 :
5308 313204 : result = TM_Ok;
5309 :
5310 313784 : out_locked:
5311 313784 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
5312 :
5313 316796 : out_unlocked:
5314 316796 : if (BufferIsValid(vmbuffer))
5315 3332 : ReleaseBuffer(vmbuffer);
5316 :
5317 : /*
5318 : * Don't update the visibility map here. Locking a tuple doesn't change
5319 : * visibility info.
5320 : */
5321 :
5322 : /*
5323 : * Now that we have successfully marked the tuple as locked, we can
5324 : * release the lmgr tuple lock, if we had it.
5325 : */
5326 316796 : if (have_tuple_lock)
5327 322 : UnlockTupleTuplock(relation, tid, mode);
5328 :
5329 316796 : return result;
5330 : }
5331 :
5332 : /*
5333 : * Acquire heavyweight lock on the given tuple, in preparation for acquiring
5334 : * its normal, Xmax-based tuple lock.
5335 : *
5336 : * have_tuple_lock is an input and output parameter: on input, it indicates
5337 : * whether the lock has previously been acquired (and this function does
5338 : * nothing in that case). If this function returns success, have_tuple_lock
5339 : * has been flipped to true.
5340 : *
5341 : * Returns false if it was unable to obtain the lock; this can only happen if
5342 : * wait_policy is Skip.
5343 : */
5344 : static bool
5345 624 : heap_acquire_tuplock(Relation relation, const ItemPointerData *tid, LockTupleMode mode,
5346 : LockWaitPolicy wait_policy, bool *have_tuple_lock)
5347 : {
5348 624 : if (*have_tuple_lock)
5349 18 : return true;
5350 :
5351 606 : switch (wait_policy)
5352 : {
5353 524 : case LockWaitBlock:
5354 524 : LockTupleTuplock(relation, tid, mode);
5355 524 : break;
5356 :
5357 68 : case LockWaitSkip:
5358 68 : if (!ConditionalLockTupleTuplock(relation, tid, mode, false))
5359 2 : return false;
5360 66 : break;
5361 :
5362 14 : case LockWaitError:
5363 14 : if (!ConditionalLockTupleTuplock(relation, tid, mode, log_lock_failures))
5364 2 : ereport(ERROR,
5365 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5366 : errmsg("could not obtain lock on row in relation \"%s\"",
5367 : RelationGetRelationName(relation))));
5368 12 : break;
5369 : }
5370 602 : *have_tuple_lock = true;
5371 :
5372 602 : return true;
5373 : }
5374 :
5375 : /*
5376 : * Given an original set of Xmax and infomask, and a transaction (identified by
5377 : * add_to_xmax) acquiring a new lock of some mode, compute the new Xmax and
5378 : * corresponding infomasks to use on the tuple.
5379 : *
5380 : * Note that this might have side effects such as creating a new MultiXactId.
5381 : *
5382 : * Most callers will have called HeapTupleSatisfiesUpdate before this function;
5383 : * that will have set the HEAP_XMAX_INVALID bit if the xmax was a MultiXactId
5384 : * but it was not running anymore. There is a race condition, which is that the
5385 : * MultiXactId may have finished since then, but that uncommon case is handled
5386 : * either here, or within MultiXactIdExpand.
5387 : *
5388 : * There is a similar race condition possible when the old xmax was a regular
5389 : * TransactionId. We test TransactionIdIsInProgress again just to narrow the
5390 : * window, but it's still possible to end up creating an unnecessary
5391 : * MultiXactId. Fortunately this is harmless.
5392 : */
5393 : static void
5394 4297138 : compute_new_xmax_infomask(TransactionId xmax, uint16 old_infomask,
5395 : uint16 old_infomask2, TransactionId add_to_xmax,
5396 : LockTupleMode mode, bool is_update,
5397 : TransactionId *result_xmax, uint16 *result_infomask,
5398 : uint16 *result_infomask2)
5399 : {
5400 : TransactionId new_xmax;
5401 : uint16 new_infomask,
5402 : new_infomask2;
5403 :
5404 : Assert(TransactionIdIsCurrentTransactionId(add_to_xmax));
5405 :
5406 208140 : l5:
5407 4505278 : new_infomask = 0;
5408 4505278 : new_infomask2 = 0;
5409 4505278 : if (old_infomask & HEAP_XMAX_INVALID)
5410 : {
5411 : /*
5412 : * No previous locker; we just insert our own TransactionId.
5413 : *
5414 : * Note that it's critical that this case be the first one checked,
5415 : * because there are several blocks below that come back to this one
5416 : * to implement certain optimizations; old_infomask might contain
5417 : * other dirty bits in those cases, but we don't really care.
5418 : */
5419 4143882 : if (is_update)
5420 : {
5421 3675370 : new_xmax = add_to_xmax;
5422 3675370 : if (mode == LockTupleExclusive)
5423 3124538 : new_infomask2 |= HEAP_KEYS_UPDATED;
5424 : }
5425 : else
5426 : {
5427 468512 : new_infomask |= HEAP_XMAX_LOCK_ONLY;
5428 468512 : switch (mode)
5429 : {
5430 5280 : case LockTupleKeyShare:
5431 5280 : new_xmax = add_to_xmax;
5432 5280 : new_infomask |= HEAP_XMAX_KEYSHR_LOCK;
5433 5280 : break;
5434 1478 : case LockTupleShare:
5435 1478 : new_xmax = add_to_xmax;
5436 1478 : new_infomask |= HEAP_XMAX_SHR_LOCK;
5437 1478 : break;
5438 270174 : case LockTupleNoKeyExclusive:
5439 270174 : new_xmax = add_to_xmax;
5440 270174 : new_infomask |= HEAP_XMAX_EXCL_LOCK;
5441 270174 : break;
5442 191580 : case LockTupleExclusive:
5443 191580 : new_xmax = add_to_xmax;
5444 191580 : new_infomask |= HEAP_XMAX_EXCL_LOCK;
5445 191580 : new_infomask2 |= HEAP_KEYS_UPDATED;
5446 191580 : break;
5447 0 : default:
5448 0 : new_xmax = InvalidTransactionId; /* silence compiler */
5449 0 : elog(ERROR, "invalid lock mode");
5450 : }
5451 : }
5452 : }
5453 361396 : else if (old_infomask & HEAP_XMAX_IS_MULTI)
5454 : {
5455 : MultiXactStatus new_status;
5456 :
5457 : /*
5458 : * Currently we don't allow XMAX_COMMITTED to be set for multis, so
5459 : * cross-check.
5460 : */
5461 : Assert(!(old_infomask & HEAP_XMAX_COMMITTED));
5462 :
5463 : /*
5464 : * A multixact together with LOCK_ONLY set but neither lock bit set
5465 : * (i.e. a pg_upgraded share locked tuple) cannot possibly be running
5466 : * anymore. This check is critical for databases upgraded by
5467 : * pg_upgrade; both MultiXactIdIsRunning and MultiXactIdExpand assume
5468 : * that such multis are never passed.
5469 : */
5470 151124 : if (HEAP_LOCKED_UPGRADED(old_infomask))
5471 : {
5472 0 : old_infomask &= ~HEAP_XMAX_IS_MULTI;
5473 0 : old_infomask |= HEAP_XMAX_INVALID;
5474 0 : goto l5;
5475 : }
5476 :
5477 : /*
5478 : * If the XMAX is already a MultiXactId, then we need to expand it to
5479 : * include add_to_xmax; but if all the members were lockers and are
5480 : * all gone, we can do away with the IS_MULTI bit and just set
5481 : * add_to_xmax as the only locker/updater. If all lockers are gone
5482 : * and we have an updater that aborted, we can also do without a
5483 : * multi.
5484 : *
5485 : * The cost of doing GetMultiXactIdMembers would be paid by
5486 : * MultiXactIdExpand if we weren't to do this, so this check is not
5487 : * incurring extra work anyhow.
5488 : */
5489 151124 : if (!MultiXactIdIsRunning(xmax, HEAP_XMAX_IS_LOCKED_ONLY(old_infomask)))
5490 : {
5491 48 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) ||
5492 16 : !TransactionIdDidCommit(MultiXactIdGetUpdateXid(xmax,
5493 : old_infomask)))
5494 : {
5495 : /*
5496 : * Reset these bits and restart; otherwise fall through to
5497 : * create a new multi below.
5498 : */
5499 48 : old_infomask &= ~HEAP_XMAX_IS_MULTI;
5500 48 : old_infomask |= HEAP_XMAX_INVALID;
5501 48 : goto l5;
5502 : }
5503 : }
5504 :
5505 151076 : new_status = get_mxact_status_for_lock(mode, is_update);
5506 :
5507 151076 : new_xmax = MultiXactIdExpand((MultiXactId) xmax, add_to_xmax,
5508 : new_status);
5509 151076 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5510 : }
5511 210272 : else if (old_infomask & HEAP_XMAX_COMMITTED)
5512 : {
5513 : /*
5514 : * It's a committed update, so we need to preserve him as updater of
5515 : * the tuple.
5516 : */
5517 : MultiXactStatus status;
5518 : MultiXactStatus new_status;
5519 :
5520 26 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5521 0 : status = MultiXactStatusUpdate;
5522 : else
5523 26 : status = MultiXactStatusNoKeyUpdate;
5524 :
5525 26 : new_status = get_mxact_status_for_lock(mode, is_update);
5526 :
5527 : /*
5528 : * since it's not running, it's obviously impossible for the old
5529 : * updater to be identical to the current one, so we need not check
5530 : * for that case as we do in the block above.
5531 : */
5532 26 : new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5533 26 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5534 : }
5535 210246 : else if (TransactionIdIsInProgress(xmax))
5536 : {
5537 : /*
5538 : * If the XMAX is a valid, in-progress TransactionId, then we need to
5539 : * create a new MultiXactId that includes both the old locker or
5540 : * updater and our own TransactionId.
5541 : */
5542 : MultiXactStatus new_status;
5543 : MultiXactStatus old_status;
5544 : LockTupleMode old_mode;
5545 :
5546 210228 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
5547 : {
5548 210176 : if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
5549 11354 : old_status = MultiXactStatusForKeyShare;
5550 198822 : else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
5551 866 : old_status = MultiXactStatusForShare;
5552 197956 : else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
5553 : {
5554 197956 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5555 185646 : old_status = MultiXactStatusForUpdate;
5556 : else
5557 12310 : old_status = MultiXactStatusForNoKeyUpdate;
5558 : }
5559 : else
5560 : {
5561 : /*
5562 : * LOCK_ONLY can be present alone only when a page has been
5563 : * upgraded by pg_upgrade. But in that case,
5564 : * TransactionIdIsInProgress() should have returned false. We
5565 : * assume it's no longer locked in this case.
5566 : */
5567 0 : elog(WARNING, "LOCK_ONLY found for Xid in progress %u", xmax);
5568 0 : old_infomask |= HEAP_XMAX_INVALID;
5569 0 : old_infomask &= ~HEAP_XMAX_LOCK_ONLY;
5570 0 : goto l5;
5571 : }
5572 : }
5573 : else
5574 : {
5575 : /* it's an update, but which kind? */
5576 52 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5577 0 : old_status = MultiXactStatusUpdate;
5578 : else
5579 52 : old_status = MultiXactStatusNoKeyUpdate;
5580 : }
5581 :
5582 210228 : old_mode = TUPLOCK_from_mxstatus(old_status);
5583 :
5584 : /*
5585 : * If the lock to be acquired is for the same TransactionId as the
5586 : * existing lock, there's an optimization possible: consider only the
5587 : * strongest of both locks as the only one present, and restart.
5588 : */
5589 210228 : if (xmax == add_to_xmax)
5590 : {
5591 : /*
5592 : * Note that it's not possible for the original tuple to be
5593 : * updated: we wouldn't be here because the tuple would have been
5594 : * invisible and we wouldn't try to update it. As a subtlety,
5595 : * this code can also run when traversing an update chain to lock
5596 : * future versions of a tuple. But we wouldn't be here either,
5597 : * because the add_to_xmax would be different from the original
5598 : * updater.
5599 : */
5600 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
5601 :
5602 : /* acquire the strongest of both */
5603 208076 : if (mode < old_mode)
5604 104510 : mode = old_mode;
5605 : /* mustn't touch is_update */
5606 :
5607 208076 : old_infomask |= HEAP_XMAX_INVALID;
5608 208076 : goto l5;
5609 : }
5610 :
5611 : /* otherwise, just fall back to creating a new multixact */
5612 2152 : new_status = get_mxact_status_for_lock(mode, is_update);
5613 2152 : new_xmax = MultiXactIdCreate(xmax, old_status,
5614 : add_to_xmax, new_status);
5615 2152 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5616 : }
5617 28 : else if (!HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) &&
5618 10 : TransactionIdDidCommit(xmax))
5619 2 : {
5620 : /*
5621 : * It's a committed update, so we gotta preserve him as updater of the
5622 : * tuple.
5623 : */
5624 : MultiXactStatus status;
5625 : MultiXactStatus new_status;
5626 :
5627 2 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5628 0 : status = MultiXactStatusUpdate;
5629 : else
5630 2 : status = MultiXactStatusNoKeyUpdate;
5631 :
5632 2 : new_status = get_mxact_status_for_lock(mode, is_update);
5633 :
5634 : /*
5635 : * since it's not running, it's obviously impossible for the old
5636 : * updater to be identical to the current one, so we need not check
5637 : * for that case as we do in the block above.
5638 : */
5639 2 : new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5640 2 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5641 : }
5642 : else
5643 : {
5644 : /*
5645 : * Can get here iff the locking/updating transaction was running when
5646 : * the infomask was extracted from the tuple, but finished before
5647 : * TransactionIdIsInProgress got to run. Deal with it as if there was
5648 : * no locker at all in the first place.
5649 : */
5650 16 : old_infomask |= HEAP_XMAX_INVALID;
5651 16 : goto l5;
5652 : }
5653 :
5654 4297138 : *result_infomask = new_infomask;
5655 4297138 : *result_infomask2 = new_infomask2;
5656 4297138 : *result_xmax = new_xmax;
5657 4297138 : }
5658 :
5659 : /*
5660 : * Subroutine for heap_lock_updated_tuple_rec.
5661 : *
5662 : * Given a hypothetical multixact status held by the transaction identified
5663 : * with the given xid, does the current transaction need to wait, fail, or can
5664 : * it continue if it wanted to acquire a lock of the given mode? "needwait"
5665 : * is set to true if waiting is necessary; if it can continue, then TM_Ok is
5666 : * returned. If the lock is already held by the current transaction, return
5667 : * TM_SelfModified. In case of a conflict with another transaction, a
5668 : * different HeapTupleSatisfiesUpdate return code is returned.
5669 : *
5670 : * The held status is said to be hypothetical because it might correspond to a
5671 : * lock held by a single Xid, i.e. not a real MultiXactId; we express it this
5672 : * way for simplicity of API.
5673 : */
5674 : static TM_Result
5675 77548 : test_lockmode_for_conflict(MultiXactStatus status, TransactionId xid,
5676 : LockTupleMode mode, HeapTuple tup,
5677 : bool *needwait)
5678 : {
5679 : MultiXactStatus wantedstatus;
5680 :
5681 77548 : *needwait = false;
5682 77548 : wantedstatus = get_mxact_status_for_lock(mode, false);
5683 :
5684 : /*
5685 : * Note: we *must* check TransactionIdIsInProgress before
5686 : * TransactionIdDidAbort/Commit; see comment at top of heapam_visibility.c
5687 : * for an explanation.
5688 : */
5689 77548 : if (TransactionIdIsCurrentTransactionId(xid))
5690 : {
5691 : /*
5692 : * The tuple has already been locked by our own transaction. This is
5693 : * very rare but can happen if multiple transactions are trying to
5694 : * lock an ancient version of the same tuple.
5695 : */
5696 0 : return TM_SelfModified;
5697 : }
5698 77548 : else if (TransactionIdIsInProgress(xid))
5699 : {
5700 : /*
5701 : * If the locking transaction is running, what we do depends on
5702 : * whether the lock modes conflict: if they do, then we must wait for
5703 : * it to finish; otherwise we can fall through to lock this tuple
5704 : * version without waiting.
5705 : */
5706 73078 : if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
5707 73078 : LOCKMODE_from_mxstatus(wantedstatus)))
5708 : {
5709 16 : *needwait = true;
5710 : }
5711 :
5712 : /*
5713 : * If we set needwait above, then this value doesn't matter;
5714 : * otherwise, this value signals to caller that it's okay to proceed.
5715 : */
5716 73078 : return TM_Ok;
5717 : }
5718 4470 : else if (TransactionIdDidAbort(xid))
5719 412 : return TM_Ok;
5720 4058 : else if (TransactionIdDidCommit(xid))
5721 : {
5722 : /*
5723 : * The other transaction committed. If it was only a locker, then the
5724 : * lock is completely gone now and we can return success; but if it
5725 : * was an update, then what we do depends on whether the two lock
5726 : * modes conflict. If they conflict, then we must report error to
5727 : * caller. But if they don't, we can fall through to allow the current
5728 : * transaction to lock the tuple.
5729 : *
5730 : * Note: the reason we worry about ISUPDATE here is because as soon as
5731 : * a transaction ends, all its locks are gone and meaningless, and
5732 : * thus we can ignore them; whereas its updates persist. In the
5733 : * TransactionIdIsInProgress case, above, we don't need to check
5734 : * because we know the lock is still "alive" and thus a conflict needs
5735 : * always be checked.
5736 : */
5737 4058 : if (!ISUPDATE_from_mxstatus(status))
5738 4040 : return TM_Ok;
5739 :
5740 18 : if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
5741 18 : LOCKMODE_from_mxstatus(wantedstatus)))
5742 : {
5743 : /* bummer */
5744 16 : if (!ItemPointerEquals(&tup->t_self, &tup->t_data->t_ctid))
5745 12 : return TM_Updated;
5746 : else
5747 4 : return TM_Deleted;
5748 : }
5749 :
5750 2 : return TM_Ok;
5751 : }
5752 :
5753 : /* Not in progress, not aborted, not committed -- must have crashed */
5754 0 : return TM_Ok;
5755 : }
5756 :
5757 :
5758 : /*
5759 : * Recursive part of heap_lock_updated_tuple
5760 : *
5761 : * Fetch the tuple pointed to by tid in rel, and mark it as locked by the given
5762 : * xid with the given mode; if this tuple is updated, recurse to lock the new
5763 : * version as well.
5764 : */
5765 : static TM_Result
5766 4424 : heap_lock_updated_tuple_rec(Relation rel, TransactionId priorXmax,
5767 : const ItemPointerData *tid, TransactionId xid,
5768 : LockTupleMode mode)
5769 : {
5770 : TM_Result result;
5771 : ItemPointerData tupid;
5772 : HeapTupleData mytup;
5773 : Buffer buf;
5774 : uint16 new_infomask,
5775 : new_infomask2,
5776 : old_infomask,
5777 : old_infomask2;
5778 : TransactionId xmax,
5779 : new_xmax;
5780 4424 : bool cleared_all_frozen = false;
5781 : bool pinned_desired_page;
5782 4424 : Buffer vmbuffer = InvalidBuffer;
5783 : BlockNumber block;
5784 :
5785 4424 : ItemPointerCopy(tid, &tupid);
5786 :
5787 : for (;;)
5788 : {
5789 4430 : new_infomask = 0;
5790 4430 : new_xmax = InvalidTransactionId;
5791 4430 : block = ItemPointerGetBlockNumber(&tupid);
5792 4430 : ItemPointerCopy(&tupid, &(mytup.t_self));
5793 :
5794 4430 : if (!heap_fetch(rel, SnapshotAny, &mytup, &buf, false))
5795 : {
5796 : /*
5797 : * if we fail to find the updated version of the tuple, it's
5798 : * because it was vacuumed/pruned away after its creator
5799 : * transaction aborted. So behave as if we got to the end of the
5800 : * chain, and there's no further tuple to lock: return success to
5801 : * caller.
5802 : */
5803 0 : result = TM_Ok;
5804 0 : goto out_unlocked;
5805 : }
5806 :
5807 4430 : l4:
5808 4446 : CHECK_FOR_INTERRUPTS();
5809 :
5810 : /*
5811 : * Before locking the buffer, pin the visibility map page if it
5812 : * appears to be necessary. Since we haven't got the lock yet,
5813 : * someone else might be in the middle of changing this, so we'll need
5814 : * to recheck after we have the lock.
5815 : */
5816 4446 : if (PageIsAllVisible(BufferGetPage(buf)))
5817 : {
5818 0 : visibilitymap_pin(rel, block, &vmbuffer);
5819 0 : pinned_desired_page = true;
5820 : }
5821 : else
5822 4446 : pinned_desired_page = false;
5823 :
5824 4446 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
5825 :
5826 : /*
5827 : * If we didn't pin the visibility map page and the page has become
5828 : * all visible while we were busy locking the buffer, we'll have to
5829 : * unlock and re-lock, to avoid holding the buffer lock across I/O.
5830 : * That's a bit unfortunate, but hopefully shouldn't happen often.
5831 : *
5832 : * Note: in some paths through this function, we will reach here
5833 : * holding a pin on a vm page that may or may not be the one matching
5834 : * this page. If this page isn't all-visible, we won't use the vm
5835 : * page, but we hold onto such a pin till the end of the function.
5836 : */
5837 4446 : if (!pinned_desired_page && PageIsAllVisible(BufferGetPage(buf)))
5838 : {
5839 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5840 0 : visibilitymap_pin(rel, block, &vmbuffer);
5841 0 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
5842 : }
5843 :
5844 : /*
5845 : * Check the tuple XMIN against prior XMAX, if any. If we reached the
5846 : * end of the chain, we're done, so return success.
5847 : */
5848 8892 : if (TransactionIdIsValid(priorXmax) &&
5849 4446 : !TransactionIdEquals(HeapTupleHeaderGetXmin(mytup.t_data),
5850 : priorXmax))
5851 : {
5852 4 : result = TM_Ok;
5853 4 : goto out_locked;
5854 : }
5855 :
5856 : /*
5857 : * Also check Xmin: if this tuple was created by an aborted
5858 : * (sub)transaction, then we already locked the last live one in the
5859 : * chain, thus we're done, so return success.
