Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * heapam.c
4 : * heap access method code
5 : *
6 : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/access/heap/heapam.c
12 : *
13 : *
14 : * INTERFACE ROUTINES
15 : * heap_beginscan - begin relation scan
16 : * heap_rescan - restart a relation scan
17 : * heap_endscan - end relation scan
18 : * heap_getnext - retrieve next tuple in scan
19 : * heap_fetch - retrieve tuple with given tid
20 : * heap_insert - insert tuple into a relation
21 : * heap_multi_insert - insert multiple tuples into a relation
22 : * heap_delete - delete a tuple from a relation
23 : * heap_update - replace a tuple in a relation with another tuple
24 : *
25 : * NOTES
26 : * This file contains the heap_ routines which implement
27 : * the POSTGRES heap access method used for all POSTGRES
28 : * relations.
29 : *
30 : *-------------------------------------------------------------------------
31 : */
32 : #include "postgres.h"
33 :
34 : #include "access/heapam.h"
35 : #include "access/heaptoast.h"
36 : #include "access/hio.h"
37 : #include "access/multixact.h"
38 : #include "access/subtrans.h"
39 : #include "access/syncscan.h"
40 : #include "access/valid.h"
41 : #include "access/visibilitymap.h"
42 : #include "access/xloginsert.h"
43 : #include "catalog/pg_database.h"
44 : #include "catalog/pg_database_d.h"
45 : #include "commands/vacuum.h"
46 : #include "pgstat.h"
47 : #include "port/pg_bitutils.h"
48 : #include "storage/lmgr.h"
49 : #include "storage/predicate.h"
50 : #include "storage/proc.h"
51 : #include "storage/procarray.h"
52 : #include "utils/datum.h"
53 : #include "utils/injection_point.h"
54 : #include "utils/inval.h"
55 : #include "utils/spccache.h"
56 : #include "utils/syscache.h"
57 :
58 :
59 : static HeapTuple heap_prepare_insert(Relation relation, HeapTuple tup,
60 : TransactionId xid, CommandId cid, uint32 options);
61 : static XLogRecPtr log_heap_update(Relation reln, Buffer oldbuf,
62 : Buffer newbuf, HeapTuple oldtup,
63 : HeapTuple newtup, HeapTuple old_key_tuple,
64 : bool all_visible_cleared, bool new_all_visible_cleared,
65 : bool walLogical);
66 : #ifdef USE_ASSERT_CHECKING
67 : static void check_lock_if_inplace_updateable_rel(Relation relation,
68 : const ItemPointerData *otid,
69 : HeapTuple newtup);
70 : static void check_inplace_rel_lock(HeapTuple oldtup);
71 : #endif
72 : static Bitmapset *HeapDetermineColumnsInfo(Relation relation,
73 : Bitmapset *interesting_cols,
74 : Bitmapset *external_cols,
75 : HeapTuple oldtup, HeapTuple newtup,
76 : bool *has_external);
77 : static bool heap_acquire_tuplock(Relation relation, const ItemPointerData *tid,
78 : LockTupleMode mode, LockWaitPolicy wait_policy,
79 : bool *have_tuple_lock);
80 : static inline BlockNumber heapgettup_advance_block(HeapScanDesc scan,
81 : BlockNumber block,
82 : ScanDirection dir);
83 : static pg_noinline BlockNumber heapgettup_initial_block(HeapScanDesc scan,
84 : ScanDirection dir);
85 : static void compute_new_xmax_infomask(TransactionId xmax, uint16 old_infomask,
86 : uint16 old_infomask2, TransactionId add_to_xmax,
87 : LockTupleMode mode, bool is_update,
88 : TransactionId *result_xmax, uint16 *result_infomask,
89 : uint16 *result_infomask2);
90 : static TM_Result heap_lock_updated_tuple(Relation rel,
91 : uint16 prior_infomask,
92 : TransactionId prior_raw_xmax,
93 : const ItemPointerData *prior_ctid,
94 : TransactionId xid,
95 : LockTupleMode mode);
96 : static void GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
97 : uint16 *new_infomask2);
98 : static TransactionId MultiXactIdGetUpdateXid(TransactionId xmax,
99 : uint16 t_infomask);
100 : static bool DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
101 : LockTupleMode lockmode, bool *current_is_member);
102 : static void MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
103 : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
104 : int *remaining);
105 : static bool ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status,
106 : uint16 infomask, Relation rel, int *remaining,
107 : bool logLockFailure);
108 : static void index_delete_sort(TM_IndexDeleteOp *delstate);
109 : static int bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate);
110 : static XLogRecPtr log_heap_new_cid(Relation relation, HeapTuple tup);
111 : static HeapTuple ExtractReplicaIdentity(Relation relation, HeapTuple tp, bool key_required,
112 : bool *copy);
113 :
114 :
115 : /*
116 : * This table lists the heavyweight lock mode that corresponds to each tuple
117 : * lock mode, as well as one or two corresponding MultiXactStatus values:
118 : * .lockstatus to merely lock tuples, and .updstatus to update them. The
119 : * latter is set to -1 if the corresponding tuple lock mode does not allow
120 : * updating tuples -- see get_mxact_status_for_lock().
121 : *
122 : * These interact with InplaceUpdateTupleLock, an alias for ExclusiveLock.
123 : *
124 : * Don't look at lockstatus/updstatus directly! Use get_mxact_status_for_lock
125 : * instead.