5860 : */
5861 4442 : if (TransactionIdDidAbort(HeapTupleHeaderGetXmin(mytup.t_data)))
5862 : {
5863 48 : result = TM_Ok;
5864 48 : goto out_locked;
5865 : }
5866 :
5867 4394 : old_infomask = mytup.t_data->t_infomask;
5868 4394 : old_infomask2 = mytup.t_data->t_infomask2;
5869 4394 : xmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
5870 :
5871 : /*
5872 : * If this tuple version has been updated or locked by some concurrent
5873 : * transaction(s), what we do depends on whether our lock mode
5874 : * conflicts with what those other transactions hold, and also on the
5875 : * status of them.
5876 : */
5877 4394 : if (!(old_infomask & HEAP_XMAX_INVALID))
5878 : {
5879 : TransactionId rawxmax;
5880 : bool needwait;
5881 :
5882 4276 : rawxmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
5883 4276 : if (old_infomask & HEAP_XMAX_IS_MULTI)
5884 : {
5885 : int nmembers;
5886 : int i;
5887 : MultiXactMember *members;
5888 :
5889 : /*
5890 : * We don't need a test for pg_upgrade'd tuples: this is only
5891 : * applied to tuples after the first in an update chain. Said
5892 : * first tuple in the chain may well be locked-in-9.2-and-
5893 : * pg_upgraded, but that one was already locked by our caller,
5894 : * not us; and any subsequent ones cannot be because our
5895 : * caller must necessarily have obtained a snapshot later than
5896 : * the pg_upgrade itself.
5897 : */
5898 : Assert(!HEAP_LOCKED_UPGRADED(mytup.t_data->t_infomask));
5899 :
5900 4218 : nmembers = GetMultiXactIdMembers(rawxmax, &members, false,
5901 4218 : HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
5902 81708 : for (i = 0; i < nmembers; i++)
5903 : {
5904 77490 : result = test_lockmode_for_conflict(members[i].status,
5905 77490 : members[i].xid,
5906 : mode,
5907 : &mytup,
5908 : &needwait);
5909 :
5910 : /*
5911 : * If the tuple was already locked by ourselves in a
5912 : * previous iteration of this (say heap_lock_tuple was
5913 : * forced to restart the locking loop because of a change
5914 : * in xmax), then we hold the lock already on this tuple
5915 : * version and we don't need to do anything; and this is
5916 : * not an error condition either. We just need to skip
5917 : * this tuple and continue locking the next version in the
5918 : * update chain.
5919 : */
5920 77490 : if (result == TM_SelfModified)
5921 : {
5922 0 : pfree(members);
5923 0 : goto next;
5924 : }
5925 :
5926 77490 : if (needwait)
5927 : {
5928 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5929 0 : XactLockTableWait(members[i].xid, rel,
5930 : &mytup.t_self,
5931 : XLTW_LockUpdated);
5932 0 : pfree(members);
5933 0 : goto l4;
5934 : }
5935 77490 : if (result != TM_Ok)
5936 : {
5937 0 : pfree(members);
5938 0 : goto out_locked;
5939 : }
5940 : }
5941 4218 : if (members)
5942 4218 : pfree(members);
5943 : }
5944 : else
5945 : {
5946 : MultiXactStatus status;
5947 :
5948 : /*
5949 : * For a non-multi Xmax, we first need to compute the
5950 : * corresponding MultiXactStatus by using the infomask bits.
5951 : */
5952 58 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
5953 : {
5954 20 : if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
5955 20 : status = MultiXactStatusForKeyShare;
5956 0 : else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
5957 0 : status = MultiXactStatusForShare;
5958 0 : else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
5959 : {
5960 0 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5961 0 : status = MultiXactStatusForUpdate;
5962 : else
5963 0 : status = MultiXactStatusForNoKeyUpdate;
5964 : }
5965 : else
5966 : {
5967 : /*
5968 : * LOCK_ONLY present alone (a pg_upgraded tuple marked
5969 : * as share-locked in the old cluster) shouldn't be
5970 : * seen in the middle of an update chain.
5971 : */
5972 0 : elog(ERROR, "invalid lock status in tuple");
5973 : }
5974 : }
5975 : else
5976 : {
5977 : /* it's an update, but which kind? */
5978 38 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5979 28 : status = MultiXactStatusUpdate;
5980 : else
5981 10 : status = MultiXactStatusNoKeyUpdate;
5982 : }
5983 :
5984 58 : result = test_lockmode_for_conflict(status, rawxmax, mode,
5985 : &mytup, &needwait);
5986 :
5987 : /*
5988 : * If the tuple was already locked by ourselves in a previous
5989 : * iteration of this (say heap_lock_tuple was forced to
5990 : * restart the locking loop because of a change in xmax), then
5991 : * we hold the lock already on this tuple version and we don't
5992 : * need to do anything; and this is not an error condition
5993 : * either. We just need to skip this tuple and continue
5994 : * locking the next version in the update chain.
5995 : */
5996 58 : if (result == TM_SelfModified)
5997 0 : goto next;
5998 :
5999 58 : if (needwait)
6000 : {
6001 16 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
6002 16 : XactLockTableWait(rawxmax, rel, &mytup.t_self,
6003 : XLTW_LockUpdated);
6004 16 : goto l4;
6005 : }
6006 42 : if (result != TM_Ok)
6007 : {
6008 16 : goto out_locked;
6009 : }
6010 : }
6011 : }
6012 :
6013 : /* compute the new Xmax and infomask values for the tuple ... */
6014 4362 : compute_new_xmax_infomask(xmax, old_infomask, mytup.t_data->t_infomask2,
6015 : xid, mode, false,
6016 : &new_xmax, &new_infomask, &new_infomask2);
6017 :
6018 4362 : if (PageIsAllVisible(BufferGetPage(buf)) &&
6019 0 : visibilitymap_clear(rel, block, vmbuffer,
6020 : VISIBILITYMAP_ALL_FROZEN))
6021 0 : cleared_all_frozen = true;
6022 :
6023 4362 : START_CRIT_SECTION();
6024 :
6025 : /* ... and set them */
6026 4362 : HeapTupleHeaderSetXmax(mytup.t_data, new_xmax);
6027 4362 : mytup.t_data->t_infomask &= ~HEAP_XMAX_BITS;
6028 4362 : mytup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
6029 4362 : mytup.t_data->t_infomask |= new_infomask;
6030 4362 : mytup.t_data->t_infomask2 |= new_infomask2;
6031 :
6032 4362 : MarkBufferDirty(buf);
6033 :
6034 : /* XLOG stuff */
6035 4362 : if (RelationNeedsWAL(rel))
6036 : {
6037 : xl_heap_lock_updated xlrec;
6038 : XLogRecPtr recptr;
6039 4362 : Page page = BufferGetPage(buf);
6040 :
6041 4362 : XLogBeginInsert();
6042 4362 : XLogRegisterBuffer(0, buf, REGBUF_STANDARD);
6043 :
6044 4362 : xlrec.offnum = ItemPointerGetOffsetNumber(&mytup.t_self);
6045 4362 : xlrec.xmax = new_xmax;
6046 4362 : xlrec.infobits_set = compute_infobits(new_infomask, new_infomask2);
6047 4362 : xlrec.flags =
6048 4362 : cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
6049 :
6050 4362 : XLogRegisterData(&xlrec, SizeOfHeapLockUpdated);
6051 :
6052 4362 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_LOCK_UPDATED);
6053 :
6054 4362 : PageSetLSN(page, recptr);
6055 : }
6056 :
6057 4362 : END_CRIT_SECTION();
6058 :
6059 4362 : next:
6060 : /* if we find the end of update chain, we're done. */
6061 8724 : if (mytup.t_data->t_infomask & HEAP_XMAX_INVALID ||
6062 8724 : HeapTupleHeaderIndicatesMovedPartitions(mytup.t_data) ||
6063 4370 : ItemPointerEquals(&mytup.t_self, &mytup.t_data->t_ctid) ||
6064 8 : HeapTupleHeaderIsOnlyLocked(mytup.t_data))
6065 : {
6066 4356 : result = TM_Ok;
6067 4356 : goto out_locked;
6068 : }
6069 :
6070 : /* tail recursion */
6071 6 : priorXmax = HeapTupleHeaderGetUpdateXid(mytup.t_data);
6072 6 : ItemPointerCopy(&(mytup.t_data->t_ctid), &tupid);
6073 6 : UnlockReleaseBuffer(buf);
6074 : }
6075 :
6076 : result = TM_Ok;
6077 :
6078 4424 : out_locked:
6079 4424 : UnlockReleaseBuffer(buf);
6080 :
6081 4424 : out_unlocked:
6082 4424 : if (vmbuffer != InvalidBuffer)
6083 0 : ReleaseBuffer(vmbuffer);
6084 :
6085 4424 : return result;
6086 : }
6087 :
6088 : /*
6089 : * heap_lock_updated_tuple
6090 : * Follow update chain when locking an updated tuple, acquiring locks (row
6091 : * marks) on the updated versions.
6092 : *
6093 : * 'prior_infomask', 'prior_raw_xmax' and 'prior_ctid' are the corresponding
6094 : * fields from the initial tuple. We will lock the tuples starting from the
6095 : * one that 'prior_ctid' points to. Note: This function does not lock the
6096 : * initial tuple itself.
6097 : *
6098 : * This function doesn't check visibility, it just unconditionally marks the
6099 : * tuple(s) as locked. If any tuple in the updated chain is being deleted
6100 : * concurrently (or updated with the key being modified), sleep until the
6101 : * transaction doing it is finished.
6102 : *
6103 : * Note that we don't acquire heavyweight tuple locks on the tuples we walk
6104 : * when we have to wait for other transactions to release them, as opposed to
6105 : * what heap_lock_tuple does. The reason is that having more than one
6106 : * transaction walking the chain is probably uncommon enough that risk of
6107 : * starvation is not likely: one of the preconditions for being here is that
6108 : * the snapshot in use predates the update that created this tuple (because we
6109 : * started at an earlier version of the tuple), but at the same time such a
6110 : * transaction cannot be using repeatable read or serializable isolation
6111 : * levels, because that would lead to a serializability failure.
6112 : */
6113 : static TM_Result
6114 4428 : heap_lock_updated_tuple(Relation rel,
6115 : uint16 prior_infomask,
6116 : TransactionId prior_raw_xmax,
6117 : const ItemPointerData *prior_ctid,
6118 : TransactionId xid, LockTupleMode mode)
6119 : {
6120 4428 : INJECTION_POINT("heap_lock_updated_tuple", NULL);
6121 :
6122 : /*
6123 : * If the tuple has moved into another partition (effectively a delete)
6124 : * stop here.
6125 : */
6126 4428 : if (!ItemPointerIndicatesMovedPartitions(prior_ctid))
6127 : {
6128 : TransactionId prior_xmax;
6129 :
6130 : /*
6131 : * If this is the first possibly-multixact-able operation in the
6132 : * current transaction, set my per-backend OldestMemberMXactId
6133 : * setting. We can be certain that the transaction will never become a
6134 : * member of any older MultiXactIds than that. (We have to do this
6135 : * even if we end up just using our own TransactionId below, since
6136 : * some other backend could incorporate our XID into a MultiXact
6137 : * immediately afterwards.)
6138 : */
6139 4424 : MultiXactIdSetOldestMember();
6140 :
6141 8848 : prior_xmax = (prior_infomask & HEAP_XMAX_IS_MULTI) ?
6142 4424 : MultiXactIdGetUpdateXid(prior_raw_xmax, prior_infomask) : prior_raw_xmax;
6143 4424 : return heap_lock_updated_tuple_rec(rel, prior_xmax, prior_ctid, xid, mode);
6144 : }
6145 :
6146 : /* nothing to lock */
6147 4 : return TM_Ok;
6148 : }
6149 :
6150 : /*
6151 : * heap_finish_speculative - mark speculative insertion as successful
6152 : *
6153 : * To successfully finish a speculative insertion we have to clear speculative
6154 : * token from tuple. To do so the t_ctid field, which will contain a
6155 : * speculative token value, is modified in place to point to the tuple itself,
6156 : * which is characteristic of a newly inserted ordinary tuple.
6157 : *
6158 : * NB: It is not ok to commit without either finishing or aborting a
6159 : * speculative insertion. We could treat speculative tuples of committed
6160 : * transactions implicitly as completed, but then we would have to be prepared
6161 : * to deal with speculative tokens on committed tuples. That wouldn't be
6162 : * difficult - no-one looks at the ctid field of a tuple with invalid xmax -
6163 : * but clearing the token at completion isn't very expensive either.
6164 : * An explicit confirmation WAL record also makes logical decoding simpler.
6165 : */
6166 : void
6167 4176 : heap_finish_speculative(Relation relation, const ItemPointerData *tid)
6168 : {
6169 : Buffer buffer;
6170 : Page page;
6171 : OffsetNumber offnum;
6172 : ItemId lp;
6173 : HeapTupleHeader htup;
6174 :
6175 4176 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
6176 4176 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6177 4176 : page = BufferGetPage(buffer);
6178 :
6179 4176 : offnum = ItemPointerGetOffsetNumber(tid);
6180 4176 : if (offnum < 1 || offnum > PageGetMaxOffsetNumber(page))
6181 0 : elog(ERROR, "offnum out of range");
6182 4176 : lp = PageGetItemId(page, offnum);
6183 4176 : if (!ItemIdIsNormal(lp))
6184 0 : elog(ERROR, "invalid lp");
6185 :
6186 4176 : htup = (HeapTupleHeader) PageGetItem(page, lp);
6187 :
6188 : /* NO EREPORT(ERROR) from here till changes are logged */
6189 4176 : START_CRIT_SECTION();
6190 :
6191 : Assert(HeapTupleHeaderIsSpeculative(htup));
6192 :
6193 4176 : MarkBufferDirty(buffer);
6194 :
6195 : /*
6196 : * Replace the speculative insertion token with a real t_ctid, pointing to
6197 : * itself like it does on regular tuples.
6198 : */
6199 4176 : htup->t_ctid = *tid;
6200 :
6201 : /* XLOG stuff */
6202 4176 : if (RelationNeedsWAL(relation))
6203 : {
6204 : xl_heap_confirm xlrec;
6205 : XLogRecPtr recptr;
6206 :
6207 4144 : xlrec.offnum = ItemPointerGetOffsetNumber(tid);
6208 :
6209 4144 : XLogBeginInsert();
6210 :
6211 : /* We want the same filtering on this as on a plain insert */
6212 4144 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
6213 :
6214 4144 : XLogRegisterData(&xlrec, SizeOfHeapConfirm);
6215 4144 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
6216 :
6217 4144 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_CONFIRM);
6218 :
6219 4144 : PageSetLSN(page, recptr);
6220 : }
6221 :
6222 4176 : END_CRIT_SECTION();
6223 :
6224 4176 : UnlockReleaseBuffer(buffer);
6225 4176 : }
6226 :
6227 : /*
6228 : * heap_abort_speculative - kill a speculatively inserted tuple
6229 : *
6230 : * Marks a tuple that was speculatively inserted in the same command as dead,
6231 : * by setting its xmin as invalid. That makes it immediately appear as dead
6232 : * to all transactions, including our own. In particular, it makes
6233 : * HeapTupleSatisfiesDirty() regard the tuple as dead, so that another backend
6234 : * inserting a duplicate key value won't unnecessarily wait for our whole
6235 : * transaction to finish (it'll just wait for our speculative insertion to
6236 : * finish).
6237 : *
6238 : * Killing the tuple prevents "unprincipled deadlocks", which are deadlocks
6239 : * that arise due to a mutual dependency that is not user visible. By
6240 : * definition, unprincipled deadlocks cannot be prevented by the user
6241 : * reordering lock acquisition in client code, because the implementation level
6242 : * lock acquisitions are not under the user's direct control. If speculative
6243 : * inserters did not take this precaution, then under high concurrency they
6244 : * could deadlock with each other, which would not be acceptable.
6245 : *
6246 : * This is somewhat redundant with heap_delete, but we prefer to have a
6247 : * dedicated routine with stripped down requirements. Note that this is also
6248 : * used to delete the TOAST tuples created during speculative insertion.
6249 : *
6250 : * This routine does not affect logical decoding as it only looks at
6251 : * confirmation records.
6252 : */
6253 : void
6254 32 : heap_abort_speculative(Relation relation, const ItemPointerData *tid)
6255 : {
6256 32 : TransactionId xid = GetCurrentTransactionId();
6257 : ItemId lp;
6258 : HeapTupleData tp;
6259 : Page page;
6260 : BlockNumber block;
6261 : Buffer buffer;
6262 :
6263 : Assert(ItemPointerIsValid(tid));
6264 :
6265 32 : block = ItemPointerGetBlockNumber(tid);
6266 32 : buffer = ReadBuffer(relation, block);
6267 32 : page = BufferGetPage(buffer);
6268 :
6269 32 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6270 :
6271 : /*
6272 : * Page can't be all visible, we just inserted into it, and are still
6273 : * running.
6274 : */
6275 : Assert(!PageIsAllVisible(page));
6276 :
6277 32 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
6278 : Assert(ItemIdIsNormal(lp));
6279 :
6280 32 : tp.t_tableOid = RelationGetRelid(relation);
6281 32 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
6282 32 : tp.t_len = ItemIdGetLength(lp);
6283 32 : tp.t_self = *tid;
6284 :
6285 : /*
6286 : * Sanity check that the tuple really is a speculatively inserted tuple,
6287 : * inserted by us.
6288 : */
6289 32 : if (tp.t_data->t_choice.t_heap.t_xmin != xid)
6290 0 : elog(ERROR, "attempted to kill a tuple inserted by another transaction");
6291 32 : if (!(IsToastRelation(relation) || HeapTupleHeaderIsSpeculative(tp.t_data)))
6292 0 : elog(ERROR, "attempted to kill a non-speculative tuple");
6293 : Assert(!HeapTupleHeaderIsHeapOnly(tp.t_data));
6294 :
6295 : /*
6296 : * No need to check for serializable conflicts here. There is never a
6297 : * need for a combo CID, either. No need to extract replica identity, or
6298 : * do anything special with infomask bits.
6299 : */
6300 :
6301 32 : START_CRIT_SECTION();
6302 :
6303 : /*
6304 : * The tuple will become DEAD immediately. Flag that this page is a
6305 : * candidate for pruning by setting xmin to TransactionXmin. While not
6306 : * immediately prunable, it is the oldest xid we can cheaply determine
6307 : * that's safe against wraparound / being older than the table's
6308 : * relfrozenxid. To defend against the unlikely case of a new relation
6309 : * having a newer relfrozenxid than our TransactionXmin, use relfrozenxid
6310 : * if so (vacuum can't subsequently move relfrozenxid to beyond
6311 : * TransactionXmin, so there's no race here).
6312 : */
6313 : Assert(TransactionIdIsValid(TransactionXmin));
6314 : {
6315 32 : TransactionId relfrozenxid = relation->rd_rel->relfrozenxid;
6316 : TransactionId prune_xid;
6317 :
6318 32 : if (TransactionIdPrecedes(TransactionXmin, relfrozenxid))
6319 0 : prune_xid = relfrozenxid;
6320 : else
6321 32 : prune_xid = TransactionXmin;
6322 32 : PageSetPrunable(page, prune_xid);
6323 : }
6324 :
6325 : /* store transaction information of xact deleting the tuple */
6326 32 : tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
6327 32 : tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
6328 :
6329 : /*
6330 : * Set the tuple header xmin to InvalidTransactionId. This makes the
6331 : * tuple immediately invisible everyone. (In particular, to any
6332 : * transactions waiting on the speculative token, woken up later.)
6333 : */
6334 32 : HeapTupleHeaderSetXmin(tp.t_data, InvalidTransactionId);
6335 :
6336 : /* Clear the speculative insertion token too */
6337 32 : tp.t_data->t_ctid = tp.t_self;
6338 :
6339 32 : MarkBufferDirty(buffer);
6340 :
6341 : /*
6342 : * XLOG stuff
6343 : *
6344 : * The WAL records generated here match heap_delete(). The same recovery
6345 : * routines are used.
6346 : */
6347 32 : if (RelationNeedsWAL(relation))
6348 : {
6349 : xl_heap_delete xlrec;
6350 : XLogRecPtr recptr;
6351 :
6352 24 : xlrec.flags = XLH_DELETE_IS_SUPER;
6353 48 : xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
6354 24 : tp.t_data->t_infomask2);
6355 24 : xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
6356 24 : xlrec.xmax = xid;
6357 :
6358 24 : XLogBeginInsert();
6359 24 : XLogRegisterData(&xlrec, SizeOfHeapDelete);
6360 24 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
6361 :
6362 : /* No replica identity & replication origin logged */
6363 :
6364 24 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
6365 :
6366 24 : PageSetLSN(page, recptr);
6367 : }
6368 :
6369 32 : END_CRIT_SECTION();
6370 :
6371 32 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6372 :
6373 32 : if (HeapTupleHasExternal(&tp))
6374 : {
6375 : Assert(!IsToastRelation(relation));
6376 2 : heap_toast_delete(relation, &tp, true);
6377 : }
6378 :
6379 : /*
6380 : * Never need to mark tuple for invalidation, since catalogs don't support
6381 : * speculative insertion
6382 : */
6383 :
6384 : /* Now we can release the buffer */
6385 32 : ReleaseBuffer(buffer);
6386 :
6387 : /* count deletion, as we counted the insertion too */
6388 32 : pgstat_count_heap_delete(relation);
6389 32 : }
6390 :
6391 : /*
6392 : * heap_inplace_lock - protect inplace update from concurrent heap_update()
6393 : *
6394 : * Evaluate whether the tuple's state is compatible with a no-key update.
6395 : * Current transaction rowmarks are fine, as is KEY SHARE from any
6396 : * transaction. If compatible, return true with the buffer exclusive-locked,
6397 : * and the caller must release that by calling
6398 : * heap_inplace_update_and_unlock(), calling heap_inplace_unlock(), or raising
6399 : * an error. Otherwise, call release_callback(arg), wait for blocking
6400 : * transactions to end, and return false.
6401 : *
6402 : * Since this is intended for system catalogs and SERIALIZABLE doesn't cover
6403 : * DDL, this doesn't guarantee any particular predicate locking.
6404 : *
6405 : * heap_delete() is a rarer source of blocking transactions (xwait). We'll
6406 : * wait for such a transaction just like for the normal heap_update() case.
6407 : * Normal concurrent DROP commands won't cause that, because all inplace
6408 : * updaters take some lock that conflicts with DROP. An explicit SQL "DELETE
6409 : * FROM pg_class" can cause it. By waiting, if the concurrent transaction
6410 : * executed both "DELETE FROM pg_class" and "INSERT INTO pg_class", our caller
6411 : * can find the successor tuple.
6412 : *
6413 : * Readers of inplace-updated fields expect changes to those fields are
6414 : * durable. For example, vac_truncate_clog() reads datfrozenxid from
6415 : * pg_database tuples via catalog snapshots. A future snapshot must not
6416 : * return a lower datfrozenxid for the same database OID (lower in the
6417 : * FullTransactionIdPrecedes() sense). We achieve that since no update of a
6418 : * tuple can start while we hold a lock on its buffer. In cases like
6419 : * BEGIN;GRANT;CREATE INDEX;COMMIT we're inplace-updating a tuple visible only
6420 : * to this transaction. ROLLBACK then is one case where it's okay to lose
6421 : * inplace updates. (Restoring relhasindex=false on ROLLBACK is fine, since
6422 : * any concurrent CREATE INDEX would have blocked, then inplace-updated the
6423 : * committed tuple.)
6424 : *
6425 : * In principle, we could avoid waiting by overwriting every tuple in the
6426 : * updated tuple chain. Reader expectations permit updating a tuple only if
6427 : * it's aborted, is the tail of the chain, or we already updated the tuple
6428 : * referenced in its t_ctid. Hence, we would need to overwrite the tuples in
6429 : * order from tail to head. That would imply either (a) mutating all tuples
6430 : * in one critical section or (b) accepting a chance of partial completion.
6431 : * Partial completion of a relfrozenxid update would have the weird
6432 : * consequence that the table's next VACUUM could see the table's relfrozenxid
6433 : * move forward between vacuum_get_cutoffs() and finishing.
6434 : */
6435 : bool
6436 317364 : heap_inplace_lock(Relation relation,
6437 : HeapTuple oldtup_ptr, Buffer buffer,
6438 : void (*release_callback) (void *), void *arg)
6439 : {
6440 317364 : HeapTupleData oldtup = *oldtup_ptr; /* minimize diff vs. heap_update() */
6441 : TM_Result result;
6442 : bool ret;
6443 :
6444 : #ifdef USE_ASSERT_CHECKING
6445 : if (RelationGetRelid(relation) == RelationRelationId)
6446 : check_inplace_rel_lock(oldtup_ptr);
6447 : #endif
6448 :
6449 : Assert(BufferIsValid(buffer));
6450 :
6451 : /*
6452 : * Register shared cache invals if necessary. Other sessions may finish
6453 : * inplace updates of this tuple between this step and LockTuple(). Since
6454 : * inplace updates don't change cache keys, that's harmless.
6455 : *
6456 : * While it's tempting to register invals only after confirming we can
6457 : * return true, the following obstacle precludes reordering steps that
6458 : * way. Registering invals might reach a CatalogCacheInitializeCache()
6459 : * that locks "buffer". That would hang indefinitely if running after our
6460 : * own LockBuffer(). Hence, we must register invals before LockBuffer().
6461 : */
6462 317364 : CacheInvalidateHeapTupleInplace(relation, oldtup_ptr);
6463 :
6464 317364 : LockTuple(relation, &oldtup.t_self, InplaceUpdateTupleLock);
6465 317364 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6466 :
6467 : /*----------
6468 : * Interpret HeapTupleSatisfiesUpdate() like heap_update() does, except:
6469 : *
6470 : * - wait unconditionally
6471 : * - already locked tuple above, since inplace needs that unconditionally
6472 : * - don't recheck header after wait: simpler to defer to next iteration
6473 : * - don't try to continue even if the updater aborts: likewise
6474 : * - no crosscheck
6475 : */
6476 317364 : result = HeapTupleSatisfiesUpdate(&oldtup, GetCurrentCommandId(false),
6477 : buffer);
6478 :
6479 317364 : if (result == TM_Invisible)
6480 : {
6481 : /* no known way this can happen */
6482 0 : ereport(ERROR,
6483 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6484 : errmsg_internal("attempted to overwrite invisible tuple")));
6485 : }
6486 317364 : else if (result == TM_SelfModified)
6487 : {
6488 : /*
6489 : * CREATE INDEX might reach this if an expression is silly enough to
6490 : * call e.g. SELECT ... FROM pg_class FOR SHARE. C code of other SQL
6491 : * statements might get here after a heap_update() of the same row, in
6492 : * the absence of an intervening CommandCounterIncrement().
6493 : */
6494 0 : ereport(ERROR,
6495 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6496 : errmsg("tuple to be updated was already modified by an operation triggered by the current command")));
6497 : }
6498 317364 : else if (result == TM_BeingModified)
6499 : {
6500 : TransactionId xwait;
6501 : uint16 infomask;
6502 :
6503 34 : xwait = HeapTupleHeaderGetRawXmax(oldtup.t_data);
6504 34 : infomask = oldtup.t_data->t_infomask;
6505 :
6506 34 : if (infomask & HEAP_XMAX_IS_MULTI)
6507 : {
6508 10 : LockTupleMode lockmode = LockTupleNoKeyExclusive;
6509 10 : MultiXactStatus mxact_status = MultiXactStatusNoKeyUpdate;
6510 : int remain;
6511 :
6512 10 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
6513 : lockmode, NULL))
6514 : {
6515 4 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6516 4 : release_callback(arg);
6517 4 : ret = false;
6518 4 : MultiXactIdWait((MultiXactId) xwait, mxact_status, infomask,
6519 : relation, &oldtup.t_self, XLTW_Update,
6520 : &remain);
6521 : }
6522 : else
6523 6 : ret = true;
6524 : }
6525 24 : else if (TransactionIdIsCurrentTransactionId(xwait))
6526 2 : ret = true;
6527 22 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask))
6528 2 : ret = true;
6529 : else
6530 : {
6531 20 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6532 20 : release_callback(arg);
6533 20 : ret = false;
6534 20 : XactLockTableWait(xwait, relation, &oldtup.t_self,
6535 : XLTW_Update);
6536 : }
6537 : }
6538 : else
6539 : {
6540 317330 : ret = (result == TM_Ok);
6541 317330 : if (!ret)
6542 : {
6543 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6544 0 : release_callback(arg);
6545 : }
6546 : }
6547 :
6548 : /*
6549 : * GetCatalogSnapshot() relies on invalidation messages to know when to
6550 : * take a new snapshot. COMMIT of xwait is responsible for sending the
6551 : * invalidation. We're not acquiring heavyweight locks sufficient to
6552 : * block if not yet sent, so we must take a new snapshot to ensure a later
6553 : * attempt has a fair chance. While we don't need this if xwait aborted,
6554 : * don't bother optimizing that.
6555 : */
6556 317364 : if (!ret)
6557 : {
6558 24 : UnlockTuple(relation, &oldtup.t_self, InplaceUpdateTupleLock);
6559 24 : ForgetInplace_Inval();
6560 24 : InvalidateCatalogSnapshot();
6561 : }
6562 317364 : return ret;
6563 : }
6564 :
6565 : /*
6566 : * heap_inplace_update_and_unlock - core of systable_inplace_update_finish
6567 : *
6568 : * The tuple cannot change size, and therefore its header fields and null
6569 : * bitmap (if any) don't change either.
6570 : *
6571 : * Since we hold LOCKTAG_TUPLE, no updater has a local copy of this tuple.
6572 : */
6573 : void
6574 167530 : heap_inplace_update_and_unlock(Relation relation,
6575 : HeapTuple oldtup, HeapTuple tuple,
6576 : Buffer buffer)
6577 : {
6578 167530 : HeapTupleHeader htup = oldtup->t_data;
6579 : uint32 oldlen;
6580 : uint32 newlen;
6581 : char *dst;
6582 : char *src;
6583 167530 : int nmsgs = 0;
6584 167530 : SharedInvalidationMessage *invalMessages = NULL;
6585 167530 : bool RelcacheInitFileInval = false;
6586 :
6587 : Assert(ItemPointerEquals(&oldtup->t_self, &tuple->t_self));
6588 167530 : oldlen = oldtup->t_len - htup->t_hoff;
6589 167530 : newlen = tuple->t_len - tuple->t_data->t_hoff;
6590 167530 : if (oldlen != newlen || htup->t_hoff != tuple->t_data->t_hoff)
6591 0 : elog(ERROR, "wrong tuple length");
6592 :
6593 167530 : dst = (char *) htup + htup->t_hoff;
6594 167530 : src = (char *) tuple->t_data + tuple->t_data->t_hoff;
6595 :
6596 : /* Like RecordTransactionCommit(), log only if needed */
6597 167530 : if (XLogStandbyInfoActive())
6598 116818 : nmsgs = inplaceGetInvalidationMessages(&invalMessages,
6599 : &RelcacheInitFileInval);
6600 :
6601 : /*
6602 : * Unlink relcache init files as needed. If unlinking, acquire
6603 : * RelCacheInitLock until after associated invalidations. By doing this
6604 : * in advance, if we checkpoint and then crash between inplace
6605 : * XLogInsert() and inval, we don't rely on StartupXLOG() ->
6606 : * RelationCacheInitFileRemove(). That uses elevel==LOG, so replay would
6607 : * neglect to PANIC on EIO.
6608 : */
6609 167530 : PreInplace_Inval();
6610 :
6611 : /*----------
6612 : * NO EREPORT(ERROR) from here till changes are complete
6613 : *
6614 : * Our buffer lock won't stop a reader having already pinned and checked
6615 : * visibility for this tuple. Hence, we write WAL first, then mutate the
6616 : * buffer. Like in MarkBufferDirtyHint() or RecordTransactionCommit(),
6617 : * checkpoint delay makes that acceptable. With the usual order of
6618 : * changes, a crash after memcpy() and before XLogInsert() could allow
6619 : * datfrozenxid to overtake relfrozenxid:
6620 : *
6621 : * ["D" is a VACUUM (ONLY_DATABASE_STATS)]
6622 : * ["R" is a VACUUM tbl]
6623 : * D: vac_update_datfrozenxid() -> systable_beginscan(pg_class)
6624 : * D: systable_getnext() returns pg_class tuple of tbl
6625 : * R: memcpy() into pg_class tuple of tbl
6626 : * D: raise pg_database.datfrozenxid, XLogInsert(), finish
6627 : * [crash]
6628 : * [recovery restores datfrozenxid w/o relfrozenxid]
6629 : *
6630 : * Mimic MarkBufferDirtyHint() subroutine XLogSaveBufferForHint().
6631 : * Specifically, use DELAY_CHKPT_START, and copy the buffer to the stack.
6632 : * The stack copy facilitates a FPI of the post-mutation block before we
6633 : * accept other sessions seeing it. DELAY_CHKPT_START allows us to
6634 : * XLogInsert() before MarkBufferDirty(). Since XLogSaveBufferForHint()
6635 : * can operate under BUFFER_LOCK_SHARED, it can't avoid DELAY_CHKPT_START.
6636 : * This function, however, likely could avoid it with the following order
6637 : * of operations: MarkBufferDirty(), XLogInsert(), memcpy(). Opt to use
6638 : * DELAY_CHKPT_START here, too, as a way to have fewer distinct code
6639 : * patterns to analyze. Inplace update isn't so frequent that it should
6640 : * pursue the small optimization of skipping DELAY_CHKPT_START.
6641 : */
6642 : Assert((MyProc->delayChkptFlags & DELAY_CHKPT_START) == 0);
6643 167530 : START_CRIT_SECTION();
6644 167530 : MyProc->delayChkptFlags |= DELAY_CHKPT_START;
6645 :
6646 : /* XLOG stuff */
6647 167530 : if (RelationNeedsWAL(relation))
6648 : {
6649 : xl_heap_inplace xlrec;
6650 : PGAlignedBlock copied_buffer;
6651 167522 : char *origdata = (char *) BufferGetBlock(buffer);
6652 167522 : Page page = BufferGetPage(buffer);
6653 167522 : uint16 lower = ((PageHeader) page)->pd_lower;
6654 167522 : uint16 upper = ((PageHeader) page)->pd_upper;
6655 : uintptr_t dst_offset_in_block;
6656 : RelFileLocator rlocator;
6657 : ForkNumber forkno;
6658 : BlockNumber blkno;
6659 : XLogRecPtr recptr;
6660 :
6661 167522 : xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
6662 167522 : xlrec.dbId = MyDatabaseId;
6663 167522 : xlrec.tsId = MyDatabaseTableSpace;
6664 167522 : xlrec.relcacheInitFileInval = RelcacheInitFileInval;
6665 167522 : xlrec.nmsgs = nmsgs;
6666 :
6667 167522 : XLogBeginInsert();
6668 167522 : XLogRegisterData(&xlrec, MinSizeOfHeapInplace);
6669 167522 : if (nmsgs != 0)
6670 86932 : XLogRegisterData(invalMessages,
6671 : nmsgs * sizeof(SharedInvalidationMessage));
6672 :
6673 : /* register block matching what buffer will look like after changes */
6674 167522 : memcpy(copied_buffer.data, origdata, lower);
6675 167522 : memcpy(copied_buffer.data + upper, origdata + upper, BLCKSZ - upper);
6676 167522 : dst_offset_in_block = dst - origdata;
6677 167522 : memcpy(copied_buffer.data + dst_offset_in_block, src, newlen);
6678 167522 : BufferGetTag(buffer, &rlocator, &forkno, &blkno);
6679 : Assert(forkno == MAIN_FORKNUM);
6680 167522 : XLogRegisterBlock(0, &rlocator, forkno, blkno, copied_buffer.data,
6681 : REGBUF_STANDARD);
6682 167522 : XLogRegisterBufData(0, src, newlen);
6683 :
6684 : /* inplace updates aren't decoded atm, don't log the origin */
6685 :
6686 167522 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_INPLACE);
6687 :
6688 167522 : PageSetLSN(page, recptr);
6689 : }
6690 :
6691 167530 : memcpy(dst, src, newlen);
6692 :
6693 167530 : MarkBufferDirty(buffer);
6694 :
6695 167530 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6696 :
6697 : /*
6698 : * Send invalidations to shared queue. SearchSysCacheLocked1() assumes we
6699 : * do this before UnlockTuple().
6700 : */
6701 167530 : AtInplace_Inval();
6702 :
6703 167530 : MyProc->delayChkptFlags &= ~DELAY_CHKPT_START;
6704 167530 : END_CRIT_SECTION();
6705 167530 : UnlockTuple(relation, &tuple->t_self, InplaceUpdateTupleLock);
6706 :
6707 167530 : AcceptInvalidationMessages(); /* local processing of just-sent inval */
6708 :
6709 : /*
6710 : * Queue a transactional inval, for logical decoding and for third-party
6711 : * code that might have been relying on it since long before inplace
6712 : * update adopted immediate invalidation. See README.tuplock section
6713 : * "Reading inplace-updated columns" for logical decoding details.
6714 : */
6715 167530 : if (!IsBootstrapProcessingMode())
6716 137644 : CacheInvalidateHeapTuple(relation, tuple, NULL);
6717 167530 : }
6718 :
6719 : /*
6720 : * heap_inplace_unlock - reverse of heap_inplace_lock
6721 : */
6722 : void
6723 149810 : heap_inplace_unlock(Relation relation,
6724 : HeapTuple oldtup, Buffer buffer)
6725 : {
6726 149810 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6727 149810 : UnlockTuple(relation, &oldtup->t_self, InplaceUpdateTupleLock);
6728 149810 : ForgetInplace_Inval();
6729 149810 : }
6730 :
6731 : #define FRM_NOOP 0x0001
6732 : #define FRM_INVALIDATE_XMAX 0x0002
6733 : #define FRM_RETURN_IS_XID 0x0004
6734 : #define FRM_RETURN_IS_MULTI 0x0008
6735 : #define FRM_MARK_COMMITTED 0x0010
6736 :
6737 : /*
6738 : * FreezeMultiXactId
6739 : * Determine what to do during freezing when a tuple is marked by a
6740 : * MultiXactId.
6741 : *
6742 : * "flags" is an output value; it's used to tell caller what to do on return.
6743 : * "pagefrz" is an input/output value, used to manage page level freezing.
6744 : *
6745 : * Possible values that we can set in "flags":
6746 : * FRM_NOOP
6747 : * don't do anything -- keep existing Xmax
6748 : * FRM_INVALIDATE_XMAX
6749 : * mark Xmax as InvalidTransactionId and set XMAX_INVALID flag.
6750 : * FRM_RETURN_IS_XID
6751 : * The Xid return value is a single update Xid to set as xmax.
6752 : * FRM_MARK_COMMITTED
6753 : * Xmax can be marked as HEAP_XMAX_COMMITTED
6754 : * FRM_RETURN_IS_MULTI
6755 : * The return value is a new MultiXactId to set as new Xmax.
6756 : * (caller must obtain proper infomask bits using GetMultiXactIdHintBits)
6757 : *
6758 : * Caller delegates control of page freezing to us. In practice we always
6759 : * force freezing of caller's page unless FRM_NOOP processing is indicated.
6760 : * We help caller ensure that XIDs < FreezeLimit and MXIDs < MultiXactCutoff
6761 : * can never be left behind. We freely choose when and how to process each
6762 : * Multi, without ever violating the cutoff postconditions for freezing.
6763 : *
6764 : * It's useful to remove Multis on a proactive timeline (relative to freezing
6765 : * XIDs) to keep MultiXact member SLRU buffer misses to a minimum. It can also
6766 : * be cheaper in the short run, for us, since we too can avoid SLRU buffer
6767 : * misses through eager processing.
6768 : *
6769 : * NB: Creates a _new_ MultiXactId when FRM_RETURN_IS_MULTI is set, though only
6770 : * when FreezeLimit and/or MultiXactCutoff cutoffs leave us with no choice.
6771 : * This can usually be put off, which is usually enough to avoid it altogether.
6772 : * Allocating new multis during VACUUM should be avoided on general principle;
6773 : * only VACUUM can advance relminmxid, so allocating new Multis here comes with
6774 : * its own special risks.
6775 : *
6776 : * NB: Caller must maintain "no freeze" NewRelfrozenXid/NewRelminMxid trackers
6777 : * using heap_tuple_should_freeze when we haven't forced page-level freezing.
6778 : *
6779 : * NB: Caller should avoid needlessly calling heap_tuple_should_freeze when we
6780 : * have already forced page-level freezing, since that might incur the same
6781 : * SLRU buffer misses that we specifically intended to avoid by freezing.
6782 : */
6783 : static TransactionId
6784 12 : FreezeMultiXactId(MultiXactId multi, uint16 t_infomask,
6785 : const struct VacuumCutoffs *cutoffs, uint16 *flags,
6786 : HeapPageFreeze *pagefrz)
6787 : {
6788 : TransactionId newxmax;
6789 : MultiXactMember *members;
6790 : int nmembers;
6791 : bool need_replace;
6792 : int nnewmembers;
6793 : MultiXactMember *newmembers;
6794 : bool has_lockers;
6795 : TransactionId update_xid;
6796 : bool update_committed;
6797 : TransactionId FreezePageRelfrozenXid;
6798 :
6799 12 : *flags = 0;
6800 :
6801 : /* We should only be called in Multis */
6802 : Assert(t_infomask & HEAP_XMAX_IS_MULTI);
6803 :
6804 24 : if (!MultiXactIdIsValid(multi) ||
6805 12 : HEAP_LOCKED_UPGRADED(t_infomask))
6806 : {
6807 0 : *flags |= FRM_INVALIDATE_XMAX;
6808 0 : pagefrz->freeze_required = true;
6809 0 : return InvalidTransactionId;
6810 : }
6811 12 : else if (MultiXactIdPrecedes(multi, cutoffs->relminmxid))
6812 0 : ereport(ERROR,
6813 : (errcode(ERRCODE_DATA_CORRUPTED),
6814 : errmsg_internal("found multixact %u from before relminmxid %u",
6815 : multi, cutoffs->relminmxid)));
6816 12 : else if (MultiXactIdPrecedes(multi, cutoffs->OldestMxact))
6817 : {
6818 : TransactionId update_xact;
6819 :
6820 : /*
6821 : * This old multi cannot possibly have members still running, but
6822 : * verify just in case. If it was a locker only, it can be removed
6823 : * without any further consideration; but if it contained an update,
6824 : * we might need to preserve it.
6825 : */
6826 8 : if (MultiXactIdIsRunning(multi,
6827 8 : HEAP_XMAX_IS_LOCKED_ONLY(t_infomask)))
6828 0 : ereport(ERROR,
6829 : (errcode(ERRCODE_DATA_CORRUPTED),
6830 : errmsg_internal("multixact %u from before multi freeze cutoff %u found to be still running",
6831 : multi, cutoffs->OldestMxact)));
6832 :
6833 8 : if (HEAP_XMAX_IS_LOCKED_ONLY(t_infomask))
6834 : {
6835 8 : *flags |= FRM_INVALIDATE_XMAX;
6836 8 : pagefrz->freeze_required = true;
6837 8 : return InvalidTransactionId;
6838 : }
6839 :
6840 : /* replace multi with single XID for its updater? */
6841 0 : update_xact = MultiXactIdGetUpdateXid(multi, t_infomask);
6842 0 : if (TransactionIdPrecedes(update_xact, cutoffs->relfrozenxid))
6843 0 : ereport(ERROR,
6844 : (errcode(ERRCODE_DATA_CORRUPTED),
6845 : errmsg_internal("multixact %u contains update XID %u from before relfrozenxid %u",
6846 : multi, update_xact,
6847 : cutoffs->relfrozenxid)));
6848 0 : else if (TransactionIdPrecedes(update_xact, cutoffs->OldestXmin))
6849 : {
6850 : /*
6851 : * Updater XID has to have aborted (otherwise the tuple would have
6852 : * been pruned away instead, since updater XID is < OldestXmin).
6853 : * Just remove xmax.
6854 : */
6855 0 : if (TransactionIdDidCommit(update_xact))
6856 0 : ereport(ERROR,
6857 : (errcode(ERRCODE_DATA_CORRUPTED),
6858 : errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
6859 : multi, update_xact,
6860 : cutoffs->OldestXmin)));
6861 0 : *flags |= FRM_INVALIDATE_XMAX;
6862 0 : pagefrz->freeze_required = true;
6863 0 : return InvalidTransactionId;
6864 : }
6865 :
6866 : /* Have to keep updater XID as new xmax */
6867 0 : *flags |= FRM_RETURN_IS_XID;
6868 0 : pagefrz->freeze_required = true;
6869 0 : return update_xact;
6870 : }
6871 :
6872 : /*
6873 : * Some member(s) of this Multi may be below FreezeLimit xid cutoff, so we
6874 : * need to walk the whole members array to figure out what to do, if
6875 : * anything.