126 : */
127 : static const struct
128 : {
129 : LOCKMODE hwlock;
130 : int lockstatus;
131 : int updstatus;
132 : } tupleLockExtraInfo[] =
133 :
134 : {
135 : [LockTupleKeyShare] = {
136 : .hwlock = AccessShareLock,
137 : .lockstatus = MultiXactStatusForKeyShare,
138 : /* KeyShare does not allow updating tuples */
139 : .updstatus = -1
140 : },
141 : [LockTupleShare] = {
142 : .hwlock = RowShareLock,
143 : .lockstatus = MultiXactStatusForShare,
144 : /* Share does not allow updating tuples */
145 : .updstatus = -1
146 : },
147 : [LockTupleNoKeyExclusive] = {
148 : .hwlock = ExclusiveLock,
149 : .lockstatus = MultiXactStatusForNoKeyUpdate,
150 : .updstatus = MultiXactStatusNoKeyUpdate
151 : },
152 : [LockTupleExclusive] = {
153 : .hwlock = AccessExclusiveLock,
154 : .lockstatus = MultiXactStatusForUpdate,
155 : .updstatus = MultiXactStatusUpdate
156 : }
157 : };
158 :
159 : /* Get the LOCKMODE for a given MultiXactStatus */
160 : #define LOCKMODE_from_mxstatus(status) \
161 : (tupleLockExtraInfo[TUPLOCK_from_mxstatus((status))].hwlock)
162 :
163 : /*
164 : * Acquire heavyweight locks on tuples, using a LockTupleMode strength value.
165 : * This is more readable than having every caller translate it to lock.h's
166 : * LOCKMODE.
167 : */
168 : #define LockTupleTuplock(rel, tup, mode) \
169 : LockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
170 : #define UnlockTupleTuplock(rel, tup, mode) \
171 : UnlockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
172 : #define ConditionalLockTupleTuplock(rel, tup, mode, log) \
173 : ConditionalLockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock, (log))
174 :
175 : #ifdef USE_PREFETCH
176 : /*
177 : * heap_index_delete_tuples and index_delete_prefetch_buffer use this
178 : * structure to coordinate prefetching activity
179 : */
180 : typedef struct
181 : {
182 : BlockNumber cur_hblkno;
183 : int next_item;
184 : int ndeltids;
185 : TM_IndexDelete *deltids;
186 : } IndexDeletePrefetchState;
187 : #endif
188 :
189 : /* heap_index_delete_tuples bottom-up index deletion costing constants */
190 : #define BOTTOMUP_MAX_NBLOCKS 6
191 : #define BOTTOMUP_TOLERANCE_NBLOCKS 3
192 :
193 : /*
194 : * heap_index_delete_tuples uses this when determining which heap blocks it
195 : * must visit to help its bottom-up index deletion caller
196 : */
197 : typedef struct IndexDeleteCounts
198 : {
199 : int16 npromisingtids; /* Number of "promising" TIDs in group */
200 : int16 ntids; /* Number of TIDs in group */
201 : int16 ifirsttid; /* Offset to group's first deltid */
202 : } IndexDeleteCounts;
203 :
204 : /*
205 : * This table maps tuple lock strength values for each particular
206 : * MultiXactStatus value.
207 : */
208 : static const int MultiXactStatusLock[MaxMultiXactStatus + 1] =
209 : {
210 : LockTupleKeyShare, /* ForKeyShare */
211 : LockTupleShare, /* ForShare */
212 : LockTupleNoKeyExclusive, /* ForNoKeyUpdate */
213 : LockTupleExclusive, /* ForUpdate */
214 : LockTupleNoKeyExclusive, /* NoKeyUpdate */
215 : LockTupleExclusive /* Update */
216 : };
217 :
218 : /* Get the LockTupleMode for a given MultiXactStatus */
219 : #define TUPLOCK_from_mxstatus(status) \
220 : (MultiXactStatusLock[(status)])
221 :
222 : /*
223 : * Check that we have a valid snapshot if we might need TOAST access.
224 : */
225 : static inline void
226 16186511 : AssertHasSnapshotForToast(Relation rel)
227 : {
228 : #ifdef USE_ASSERT_CHECKING
229 :
230 : /* bootstrap mode in particular breaks this rule */
231 : if (!IsNormalProcessingMode())
232 : return;
233 :
234 : /* if the relation doesn't have a TOAST table, we are good */
235 : if (!OidIsValid(rel->rd_rel->reltoastrelid))
236 : return;
237 :
238 : Assert(HaveRegisteredOrActiveSnapshot());
239 :
240 : #endif /* USE_ASSERT_CHECKING */
241 16186511 : }
242 :
243 : /* ----------------------------------------------------------------
244 : * heap support routines
245 : * ----------------------------------------------------------------
246 : */
247 :
248 : /*
249 : * Streaming read API callback for parallel sequential scans. Returns the next
250 : * block the caller wants from the read stream or InvalidBlockNumber when done.
251 : */
252 : static BlockNumber
253 146246 : heap_scan_stream_read_next_parallel(ReadStream *stream,
254 : void *callback_private_data,
255 : void *per_buffer_data)
256 : {
257 146246 : HeapScanDesc scan = (HeapScanDesc) callback_private_data;
258 :
259 : Assert(ScanDirectionIsForward(scan->rs_dir));
260 : Assert(scan->rs_base.rs_parallel);
261 :
262 146246 : if (unlikely(!scan->rs_inited))
263 : {
264 : /* parallel scan */
265 2771 : table_block_parallelscan_startblock_init(scan->rs_base.rs_rd,
266 2771 : scan->rs_parallelworkerdata,
267 2771 : (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel,
268 : scan->rs_startblock,
269 : scan->rs_numblocks);
270 :
271 : /* may return InvalidBlockNumber if there are no more blocks */
272 5542 : scan->rs_prefetch_block = table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
273 2771 : scan->rs_parallelworkerdata,
274 2771 : (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel);
275 2771 : scan->rs_inited = true;
276 : }
277 : else
278 : {
279 143475 : scan->rs_prefetch_block = table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
280 143475 : scan->rs_parallelworkerdata, (ParallelBlockTableScanDesc)
281 143475 : scan->rs_base.rs_parallel);
282 : }
283 :
284 146246 : return scan->rs_prefetch_block;
285 : }
286 :
287 : /*
288 : * Streaming read API callback for serial sequential and TID range scans.
289 : * Returns the next block the caller wants from the read stream or
290 : * InvalidBlockNumber when done.