6876 : */
6877 : nmembers =
6878 4 : GetMultiXactIdMembers(multi, &members, false,
6879 4 : HEAP_XMAX_IS_LOCKED_ONLY(t_infomask));
6880 4 : if (nmembers <= 0)
6881 : {
6882 : /* Nothing worth keeping */
6883 0 : *flags |= FRM_INVALIDATE_XMAX;
6884 0 : pagefrz->freeze_required = true;
6885 0 : return InvalidTransactionId;
6886 : }
6887 :
6888 : /*
6889 : * The FRM_NOOP case is the only case where we might need to ratchet back
6890 : * FreezePageRelfrozenXid or FreezePageRelminMxid. It is also the only
6891 : * case where our caller might ratchet back its NoFreezePageRelfrozenXid
6892 : * or NoFreezePageRelminMxid "no freeze" trackers to deal with a multi.
6893 : * FRM_NOOP handling should result in the NewRelfrozenXid/NewRelminMxid
6894 : * trackers managed by VACUUM being ratcheting back by xmax to the degree
6895 : * required to make it safe to leave xmax undisturbed, independent of
6896 : * whether or not page freezing is triggered somewhere else.
6897 : *
6898 : * Our policy is to force freezing in every case other than FRM_NOOP,
6899 : * which obviates the need to maintain either set of trackers, anywhere.
6900 : * Every other case will reliably execute a freeze plan for xmax that
6901 : * either replaces xmax with an XID/MXID >= OldestXmin/OldestMxact, or
6902 : * sets xmax to an InvalidTransactionId XID, rendering xmax fully frozen.
6903 : * (VACUUM's NewRelfrozenXid/NewRelminMxid trackers are initialized with
6904 : * OldestXmin/OldestMxact, so later values never need to be tracked here.)
6905 : */
6906 4 : need_replace = false;
6907 4 : FreezePageRelfrozenXid = pagefrz->FreezePageRelfrozenXid;
6908 8 : for (int i = 0; i < nmembers; i++)
6909 : {
6910 6 : TransactionId xid = members[i].xid;
6911 :
6912 : Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
6913 :
6914 6 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
6915 : {
6916 : /* Can't violate the FreezeLimit postcondition */
6917 2 : need_replace = true;
6918 2 : break;
6919 : }
6920 4 : if (TransactionIdPrecedes(xid, FreezePageRelfrozenXid))
6921 0 : FreezePageRelfrozenXid = xid;
6922 : }
6923 :
6924 : /* Can't violate the MultiXactCutoff postcondition, either */
6925 4 : if (!need_replace)
6926 2 : need_replace = MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff);
6927 :
6928 4 : if (!need_replace)
6929 : {
6930 : /*
6931 : * vacuumlazy.c might ratchet back NewRelminMxid, NewRelfrozenXid, or
6932 : * both together to make it safe to retain this particular multi after
6933 : * freezing its page
6934 : */
6935 2 : *flags |= FRM_NOOP;
6936 2 : pagefrz->FreezePageRelfrozenXid = FreezePageRelfrozenXid;
6937 2 : if (MultiXactIdPrecedes(multi, pagefrz->FreezePageRelminMxid))
6938 0 : pagefrz->FreezePageRelminMxid = multi;
6939 2 : pfree(members);
6940 2 : return multi;
6941 : }
6942 :
6943 : /*
6944 : * Do a more thorough second pass over the multi to figure out which
6945 : * member XIDs actually need to be kept. Checking the precise status of
6946 : * individual members might even show that we don't need to keep anything.
6947 : * That is quite possible even though the Multi must be >= OldestMxact,
6948 : * since our second pass only keeps member XIDs when it's truly necessary;
6949 : * even member XIDs >= OldestXmin often won't be kept by second pass.
6950 : */
6951 2 : nnewmembers = 0;
6952 2 : newmembers = palloc_array(MultiXactMember, nmembers);
6953 2 : has_lockers = false;
6954 2 : update_xid = InvalidTransactionId;
6955 2 : update_committed = false;
6956 :
6957 : /*
6958 : * Determine whether to keep each member xid, or to ignore it instead
6959 : */
6960 6 : for (int i = 0; i < nmembers; i++)
6961 : {
6962 4 : TransactionId xid = members[i].xid;
6963 4 : MultiXactStatus mstatus = members[i].status;
6964 :
6965 : Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
6966 :
6967 4 : if (!ISUPDATE_from_mxstatus(mstatus))
6968 : {
6969 : /*
6970 : * Locker XID (not updater XID). We only keep lockers that are
6971 : * still running.
6972 : */
6973 8 : if (TransactionIdIsCurrentTransactionId(xid) ||
6974 4 : TransactionIdIsInProgress(xid))
6975 : {
6976 2 : if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
6977 0 : ereport(ERROR,
6978 : (errcode(ERRCODE_DATA_CORRUPTED),
6979 : errmsg_internal("multixact %u contains running locker XID %u from before removable cutoff %u",
6980 : multi, xid,
6981 : cutoffs->OldestXmin)));
6982 2 : newmembers[nnewmembers++] = members[i];
6983 2 : has_lockers = true;
6984 : }
6985 :
6986 4 : continue;
6987 : }
6988 :
6989 : /*
6990 : * Updater XID (not locker XID). Should we keep it?
6991 : *
6992 : * Since the tuple wasn't totally removed when vacuum pruned, the
6993 : * update Xid cannot possibly be older than OldestXmin cutoff unless
6994 : * the updater XID aborted. If the updater transaction is known
6995 : * aborted or crashed then it's okay to ignore it, otherwise not.
6996 : *
6997 : * In any case the Multi should never contain two updaters, whatever
6998 : * their individual commit status. Check for that first, in passing.
6999 : */
7000 0 : if (TransactionIdIsValid(update_xid))
7001 0 : ereport(ERROR,
7002 : (errcode(ERRCODE_DATA_CORRUPTED),
7003 : errmsg_internal("multixact %u has two or more updating members",
7004 : multi),
7005 : errdetail_internal("First updater XID=%u second updater XID=%u.",
7006 : update_xid, xid)));
7007 :
7008 : /*
7009 : * As with all tuple visibility routines, it's critical to test
7010 : * TransactionIdIsInProgress before TransactionIdDidCommit, because of
7011 : * race conditions explained in detail in heapam_visibility.c.
7012 : */
7013 0 : if (TransactionIdIsCurrentTransactionId(xid) ||
7014 0 : TransactionIdIsInProgress(xid))
7015 0 : update_xid = xid;
7016 0 : else if (TransactionIdDidCommit(xid))
7017 : {
7018 : /*
7019 : * The transaction committed, so we can tell caller to set
7020 : * HEAP_XMAX_COMMITTED. (We can only do this because we know the
7021 : * transaction is not running.)
7022 : */
7023 0 : update_committed = true;
7024 0 : update_xid = xid;
7025 : }
7026 : else
7027 : {
7028 : /*
7029 : * Not in progress, not committed -- must be aborted or crashed;
7030 : * we can ignore it.
7031 : */
7032 0 : continue;
7033 : }
7034 :
7035 : /*
7036 : * We determined that updater must be kept -- add it to pending new
7037 : * members list
7038 : */
7039 0 : if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
7040 0 : ereport(ERROR,
7041 : (errcode(ERRCODE_DATA_CORRUPTED),
7042 : errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
7043 : multi, xid, cutoffs->OldestXmin)));
7044 0 : newmembers[nnewmembers++] = members[i];
7045 : }
7046 :
7047 2 : pfree(members);
7048 :
7049 : /*
7050 : * Determine what to do with caller's multi based on information gathered
7051 : * during our second pass
7052 : */
7053 2 : if (nnewmembers == 0)
7054 : {
7055 : /* Nothing worth keeping */
7056 0 : *flags |= FRM_INVALIDATE_XMAX;
7057 0 : newxmax = InvalidTransactionId;
7058 : }
7059 2 : else if (TransactionIdIsValid(update_xid) && !has_lockers)
7060 : {
7061 : /*
7062 : * If there's a single member and it's an update, pass it back alone
7063 : * without creating a new Multi. (XXX we could do this when there's a
7064 : * single remaining locker, too, but that would complicate the API too
7065 : * much; moreover, the case with the single updater is more
7066 : * interesting, because those are longer-lived.)
7067 : */
7068 : Assert(nnewmembers == 1);
7069 0 : *flags |= FRM_RETURN_IS_XID;
7070 0 : if (update_committed)
7071 0 : *flags |= FRM_MARK_COMMITTED;
7072 0 : newxmax = update_xid;
7073 : }
7074 : else
7075 : {
7076 : /*
7077 : * Create a new multixact with the surviving members of the previous
7078 : * one, to set as new Xmax in the tuple
7079 : */
7080 2 : newxmax = MultiXactIdCreateFromMembers(nnewmembers, newmembers);
7081 2 : *flags |= FRM_RETURN_IS_MULTI;
7082 : }
7083 :
7084 2 : pfree(newmembers);
7085 :
7086 2 : pagefrz->freeze_required = true;
7087 2 : return newxmax;
7088 : }
7089 :
7090 : /*
7091 : * heap_prepare_freeze_tuple
7092 : *
7093 : * Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
7094 : * are older than the OldestXmin and/or OldestMxact freeze cutoffs. If so,
7095 : * setup enough state (in the *frz output argument) to enable caller to
7096 : * process this tuple as part of freezing its page, and return true. Return
7097 : * false if nothing can be changed about the tuple right now.
7098 : *
7099 : * Also sets *totally_frozen to true if the tuple will be totally frozen once
7100 : * caller executes returned freeze plan (or if the tuple was already totally
7101 : * frozen by an earlier VACUUM). This indicates that there are no remaining
7102 : * XIDs or MultiXactIds that will need to be processed by a future VACUUM.
7103 : *
7104 : * VACUUM caller must assemble HeapTupleFreeze freeze plan entries for every
7105 : * tuple that we returned true for, and then execute freezing. Caller must
7106 : * initialize pagefrz fields for page as a whole before first call here for
7107 : * each heap page.
7108 : *
7109 : * VACUUM caller decides on whether or not to freeze the page as a whole.
7110 : * We'll often prepare freeze plans for a page that caller just discards.
7111 : * However, VACUUM doesn't always get to make a choice; it must freeze when
7112 : * pagefrz.freeze_required is set, to ensure that any XIDs < FreezeLimit (and
7113 : * MXIDs < MultiXactCutoff) can never be left behind. We help to make sure
7114 : * that VACUUM always follows that rule.
7115 : *
7116 : * We sometimes force freezing of xmax MultiXactId values long before it is
7117 : * strictly necessary to do so just to ensure the FreezeLimit postcondition.
7118 : * It's worth processing MultiXactIds proactively when it is cheap to do so,
7119 : * and it's convenient to make that happen by piggy-backing it on the "force
7120 : * freezing" mechanism. Conversely, we sometimes delay freezing MultiXactIds
7121 : * because it is expensive right now (though only when it's still possible to
7122 : * do so without violating the FreezeLimit/MultiXactCutoff postcondition).
7123 : *
7124 : * It is assumed that the caller has checked the tuple with
7125 : * HeapTupleSatisfiesVacuum() and determined that it is not HEAPTUPLE_DEAD
7126 : * (else we should be removing the tuple, not freezing it).
7127 : *
7128 : * NB: This function has side effects: it might allocate a new MultiXactId.
7129 : * It will be set as tuple's new xmax when our *frz output is processed within
7130 : * heap_execute_freeze_tuple later on. If the tuple is in a shared buffer
7131 : * then caller had better have an exclusive lock on it already.
7132 : */
7133 : bool
7134 28601634 : heap_prepare_freeze_tuple(HeapTupleHeader tuple,
7135 : const struct VacuumCutoffs *cutoffs,
7136 : HeapPageFreeze *pagefrz,
7137 : HeapTupleFreeze *frz, bool *totally_frozen)
7138 : {
7139 28601634 : bool xmin_already_frozen = false,
7140 28601634 : xmax_already_frozen = false;
7141 28601634 : bool freeze_xmin = false,
7142 28601634 : replace_xvac = false,
7143 28601634 : replace_xmax = false,
7144 28601634 : freeze_xmax = false;
7145 : TransactionId xid;
7146 :
7147 28601634 : frz->xmax = HeapTupleHeaderGetRawXmax(tuple);
7148 28601634 : frz->t_infomask2 = tuple->t_infomask2;
7149 28601634 : frz->t_infomask = tuple->t_infomask;
7150 28601634 : frz->frzflags = 0;
7151 28601634 : frz->checkflags = 0;
7152 :
7153 : /*
7154 : * Process xmin, while keeping track of whether it's already frozen, or
7155 : * will become frozen iff our freeze plan is executed by caller (could be
7156 : * neither).
7157 : */
7158 28601634 : xid = HeapTupleHeaderGetXmin(tuple);
7159 28601634 : if (!TransactionIdIsNormal(xid))
7160 22750320 : xmin_already_frozen = true;
7161 : else
7162 : {
7163 5851314 : if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
7164 0 : ereport(ERROR,
7165 : (errcode(ERRCODE_DATA_CORRUPTED),
7166 : errmsg_internal("found xmin %u from before relfrozenxid %u",
7167 : xid, cutoffs->relfrozenxid)));
7168 :
7169 : /* Will set freeze_xmin flags in freeze plan below */
7170 5851314 : freeze_xmin = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
7171 :
7172 : /* Verify that xmin committed if and when freeze plan is executed */
7173 5851314 : if (freeze_xmin)
7174 4602988 : frz->checkflags |= HEAP_FREEZE_CHECK_XMIN_COMMITTED;
7175 : }
7176 :
7177 : /*
7178 : * Old-style VACUUM FULL is gone, but we have to process xvac for as long
7179 : * as we support having MOVED_OFF/MOVED_IN tuples in the database
7180 : */
7181 28601634 : xid = HeapTupleHeaderGetXvac(tuple);
7182 28601634 : if (TransactionIdIsNormal(xid))
7183 : {
7184 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7185 : Assert(TransactionIdPrecedes(xid, cutoffs->OldestXmin));
7186 :
7187 : /*
7188 : * For Xvac, we always freeze proactively. This allows totally_frozen
7189 : * tracking to ignore xvac.
7190 : */
7191 0 : replace_xvac = pagefrz->freeze_required = true;
7192 :
7193 : /* Will set replace_xvac flags in freeze plan below */
7194 : }
7195 :
7196 : /* Now process xmax */
7197 28601634 : xid = frz->xmax;
7198 28601634 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7199 : {
7200 : /* Raw xmax is a MultiXactId */
7201 : TransactionId newxmax;
7202 : uint16 flags;
7203 :
7204 : /*
7205 : * We will either remove xmax completely (in the "freeze_xmax" path),
7206 : * process xmax by replacing it (in the "replace_xmax" path), or
7207 : * perform no-op xmax processing. The only constraint is that the
7208 : * FreezeLimit/MultiXactCutoff postcondition must never be violated.
7209 : */
7210 12 : newxmax = FreezeMultiXactId(xid, tuple->t_infomask, cutoffs,
7211 : &flags, pagefrz);
7212 :
7213 12 : if (flags & FRM_NOOP)
7214 : {
7215 : /*
7216 : * xmax is a MultiXactId, and nothing about it changes for now.
7217 : * This is the only case where 'freeze_required' won't have been
7218 : * set for us by FreezeMultiXactId, as well as the only case where
7219 : * neither freeze_xmax nor replace_xmax are set (given a multi).
7220 : *
7221 : * This is a no-op, but the call to FreezeMultiXactId might have
7222 : * ratcheted back NewRelfrozenXid and/or NewRelminMxid trackers
7223 : * for us (the "freeze page" variants, specifically). That'll
7224 : * make it safe for our caller to freeze the page later on, while
7225 : * leaving this particular xmax undisturbed.
7226 : *
7227 : * FreezeMultiXactId is _not_ responsible for the "no freeze"
7228 : * NewRelfrozenXid/NewRelminMxid trackers, though -- that's our
7229 : * job. A call to heap_tuple_should_freeze for this same tuple
7230 : * will take place below if 'freeze_required' isn't set already.
7231 : * (This repeats work from FreezeMultiXactId, but allows "no
7232 : * freeze" tracker maintenance to happen in only one place.)
7233 : */
7234 : Assert(!MultiXactIdPrecedes(newxmax, cutoffs->MultiXactCutoff));
7235 : Assert(MultiXactIdIsValid(newxmax) && xid == newxmax);
7236 : }
7237 10 : else if (flags & FRM_RETURN_IS_XID)
7238 : {
7239 : /*
7240 : * xmax will become an updater Xid (original MultiXact's updater
7241 : * member Xid will be carried forward as a simple Xid in Xmax).
7242 : */
7243 : Assert(!TransactionIdPrecedes(newxmax, cutoffs->OldestXmin));
7244 :
7245 : /*
7246 : * NB -- some of these transformations are only valid because we
7247 : * know the return Xid is a tuple updater (i.e. not merely a
7248 : * locker.) Also note that the only reason we don't explicitly
7249 : * worry about HEAP_KEYS_UPDATED is because it lives in
7250 : * t_infomask2 rather than t_infomask.
7251 : */
7252 0 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7253 0 : frz->xmax = newxmax;
7254 0 : if (flags & FRM_MARK_COMMITTED)
7255 0 : frz->t_infomask |= HEAP_XMAX_COMMITTED;
7256 0 : replace_xmax = true;
7257 : }
7258 10 : else if (flags & FRM_RETURN_IS_MULTI)
7259 : {
7260 : uint16 newbits;
7261 : uint16 newbits2;
7262 :
7263 : /*
7264 : * xmax is an old MultiXactId that we have to replace with a new
7265 : * MultiXactId, to carry forward two or more original member XIDs.
7266 : */
7267 : Assert(!MultiXactIdPrecedes(newxmax, cutoffs->OldestMxact));
7268 :
7269 : /*
7270 : * We can't use GetMultiXactIdHintBits directly on the new multi
7271 : * here; that routine initializes the masks to all zeroes, which
7272 : * would lose other bits we need. Doing it this way ensures all
7273 : * unrelated bits remain untouched.
7274 : */
7275 2 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7276 2 : frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
7277 2 : GetMultiXactIdHintBits(newxmax, &newbits, &newbits2);
7278 2 : frz->t_infomask |= newbits;
7279 2 : frz->t_infomask2 |= newbits2;
7280 2 : frz->xmax = newxmax;
7281 2 : replace_xmax = true;
7282 : }
7283 : else
7284 : {
7285 : /*
7286 : * Freeze plan for tuple "freezes xmax" in the strictest sense:
7287 : * it'll leave nothing in xmax (neither an Xid nor a MultiXactId).
7288 : */
7289 : Assert(flags & FRM_INVALIDATE_XMAX);
7290 : Assert(!TransactionIdIsValid(newxmax));
7291 :
7292 : /* Will set freeze_xmax flags in freeze plan below */
7293 8 : freeze_xmax = true;
7294 : }
7295 :
7296 : /* MultiXactId processing forces freezing (barring FRM_NOOP case) */
7297 : Assert(pagefrz->freeze_required || (!freeze_xmax && !replace_xmax));
7298 : }
7299 28601622 : else if (TransactionIdIsNormal(xid))
7300 : {
7301 : /* Raw xmax is normal XID */
7302 6634342 : if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
7303 0 : ereport(ERROR,
7304 : (errcode(ERRCODE_DATA_CORRUPTED),
7305 : errmsg_internal("found xmax %u from before relfrozenxid %u",
7306 : xid, cutoffs->relfrozenxid)));
7307 :
7308 : /* Will set freeze_xmax flags in freeze plan below */
7309 6634342 : freeze_xmax = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
7310 :
7311 : /*
7312 : * Verify that xmax aborted if and when freeze plan is executed,
7313 : * provided it's from an update. (A lock-only xmax can be removed
7314 : * independent of this, since the lock is released at xact end.)
7315 : */
7316 6634342 : if (freeze_xmax && !HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask))
7317 2012 : frz->checkflags |= HEAP_FREEZE_CHECK_XMAX_ABORTED;
7318 : }
7319 21967280 : else if (!TransactionIdIsValid(xid))
7320 : {
7321 : /* Raw xmax is InvalidTransactionId XID */
7322 : Assert((tuple->t_infomask & HEAP_XMAX_IS_MULTI) == 0);
7323 21967280 : xmax_already_frozen = true;
7324 : }
7325 : else
7326 0 : ereport(ERROR,
7327 : (errcode(ERRCODE_DATA_CORRUPTED),
7328 : errmsg_internal("found raw xmax %u (infomask 0x%04x) not invalid and not multi",
7329 : xid, tuple->t_infomask)));
7330 :
7331 28601634 : if (freeze_xmin)
7332 : {
7333 : Assert(!xmin_already_frozen);
7334 :
7335 4602988 : frz->t_infomask |= HEAP_XMIN_FROZEN;
7336 : }
7337 28601634 : if (replace_xvac)
7338 : {
7339 : /*
7340 : * If a MOVED_OFF tuple is not dead, the xvac transaction must have
7341 : * failed; whereas a non-dead MOVED_IN tuple must mean the xvac
7342 : * transaction succeeded.
7343 : */
7344 : Assert(pagefrz->freeze_required);
7345 0 : if (tuple->t_infomask & HEAP_MOVED_OFF)
7346 0 : frz->frzflags |= XLH_INVALID_XVAC;
7347 : else
7348 0 : frz->frzflags |= XLH_FREEZE_XVAC;
7349 : }
7350 : if (replace_xmax)
7351 : {
7352 : Assert(!xmax_already_frozen && !freeze_xmax);
7353 : Assert(pagefrz->freeze_required);
7354 :
7355 : /* Already set replace_xmax flags in freeze plan earlier */
7356 : }
7357 28601634 : if (freeze_xmax)
7358 : {
7359 : Assert(!xmax_already_frozen && !replace_xmax);
7360 :
7361 3964 : frz->xmax = InvalidTransactionId;
7362 :
7363 : /*
7364 : * The tuple might be marked either XMAX_INVALID or XMAX_COMMITTED +
7365 : * LOCKED. Normalize to INVALID just to be sure no one gets confused.
7366 : * Also get rid of the HEAP_KEYS_UPDATED bit.