291 : */
292 : static BlockNumber
293 5092592 : heap_scan_stream_read_next_serial(ReadStream *stream,
294 : void *callback_private_data,
295 : void *per_buffer_data)
296 : {
297 5092592 : HeapScanDesc scan = (HeapScanDesc) callback_private_data;
298 :
299 5092592 : if (unlikely(!scan->rs_inited))
300 : {
301 1326560 : scan->rs_prefetch_block = heapgettup_initial_block(scan, scan->rs_dir);
302 1326560 : scan->rs_inited = true;
303 : }
304 : else
305 3766032 : scan->rs_prefetch_block = heapgettup_advance_block(scan,
306 : scan->rs_prefetch_block,
307 : scan->rs_dir);
308 :
309 5092592 : return scan->rs_prefetch_block;
310 : }
311 :
312 : /*
313 : * Read stream API callback for bitmap heap scans.
314 : * Returns the next block the caller wants from the read stream or
315 : * InvalidBlockNumber when done.
316 : */
317 : static BlockNumber
318 263892 : bitmapheap_stream_read_next(ReadStream *pgsr, void *private_data,
319 : void *per_buffer_data)
320 : {
321 263892 : TBMIterateResult *tbmres = per_buffer_data;
322 263892 : BitmapHeapScanDesc bscan = (BitmapHeapScanDesc) private_data;
323 263892 : HeapScanDesc hscan = (HeapScanDesc) bscan;
324 263892 : TableScanDesc sscan = &hscan->rs_base;
325 :
326 : for (;;)
327 : {
328 263892 : CHECK_FOR_INTERRUPTS();
329 :
330 : /* no more entries in the bitmap */
331 263892 : if (!tbm_iterate(&sscan->st.rs_tbmiterator, tbmres))
332 15932 : return InvalidBlockNumber;
333 :
334 : /*
335 : * Ignore any claimed entries past what we think is the end of the
336 : * relation. It may have been extended after the start of our scan (we
337 : * only hold an AccessShareLock, and it could be inserts from this
338 : * backend). We don't take this optimization in SERIALIZABLE
339 : * isolation though, as we need to examine all invisible tuples
340 : * reachable by the index.
341 : */
342 247960 : if (!IsolationIsSerializable() &&
343 247851 : tbmres->blockno >= hscan->rs_nblocks)
344 0 : continue;
345 :
346 247960 : return tbmres->blockno;
347 : }
348 :
349 : /* not reachable */
350 : Assert(false);
351 : }
352 :
353 : /* ----------------
354 : * initscan - scan code common to heap_beginscan and heap_rescan
355 : * ----------------
356 : */
357 : static void
358 1358827 : initscan(HeapScanDesc scan, ScanKey key, bool keep_startblock)
359 : {
360 1358827 : ParallelBlockTableScanDesc bpscan = NULL;
361 : bool allow_strat;
362 : bool allow_sync;
363 :
364 : /*
365 : * Determine the number of blocks we have to scan.
366 : *
367 : * It is sufficient to do this once at scan start, since any tuples added
368 : * while the scan is in progress will be invisible to my snapshot anyway.
369 : * (That is not true when using a non-MVCC snapshot. However, we couldn't
370 : * guarantee to return tuples added after scan start anyway, since they
371 : * might go into pages we already scanned. To guarantee consistent
372 : * results for a non-MVCC snapshot, the caller must hold some higher-level
373 : * lock that ensures the interesting tuple(s) won't change.)
374 : */
375 1358827 : if (scan->rs_base.rs_parallel != NULL)
376 : {
377 4672 : bpscan = (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel;
378 4672 : scan->rs_nblocks = bpscan->phs_nblocks;
379 : }
380 : else
381 1354155 : scan->rs_nblocks = RelationGetNumberOfBlocks(scan->rs_base.rs_rd);
382 :
383 : /*
384 : * If the table is large relative to NBuffers, use a bulk-read access
385 : * strategy and enable synchronized scanning (see syncscan.c). Although
386 : * the thresholds for these features could be different, we make them the
387 : * same so that there are only two behaviors to tune rather than four.
388 : * (However, some callers need to be able to disable one or both of these
389 : * behaviors, independently of the size of the table; also there is a GUC
390 : * variable that can disable synchronized scanning.)
391 : *
392 : * Note that table_block_parallelscan_initialize has a very similar test;
393 : * if you change this, consider changing that one, too.
394 : */
395 1358825 : if (!RelationUsesLocalBuffers(scan->rs_base.rs_rd) &&
396 1349045 : scan->rs_nblocks > NBuffers / 4)
397 : {
398 14997 : allow_strat = (scan->rs_base.rs_flags & SO_ALLOW_STRAT) != 0;
399 14997 : allow_sync = (scan->rs_base.rs_flags & SO_ALLOW_SYNC) != 0;
400 : }
401 : else
402 1343828 : allow_strat = allow_sync = false;
403 :
404 1358825 : if (allow_strat)
405 : {
406 : /* During a rescan, keep the previous strategy object. */
407 13642 : if (scan->rs_strategy == NULL)
408 13447 : scan->rs_strategy = GetAccessStrategy(BAS_BULKREAD);
409 : }
410 : else
411 : {
412 1345183 : if (scan->rs_strategy != NULL)
413 0 : FreeAccessStrategy(scan->rs_strategy);
414 1345183 : scan->rs_strategy = NULL;
415 : }
416 :
417 1358825 : if (scan->rs_base.rs_parallel != NULL)
418 : {
419 : /* For parallel scan, believe whatever ParallelTableScanDesc says. */
420 4672 : if (scan->rs_base.rs_parallel->phs_syncscan)
421 5 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
422 : else
423 4667 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
424 :
425 : /*
426 : * If not rescanning, initialize the startblock. Finding the actual
427 : * start location is done in table_block_parallelscan_startblock_init,
428 : * based on whether an alternative start location has been set with
429 : * heap_setscanlimits, or using the syncscan location, when syncscan
430 : * is enabled.