7367 : */
7368 3964 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7369 3964 : frz->t_infomask |= HEAP_XMAX_INVALID;
7370 3964 : frz->t_infomask2 &= ~HEAP_HOT_UPDATED;
7371 3964 : frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
7372 : }
7373 :
7374 : /*
7375 : * Determine if this tuple is already totally frozen, or will become
7376 : * totally frozen (provided caller executes freeze plans for the page)
7377 : */
7378 55950978 : *totally_frozen = ((freeze_xmin || xmin_already_frozen) &&
7379 27349344 : (freeze_xmax || xmax_already_frozen));
7380 :
7381 28601634 : if (!pagefrz->freeze_required && !(xmin_already_frozen &&
7382 : xmax_already_frozen))
7383 : {
7384 : /*
7385 : * So far no previous tuple from the page made freezing mandatory.
7386 : * Does this tuple force caller to freeze the entire page?
7387 : */
7388 9454150 : pagefrz->freeze_required =
7389 9454150 : heap_tuple_should_freeze(tuple, cutoffs,
7390 : &pagefrz->NoFreezePageRelfrozenXid,
7391 : &pagefrz->NoFreezePageRelminMxid);
7392 : }
7393 :
7394 : /* Tell caller if this tuple has a usable freeze plan set in *frz */
7395 28601634 : return freeze_xmin || replace_xvac || replace_xmax || freeze_xmax;
7396 : }
7397 :
7398 : /*
7399 : * Perform xmin/xmax XID status sanity checks before actually executing freeze
7400 : * plans.
7401 : *
7402 : * heap_prepare_freeze_tuple doesn't perform these checks directly because
7403 : * pg_xact lookups are relatively expensive. They shouldn't be repeated by
7404 : * successive VACUUMs that each decide against freezing the same page.
7405 : */
7406 : void
7407 45544 : heap_pre_freeze_checks(Buffer buffer,
7408 : HeapTupleFreeze *tuples, int ntuples)
7409 : {
7410 45544 : Page page = BufferGetPage(buffer);
7411 :
7412 1962796 : for (int i = 0; i < ntuples; i++)
7413 : {
7414 1917252 : HeapTupleFreeze *frz = tuples + i;
7415 1917252 : ItemId itemid = PageGetItemId(page, frz->offset);
7416 : HeapTupleHeader htup;
7417 :
7418 1917252 : htup = (HeapTupleHeader) PageGetItem(page, itemid);
7419 :
7420 : /* Deliberately avoid relying on tuple hint bits here */
7421 1917252 : if (frz->checkflags & HEAP_FREEZE_CHECK_XMIN_COMMITTED)
7422 : {
7423 1917250 : TransactionId xmin = HeapTupleHeaderGetRawXmin(htup);
7424 :
7425 : Assert(!HeapTupleHeaderXminFrozen(htup));
7426 1917250 : if (unlikely(!TransactionIdDidCommit(xmin)))
7427 0 : ereport(ERROR,
7428 : (errcode(ERRCODE_DATA_CORRUPTED),
7429 : errmsg_internal("uncommitted xmin %u needs to be frozen",
7430 : xmin)));
7431 : }
7432 :
7433 : /*
7434 : * TransactionIdDidAbort won't work reliably in the presence of XIDs
7435 : * left behind by transactions that were in progress during a crash,
7436 : * so we can only check that xmax didn't commit
7437 : */
7438 1917252 : if (frz->checkflags & HEAP_FREEZE_CHECK_XMAX_ABORTED)
7439 : {
7440 630 : TransactionId xmax = HeapTupleHeaderGetRawXmax(htup);
7441 :
7442 : Assert(TransactionIdIsNormal(xmax));
7443 630 : if (unlikely(TransactionIdDidCommit(xmax)))
7444 0 : ereport(ERROR,
7445 : (errcode(ERRCODE_DATA_CORRUPTED),
7446 : errmsg_internal("cannot freeze committed xmax %u",
7447 : xmax)));
7448 : }
7449 : }
7450 45544 : }
7451 :
7452 : /*
7453 : * Helper which executes freezing of one or more heap tuples on a page on
7454 : * behalf of caller. Caller passes an array of tuple plans from
7455 : * heap_prepare_freeze_tuple. Caller must set 'offset' in each plan for us.
7456 : * Must be called in a critical section that also marks the buffer dirty and,
7457 : * if needed, emits WAL.
7458 : */
7459 : void
7460 45544 : heap_freeze_prepared_tuples(Buffer buffer, HeapTupleFreeze *tuples, int ntuples)
7461 : {
7462 45544 : Page page = BufferGetPage(buffer);
7463 :
7464 1962796 : for (int i = 0; i < ntuples; i++)
7465 : {
7466 1917252 : HeapTupleFreeze *frz = tuples + i;
7467 1917252 : ItemId itemid = PageGetItemId(page, frz->offset);
7468 : HeapTupleHeader htup;
7469 :
7470 1917252 : htup = (HeapTupleHeader) PageGetItem(page, itemid);
7471 1917252 : heap_execute_freeze_tuple(htup, frz);
7472 : }
7473 45544 : }
7474 :
7475 : /*
7476 : * heap_freeze_tuple
7477 : * Freeze tuple in place, without WAL logging.
7478 : *
7479 : * Useful for callers like CLUSTER that perform their own WAL logging.
7480 : */
7481 : bool
7482 736804 : heap_freeze_tuple(HeapTupleHeader tuple,
7483 : TransactionId relfrozenxid, TransactionId relminmxid,
7484 : TransactionId FreezeLimit, TransactionId MultiXactCutoff)
7485 : {
7486 : HeapTupleFreeze frz;
7487 : bool do_freeze;
7488 : bool totally_frozen;
7489 : struct VacuumCutoffs cutoffs;
7490 : HeapPageFreeze pagefrz;
7491 :
7492 736804 : cutoffs.relfrozenxid = relfrozenxid;
7493 736804 : cutoffs.relminmxid = relminmxid;
7494 736804 : cutoffs.OldestXmin = FreezeLimit;
7495 736804 : cutoffs.OldestMxact = MultiXactCutoff;
7496 736804 : cutoffs.FreezeLimit = FreezeLimit;
7497 736804 : cutoffs.MultiXactCutoff = MultiXactCutoff;
7498 :
7499 736804 : pagefrz.freeze_required = true;
7500 736804 : pagefrz.FreezePageRelfrozenXid = FreezeLimit;
7501 736804 : pagefrz.FreezePageRelminMxid = MultiXactCutoff;
7502 736804 : pagefrz.NoFreezePageRelfrozenXid = FreezeLimit;
7503 736804 : pagefrz.NoFreezePageRelminMxid = MultiXactCutoff;
7504 :
7505 736804 : do_freeze = heap_prepare_freeze_tuple(tuple, &cutoffs,
7506 : &pagefrz, &frz, &totally_frozen);
7507 :
7508 : /*
7509 : * Note that because this is not a WAL-logged operation, we don't need to
7510 : * fill in the offset in the freeze record.
7511 : */
7512 :
7513 736804 : if (do_freeze)
7514 506538 : heap_execute_freeze_tuple(tuple, &frz);
7515 736804 : return do_freeze;
7516 : }
7517 :
7518 : /*
7519 : * For a given MultiXactId, return the hint bits that should be set in the
7520 : * tuple's infomask.
7521 : *
7522 : * Normally this should be called for a multixact that was just created, and
7523 : * so is on our local cache, so the GetMembers call is fast.
7524 : */
7525 : static void
7526 153602 : GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
7527 : uint16 *new_infomask2)
7528 : {
7529 : int nmembers;
7530 : MultiXactMember *members;
7531 : int i;
7532 153602 : uint16 bits = HEAP_XMAX_IS_MULTI;
7533 153602 : uint16 bits2 = 0;
7534 153602 : bool has_update = false;
7535 153602 : LockTupleMode strongest = LockTupleKeyShare;
7536 :
7537 : /*
7538 : * We only use this in multis we just created, so they cannot be values
7539 : * pre-pg_upgrade.
7540 : */
7541 153602 : nmembers = GetMultiXactIdMembers(multi, &members, false, false);
7542 :
7543 2945222 : for (i = 0; i < nmembers; i++)
7544 : {
7545 : LockTupleMode mode;
7546 :
7547 : /*
7548 : * Remember the strongest lock mode held by any member of the
7549 : * multixact.
7550 : */
7551 2791620 : mode = TUPLOCK_from_mxstatus(members[i].status);
7552 2791620 : if (mode > strongest)
7553 5776 : strongest = mode;
7554 :
7555 : /* See what other bits we need */
7556 2791620 : switch (members[i].status)
7557 : {
7558 2786784 : case MultiXactStatusForKeyShare:
7559 : case MultiXactStatusForShare:
7560 : case MultiXactStatusForNoKeyUpdate:
7561 2786784 : break;
7562 :
7563 104 : case MultiXactStatusForUpdate:
7564 104 : bits2 |= HEAP_KEYS_UPDATED;
7565 104 : break;
7566 :
7567 4712 : case MultiXactStatusNoKeyUpdate:
7568 4712 : has_update = true;
7569 4712 : break;
7570 :
7571 20 : case MultiXactStatusUpdate:
7572 20 : bits2 |= HEAP_KEYS_UPDATED;
7573 20 : has_update = true;
7574 20 : break;
7575 : }
7576 : }
7577 :
7578 153602 : if (strongest == LockTupleExclusive ||
7579 : strongest == LockTupleNoKeyExclusive)
7580 4892 : bits |= HEAP_XMAX_EXCL_LOCK;
7581 148710 : else if (strongest == LockTupleShare)
7582 878 : bits |= HEAP_XMAX_SHR_LOCK;
7583 147832 : else if (strongest == LockTupleKeyShare)
7584 147832 : bits |= HEAP_XMAX_KEYSHR_LOCK;
7585 :
7586 153602 : if (!has_update)
7587 148870 : bits |= HEAP_XMAX_LOCK_ONLY;
7588 :
7589 153602 : if (nmembers > 0)
7590 153602 : pfree(members);
7591 :
7592 153602 : *new_infomask = bits;
7593 153602 : *new_infomask2 = bits2;
7594 153602 : }
7595 :
7596 : /*
7597 : * MultiXactIdGetUpdateXid
7598 : *
7599 : * Given a multixact Xmax and corresponding infomask, which does not have the
7600 : * HEAP_XMAX_LOCK_ONLY bit set, obtain and return the Xid of the updating
7601 : * transaction.
7602 : *
7603 : * Caller is expected to check the status of the updating transaction, if
7604 : * necessary.
7605 : */
7606 : static TransactionId
7607 324038 : MultiXactIdGetUpdateXid(TransactionId xmax, uint16 t_infomask)
7608 : {
7609 324038 : TransactionId update_xact = InvalidTransactionId;
7610 : MultiXactMember *members;
7611 : int nmembers;
7612 :
7613 : Assert(!(t_infomask & HEAP_XMAX_LOCK_ONLY));
7614 : Assert(t_infomask & HEAP_XMAX_IS_MULTI);
7615 :
7616 : /*
7617 : * Since we know the LOCK_ONLY bit is not set, this cannot be a multi from
7618 : * pre-pg_upgrade.
7619 : */
7620 324038 : nmembers = GetMultiXactIdMembers(xmax, &members, false, false);
7621 :
7622 324038 : if (nmembers > 0)
7623 : {
7624 : int i;
7625 :
7626 491560 : for (i = 0; i < nmembers; i++)
7627 : {
7628 : /* Ignore lockers */
7629 491560 : if (!ISUPDATE_from_mxstatus(members[i].status))
7630 167522 : continue;
7631 :
7632 : /* there can be at most one updater */
7633 : Assert(update_xact == InvalidTransactionId);
7634 324038 : update_xact = members[i].xid;
7635 : #ifndef USE_ASSERT_CHECKING
7636 :
7637 : /*
7638 : * in an assert-enabled build, walk the whole array to ensure
7639 : * there's no other updater.
7640 : */
7641 324038 : break;
7642 : #endif
7643 : }
7644 :
7645 324038 : pfree(members);
7646 : }
7647 :
7648 324038 : return update_xact;
7649 : }
7650 :
7651 : /*
7652 : * HeapTupleGetUpdateXid
7653 : * As above, but use a HeapTupleHeader
7654 : *
7655 : * See also HeapTupleHeaderGetUpdateXid, which can be used without previously
7656 : * checking the hint bits.
7657 : */
7658 : TransactionId
7659 319762 : HeapTupleGetUpdateXid(const HeapTupleHeaderData *tup)
7660 : {
7661 319762 : return MultiXactIdGetUpdateXid(HeapTupleHeaderGetRawXmax(tup),
7662 319762 : tup->t_infomask);
7663 : }
7664 :
7665 : /*
7666 : * Does the given multixact conflict with the current transaction grabbing a
7667 : * tuple lock of the given strength?
7668 : *
7669 : * The passed infomask pairs up with the given multixact in the tuple header.
7670 : *
7671 : * If current_is_member is not NULL, it is set to 'true' if the current
7672 : * transaction is a member of the given multixact.
7673 : */
7674 : static bool
7675 436 : DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
7676 : LockTupleMode lockmode, bool *current_is_member)
7677 : {
7678 : int nmembers;
7679 : MultiXactMember *members;
7680 436 : bool result = false;
7681 436 : LOCKMODE wanted = tupleLockExtraInfo[lockmode].hwlock;
7682 :
7683 436 : if (HEAP_LOCKED_UPGRADED(infomask))
7684 0 : return false;
7685 :
7686 436 : nmembers = GetMultiXactIdMembers(multi, &members, false,
7687 436 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
7688 436 : if (nmembers >= 0)
7689 : {
7690 : int i;
7691 :
7692 5364 : for (i = 0; i < nmembers; i++)
7693 : {
7694 : TransactionId memxid;
7695 : LOCKMODE memlockmode;
7696 :
7697 4942 : if (result && (current_is_member == NULL || *current_is_member))
7698 : break;
7699 :
7700 4928 : memlockmode = LOCKMODE_from_mxstatus(members[i].status);
7701 :
7702 : /* ignore members from current xact (but track their presence) */
7703 4928 : memxid = members[i].xid;
7704 4928 : if (TransactionIdIsCurrentTransactionId(memxid))
7705 : {
7706 184 : if (current_is_member != NULL)
7707 156 : *current_is_member = true;
7708 184 : continue;
7709 : }
7710 4744 : else if (result)
7711 16 : continue;
7712 :
7713 : /* ignore members that don't conflict with the lock we want */
7714 4728 : if (!DoLockModesConflict(memlockmode, wanted))
7715 4650 : continue;
7716 :
7717 78 : if (ISUPDATE_from_mxstatus(members[i].status))
7718 : {
7719 : /* ignore aborted updaters */
7720 34 : if (TransactionIdDidAbort(memxid))
7721 2 : continue;
7722 : }
7723 : else
7724 : {
7725 : /* ignore lockers-only that are no longer in progress */
7726 44 : if (!TransactionIdIsInProgress(memxid))
7727 14 : continue;
7728 : }
7729 :
7730 : /*
7731 : * Whatever remains are either live lockers that conflict with our
7732 : * wanted lock, and updaters that are not aborted. Those conflict
7733 : * with what we want. Set up to return true, but keep going to
7734 : * look for the current transaction among the multixact members,
7735 : * if needed.
7736 : */
7737 62 : result = true;
7738 : }
7739 436 : pfree(members);
7740 : }
7741 :
7742 436 : return result;
7743 : }
7744 :
7745 : /*
7746 : * Do_MultiXactIdWait
7747 : * Actual implementation for the two functions below.
7748 : *
7749 : * 'multi', 'status' and 'infomask' indicate what to sleep on (the status is
7750 : * needed to ensure we only sleep on conflicting members, and the infomask is
7751 : * used to optimize multixact access in case it's a lock-only multi); 'nowait'
7752 : * indicates whether to use conditional lock acquisition, to allow callers to
7753 : * fail if lock is unavailable. 'rel', 'ctid' and 'oper' are used to set up
7754 : * context information for error messages. 'remaining', if not NULL, receives
7755 : * the number of members that are still running, including any (non-aborted)
7756 : * subtransactions of our own transaction. 'logLockFailure' indicates whether
7757 : * to log details when a lock acquisition fails with 'nowait' enabled.
7758 : *
7759 : * We do this by sleeping on each member using XactLockTableWait. Any
7760 : * members that belong to the current backend are *not* waited for, however;
7761 : * this would not merely be useless but would lead to Assert failure inside
7762 : * XactLockTableWait. By the time this returns, it is certain that all
7763 : * transactions *of other backends* that were members of the MultiXactId
7764 : * that conflict with the requested status are dead (and no new ones can have
7765 : * been added, since it is not legal to add members to an existing
7766 : * MultiXactId).
7767 : *
7768 : * But by the time we finish sleeping, someone else may have changed the Xmax
7769 : * of the containing tuple, so the caller needs to iterate on us somehow.
7770 : *
7771 : * Note that in case we return false, the number of remaining members is
7772 : * not to be trusted.
7773 : */
7774 : static bool
7775 116 : Do_MultiXactIdWait(MultiXactId multi, MultiXactStatus status,
7776 : uint16 infomask, bool nowait,
7777 : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
7778 : int *remaining, bool logLockFailure)
7779 : {
7780 116 : bool result = true;
7781 : MultiXactMember *members;
7782 : int nmembers;
7783 116 : int remain = 0;
7784 :
7785 : /* for pre-pg_upgrade tuples, no need to sleep at all */
7786 116 : nmembers = HEAP_LOCKED_UPGRADED(infomask) ? -1 :
7787 116 : GetMultiXactIdMembers(multi, &members, false,
7788 116 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
7789 :
7790 116 : if (nmembers >= 0)
7791 : {
7792 : int i;
7793 :
7794 374 : for (i = 0; i < nmembers; i++)
7795 : {
7796 266 : TransactionId memxid = members[i].xid;
7797 266 : MultiXactStatus memstatus = members[i].status;
7798 :
7799 266 : if (TransactionIdIsCurrentTransactionId(memxid))
7800 : {
7801 48 : remain++;
7802 48 : continue;
7803 : }
7804 :
7805 218 : if (!DoLockModesConflict(LOCKMODE_from_mxstatus(memstatus),
7806 218 : LOCKMODE_from_mxstatus(status)))
7807 : {
7808 44 : if (remaining && TransactionIdIsInProgress(memxid))
7809 16 : remain++;
7810 44 : continue;
7811 : }
7812 :
7813 : /*
7814 : * This member conflicts with our multi, so we have to sleep (or
7815 : * return failure, if asked to avoid waiting.)
7816 : *
7817 : * Note that we don't set up an error context callback ourselves,
7818 : * but instead we pass the info down to XactLockTableWait. This
7819 : * might seem a bit wasteful because the context is set up and
7820 : * tore down for each member of the multixact, but in reality it
7821 : * should be barely noticeable, and it avoids duplicate code.
7822 : */
7823 174 : if (nowait)
7824 : {
7825 8 : result = ConditionalXactLockTableWait(memxid, logLockFailure);
7826 8 : if (!result)
7827 8 : break;
7828 : }
7829 : else
7830 166 : XactLockTableWait(memxid, rel, ctid, oper);
7831 : }
7832 :
7833 116 : pfree(members);
7834 : }
7835 :
7836 116 : if (remaining)
7837 20 : *remaining = remain;
7838 :
7839 116 : return result;
7840 : }
7841 :
7842 : /*
7843 : * MultiXactIdWait
7844 : * Sleep on a MultiXactId.
7845 : *
7846 : * By the time we finish sleeping, someone else may have changed the Xmax
7847 : * of the containing tuple, so the caller needs to iterate on us somehow.
7848 : *
7849 : * We return (in *remaining, if not NULL) the number of members that are still
7850 : * running, including any (non-aborted) subtransactions of our own transaction.
7851 : */
7852 : static void
7853 108 : MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
7854 : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
7855 : int *remaining)
7856 : {
7857 108 : (void) Do_MultiXactIdWait(multi, status, infomask, false,
7858 : rel, ctid, oper, remaining, false);
7859 108 : }
7860 :
7861 : /*
7862 : * ConditionalMultiXactIdWait
7863 : * As above, but only lock if we can get the lock without blocking.
7864 : *
7865 : * By the time we finish sleeping, someone else may have changed the Xmax
7866 : * of the containing tuple, so the caller needs to iterate on us somehow.
7867 : *
7868 : * If the multixact is now all gone, return true. Returns false if some
7869 : * transactions might still be running.
7870 : *
7871 : * We return (in *remaining, if not NULL) the number of members that are still
7872 : * running, including any (non-aborted) subtransactions of our own transaction.
7873 : */
7874 : static bool
7875 8 : ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status,
7876 : uint16 infomask, Relation rel, int *remaining,
7877 : bool logLockFailure)
7878 : {
7879 8 : return Do_MultiXactIdWait(multi, status, infomask, true,
7880 : rel, NULL, XLTW_None, remaining, logLockFailure);
7881 : }
7882 :
7883 : /*
7884 : * heap_tuple_needs_eventual_freeze
7885 : *
7886 : * Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
7887 : * will eventually require freezing (if tuple isn't removed by pruning first).
7888 : */
7889 : bool
7890 274804 : heap_tuple_needs_eventual_freeze(HeapTupleHeader tuple)
7891 : {
7892 : TransactionId xid;
7893 :
7894 : /*
7895 : * If xmin is a normal transaction ID, this tuple is definitely not
7896 : * frozen.
7897 : */
7898 274804 : xid = HeapTupleHeaderGetXmin(tuple);
7899 274804 : if (TransactionIdIsNormal(xid))
7900 5470 : return true;
7901 :
7902 : /*
7903 : * If xmax is a valid xact or multixact, this tuple is also not frozen.
7904 : */
7905 269334 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7906 : {
7907 : MultiXactId multi;
7908 :
7909 0 : multi = HeapTupleHeaderGetRawXmax(tuple);
7910 0 : if (MultiXactIdIsValid(multi))
7911 0 : return true;
7912 : }
7913 : else
7914 : {
7915 269334 : xid = HeapTupleHeaderGetRawXmax(tuple);
7916 269334 : if (TransactionIdIsNormal(xid))
7917 14 : return true;
7918 : }
7919 :
7920 269320 : if (tuple->t_infomask & HEAP_MOVED)
7921 : {
7922 0 : xid = HeapTupleHeaderGetXvac(tuple);
7923 0 : if (TransactionIdIsNormal(xid))
7924 0 : return true;
7925 : }
7926 :
7927 269320 : return false;
7928 : }
7929 :
7930 : /*
7931 : * heap_tuple_should_freeze
7932 : *
7933 : * Return value indicates if heap_prepare_freeze_tuple sibling function would
7934 : * (or should) force freezing of the heap page that contains caller's tuple.
7935 : * Tuple header XIDs/MXIDs < FreezeLimit/MultiXactCutoff trigger freezing.
7936 : * This includes (xmin, xmax, xvac) fields, as well as MultiXact member XIDs.
7937 : *
7938 : * The *NoFreezePageRelfrozenXid and *NoFreezePageRelminMxid input/output
7939 : * arguments help VACUUM track the oldest extant XID/MXID remaining in rel.
7940 : * Our working assumption is that caller won't decide to freeze this tuple.
7941 : * It's up to caller to only ratchet back its own top-level trackers after the
7942 : * point that it fully commits to not freezing the tuple/page in question.
7943 : */
7944 : bool
7945 9457146 : heap_tuple_should_freeze(HeapTupleHeader tuple,
7946 : const struct VacuumCutoffs *cutoffs,
7947 : TransactionId *NoFreezePageRelfrozenXid,
7948 : MultiXactId *NoFreezePageRelminMxid)
7949 : {
7950 : TransactionId xid;
7951 : MultiXactId multi;
7952 9457146 : bool freeze = false;
7953 :
7954 : /* First deal with xmin */
7955 9457146 : xid = HeapTupleHeaderGetXmin(tuple);
7956 9457146 : if (TransactionIdIsNormal(xid))
7957 : {
7958 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7959 3465136 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7960 45904 : *NoFreezePageRelfrozenXid = xid;
7961 3465136 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7962 42758 : freeze = true;
7963 : }
7964 :
7965 : /* Now deal with xmax */
7966 9457146 : xid = InvalidTransactionId;
7967 9457146 : multi = InvalidMultiXactId;
7968 9457146 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7969 4 : multi = HeapTupleHeaderGetRawXmax(tuple);
7970 : else
7971 9457142 : xid = HeapTupleHeaderGetRawXmax(tuple);
7972 :
7973 9457146 : if (TransactionIdIsNormal(xid))
7974 : {
7975 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7976 : /* xmax is a non-permanent XID */
7977 6482022 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7978 8 : *NoFreezePageRelfrozenXid = xid;
7979 6482022 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7980 54 : freeze = true;
7981 : }
7982 2975124 : else if (!MultiXactIdIsValid(multi))
7983 : {
7984 : /* xmax is a permanent XID or invalid MultiXactId/XID */
7985 : }
7986 4 : else if (HEAP_LOCKED_UPGRADED(tuple->t_infomask))
7987 : {
7988 : /* xmax is a pg_upgrade'd MultiXact, which can't have updater XID */
7989 0 : if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
7990 0 : *NoFreezePageRelminMxid = multi;
7991 : /* heap_prepare_freeze_tuple always freezes pg_upgrade'd xmax */
7992 0 : freeze = true;
7993 : }
7994 : else
7995 : {
7996 : /* xmax is a MultiXactId that may have an updater XID */
7997 : MultiXactMember *members;
7998 : int nmembers;
7999 :
8000 : Assert(MultiXactIdPrecedesOrEquals(cutoffs->relminmxid, multi));
8001 4 : if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
8002 4 : *NoFreezePageRelminMxid = multi;
8003 4 : if (MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff))
8004 4 : freeze = true;
8005 :
8006 : /* need to check whether any member of the mxact is old */
8007 4 : nmembers = GetMultiXactIdMembers(multi, &members, false,
8008 4 : HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask));
8009 :
8010 10 : for (int i = 0; i < nmembers; i++)
8011 : {
8012 6 : xid = members[i].xid;
8013 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
8014 6 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
8015 0 : *NoFreezePageRelfrozenXid = xid;
8016 6 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
8017 0 : freeze = true;
8018 : }
8019 4 : if (nmembers > 0)
8020 2 : pfree(members);
8021 : }
8022 :
8023 9457146 : if (tuple->t_infomask & HEAP_MOVED)
8024 : {
8025 0 : xid = HeapTupleHeaderGetXvac(tuple);
8026 0 : if (TransactionIdIsNormal(xid))
8027 : {
8028 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
8029 0 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
8030 0 : *NoFreezePageRelfrozenXid = xid;
8031 : /* heap_prepare_freeze_tuple forces xvac freezing */
8032 0 : freeze = true;
8033 : }
8034 : }
8035 :
8036 9457146 : return freeze;
8037 : }
8038 :
8039 : /*
8040 : * Maintain snapshotConflictHorizon for caller by ratcheting forward its value
8041 : * using any committed XIDs contained in 'tuple', an obsolescent heap tuple
8042 : * that caller is in the process of physically removing, e.g. via HOT pruning
8043 : * or index deletion.
8044 : *
8045 : * Caller must initialize its value to InvalidTransactionId, which is
8046 : * generally interpreted as "definitely no need for a recovery conflict".
8047 : * Final value must reflect all heap tuples that caller will physically remove
8048 : * (or remove TID references to) via its ongoing pruning/deletion operation.
8049 : * ResolveRecoveryConflictWithSnapshot() is passed the final value (taken from
8050 : * caller's WAL record) by REDO routine when it replays caller's operation.
8051 : */
8052 : void
8053 3148698 : HeapTupleHeaderAdvanceConflictHorizon(HeapTupleHeader tuple,
8054 : TransactionId *snapshotConflictHorizon)
8055 : {
8056 3148698 : TransactionId xmin = HeapTupleHeaderGetXmin(tuple);
8057 3148698 : TransactionId xmax = HeapTupleHeaderGetUpdateXid(tuple);
8058 3148698 : TransactionId xvac = HeapTupleHeaderGetXvac(tuple);
8059 :
8060 3148698 : if (tuple->t_infomask & HEAP_MOVED)
8061 : {
8062 0 : if (TransactionIdPrecedes(*snapshotConflictHorizon, xvac))
8063 0 : *snapshotConflictHorizon = xvac;
8064 : }
8065 :
8066 : /*
8067 : * Ignore tuples inserted by an aborted transaction or if the tuple was
8068 : * updated/deleted by the inserting transaction.
8069 : *
8070 : * Look for a committed hint bit, or if no xmin bit is set, check clog.
8071 : */
8072 3148698 : if (HeapTupleHeaderXminCommitted(tuple) ||
8073 187214 : (!HeapTupleHeaderXminInvalid(tuple) && TransactionIdDidCommit(xmin)))
8074 : {
8075 5662168 : if (xmax != xmin &&
8076 2671774 : TransactionIdFollows(xmax, *snapshotConflictHorizon))
8077 200946 : *snapshotConflictHorizon = xmax;
8078 : }
8079 3148698 : }
8080 :
8081 : #ifdef USE_PREFETCH
8082 : /*
8083 : * Helper function for heap_index_delete_tuples. Issues prefetch requests for
8084 : * prefetch_count buffers. The prefetch_state keeps track of all the buffers
8085 : * we can prefetch, and which have already been prefetched; each call to this
8086 : * function picks up where the previous call left off.
8087 : *
8088 : * Note: we expect the deltids array to be sorted in an order that groups TIDs
8089 : * by heap block, with all TIDs for each block appearing together in exactly
8090 : * one group.
8091 : */
8092 : static void
8093 39442 : index_delete_prefetch_buffer(Relation rel,
8094 : IndexDeletePrefetchState *prefetch_state,
8095 : int prefetch_count)
8096 : {
8097 39442 : BlockNumber cur_hblkno = prefetch_state->cur_hblkno;
8098 39442 : int count = 0;
8099 : int i;
8100 39442 : int ndeltids = prefetch_state->ndeltids;
8101 39442 : TM_IndexDelete *deltids = prefetch_state->deltids;
8102 :
8103 39442 : for (i = prefetch_state->next_item;
8104 1372692 : i < ndeltids && count < prefetch_count;
8105 1333250 : i++)
8106 : {
8107 1333250 : ItemPointer htid = &deltids[i].tid;
8108 :
8109 2654674 : if (cur_hblkno == InvalidBlockNumber ||
8110 1321424 : ItemPointerGetBlockNumber(htid) != cur_hblkno)
8111 : {
8112 36186 : cur_hblkno = ItemPointerGetBlockNumber(htid);
8113 36186 : PrefetchBuffer(rel, MAIN_FORKNUM, cur_hblkno);
8114 36186 : count++;
8115 : }
8116 : }
8117 :
8118 : /*
8119 : * Save the prefetch position so that next time we can continue from that
8120 : * position.
8121 : */
8122 39442 : prefetch_state->next_item = i;
8123 39442 : prefetch_state->cur_hblkno = cur_hblkno;
8124 39442 : }
8125 : #endif
8126 :
8127 : /*
8128 : * Helper function for heap_index_delete_tuples. Checks for index corruption
8129 : * involving an invalid TID in index AM caller's index page.
8130 : *
8131 : * This is an ideal place for these checks. The index AM must hold a buffer
8132 : * lock on the index page containing the TIDs we examine here, so we don't
8133 : * have to worry about concurrent VACUUMs at all. We can be sure that the
8134 : * index is corrupt when htid points directly to an LP_UNUSED item or
8135 : * heap-only tuple, which is not the case during standard index scans.
8136 : */
8137 : static inline void
8138 1100810 : index_delete_check_htid(TM_IndexDeleteOp *delstate,
8139 : Page page, OffsetNumber maxoff,
8140 : const ItemPointerData *htid, TM_IndexStatus *istatus)
8141 : {
8142 1100810 : OffsetNumber indexpagehoffnum = ItemPointerGetOffsetNumber(htid);
8143 : ItemId iid;
8144 :
8145 : Assert(OffsetNumberIsValid(istatus->idxoffnum));
8146 :
8147 1100810 : if (unlikely(indexpagehoffnum > maxoff))
8148 0 : ereport(ERROR,
8149 : (errcode(ERRCODE_INDEX_CORRUPTED),
8150 : 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\"",
8151 : ItemPointerGetBlockNumber(htid),
8152 : indexpagehoffnum,
8153 : istatus->idxoffnum, delstate->iblknum,
8154 : RelationGetRelationName(delstate->irel))));
8155 :
8156 1100810 : iid = PageGetItemId(page, indexpagehoffnum);
8157 1100810 : if (unlikely(!ItemIdIsUsed(iid)))
8158 0 : ereport(ERROR,
8159 : (errcode(ERRCODE_INDEX_CORRUPTED),
8160 : errmsg_internal("heap tid from index tuple (%u,%u) points to unused heap page item at offset %u of block %u in index \"%s\"",
8161 : ItemPointerGetBlockNumber(htid),
8162 : indexpagehoffnum,
8163 : istatus->idxoffnum, delstate->iblknum,
8164 : RelationGetRelationName(delstate->irel))));
8165 :
8166 1100810 : if (ItemIdHasStorage(iid))
8167 : {
8168 : HeapTupleHeader htup;
8169 :
8170 : Assert(ItemIdIsNormal(iid));
8171 649158 : htup = (HeapTupleHeader) PageGetItem(page, iid);
8172 :
8173 649158 : if (unlikely(HeapTupleHeaderIsHeapOnly(htup)))
8174 0 : ereport(ERROR,
8175 : (errcode(ERRCODE_INDEX_CORRUPTED),
8176 : errmsg_internal("heap tid from index tuple (%u,%u) points to heap-only tuple at offset %u of block %u in index \"%s\"",
8177 : ItemPointerGetBlockNumber(htid),
8178 : indexpagehoffnum,
8179 : istatus->idxoffnum, delstate->iblknum,
8180 : RelationGetRelationName(delstate->irel))));
8181 : }
8182 1100810 : }
8183 :
8184 : /*
8185 : * heapam implementation of tableam's index_delete_tuples interface.
8186 : *
8187 : * This helper function is called by index AMs during index tuple deletion.
8188 : * See tableam header comments for an explanation of the interface implemented
8189 : * here and a general theory of operation. Note that each call here is either
8190 : * a simple index deletion call, or a bottom-up index deletion call.
8191 : *
8192 : * It's possible for this to generate a fair amount of I/O, since we may be
8193 : * deleting hundreds of tuples from a single index block. To amortize that
8194 : * cost to some degree, this uses prefetching and combines repeat accesses to
8195 : * the same heap block.
8196 : */
8197 : TransactionId
8198 11826 : heap_index_delete_tuples(Relation rel, TM_IndexDeleteOp *delstate)
8199 : {
8200 : /* Initial assumption is that earlier pruning took care of conflict */
8201 11826 : TransactionId snapshotConflictHorizon = InvalidTransactionId;
8202 11826 : BlockNumber blkno = InvalidBlockNumber;
8203 11826 : Buffer buf = InvalidBuffer;
8204 11826 : Page page = NULL;
8205 11826 : OffsetNumber maxoff = InvalidOffsetNumber;
8206 : TransactionId priorXmax;
8207 : #ifdef USE_PREFETCH
8208 : IndexDeletePrefetchState prefetch_state;
8209 : int prefetch_distance;
8210 : #endif
8211 : SnapshotData SnapshotNonVacuumable;
8212 11826 : int finalndeltids = 0,
8213 11826 : nblocksaccessed = 0;
8214 :
8215 : /* State that's only used in bottom-up index deletion case */
8216 11826 : int nblocksfavorable = 0;
8217 11826 : int curtargetfreespace = delstate->bottomupfreespace,
8218 11826 : lastfreespace = 0,
8219 11826 : actualfreespace = 0;
8220 11826 : bool bottomup_final_block = false;
8221 :
8222 11826 : InitNonVacuumableSnapshot(SnapshotNonVacuumable, GlobalVisTestFor(rel));
8223 :
8224 : /* Sort caller's deltids array by TID for further processing */
8225 11826 : index_delete_sort(delstate);
8226 :
8227 : /*
8228 : * Bottom-up case: resort deltids array in an order attuned to where the
8229 : * greatest number of promising TIDs are to be found, and determine how
8230 : * many blocks from the start of sorted array should be considered
8231 : * favorable. This will also shrink the deltids array in order to
8232 : * eliminate completely unfavorable blocks up front.
8233 : */
8234 11826 : if (delstate->bottomup)
8235 4138 : nblocksfavorable = bottomup_sort_and_shrink(delstate);
8236 :
8237 : #ifdef USE_PREFETCH
8238 : /* Initialize prefetch state. */
8239 11826 : prefetch_state.cur_hblkno = InvalidBlockNumber;
8240 11826 : prefetch_state.next_item = 0;
8241 11826 : prefetch_state.ndeltids = delstate->ndeltids;
8242 11826 : prefetch_state.deltids = delstate->deltids;
8243 :
8244 : /*
8245 : * Determine the prefetch distance that we will attempt to maintain.
8246 : *
8247 : * Since the caller holds a buffer lock somewhere in rel, we'd better make
8248 : * sure that isn't a catalog relation before we call code that does
8249 : * syscache lookups, to avoid risk of deadlock.
8250 : */
8251 11826 : if (IsCatalogRelation(rel))
8252 8634 : prefetch_distance = maintenance_io_concurrency;
8253 : else
8254 : prefetch_distance =
8255 3192 : get_tablespace_maintenance_io_concurrency(rel->rd_rel->reltablespace);
8256 :
8257 : /* Cap initial prefetch distance for bottom-up deletion caller */
8258 11826 : if (delstate->bottomup)
8259 : {
8260 : Assert(nblocksfavorable >= 1);
8261 : Assert(nblocksfavorable <= BOTTOMUP_MAX_NBLOCKS);
8262 4138 : prefetch_distance = Min(prefetch_distance, nblocksfavorable);
8263 : }
8264 :
8265 : /* Start prefetching. */
8266 11826 : index_delete_prefetch_buffer(rel, &prefetch_state, prefetch_distance);
8267 : #endif
8268 :
8269 : /* Iterate over deltids, determine which to delete, check their horizon */
8270 : Assert(delstate->ndeltids > 0);
8271 1112636 : for (int i = 0; i < delstate->ndeltids; i++)
8272 : {
8273 1104948 : TM_IndexDelete *ideltid = &delstate->deltids[i];
8274 1104948 : TM_IndexStatus *istatus = delstate->status + ideltid->id;
8275 1104948 : ItemPointer htid = &ideltid->tid;
8276 : OffsetNumber offnum;
8277 :
8278 : /*
8279 : * Read buffer, and perform required extra steps each time a new block
8280 : * is encountered. Avoid refetching if it's the same block as the one
8281 : * from the last htid.
8282 : */
8283 2198070 : if (blkno == InvalidBlockNumber ||
8284 1093122 : ItemPointerGetBlockNumber(htid) != blkno)
8285 : {
8286 : /*
8287 : * Consider giving up early for bottom-up index deletion caller
8288 : * first. (Only prefetch next-next block afterwards, when it
8289 : * becomes clear that we're at least going to access the next
8290 : * block in line.)
8291 : *
8292 : * Sometimes the first block frees so much space for bottom-up
8293 : * caller that the deletion process can end without accessing any
8294 : * more blocks. It is usually necessary to access 2 or 3 blocks
8295 : * per bottom-up deletion operation, though.
8296 : */
8297 31754 : if (delstate->bottomup)
8298 : {
8299 : /*
8300 : * We often allow caller to delete a few additional items
8301 : * whose entries we reached after the point that space target
8302 : * from caller was satisfied. The cost of accessing the page
8303 : * was already paid at that point, so it made sense to finish
8304 : * it off. When that happened, we finalize everything here
8305 : * (by finishing off the whole bottom-up deletion operation
8306 : * without needlessly paying the cost of accessing any more
8307 : * blocks).
8308 : */
8309 9274 : if (bottomup_final_block)
8310 326 : break;
8311 :
8312 : /*
8313 : * Give up when we didn't enable our caller to free any
8314 : * additional space as a result of processing the page that we
8315 : * just finished up with. This rule is the main way in which
8316 : * we keep the cost of bottom-up deletion under control.
8317 : */
8318 8948 : if (nblocksaccessed >= 1 && actualfreespace == lastfreespace)
8319 3812 : break;
8320 5136 : lastfreespace = actualfreespace; /* for next time */
8321 :
8322 : /*
8323 : * Deletion operation (which is bottom-up) will definitely
8324 : * access the next block in line. Prepare for that now.
8325 : *
8326 : * Decay target free space so that we don't hang on for too
8327 : * long with a marginal case. (Space target is only truly
8328 : * helpful when it allows us to recognize that we don't need
8329 : * to access more than 1 or 2 blocks to satisfy caller due to
8330 : * agreeable workload characteristics.)
8331 : *
8332 : * We are a bit more patient when we encounter contiguous
8333 : * blocks, though: these are treated as favorable blocks. The
8334 : * decay process is only applied when the next block in line
8335 : * is not a favorable/contiguous block. This is not an
8336 : * exception to the general rule; we still insist on finding
8337 : * at least one deletable item per block accessed. See
8338 : * bottomup_nblocksfavorable() for full details of the theory
8339 : * behind favorable blocks and heap block locality in general.
8340 : *
8341 : * Note: The first block in line is always treated as a
8342 : * favorable block, so the earliest possible point that the
8343 : * decay can be applied is just before we access the second
8344 : * block in line. The Assert() verifies this for us.
8345 : */
8346 : Assert(nblocksaccessed > 0 || nblocksfavorable > 0);
8347 5136 : if (nblocksfavorable > 0)
8348 4636 : nblocksfavorable--;
8349 : else
8350 500 : curtargetfreespace /= 2;
8351 : }
8352 :
8353 : /* release old buffer */
8354 27616 : if (BufferIsValid(buf))
8355 15790 : UnlockReleaseBuffer(buf);
8356 :
8357 27616 : blkno = ItemPointerGetBlockNumber(htid);
8358 27616 : buf = ReadBuffer(rel, blkno);
8359 27616 : nblocksaccessed++;
8360 : Assert(!delstate->bottomup ||
8361 : nblocksaccessed <= BOTTOMUP_MAX_NBLOCKS);
8362 :
8363 : #ifdef USE_PREFETCH
8364 :
8365 : /*
8366 : * To maintain the prefetch distance, prefetch one more page for
8367 : * each page we read.
8368 : */
8369 27616 : index_delete_prefetch_buffer(rel, &prefetch_state, 1);
8370 : #endif
8371 :
8372 27616 : LockBuffer(buf, BUFFER_LOCK_SHARE);
8373 :
8374 27616 : page = BufferGetPage(buf);
8375 27616 : maxoff = PageGetMaxOffsetNumber(page);
8376 : }
8377 :
8378 : /*
8379 : * In passing, detect index corruption involving an index page with a
8380 : * TID that points to a location in the heap that couldn't possibly be
8381 : * correct. We only do this with actual TIDs from caller's index page
8382 : * (not items reached by traversing through a HOT chain).
8383 : */
8384 1100810 : index_delete_check_htid(delstate, page, maxoff, htid, istatus);
8385 :
8386 1100810 : if (istatus->knowndeletable)
8387 : Assert(!delstate->bottomup && !istatus->promising);
8388 : else
8389 : {
8390 830350 : ItemPointerData tmp = *htid;
8391 : HeapTupleData heapTuple;
8392 :
8393 : /* Are any tuples from this HOT chain non-vacuumable? */
8394 830350 : if (heap_hot_search_buffer(&tmp, rel, buf, &SnapshotNonVacuumable,
8395 : &heapTuple, NULL, true))
8396 498002 : continue; /* can't delete entry */
8397 :
8398 : /* Caller will delete, since whole HOT chain is vacuumable */
8399 332348 : istatus->knowndeletable = true;
8400 :
8401 : /* Maintain index free space info for bottom-up deletion case */
8402 332348 : if (delstate->bottomup)
8403 : {
8404 : Assert(istatus->freespace > 0);
8405 18618 : actualfreespace += istatus->freespace;
8406 18618 : if (actualfreespace >= curtargetfreespace)
8407 4542 : bottomup_final_block = true;
8408 : }
8409 : }
8410 :
8411 : /*
8412 : * Maintain snapshotConflictHorizon value for deletion operation as a
8413 : * whole by advancing current value using heap tuple headers. This is
8414 : * loosely based on the logic for pruning a HOT chain.
8415 : */
8416 602808 : offnum = ItemPointerGetOffsetNumber(htid);
8417 602808 : priorXmax = InvalidTransactionId; /* cannot check first XMIN */
8418 : for (;;)
8419 41262 : {
8420 : ItemId lp;
8421 : HeapTupleHeader htup;
8422 :
8423 : /* Sanity check (pure paranoia) */
8424 644070 : if (offnum < FirstOffsetNumber)
8425 0 : break;
8426 :
8427 : /*
8428 : * An offset past the end of page's line pointer array is possible
8429 : * when the array was truncated
8430 : */
8431 644070 : if (offnum > maxoff)
8432 0 : break;
8433 :
8434 644070 : lp = PageGetItemId(page, offnum);
8435 644070 : if (ItemIdIsRedirected(lp))
8436 : {
8437 18928 : offnum = ItemIdGetRedirect(lp);
8438 18928 : continue;
8439 : }
8440 :
8441 : /*
8442 : * We'll often encounter LP_DEAD line pointers (especially with an
8443 : * entry marked knowndeletable by our caller up front). No heap
8444 : * tuple headers get examined for an htid that leads us to an
8445 : * LP_DEAD item. This is okay because the earlier pruning
8446 : * operation that made the line pointer LP_DEAD in the first place
8447 : * must have considered the original tuple header as part of
8448 : * generating its own snapshotConflictHorizon value.