431 : */
432 4672 : if (!keep_startblock)
433 4520 : scan->rs_startblock = InvalidBlockNumber;
434 : }
435 : else
436 : {
437 1354153 : if (keep_startblock)
438 : {
439 : /*
440 : * When rescanning, we want to keep the previous startblock
441 : * setting, so that rewinding a cursor doesn't generate surprising
442 : * results. Reset the active syncscan setting, though.
443 : */
444 870610 : if (allow_sync && synchronize_seqscans)
445 50 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
446 : else
447 870560 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
448 : }
449 483543 : else if (allow_sync && synchronize_seqscans)
450 : {
451 78 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
452 78 : scan->rs_startblock = ss_get_location(scan->rs_base.rs_rd, scan->rs_nblocks);
453 : }
454 : else
455 : {
456 483465 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
457 483465 : scan->rs_startblock = 0;
458 : }
459 : }
460 :
461 1358825 : scan->rs_numblocks = InvalidBlockNumber;
462 1358825 : scan->rs_inited = false;
463 1358825 : scan->rs_ctup.t_data = NULL;
464 1358825 : ItemPointerSetInvalid(&scan->rs_ctup.t_self);
465 1358825 : scan->rs_cbuf = InvalidBuffer;
466 1358825 : scan->rs_cblock = InvalidBlockNumber;
467 1358825 : scan->rs_ntuples = 0;
468 1358825 : scan->rs_cindex = 0;
469 :
470 : /*
471 : * Initialize to ForwardScanDirection because it is most common and
472 : * because heap scans go forward before going backward (e.g. CURSORs).
473 : */
474 1358825 : scan->rs_dir = ForwardScanDirection;
475 1358825 : scan->rs_prefetch_block = InvalidBlockNumber;
476 :
477 : /* page-at-a-time fields are always invalid when not rs_inited */
478 :
479 : /*
480 : * copy the scan key, if appropriate
481 : */
482 1358825 : if (key != NULL && scan->rs_base.rs_nkeys > 0)
483 267671 : memcpy(scan->rs_base.rs_key, key, scan->rs_base.rs_nkeys * sizeof(ScanKeyData));
484 :
485 : /*
486 : * Currently, we only have a stats counter for sequential heap scans (but
487 : * e.g for bitmap scans the underlying bitmap index scans will be counted,
488 : * and for sample scans we update stats for tuple fetches).
489 : */
490 1358825 : if (scan->rs_base.rs_flags & SO_TYPE_SEQSCAN)
491 1329897 : pgstat_count_heap_scan(scan->rs_base.rs_rd);
492 1358825 : }
493 :
494 : /*
495 : * heap_setscanlimits - restrict range of a heapscan
496 : *
497 : * startBlk is the page to start at
498 : * numBlks is number of pages to scan (InvalidBlockNumber means "all")
499 : */
500 : void
501 3274 : heap_setscanlimits(TableScanDesc sscan, BlockNumber startBlk, BlockNumber numBlks)
502 : {
503 3274 : HeapScanDesc scan = (HeapScanDesc) sscan;
504 :
505 : Assert(!scan->rs_inited); /* else too late to change */
506 : /* else rs_startblock is significant */
507 : Assert(!(scan->rs_base.rs_flags & SO_ALLOW_SYNC));
508 :
509 : /* Check startBlk is valid (but allow case of zero blocks...) */
510 : Assert(startBlk == 0 || startBlk < scan->rs_nblocks);
511 :
512 3274 : scan->rs_startblock = startBlk;
513 3274 : scan->rs_numblocks = numBlks;
514 3274 : }
515 :
516 : /*
517 : * Per-tuple loop for heap_prepare_pagescan(). Pulled out so it can be called
518 : * multiple times, with constant arguments for all_visible,
519 : * check_serializable.
520 : */
521 : pg_attribute_always_inline
522 : static int
523 3778545 : page_collect_tuples(HeapScanDesc scan, Snapshot snapshot,
524 : Page page, Buffer buffer,
525 : BlockNumber block, int lines,
526 : bool all_visible, bool check_serializable)
527 : {
528 3778545 : Oid relid = RelationGetRelid(scan->rs_base.rs_rd);
529 3778545 : int ntup = 0;
530 3778545 : int nvis = 0;
531 : BatchMVCCState batchmvcc;
532 :
533 : /* page at a time should have been disabled otherwise */
534 : Assert(IsMVCCSnapshot(snapshot));
535 :
536 : /* first find all tuples on the page */
537 202750219 : for (OffsetNumber lineoff = FirstOffsetNumber; lineoff <= lines; lineoff++)
538 : {
539 198971674 : ItemId lpp = PageGetItemId(page, lineoff);
540 : HeapTuple tup;
541 :
542 198971674 : if (unlikely(!ItemIdIsNormal(lpp)))
543 52356454 : continue;
544 :
545 : /*
546 : * If the page is not all-visible or we need to check serializability,
547 : * maintain enough state to be able to refind the tuple efficiently,
548 : * without again first needing to fetch the item and then via that the
549 : * tuple.
550 : */
551 146615220 : if (!all_visible || check_serializable)
552 : {
553 83999474 : tup = &batchmvcc.tuples[ntup];
554 :
555 83999474 : tup->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
556 83999474 : tup->t_len = ItemIdGetLength(lpp);
557 83999474 : tup->t_tableOid = relid;
558 83999474 : ItemPointerSet(&(tup->t_self), block, lineoff);
559 : }
560 :
561 : /*
562 : * If the page is all visible, these fields otherwise won't be
563 : * populated in loop below.
564 : */
565 146615220 : if (all_visible)
566 : {
567 62615746 : if (check_serializable)
568 : {
569 0 : batchmvcc.visible[ntup] = true;
570 : }
571 62615746 : scan->rs_vistuples[ntup] = lineoff;
572 : }
573 :
574 146615220 : ntup++;
575 : }
576 :
577 : Assert(ntup <= MaxHeapTuplesPerPage);
578 :
579 : /*
580 : * Unless the page is all visible, test visibility for all tuples one go.
581 : * That is considerably more efficient than calling
582 : * HeapTupleSatisfiesMVCC() one-by-one.