8449 : *
8450 : * Relying on XLOG_HEAP2_PRUNE_VACUUM_SCAN records like this is
8451 : * the same strategy that index vacuuming uses in all cases. Index
8452 : * VACUUM WAL records don't even have a snapshotConflictHorizon
8453 : * field of their own for this reason.
8454 : */
8455 625142 : if (!ItemIdIsNormal(lp))
8456 401434 : break;
8457 :
8458 223708 : htup = (HeapTupleHeader) PageGetItem(page, lp);
8459 :
8460 : /*
8461 : * Check the tuple XMIN against prior XMAX, if any
8462 : */
8463 246042 : if (TransactionIdIsValid(priorXmax) &&
8464 22334 : !TransactionIdEquals(HeapTupleHeaderGetXmin(htup), priorXmax))
8465 0 : break;
8466 :
8467 223708 : HeapTupleHeaderAdvanceConflictHorizon(htup,
8468 : &snapshotConflictHorizon);
8469 :
8470 : /*
8471 : * If the tuple is not HOT-updated, then we are at the end of this
8472 : * HOT-chain. No need to visit later tuples from the same update
8473 : * chain (they get their own index entries) -- just move on to
8474 : * next htid from index AM caller.
8475 : */
8476 223708 : if (!HeapTupleHeaderIsHotUpdated(htup))
8477 201374 : break;
8478 :
8479 : /* Advance to next HOT chain member */
8480 : Assert(ItemPointerGetBlockNumber(&htup->t_ctid) == blkno);
8481 22334 : offnum = ItemPointerGetOffsetNumber(&htup->t_ctid);
8482 22334 : priorXmax = HeapTupleHeaderGetUpdateXid(htup);
8483 : }
8484 :
8485 : /* Enable further/final shrinking of deltids for caller */
8486 602808 : finalndeltids = i + 1;
8487 : }
8488 :
8489 11826 : UnlockReleaseBuffer(buf);
8490 :
8491 : /*
8492 : * Shrink deltids array to exclude non-deletable entries at the end. This
8493 : * is not just a minor optimization. Final deltids array size might be
8494 : * zero for a bottom-up caller. Index AM is explicitly allowed to rely on
8495 : * ndeltids being zero in all cases with zero total deletable entries.
8496 : */
8497 : Assert(finalndeltids > 0 || delstate->bottomup);
8498 11826 : delstate->ndeltids = finalndeltids;
8499 :
8500 11826 : return snapshotConflictHorizon;
8501 : }
8502 :
8503 : /*
8504 : * Specialized inlineable comparison function for index_delete_sort()
8505 : */
8506 : static inline int
8507 26455812 : index_delete_sort_cmp(TM_IndexDelete *deltid1, TM_IndexDelete *deltid2)
8508 : {
8509 26455812 : ItemPointer tid1 = &deltid1->tid;
8510 26455812 : ItemPointer tid2 = &deltid2->tid;
8511 :
8512 : {
8513 26455812 : BlockNumber blk1 = ItemPointerGetBlockNumber(tid1);
8514 26455812 : BlockNumber blk2 = ItemPointerGetBlockNumber(tid2);
8515 :
8516 26455812 : if (blk1 != blk2)
8517 10886544 : return (blk1 < blk2) ? -1 : 1;
8518 : }
8519 : {
8520 15569268 : OffsetNumber pos1 = ItemPointerGetOffsetNumber(tid1);
8521 15569268 : OffsetNumber pos2 = ItemPointerGetOffsetNumber(tid2);
8522 :
8523 15569268 : if (pos1 != pos2)
8524 15569268 : return (pos1 < pos2) ? -1 : 1;
8525 : }
8526 :
8527 : Assert(false);
8528 :
8529 0 : return 0;
8530 : }
8531 :
8532 : /*
8533 : * Sort deltids array from delstate by TID. This prepares it for further
8534 : * processing by heap_index_delete_tuples().
8535 : *
8536 : * This operation becomes a noticeable consumer of CPU cycles with some
8537 : * workloads, so we go to the trouble of specialization/micro optimization.
8538 : * We use shellsort for this because it's easy to specialize, compiles to
8539 : * relatively few instructions, and is adaptive to presorted inputs/subsets
8540 : * (which are typical here).
8541 : */
8542 : static void
8543 11826 : index_delete_sort(TM_IndexDeleteOp *delstate)
8544 : {
8545 11826 : TM_IndexDelete *deltids = delstate->deltids;
8546 11826 : int ndeltids = delstate->ndeltids;
8547 :
8548 : /*
8549 : * Shellsort gap sequence (taken from Sedgewick-Incerpi paper).
8550 : *
8551 : * This implementation is fast with array sizes up to ~4500. This covers
8552 : * all supported BLCKSZ values.
8553 : */
8554 11826 : const int gaps[9] = {1968, 861, 336, 112, 48, 21, 7, 3, 1};
8555 :
8556 : /* Think carefully before changing anything here -- keep swaps cheap */
8557 : StaticAssertDecl(sizeof(TM_IndexDelete) <= 8,
8558 : "element size exceeds 8 bytes");
8559 :
8560 118260 : for (int g = 0; g < lengthof(gaps); g++)
8561 : {
8562 15735374 : for (int hi = gaps[g], i = hi; i < ndeltids; i++)
8563 : {
8564 15628940 : TM_IndexDelete d = deltids[i];
8565 15628940 : int j = i;
8566 :
8567 27237124 : while (j >= hi && index_delete_sort_cmp(&deltids[j - hi], &d) >= 0)
8568 : {
8569 11608184 : deltids[j] = deltids[j - hi];
8570 11608184 : j -= hi;
8571 : }
8572 15628940 : deltids[j] = d;
8573 : }
8574 : }
8575 11826 : }
8576 :
8577 : /*
8578 : * Returns how many blocks should be considered favorable/contiguous for a
8579 : * bottom-up index deletion pass. This is a number of heap blocks that starts
8580 : * from and includes the first block in line.
8581 : *
8582 : * There is always at least one favorable block during bottom-up index
8583 : * deletion. In the worst case (i.e. with totally random heap blocks) the
8584 : * first block in line (the only favorable block) can be thought of as a
8585 : * degenerate array of contiguous blocks that consists of a single block.
8586 : * heap_index_delete_tuples() will expect this.
8587 : *
8588 : * Caller passes blockgroups, a description of the final order that deltids
8589 : * will be sorted in for heap_index_delete_tuples() bottom-up index deletion
8590 : * processing. Note that deltids need not actually be sorted just yet (caller
8591 : * only passes deltids to us so that we can interpret blockgroups).
8592 : *
8593 : * You might guess that the existence of contiguous blocks cannot matter much,
8594 : * since in general the main factor that determines which blocks we visit is
8595 : * the number of promising TIDs, which is a fixed hint from the index AM.
8596 : * We're not really targeting the general case, though -- the actual goal is
8597 : * to adapt our behavior to a wide variety of naturally occurring conditions.
8598 : * The effects of most of the heuristics we apply are only noticeable in the
8599 : * aggregate, over time and across many _related_ bottom-up index deletion
8600 : * passes.
8601 : *
8602 : * Deeming certain blocks favorable allows heapam to recognize and adapt to
8603 : * workloads where heap blocks visited during bottom-up index deletion can be
8604 : * accessed contiguously, in the sense that each newly visited block is the
8605 : * neighbor of the block that bottom-up deletion just finished processing (or
8606 : * close enough to it). It will likely be cheaper to access more favorable
8607 : * blocks sooner rather than later (e.g. in this pass, not across a series of
8608 : * related bottom-up passes). Either way it is probably only a matter of time
8609 : * (or a matter of further correlated version churn) before all blocks that
8610 : * appear together as a single large batch of favorable blocks get accessed by
8611 : * _some_ bottom-up pass. Large batches of favorable blocks tend to either
8612 : * appear almost constantly or not even once (it all depends on per-index
8613 : * workload characteristics).
8614 : *
8615 : * Note that the blockgroups sort order applies a power-of-two bucketing
8616 : * scheme that creates opportunities for contiguous groups of blocks to get
8617 : * batched together, at least with workloads that are naturally amenable to
8618 : * being driven by heap block locality. This doesn't just enhance the spatial
8619 : * locality of bottom-up heap block processing in the obvious way. It also
8620 : * enables temporal locality of access, since sorting by heap block number
8621 : * naturally tends to make the bottom-up processing order deterministic.
8622 : *
8623 : * Consider the following example to get a sense of how temporal locality
8624 : * might matter: There is a heap relation with several indexes, each of which
8625 : * is low to medium cardinality. It is subject to constant non-HOT updates.
8626 : * The updates are skewed (in one part of the primary key, perhaps). None of
8627 : * the indexes are logically modified by the UPDATE statements (if they were
8628 : * then bottom-up index deletion would not be triggered in the first place).
8629 : * Naturally, each new round of index tuples (for each heap tuple that gets a
8630 : * heap_update() call) will have the same heap TID in each and every index.
8631 : * Since these indexes are low cardinality and never get logically modified,
8632 : * heapam processing during bottom-up deletion passes will access heap blocks
8633 : * in approximately sequential order. Temporal locality of access occurs due
8634 : * to bottom-up deletion passes behaving very similarly across each of the
8635 : * indexes at any given moment. This keeps the number of buffer misses needed
8636 : * to visit heap blocks to a minimum.
8637 : */
8638 : static int
8639 4138 : bottomup_nblocksfavorable(IndexDeleteCounts *blockgroups, int nblockgroups,
8640 : TM_IndexDelete *deltids)
8641 : {
8642 4138 : int64 lastblock = -1;
8643 4138 : int nblocksfavorable = 0;
8644 :
8645 : Assert(nblockgroups >= 1);
8646 : Assert(nblockgroups <= BOTTOMUP_MAX_NBLOCKS);
8647 :
8648 : /*
8649 : * We tolerate heap blocks that will be accessed only slightly out of
8650 : * physical order. Small blips occur when a pair of almost-contiguous
8651 : * blocks happen to fall into different buckets (perhaps due only to a
8652 : * small difference in npromisingtids that the bucketing scheme didn't
8653 : * quite manage to ignore). We effectively ignore these blips by applying
8654 : * a small tolerance. The precise tolerance we use is a little arbitrary,
8655 : * but it works well enough in practice.
8656 : */
8657 13438 : for (int b = 0; b < nblockgroups; b++)
8658 : {
8659 12802 : IndexDeleteCounts *group = blockgroups + b;
8660 12802 : TM_IndexDelete *firstdtid = deltids + group->ifirsttid;
8661 12802 : BlockNumber block = ItemPointerGetBlockNumber(&firstdtid->tid);
8662 :
8663 12802 : if (lastblock != -1 &&
8664 8664 : ((int64) block < lastblock - BOTTOMUP_TOLERANCE_NBLOCKS ||
8665 7666 : (int64) block > lastblock + BOTTOMUP_TOLERANCE_NBLOCKS))
8666 : break;
8667 :
8668 9300 : nblocksfavorable++;
8669 9300 : lastblock = block;
8670 : }
8671 :
8672 : /* Always indicate that there is at least 1 favorable block */
8673 : Assert(nblocksfavorable >= 1);
8674 :
8675 4138 : return nblocksfavorable;
8676 : }
8677 :
8678 : /*
8679 : * qsort comparison function for bottomup_sort_and_shrink()
8680 : */
8681 : static int
8682 415150 : bottomup_sort_and_shrink_cmp(const void *arg1, const void *arg2)
8683 : {
8684 415150 : const IndexDeleteCounts *group1 = (const IndexDeleteCounts *) arg1;
8685 415150 : const IndexDeleteCounts *group2 = (const IndexDeleteCounts *) arg2;
8686 :
8687 : /*
8688 : * Most significant field is npromisingtids (which we invert the order of
8689 : * so as to sort in desc order).
8690 : *
8691 : * Caller should have already normalized npromisingtids fields into
8692 : * power-of-two values (buckets).
8693 : */
8694 415150 : if (group1->npromisingtids > group2->npromisingtids)
8695 19710 : return -1;
8696 395440 : if (group1->npromisingtids < group2->npromisingtids)
8697 21192 : return 1;
8698 :
8699 : /*
8700 : * Tiebreak: desc ntids sort order.
8701 : *
8702 : * We cannot expect power-of-two values for ntids fields. We should
8703 : * behave as if they were already rounded up for us instead.
8704 : */
8705 374248 : if (group1->ntids != group2->ntids)
8706 : {
8707 269486 : uint32 ntids1 = pg_nextpower2_32((uint32) group1->ntids);
8708 269486 : uint32 ntids2 = pg_nextpower2_32((uint32) group2->ntids);
8709 :
8710 269486 : if (ntids1 > ntids2)
8711 42298 : return -1;
8712 227188 : if (ntids1 < ntids2)
8713 52322 : return 1;
8714 : }
8715 :
8716 : /*
8717 : * Tiebreak: asc offset-into-deltids-for-block (offset to first TID for
8718 : * block in deltids array) order.
8719 : *
8720 : * This is equivalent to sorting in ascending heap block number order
8721 : * (among otherwise equal subsets of the array). This approach allows us
8722 : * to avoid accessing the out-of-line TID. (We rely on the assumption
8723 : * that the deltids array was sorted in ascending heap TID order when
8724 : * these offsets to the first TID from each heap block group were formed.)
8725 : */
8726 279628 : if (group1->ifirsttid > group2->ifirsttid)
8727 136636 : return 1;
8728 142992 : if (group1->ifirsttid < group2->ifirsttid)
8729 142992 : return -1;
8730 :
8731 0 : pg_unreachable();
8732 :
8733 : return 0;
8734 : }
8735 :
8736 : /*
8737 : * heap_index_delete_tuples() helper function for bottom-up deletion callers.
8738 : *
8739 : * Sorts deltids array in the order needed for useful processing by bottom-up
8740 : * deletion. The array should already be sorted in TID order when we're
8741 : * called. The sort process groups heap TIDs from deltids into heap block
8742 : * groupings. Earlier/more-promising groups/blocks are usually those that are
8743 : * known to have the most "promising" TIDs.
8744 : *
8745 : * Sets new size of deltids array (ndeltids) in state. deltids will only have
8746 : * TIDs from the BOTTOMUP_MAX_NBLOCKS most promising heap blocks when we
8747 : * return. This often means that deltids will be shrunk to a small fraction
8748 : * of its original size (we eliminate many heap blocks from consideration for
8749 : * caller up front).
8750 : *
8751 : * Returns the number of "favorable" blocks. See bottomup_nblocksfavorable()
8752 : * for a definition and full details.
8753 : */
8754 : static int
8755 4138 : bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate)
8756 : {
8757 : IndexDeleteCounts *blockgroups;
8758 : TM_IndexDelete *reordereddeltids;
8759 4138 : BlockNumber curblock = InvalidBlockNumber;
8760 4138 : int nblockgroups = 0;
8761 4138 : int ncopied = 0;
8762 4138 : int nblocksfavorable = 0;
8763 :
8764 : Assert(delstate->bottomup);
8765 : Assert(delstate->ndeltids > 0);
8766 :
8767 : /* Calculate per-heap-block count of TIDs */
8768 4138 : blockgroups = palloc_array(IndexDeleteCounts, delstate->ndeltids);
8769 1972606 : for (int i = 0; i < delstate->ndeltids; i++)
8770 : {
8771 1968468 : TM_IndexDelete *ideltid = &delstate->deltids[i];
8772 1968468 : TM_IndexStatus *istatus = delstate->status + ideltid->id;
8773 1968468 : ItemPointer htid = &ideltid->tid;
8774 1968468 : bool promising = istatus->promising;
8775 :
8776 1968468 : if (curblock != ItemPointerGetBlockNumber(htid))
8777 : {
8778 : /* New block group */
8779 81752 : nblockgroups++;
8780 :
8781 : Assert(curblock < ItemPointerGetBlockNumber(htid) ||
8782 : !BlockNumberIsValid(curblock));
8783 :
8784 81752 : curblock = ItemPointerGetBlockNumber(htid);
8785 81752 : blockgroups[nblockgroups - 1].ifirsttid = i;
8786 81752 : blockgroups[nblockgroups - 1].ntids = 1;
8787 81752 : blockgroups[nblockgroups - 1].npromisingtids = 0;
8788 : }
8789 : else
8790 : {
8791 1886716 : blockgroups[nblockgroups - 1].ntids++;
8792 : }
8793 :
8794 1968468 : if (promising)
8795 247944 : blockgroups[nblockgroups - 1].npromisingtids++;
8796 : }
8797 :
8798 : /*
8799 : * We're about ready to sort block groups to determine the optimal order
8800 : * for visiting heap blocks. But before we do, round the number of
8801 : * promising tuples for each block group up to the next power-of-two,
8802 : * unless it is very low (less than 4), in which case we round up to 4.
8803 : * npromisingtids is far too noisy to trust when choosing between a pair
8804 : * of block groups that both have very low values.
8805 : *
8806 : * This scheme divides heap blocks/block groups into buckets. Each bucket
8807 : * contains blocks that have _approximately_ the same number of promising
8808 : * TIDs as each other. The goal is to ignore relatively small differences
8809 : * in the total number of promising entries, so that the whole process can
8810 : * give a little weight to heapam factors (like heap block locality)
8811 : * instead. This isn't a trade-off, really -- we have nothing to lose. It
8812 : * would be foolish to interpret small differences in npromisingtids
8813 : * values as anything more than noise.
8814 : *
8815 : * We tiebreak on nhtids when sorting block group subsets that have the
8816 : * same npromisingtids, but this has the same issues as npromisingtids,
8817 : * and so nhtids is subject to the same power-of-two bucketing scheme. The
8818 : * only reason that we don't fix nhtids in the same way here too is that
8819 : * we'll need accurate nhtids values after the sort. We handle nhtids
8820 : * bucketization dynamically instead (in the sort comparator).
8821 : *
8822 : * See bottomup_nblocksfavorable() for a full explanation of when and how
8823 : * heap locality/favorable blocks can significantly influence when and how
8824 : * heap blocks are accessed.
8825 : */
8826 85890 : for (int b = 0; b < nblockgroups; b++)
8827 : {
8828 81752 : IndexDeleteCounts *group = blockgroups + b;
8829 :
8830 : /* Better off falling back on nhtids with low npromisingtids */
8831 81752 : if (group->npromisingtids <= 4)
8832 70324 : group->npromisingtids = 4;
8833 : else
8834 11428 : group->npromisingtids =
8835 11428 : pg_nextpower2_32((uint32) group->npromisingtids);
8836 : }
8837 :
8838 : /* Sort groups and rearrange caller's deltids array */
8839 4138 : qsort(blockgroups, nblockgroups, sizeof(IndexDeleteCounts),
8840 : bottomup_sort_and_shrink_cmp);
8841 4138 : reordereddeltids = palloc(delstate->ndeltids * sizeof(TM_IndexDelete));
8842 :
8843 4138 : nblockgroups = Min(BOTTOMUP_MAX_NBLOCKS, nblockgroups);
8844 : /* Determine number of favorable blocks at the start of final deltids */
8845 4138 : nblocksfavorable = bottomup_nblocksfavorable(blockgroups, nblockgroups,
8846 : delstate->deltids);
8847 :
8848 27748 : for (int b = 0; b < nblockgroups; b++)
8849 : {
8850 23610 : IndexDeleteCounts *group = blockgroups + b;
8851 23610 : TM_IndexDelete *firstdtid = delstate->deltids + group->ifirsttid;
8852 :
8853 23610 : memcpy(reordereddeltids + ncopied, firstdtid,
8854 23610 : sizeof(TM_IndexDelete) * group->ntids);
8855 23610 : ncopied += group->ntids;
8856 : }
8857 :
8858 : /* Copy final grouped and sorted TIDs back into start of caller's array */
8859 4138 : memcpy(delstate->deltids, reordereddeltids,
8860 : sizeof(TM_IndexDelete) * ncopied);
8861 4138 : delstate->ndeltids = ncopied;
8862 :
8863 4138 : pfree(reordereddeltids);
8864 4138 : pfree(blockgroups);
8865 :
8866 4138 : return nblocksfavorable;
8867 : }
8868 :
8869 : /*
8870 : * Perform XLogInsert for a heap-visible operation. 'block' is the block
8871 : * being marked all-visible, and vm_buffer is the buffer containing the
8872 : * corresponding visibility map block. Both should have already been modified
8873 : * and dirtied.
8874 : *
8875 : * snapshotConflictHorizon comes from the largest xmin on the page being
8876 : * marked all-visible. REDO routine uses it to generate recovery conflicts.
8877 : *
8878 : * If checksums or wal_log_hints are enabled, we may also generate a full-page
8879 : * image of heap_buffer. Otherwise, we optimize away the FPI (by specifying
8880 : * REGBUF_NO_IMAGE for the heap buffer), in which case the caller should *not*
8881 : * update the heap page's LSN.
8882 : */
8883 : XLogRecPtr
8884 68416 : log_heap_visible(Relation rel, Buffer heap_buffer, Buffer vm_buffer,
8885 : TransactionId snapshotConflictHorizon, uint8 vmflags)
8886 : {
8887 : xl_heap_visible xlrec;
8888 : XLogRecPtr recptr;
8889 : uint8 flags;
8890 :
8891 : Assert(BufferIsValid(heap_buffer));
8892 : Assert(BufferIsValid(vm_buffer));
8893 :
8894 68416 : xlrec.snapshotConflictHorizon = snapshotConflictHorizon;
8895 68416 : xlrec.flags = vmflags;
8896 68416 : if (RelationIsAccessibleInLogicalDecoding(rel))
8897 112 : xlrec.flags |= VISIBILITYMAP_XLOG_CATALOG_REL;
8898 68416 : XLogBeginInsert();
8899 68416 : XLogRegisterData(&xlrec, SizeOfHeapVisible);
8900 :
8901 68416 : XLogRegisterBuffer(0, vm_buffer, 0);
8902 :
8903 68416 : flags = REGBUF_STANDARD;
8904 68416 : if (!XLogHintBitIsNeeded())
8905 6170 : flags |= REGBUF_NO_IMAGE;
8906 68416 : XLogRegisterBuffer(1, heap_buffer, flags);
8907 :
8908 68416 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_VISIBLE);
8909 :
8910 68416 : return recptr;
8911 : }
8912 :
8913 : /*
8914 : * Perform XLogInsert for a heap-update operation. Caller must already
8915 : * have modified the buffer(s) and marked them dirty.