583 : */
584 3778545 : if (all_visible)
585 1314806 : nvis = ntup;
586 : else
587 2463739 : nvis = HeapTupleSatisfiesMVCCBatch(snapshot, buffer,
588 : ntup,
589 : &batchmvcc,
590 2463739 : scan->rs_vistuples);
591 :
592 : /*
593 : * So far we don't have batch API for testing serializabilty, so do so
594 : * one-by-one.
595 : */
596 3778545 : if (check_serializable)
597 : {
598 2087 : for (int i = 0; i < ntup; i++)
599 : {
600 1464 : HeapCheckForSerializableConflictOut(batchmvcc.visible[i],
601 : scan->rs_base.rs_rd,
602 : &batchmvcc.tuples[i],
603 : buffer, snapshot);
604 : }
605 : }
606 :
607 3778537 : return nvis;
608 : }
609 :
610 : /*
611 : * heap_prepare_pagescan - Prepare current scan page to be scanned in pagemode
612 : *
613 : * Preparation currently consists of 1. prune the scan's rs_cbuf page, and 2.
614 : * fill the rs_vistuples[] array with the OffsetNumbers of visible tuples.
615 : */
616 : void
617 3778545 : heap_prepare_pagescan(TableScanDesc sscan)
618 : {
619 3778545 : HeapScanDesc scan = (HeapScanDesc) sscan;
620 3778545 : Buffer buffer = scan->rs_cbuf;
621 3778545 : BlockNumber block = scan->rs_cblock;
622 : Snapshot snapshot;
623 : Page page;
624 : int lines;
625 : bool all_visible;
626 : bool check_serializable;
627 :
628 : Assert(BufferGetBlockNumber(buffer) == block);
629 :
630 : /* ensure we're not accidentally being used when not in pagemode */
631 : Assert(scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE);
632 3778545 : snapshot = scan->rs_base.rs_snapshot;
633 :
634 : /*
635 : * Prune and repair fragmentation for the whole page, if possible.
636 : */
637 3778545 : heap_page_prune_opt(scan->rs_base.rs_rd, buffer, &scan->rs_vmbuffer,
638 3778545 : sscan->rs_flags & SO_HINT_REL_READ_ONLY);
639 :
640 : /*
641 : * We must hold share lock on the buffer content while examining tuple
642 : * visibility. Afterwards, however, the tuples we have found to be
643 : * visible are guaranteed good as long as we hold the buffer pin.
644 : */
645 3778545 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
646 :
647 3778545 : page = BufferGetPage(buffer);
648 3778545 : lines = PageGetMaxOffsetNumber(page);
649 :
650 : /*
651 : * If the all-visible flag indicates that all tuples on the page are
652 : * visible to everyone, we can skip the per-tuple visibility tests.
653 : *
654 : * Note: In hot standby, a tuple that's already visible to all
655 : * transactions on the primary might still be invisible to a read-only
656 : * transaction in the standby. We partly handle this problem by tracking
657 : * the minimum xmin of visible tuples as the cut-off XID while marking a
658 : * page all-visible on the primary and WAL log that along with the
659 : * visibility map SET operation. In hot standby, we wait for (or abort)
660 : * all transactions that can potentially may not see one or more tuples on
661 : * the page. That's how index-only scans work fine in hot standby. A
662 : * crucial difference between index-only scans and heap scans is that the
663 : * index-only scan completely relies on the visibility map where as heap
664 : * scan looks at the page-level PD_ALL_VISIBLE flag. We are not sure if
665 : * the page-level flag can be trusted in the same way, because it might
666 : * get propagated somehow without being explicitly WAL-logged, e.g. via a
667 : * full page write. Until we can prove that beyond doubt, let's check each
668 : * tuple for visibility the hard way.
669 : */
670 3778545 : all_visible = PageIsAllVisible(page) && !snapshot->takenDuringRecovery;
671 : check_serializable =
672 3778545 : CheckForSerializableConflictOutNeeded(scan->rs_base.rs_rd, snapshot);
673 :
674 : /*
675 : * We call page_collect_tuples() with constant arguments, to get the
676 : * compiler to constant fold the constant arguments. Separate calls with
677 : * constant arguments, rather than variables, are needed on several
678 : * compilers to actually perform constant folding.
679 : */
680 3778545 : if (likely(all_visible))
681 : {
682 1314806 : if (likely(!check_serializable))
683 1314806 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
684 : block, lines, true, false);
685 : else
686 0 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
687 : block, lines, true, true);
688 : }
689 : else
690 : {
691 2463739 : if (likely(!check_serializable))
692 2463108 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
693 : block, lines, false, false);
694 : else
695 631 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
696 : block, lines, false, true);
697 : }
698 :
699 3778537 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
700 3778537 : }
701 :
702 : /*
703 : * heap_fetch_next_buffer - read and pin the next block from MAIN_FORKNUM.
704 : *
705 : * Read the next block of the scan relation from the read stream and save it
706 : * in the scan descriptor. It is already pinned.
707 : */
708 : static inline void
709 5021919 : heap_fetch_next_buffer(HeapScanDesc scan, ScanDirection dir)
710 : {
711 : Assert(scan->rs_read_stream);
712 :
713 : /* release previous scan buffer, if any */
714 5021919 : if (BufferIsValid(scan->rs_cbuf))
715 : {
716 3692586 : ReleaseBuffer(scan->rs_cbuf);
717 3692586 : scan->rs_cbuf = InvalidBuffer;
718 : }
719 :
720 : /*
721 : * Be sure to check for interrupts at least once per page. Checks at
722 : * higher code levels won't be able to stop a seqscan that encounters many
723 : * pages' worth of consecutive dead tuples.
724 : */
725 5021919 : CHECK_FOR_INTERRUPTS();
726 :
727 : /*
728 : * If the scan direction is changing, reset the prefetch block to the
729 : * current block. Otherwise, we will incorrectly prefetch the blocks
730 : * between the prefetch block and the current block again before
731 : * prefetching blocks in the new, correct scan direction.
732 : */
733 5021916 : if (unlikely(scan->rs_dir != dir))
734 : {
735 102 : scan->rs_prefetch_block = scan->rs_cblock;
736 102 : read_stream_reset(scan->rs_read_stream);
737 : }
738 :
739 5021916 : scan->rs_dir = dir;
740 :
741 5021916 : scan->rs_cbuf = read_stream_next_buffer(scan->rs_read_stream, NULL);
742 5021888 : if (BufferIsValid(scan->rs_cbuf))
743 3889068 : scan->rs_cblock = BufferGetBlockNumber(scan->rs_cbuf);
744 5021888 : }
745 :
746 : /*
747 : * heapgettup_initial_block - return the first BlockNumber to scan
748 : *
749 : * Returns InvalidBlockNumber when there are no blocks to scan. This can
750 : * occur with empty tables and in parallel scans when parallel workers get all
751 : * of the pages before we can get a chance to get our first page.
752 : */
753 : static pg_noinline BlockNumber
754 1326560 : heapgettup_initial_block(HeapScanDesc scan, ScanDirection dir)
755 : {
756 : Assert(!scan->rs_inited);
757 : Assert(scan->rs_base.rs_parallel == NULL);
758 :
759 : /* When there are no pages to scan, return InvalidBlockNumber */
760 1326560 : if (scan->rs_nblocks == 0 || scan->rs_numblocks == 0)
761 766493 : return InvalidBlockNumber;
762 :
763 560067 : if (ScanDirectionIsForward(dir))
764 : {
765 560025 : return scan->rs_startblock;
766 : }
767 : else
768 : {
769 : /*
770 : * Disable reporting to syncscan logic in a backwards scan; it's not
771 : * very likely anyone else is doing the same thing at the same time,
772 : * and much more likely that we'll just bollix things for forward
773 : * scanners.
774 : */
775 42 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
776 :
777 : /*
778 : * Start from last page of the scan. Ensure we take into account
779 : * rs_numblocks if it's been adjusted by heap_setscanlimits().
780 : */
781 42 : if (scan->rs_numblocks != InvalidBlockNumber)
782 4 : return (scan->rs_startblock + scan->rs_numblocks - 1) % scan->rs_nblocks;
783 :
784 38 : if (scan->rs_startblock > 0)
785 0 : return scan->rs_startblock - 1;
786 :
787 38 : return scan->rs_nblocks - 1;
788 : }
789 : }
790 :
791 :
792 : /*
793 : * heapgettup_start_page - helper function for heapgettup()
794 : *
795 : * Return the next page to scan based on the scan->rs_cbuf and set *linesleft
796 : * to the number of tuples on this page. Also set *lineoff to the first
797 : * offset to scan with forward scans getting the first offset and backward
798 : * getting the final offset on the page.
799 : */
800 : static Page
801 116209 : heapgettup_start_page(HeapScanDesc scan, ScanDirection dir, int *linesleft,
802 : OffsetNumber *lineoff)
803 : {
804 : Page page;
805 :
806 : Assert(scan->rs_inited);
807 : Assert(BufferIsValid(scan->rs_cbuf));
808 :
809 : /* Caller is responsible for ensuring buffer is locked if needed */
810 116209 : page = BufferGetPage(scan->rs_cbuf);
811 :
812 116209 : *linesleft = PageGetMaxOffsetNumber(page) - FirstOffsetNumber + 1;
813 :
814 116209 : if (ScanDirectionIsForward(dir))
815 116209 : *lineoff = FirstOffsetNumber;
816 : else
817 0 : *lineoff = (OffsetNumber) (*linesleft);
818 :
819 : /* lineoff now references the physically previous or next tid */
820 116209 : return page;
821 : }
822 :
823 :
824 : /*
825 : * heapgettup_continue_page - helper function for heapgettup()
826 : *
827 : * Return the next page to scan based on the scan->rs_cbuf and set *linesleft
828 : * to the number of tuples left to scan on this page. Also set *lineoff to
829 : * the next offset to scan according to the ScanDirection in 'dir'.
830 : */
831 : static inline Page
832 9266823 : heapgettup_continue_page(HeapScanDesc scan, ScanDirection dir, int *linesleft,
833 : OffsetNumber *lineoff)
834 : {
835 : Page page;
836 :
837 : Assert(scan->rs_inited);
838 : Assert(BufferIsValid(scan->rs_cbuf));
839 :
840 : /* Caller is responsible for ensuring buffer is locked if needed */
841 9266823 : page = BufferGetPage(scan->rs_cbuf);
842 :
843 9266823 : if (ScanDirectionIsForward(dir))
844 : {
845 9266823 : *lineoff = OffsetNumberNext(scan->rs_coffset);
846 9266823 : *linesleft = PageGetMaxOffsetNumber(page) - (*lineoff) + 1;
847 : }
848 : else
849 : {
850 : /*
851 : * The previous returned tuple may have been vacuumed since the
852 : * previous scan when we use a non-MVCC snapshot, so we must
853 : * re-establish the lineoff <= PageGetMaxOffsetNumber(page) invariant
854 : */
855 0 : *lineoff = Min(PageGetMaxOffsetNumber(page), OffsetNumberPrev(scan->rs_coffset));
856 0 : *linesleft = *lineoff;
857 : }
858 :
859 : /* lineoff now references the physically previous or next tid */
860 9266823 : return page;
861 : }
862 :
863 : /*
864 : * heapgettup_advance_block - helper for heap_fetch_next_buffer()
865 : *
866 : * Given the current block number, the scan direction, and various information
867 : * contained in the scan descriptor, calculate the BlockNumber to scan next
868 : * and return it. If there are no further blocks to scan, return
869 : * InvalidBlockNumber to indicate this fact to the caller.
870 : *
871 : * This should not be called to determine the initial block number -- only for
872 : * subsequent blocks.
873 : *
874 : * This also adjusts rs_numblocks when a limit has been imposed by
875 : * heap_setscanlimits().