8916 : */
8917 : static XLogRecPtr
8918 601206 : log_heap_update(Relation reln, Buffer oldbuf,
8919 : Buffer newbuf, HeapTuple oldtup, HeapTuple newtup,
8920 : HeapTuple old_key_tuple,
8921 : bool all_visible_cleared, bool new_all_visible_cleared)
8922 : {
8923 : xl_heap_update xlrec;
8924 : xl_heap_header xlhdr;
8925 : xl_heap_header xlhdr_idx;
8926 : uint8 info;
8927 : uint16 prefix_suffix[2];
8928 601206 : uint16 prefixlen = 0,
8929 601206 : suffixlen = 0;
8930 : XLogRecPtr recptr;
8931 601206 : Page page = BufferGetPage(newbuf);
8932 601206 : bool need_tuple_data = RelationIsLogicallyLogged(reln);
8933 : bool init;
8934 : int bufflags;
8935 :
8936 : /* Caller should not call me on a non-WAL-logged relation */
8937 : Assert(RelationNeedsWAL(reln));
8938 :
8939 601206 : XLogBeginInsert();
8940 :
8941 601206 : if (HeapTupleIsHeapOnly(newtup))
8942 294274 : info = XLOG_HEAP_HOT_UPDATE;
8943 : else
8944 306932 : info = XLOG_HEAP_UPDATE;
8945 :
8946 : /*
8947 : * If the old and new tuple are on the same page, we only need to log the
8948 : * parts of the new tuple that were changed. That saves on the amount of
8949 : * WAL we need to write. Currently, we just count any unchanged bytes in
8950 : * the beginning and end of the tuple. That's quick to check, and
8951 : * perfectly covers the common case that only one field is updated.
8952 : *
8953 : * We could do this even if the old and new tuple are on different pages,
8954 : * but only if we don't make a full-page image of the old page, which is
8955 : * difficult to know in advance. Also, if the old tuple is corrupt for
8956 : * some reason, it would allow the corruption to propagate the new page,
8957 : * so it seems best to avoid. Under the general assumption that most
8958 : * updates tend to create the new tuple version on the same page, there
8959 : * isn't much to be gained by doing this across pages anyway.
8960 : *
8961 : * Skip this if we're taking a full-page image of the new page, as we
8962 : * don't include the new tuple in the WAL record in that case. Also
8963 : * disable if effective_wal_level='logical', as logical decoding needs to
8964 : * be able to read the new tuple in whole from the WAL record alone.
8965 : */
8966 601206 : if (oldbuf == newbuf && !need_tuple_data &&
8967 294928 : !XLogCheckBufferNeedsBackup(newbuf))
8968 : {
8969 293666 : char *oldp = (char *) oldtup->t_data + oldtup->t_data->t_hoff;
8970 293666 : char *newp = (char *) newtup->t_data + newtup->t_data->t_hoff;
8971 293666 : int oldlen = oldtup->t_len - oldtup->t_data->t_hoff;
8972 293666 : int newlen = newtup->t_len - newtup->t_data->t_hoff;
8973 :
8974 : /* Check for common prefix between old and new tuple */
8975 24665908 : for (prefixlen = 0; prefixlen < Min(oldlen, newlen); prefixlen++)
8976 : {
8977 24613916 : if (newp[prefixlen] != oldp[prefixlen])
8978 241674 : break;
8979 : }
8980 :
8981 : /*
8982 : * Storing the length of the prefix takes 2 bytes, so we need to save
8983 : * at least 3 bytes or there's no point.
8984 : */
8985 293666 : if (prefixlen < 3)
8986 44210 : prefixlen = 0;
8987 :
8988 : /* Same for suffix */
8989 9599876 : for (suffixlen = 0; suffixlen < Min(oldlen, newlen) - prefixlen; suffixlen++)
8990 : {
8991 9547376 : if (newp[newlen - suffixlen - 1] != oldp[oldlen - suffixlen - 1])
8992 241166 : break;
8993 : }
8994 293666 : if (suffixlen < 3)
8995 73668 : suffixlen = 0;
8996 : }
8997 :
8998 : /* Prepare main WAL data chain */
8999 601206 : xlrec.flags = 0;
9000 601206 : if (all_visible_cleared)
9001 3054 : xlrec.flags |= XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED;
9002 601206 : if (new_all_visible_cleared)
9003 1616 : xlrec.flags |= XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED;
9004 601206 : if (prefixlen > 0)
9005 249456 : xlrec.flags |= XLH_UPDATE_PREFIX_FROM_OLD;
9006 601206 : if (suffixlen > 0)
9007 219998 : xlrec.flags |= XLH_UPDATE_SUFFIX_FROM_OLD;
9008 601206 : if (need_tuple_data)
9009 : {
9010 94044 : xlrec.flags |= XLH_UPDATE_CONTAINS_NEW_TUPLE;
9011 94044 : if (old_key_tuple)
9012 : {
9013 292 : if (reln->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
9014 130 : xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_TUPLE;
9015 : else
9016 162 : xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_KEY;
9017 : }
9018 : }
9019 :
9020 : /* If new tuple is the single and first tuple on page... */
9021 608452 : if (ItemPointerGetOffsetNumber(&(newtup->t_self)) == FirstOffsetNumber &&
9022 7246 : PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
9023 : {
9024 6868 : info |= XLOG_HEAP_INIT_PAGE;
9025 6868 : init = true;
9026 : }
9027 : else
9028 594338 : init = false;
9029 :
9030 : /* Prepare WAL data for the old page */
9031 601206 : xlrec.old_offnum = ItemPointerGetOffsetNumber(&oldtup->t_self);
9032 601206 : xlrec.old_xmax = HeapTupleHeaderGetRawXmax(oldtup->t_data);
9033 1202412 : xlrec.old_infobits_set = compute_infobits(oldtup->t_data->t_infomask,
9034 601206 : oldtup->t_data->t_infomask2);
9035 :
9036 : /* Prepare WAL data for the new page */
9037 601206 : xlrec.new_offnum = ItemPointerGetOffsetNumber(&newtup->t_self);
9038 601206 : xlrec.new_xmax = HeapTupleHeaderGetRawXmax(newtup->t_data);
9039 :
9040 601206 : bufflags = REGBUF_STANDARD;
9041 601206 : if (init)
9042 6868 : bufflags |= REGBUF_WILL_INIT;
9043 601206 : if (need_tuple_data)
9044 94044 : bufflags |= REGBUF_KEEP_DATA;
9045 :
9046 601206 : XLogRegisterBuffer(0, newbuf, bufflags);
9047 601206 : if (oldbuf != newbuf)
9048 282394 : XLogRegisterBuffer(1, oldbuf, REGBUF_STANDARD);
9049 :
9050 601206 : XLogRegisterData(&xlrec, SizeOfHeapUpdate);
9051 :
9052 : /*
9053 : * Prepare WAL data for the new tuple.
9054 : */
9055 601206 : if (prefixlen > 0 || suffixlen > 0)
9056 : {
9057 292738 : if (prefixlen > 0 && suffixlen > 0)
9058 : {
9059 176716 : prefix_suffix[0] = prefixlen;
9060 176716 : prefix_suffix[1] = suffixlen;
9061 176716 : XLogRegisterBufData(0, &prefix_suffix, sizeof(uint16) * 2);
9062 : }
9063 116022 : else if (prefixlen > 0)
9064 : {
9065 72740 : XLogRegisterBufData(0, &prefixlen, sizeof(uint16));
9066 : }
9067 : else
9068 : {
9069 43282 : XLogRegisterBufData(0, &suffixlen, sizeof(uint16));
9070 : }
9071 : }
9072 :
9073 601206 : xlhdr.t_infomask2 = newtup->t_data->t_infomask2;
9074 601206 : xlhdr.t_infomask = newtup->t_data->t_infomask;
9075 601206 : xlhdr.t_hoff = newtup->t_data->t_hoff;
9076 : Assert(SizeofHeapTupleHeader + prefixlen + suffixlen <= newtup->t_len);
9077 :
9078 : /*
9079 : * PG73FORMAT: write bitmap [+ padding] [+ oid] + data
9080 : *
9081 : * The 'data' doesn't include the common prefix or suffix.
9082 : */
9083 601206 : XLogRegisterBufData(0, &xlhdr, SizeOfHeapHeader);
9084 601206 : if (prefixlen == 0)
9085 : {
9086 351750 : XLogRegisterBufData(0,
9087 351750 : (char *) newtup->t_data + SizeofHeapTupleHeader,
9088 351750 : newtup->t_len - SizeofHeapTupleHeader - suffixlen);
9089 : }
9090 : else
9091 : {
9092 : /*
9093 : * Have to write the null bitmap and data after the common prefix as
9094 : * two separate rdata entries.
9095 : */
9096 : /* bitmap [+ padding] [+ oid] */
9097 249456 : if (newtup->t_data->t_hoff - SizeofHeapTupleHeader > 0)
9098 : {
9099 249456 : XLogRegisterBufData(0,
9100 249456 : (char *) newtup->t_data + SizeofHeapTupleHeader,
9101 249456 : newtup->t_data->t_hoff - SizeofHeapTupleHeader);
9102 : }
9103 :
9104 : /* data after common prefix */
9105 249456 : XLogRegisterBufData(0,
9106 249456 : (char *) newtup->t_data + newtup->t_data->t_hoff + prefixlen,
9107 249456 : newtup->t_len - newtup->t_data->t_hoff - prefixlen - suffixlen);
9108 : }
9109 :
9110 : /* We need to log a tuple identity */
9111 601206 : if (need_tuple_data && old_key_tuple)
9112 : {
9113 : /* don't really need this, but its more comfy to decode */
9114 292 : xlhdr_idx.t_infomask2 = old_key_tuple->t_data->t_infomask2;
9115 292 : xlhdr_idx.t_infomask = old_key_tuple->t_data->t_infomask;
9116 292 : xlhdr_idx.t_hoff = old_key_tuple->t_data->t_hoff;
9117 :
9118 292 : XLogRegisterData(&xlhdr_idx, SizeOfHeapHeader);
9119 :
9120 : /* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
9121 292 : XLogRegisterData((char *) old_key_tuple->t_data + SizeofHeapTupleHeader,
9122 292 : old_key_tuple->t_len - SizeofHeapTupleHeader);
9123 : }
9124 :
9125 : /* filtering by origin on a row level is much more efficient */
9126 601206 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
9127 :
9128 601206 : recptr = XLogInsert(RM_HEAP_ID, info);
9129 :
9130 601206 : return recptr;
9131 : }
9132 :
9133 : /*
9134 : * Perform XLogInsert of an XLOG_HEAP2_NEW_CID record
9135 : *
9136 : * This is only used when effective_wal_level is logical, and only for
9137 : * catalog tuples.
9138 : */
9139 : static XLogRecPtr
9140 49322 : log_heap_new_cid(Relation relation, HeapTuple tup)
9141 : {
9142 : xl_heap_new_cid xlrec;
9143 :
9144 : XLogRecPtr recptr;
9145 49322 : HeapTupleHeader hdr = tup->t_data;
9146 :
9147 : Assert(ItemPointerIsValid(&tup->t_self));
9148 : Assert(tup->t_tableOid != InvalidOid);
9149 :
9150 49322 : xlrec.top_xid = GetTopTransactionId();
9151 49322 : xlrec.target_locator = relation->rd_locator;
9152 49322 : xlrec.target_tid = tup->t_self;
9153 :
9154 : /*
9155 : * If the tuple got inserted & deleted in the same TX we definitely have a
9156 : * combo CID, set cmin and cmax.
9157 : */
9158 49322 : if (hdr->t_infomask & HEAP_COMBOCID)
9159 : {
9160 : Assert(!(hdr->t_infomask & HEAP_XMAX_INVALID));
9161 : Assert(!HeapTupleHeaderXminInvalid(hdr));
9162 4048 : xlrec.cmin = HeapTupleHeaderGetCmin(hdr);
9163 4048 : xlrec.cmax = HeapTupleHeaderGetCmax(hdr);
9164 4048 : xlrec.combocid = HeapTupleHeaderGetRawCommandId(hdr);
9165 : }
9166 : /* No combo CID, so only cmin or cmax can be set by this TX */
9167 : else
9168 : {
9169 : /*
9170 : * Tuple inserted.
9171 : *
9172 : * We need to check for LOCK ONLY because multixacts might be
9173 : * transferred to the new tuple in case of FOR KEY SHARE updates in
9174 : * which case there will be an xmax, although the tuple just got
9175 : * inserted.
9176 : */
9177 58946 : if (hdr->t_infomask & HEAP_XMAX_INVALID ||
9178 13672 : HEAP_XMAX_IS_LOCKED_ONLY(hdr->t_infomask))
9179 : {
9180 31604 : xlrec.cmin = HeapTupleHeaderGetRawCommandId(hdr);
9181 31604 : xlrec.cmax = InvalidCommandId;
9182 : }
9183 : /* Tuple from a different tx updated or deleted. */
9184 : else
9185 : {
9186 13670 : xlrec.cmin = InvalidCommandId;
9187 13670 : xlrec.cmax = HeapTupleHeaderGetRawCommandId(hdr);
9188 : }
9189 45274 : xlrec.combocid = InvalidCommandId;
9190 : }
9191 :
9192 : /*
9193 : * Note that we don't need to register the buffer here, because this
9194 : * operation does not modify the page. The insert/update/delete that
9195 : * called us certainly did, but that's WAL-logged separately.
9196 : */
9197 49322 : XLogBeginInsert();
9198 49322 : XLogRegisterData(&xlrec, SizeOfHeapNewCid);
9199 :
9200 : /* will be looked at irrespective of origin */
9201 :
9202 49322 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_NEW_CID);
9203 :
9204 49322 : return recptr;
9205 : }
9206 :
9207 : /*
9208 : * Build a heap tuple representing the configured REPLICA IDENTITY to represent
9209 : * the old tuple in an UPDATE or DELETE.
9210 : *
9211 : * Returns NULL if there's no need to log an identity or if there's no suitable
9212 : * key defined.
9213 : *
9214 : * Pass key_required true if any replica identity columns changed value, or if
9215 : * any of them have any external data. Delete must always pass true.
9216 : *
9217 : * *copy is set to true if the returned tuple is a modified copy rather than
9218 : * the same tuple that was passed in.
9219 : */
9220 : static HeapTuple
9221 3675776 : ExtractReplicaIdentity(Relation relation, HeapTuple tp, bool key_required,
9222 : bool *copy)
9223 : {
9224 3675776 : TupleDesc desc = RelationGetDescr(relation);
9225 3675776 : char replident = relation->rd_rel->relreplident;
9226 : Bitmapset *idattrs;
9227 : HeapTuple key_tuple;
9228 : bool nulls[MaxHeapAttributeNumber];
9229 : Datum values[MaxHeapAttributeNumber];
9230 :
9231 3675776 : *copy = false;
9232 :
9233 3675776 : if (!RelationIsLogicallyLogged(relation))
9234 3475192 : return NULL;
9235 :
9236 200584 : if (replident == REPLICA_IDENTITY_NOTHING)
9237 462 : return NULL;
9238 :
9239 200122 : if (replident == REPLICA_IDENTITY_FULL)
9240 : {
9241 : /*
9242 : * When logging the entire old tuple, it very well could contain
9243 : * toasted columns. If so, force them to be inlined.
9244 : */
9245 394 : if (HeapTupleHasExternal(tp))
9246 : {
9247 8 : *copy = true;
9248 8 : tp = toast_flatten_tuple(tp, desc);
9249 : }
9250 394 : return tp;
9251 : }
9252 :
9253 : /* if the key isn't required and we're only logging the key, we're done */
9254 199728 : if (!key_required)
9255 93752 : return NULL;
9256 :
9257 : /* find out the replica identity columns */
9258 105976 : idattrs = RelationGetIndexAttrBitmap(relation,
9259 : INDEX_ATTR_BITMAP_IDENTITY_KEY);
9260 :
9261 : /*
9262 : * If there's no defined replica identity columns, treat as !key_required.
9263 : * (This case should not be reachable from heap_update, since that should
9264 : * calculate key_required accurately. But heap_delete just passes
9265 : * constant true for key_required, so we can hit this case in deletes.)
9266 : */
9267 105976 : if (bms_is_empty(idattrs))
9268 12042 : return NULL;
9269 :
9270 : /*
9271 : * Construct a new tuple containing only the replica identity columns,
9272 : * with nulls elsewhere. While we're at it, assert that the replica
9273 : * identity columns aren't null.
9274 : */
9275 93934 : heap_deform_tuple(tp, desc, values, nulls);
9276 :
9277 301790 : for (int i = 0; i < desc->natts; i++)
9278 : {
9279 207856 : if (bms_is_member(i + 1 - FirstLowInvalidHeapAttributeNumber,
9280 : idattrs))
9281 : Assert(!nulls[i]);
9282 : else
9283 113898 : nulls[i] = true;
9284 : }
9285 :
9286 93934 : key_tuple = heap_form_tuple(desc, values, nulls);
9287 93934 : *copy = true;
9288 :
9289 93934 : bms_free(idattrs);
9290 :
9291 : /*
9292 : * If the tuple, which by here only contains indexed columns, still has
9293 : * toasted columns, force them to be inlined. This is somewhat unlikely
9294 : * since there's limits on the size of indexed columns, so we don't
9295 : * duplicate toast_flatten_tuple()s functionality in the above loop over
9296 : * the indexed columns, even if it would be more efficient.
9297 : */
9298 93934 : if (HeapTupleHasExternal(key_tuple))
9299 : {
9300 8 : HeapTuple oldtup = key_tuple;
9301 :
9302 8 : key_tuple = toast_flatten_tuple(oldtup, desc);
9303 8 : heap_freetuple(oldtup);
9304 : }
9305 :
9306 93934 : return key_tuple;
9307 : }
9308 :
9309 : /*
9310 : * HeapCheckForSerializableConflictOut
9311 : * We are reading a tuple. If it's not visible, there may be a
9312 : * rw-conflict out with the inserter. Otherwise, if it is visible to us
9313 : * but has been deleted, there may be a rw-conflict out with the deleter.
9314 : *
9315 : * We will determine the top level xid of the writing transaction with which
9316 : * we may be in conflict, and ask CheckForSerializableConflictOut() to check
9317 : * for overlap with our own transaction.
9318 : *
9319 : * This function should be called just about anywhere in heapam.c where a
9320 : * tuple has been read. The caller must hold at least a shared lock on the
9321 : * buffer, because this function might set hint bits on the tuple. There is
9322 : * currently no known reason to call this function from an index AM.
9323 : */
9324 : void
9325 63037244 : HeapCheckForSerializableConflictOut(bool visible, Relation relation,
9326 : HeapTuple tuple, Buffer buffer,
9327 : Snapshot snapshot)
9328 : {
9329 : TransactionId xid;
9330 : HTSV_Result htsvResult;
9331 :
9332 63037244 : if (!CheckForSerializableConflictOutNeeded(relation, snapshot))
9333 62986496 : return;
9334 :
9335 : /*
9336 : * Check to see whether the tuple has been written to by a concurrent
9337 : * transaction, either to create it not visible to us, or to delete it
9338 : * while it is visible to us. The "visible" bool indicates whether the
9339 : * tuple is visible to us, while HeapTupleSatisfiesVacuum checks what else
9340 : * is going on with it.
9341 : *
9342 : * In the event of a concurrently inserted tuple that also happens to have
9343 : * been concurrently updated (by a separate transaction), the xmin of the
9344 : * tuple will be used -- not the updater's xid.
9345 : */
9346 50748 : htsvResult = HeapTupleSatisfiesVacuum(tuple, TransactionXmin, buffer);
9347 50748 : switch (htsvResult)
9348 : {
9349 49122 : case HEAPTUPLE_LIVE:
9350 49122 : if (visible)
9351 49096 : return;
9352 26 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9353 26 : break;
9354 722 : case HEAPTUPLE_RECENTLY_DEAD:
9355 : case HEAPTUPLE_DELETE_IN_PROGRESS:
9356 722 : if (visible)
9357 570 : xid = HeapTupleHeaderGetUpdateXid(tuple->t_data);
9358 : else
9359 152 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9360 :
9361 722 : if (TransactionIdPrecedes(xid, TransactionXmin))
9362 : {
9363 : /* This is like the HEAPTUPLE_DEAD case */
9364 : Assert(!visible);
9365 134 : return;
9366 : }
9367 588 : break;
9368 656 : case HEAPTUPLE_INSERT_IN_PROGRESS:
9369 656 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9370 656 : break;
9371 248 : case HEAPTUPLE_DEAD:
9372 : Assert(!visible);
9373 248 : return;
9374 0 : default:
9375 :
9376 : /*
9377 : * The only way to get to this default clause is if a new value is
9378 : * added to the enum type without adding it to this switch
9379 : * statement. That's a bug, so elog.
9380 : */
9381 0 : elog(ERROR, "unrecognized return value from HeapTupleSatisfiesVacuum: %u", htsvResult);
9382 :
9383 : /*
9384 : * In spite of having all enum values covered and calling elog on
9385 : * this default, some compilers think this is a code path which
9386 : * allows xid to be used below without initialization. Silence
9387 : * that warning.
9388 : */
9389 : xid = InvalidTransactionId;
9390 : }
9391 :
9392 : Assert(TransactionIdIsValid(xid));
9393 : Assert(TransactionIdFollowsOrEquals(xid, TransactionXmin));
9394 :
9395 : /*
9396 : * Find top level xid. Bail out if xid is too early to be a conflict, or
9397 : * if it's our own xid.
9398 : */
9399 1270 : if (TransactionIdEquals(xid, GetTopTransactionIdIfAny()))
9400 128 : return;
9401 1142 : xid = SubTransGetTopmostTransaction(xid);
9402 1142 : if (TransactionIdPrecedes(xid, TransactionXmin))
9403 0 : return;
9404 :
9405 1142 : CheckForSerializableConflictOut(relation, xid, snapshot);
9406 : }
|