876 : */
877 : static inline BlockNumber
878 3766032 : heapgettup_advance_block(HeapScanDesc scan, BlockNumber block, ScanDirection dir)
879 : {
880 : Assert(scan->rs_base.rs_parallel == NULL);
881 :
882 3766032 : if (likely(ScanDirectionIsForward(dir)))
883 : {
884 3765954 : block++;
885 :
886 : /* wrap back to the start of the heap */
887 3765954 : if (block >= scan->rs_nblocks)
888 438750 : block = 0;
889 :
890 : /*
891 : * Report our new scan position for synchronization purposes. We don't
892 : * do that when moving backwards, however. That would just mess up any
893 : * other forward-moving scanners.
894 : *
895 : * Note: we do this before checking for end of scan so that the final
896 : * state of the position hint is back at the start of the rel. That's
897 : * not strictly necessary, but otherwise when you run the same query
898 : * multiple times the starting position would shift a little bit
899 : * backwards on every invocation, which is confusing. We don't
900 : * guarantee any specific ordering in general, though.
901 : */
902 3765954 : if (scan->rs_base.rs_flags & SO_ALLOW_SYNC)
903 38364 : ss_report_location(scan->rs_base.rs_rd, block);
904 :
905 : /* we're done if we're back at where we started */
906 3765954 : if (block == scan->rs_startblock)
907 438700 : return InvalidBlockNumber;
908 :
909 : /* check if the limit imposed by heap_setscanlimits() is met */
910 3327254 : if (scan->rs_numblocks != InvalidBlockNumber)
911 : {
912 2818 : if (--scan->rs_numblocks == 0)
913 1590 : return InvalidBlockNumber;
914 : }
915 :
916 3325664 : return block;
917 : }
918 : else
919 : {
920 : /* we're done if the last block is the start position */
921 78 : if (block == scan->rs_startblock)
922 78 : return InvalidBlockNumber;
923 :
924 : /* check if the limit imposed by heap_setscanlimits() is met */
925 0 : if (scan->rs_numblocks != InvalidBlockNumber)
926 : {
927 0 : if (--scan->rs_numblocks == 0)
928 0 : return InvalidBlockNumber;
929 : }
930 :
931 : /* wrap to the end of the heap when the last page was page 0 */
932 0 : if (block == 0)
933 0 : block = scan->rs_nblocks;
934 :
935 0 : block--;
936 :
937 0 : return block;
938 : }
939 : }
940 :
941 : /* ----------------
942 : * heapgettup - fetch next heap tuple
943 : *
944 : * Initialize the scan if not already done; then advance to the next
945 : * tuple as indicated by "dir"; return the next tuple in scan->rs_ctup,
946 : * or set scan->rs_ctup.t_data = NULL if no more tuples.
947 : *
948 : * Note: the reason nkeys/key are passed separately, even though they are
949 : * kept in the scan descriptor, is that the caller may not want us to check
950 : * the scankeys.
951 : *
952 : * Note: when we fall off the end of the scan in either direction, we
953 : * reset rs_inited. This means that a further request with the same
954 : * scan direction will restart the scan, which is a bit odd, but a
955 : * request with the opposite scan direction will start a fresh scan
956 : * in the proper direction. The latter is required behavior for cursors,
957 : * while the former case is generally undefined behavior in Postgres
958 : * so we don't care too much.
959 : * ----------------
960 : */
961 : static void
962 9293741 : heapgettup(HeapScanDesc scan,
963 : ScanDirection dir,
964 : int nkeys,
965 : ScanKey key)
966 : {
967 9293741 : HeapTuple tuple = &(scan->rs_ctup);
968 : Page page;
969 : OffsetNumber lineoff;
970 : int linesleft;
971 :
972 9293741 : if (likely(scan->rs_inited))
973 : {
974 : /* continue from previously returned page/tuple */
975 9266823 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
976 9266823 : page = heapgettup_continue_page(scan, dir, &linesleft, &lineoff);
977 9266823 : goto continue_page;
978 : }
979 :
980 : /*
981 : * advance the scan until we find a qualifying tuple or run out of stuff
982 : * to scan
983 : */
984 : while (true)
985 : {
986 142249 : heap_fetch_next_buffer(scan, dir);
987 :
988 : /* did we run out of blocks to scan? */
989 142249 : if (!BufferIsValid(scan->rs_cbuf))
990 26040 : break;
991 :
992 : Assert(BufferGetBlockNumber(scan->rs_cbuf) == scan->rs_cblock);
993 :
994 116209 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
995 116209 : page = heapgettup_start_page(scan, dir, &linesleft, &lineoff);
996 9383032 : continue_page:
997 :
998 : /*
999 : * Only continue scanning the page while we have lines left.
1000 : *
1001 : * Note that this protects us from accessing line pointers past
1002 : * PageGetMaxOffsetNumber(); both for forward scans when we resume the
1003 : * table scan, and for when we start scanning a new page.
1004 : */
1005 9450962 : for (; linesleft > 0; linesleft--, lineoff += dir)
1006 : {
1007 : bool visible;
1008 9335631 : ItemId lpp = PageGetItemId(page, lineoff);
1009 :
1010 9335631 : if (!ItemIdIsNormal(lpp))
1011 46046 : continue;
1012 :
1013 9289585 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
1014 9289585 : tuple->t_len = ItemIdGetLength(lpp);
1015 9289585 : ItemPointerSet(&(tuple->t_self), scan->rs_cblock, lineoff);
1016 :
1017 9289585 : visible = HeapTupleSatisfiesVisibility(tuple,
1018 : scan->rs_base.rs_snapshot,
1019 : scan->rs_cbuf);
1020 :
1021 9289585 : HeapCheckForSerializableConflictOut(visible, scan->rs_base.rs_rd,
1022 : tuple, scan->rs_cbuf,
1023 : scan->rs_base.rs_snapshot);
1024 :
1025 : /* skip tuples not visible to this snapshot */
1026 9289585 : if (!visible)
1027 7493 : continue;
1028 :
1029 : /* skip any tuples that don't match the scan key */
1030 9282092 : if (key != NULL &&
1031 15116 : !HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
1032 : nkeys, key))
1033 14391 : continue;
1034 :
1035 9267701 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1036 9267701 : scan->rs_coffset = lineoff;
1037 9267701 : return;
1038 : }
1039 :
1040 : /*
1041 : * if we get here, it means we've exhausted the items on this page and
1042 : * it's time to move to the next.
1043 : */
1044 115331 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1045 : }
1046 :
1047 : /* end of scan */
1048 26040 : if (BufferIsValid(scan->rs_cbuf))
1049 0 : ReleaseBuffer(scan->rs_cbuf);
1050 :
1051 26040 : scan->rs_cbuf = InvalidBuffer;
1052 26040 : scan->rs_cblock = InvalidBlockNumber;
1053 26040 : scan->rs_prefetch_block = InvalidBlockNumber;
1054 26040 : tuple->t_data = NULL;
1055 26040 : scan->rs_inited = false;
1056 : }
1057 :
1058 : /* ----------------
1059 : * heapgettup_pagemode - fetch next heap tuple in page-at-a-time mode
1060 : *
1061 : * Same API as heapgettup, but used in page-at-a-time mode
1062 : *
1063 : * The internal logic is much the same as heapgettup's too, but there are some
1064 : * differences: we do not take the buffer content lock (that only needs to
1065 : * happen inside heap_prepare_pagescan), and we iterate through just the
1066 : * tuples listed in rs_vistuples[] rather than all tuples on the page. Notice
1067 : * that lineindex is 0-based, where the corresponding loop variable lineoff in
1068 : * heapgettup is 1-based.
1069 : * ----------------
1070 : */
1071 : static void
1072 70190181 : heapgettup_pagemode(HeapScanDesc scan,
1073 : ScanDirection dir,
1074 : int nkeys,
1075 : ScanKey key)
1076 : {
1077 70190181 : HeapTuple tuple = &(scan->rs_ctup);
1078 : Page page;
1079 : uint32 lineindex;
1080 : uint32 linesleft;
1081 :
1082 70190181 : if (likely(scan->rs_inited))
1083 : {
1084 : /* continue from previously returned page/tuple */
1085 68887766 : page = BufferGetPage(scan->rs_cbuf);
1086 :
1087 68887766 : lineindex = scan->rs_cindex + dir;
1088 68887766 : if (ScanDirectionIsForward(dir))
1089 68887328 : linesleft = scan->rs_ntuples - lineindex;
1090 : else
1091 438 : linesleft = scan->rs_cindex;
1092 : /* lineindex now references the next or previous visible tid */
1093 :
1094 68887766 : goto continue_page;
1095 : }
1096 :
1097 : /*
1098 : * advance the scan until we find a qualifying tuple or run out of stuff
1099 : * to scan
1100 : */
1101 : while (true)
1102 : {
1103 4879670 : heap_fetch_next_buffer(scan, dir);
1104 :
1105 : /* did we run out of blocks to scan? */
1106 4879639 : if (!BufferIsValid(scan->rs_cbuf))
1107 1106780 : break;
1108 :
1109 : Assert(BufferGetBlockNumber(scan->rs_cbuf) == scan->rs_cblock);
1110 :
1111 : /* prune the page and determine visible tuple offsets */
1112 3772859 : heap_prepare_pagescan((TableScanDesc) scan);
1113 3772851 : page = BufferGetPage(scan->rs_cbuf);
1114 3772851 : linesleft = scan->rs_ntuples;
1115 3772851 : lineindex = ScanDirectionIsForward(dir) ? 0 : linesleft - 1;
1116 :
1117 : /* block is the same for all tuples, set it once outside the loop */
1118 3772851 : ItemPointerSetBlockNumber(&tuple->t_self, scan->rs_cblock);
1119 :
1120 : /* lineindex now references the next or previous visible tid */
1121 72660617 : continue_page:
1122 :
1123 134207700 : for (; linesleft > 0; linesleft--, lineindex += dir)
1124 : {
1125 : ItemId lpp;
1126 : OffsetNumber lineoff;
1127 :
1128 : Assert(lineindex < scan->rs_ntuples);
1129 130630445 : lineoff = scan->rs_vistuples[lineindex];
1130 130630445 : lpp = PageGetItemId(page, lineoff);
1131 : Assert(ItemIdIsNormal(lpp));
1132 :
1133 130630445 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
1134 130630445 : tuple->t_len = ItemIdGetLength(lpp);
1135 130630445 : ItemPointerSetOffsetNumber(&tuple->t_self, lineoff);
1136 :
1137 : /* skip any tuples that don't match the scan key */
1138 130630445 : if (key != NULL &&
1139 61984102 : !HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
1140 : nkeys, key))
1141 61547083 : continue;
1142 :
1143 69083362 : scan->rs_cindex = lineindex;
1144 69083362 : return;
1145 : }
1146 : }
1147 :
1148 : /* end of scan */
1149 1106780 : if (BufferIsValid(scan->rs_cbuf))
1150 0 : ReleaseBuffer(scan->rs_cbuf);
1151 1106780 : scan->rs_cbuf = InvalidBuffer;
1152 1106780 : scan->rs_cblock = InvalidBlockNumber;
1153 1106780 : scan->rs_prefetch_block = InvalidBlockNumber;
1154 1106780 : tuple->t_data = NULL;
1155 1106780 : scan->rs_inited = false;
1156 : }
1157 :
1158 :
1159 : /* ----------------------------------------------------------------
1160 : * heap access method interface
1161 : * ----------------------------------------------------------------
1162 : */
1163 :
1164 :
1165 : TableScanDesc
1166 488065 : heap_beginscan(Relation relation, Snapshot snapshot,
1167 : int nkeys, ScanKey key,
1168 : ParallelTableScanDesc parallel_scan,
1169 : uint32 flags)
1170 : {
1171 : HeapScanDesc scan;
1172 :
1173 : /*
1174 : * increment relation ref count while scanning relation
1175 : *
1176 : * This is just to make really sure the relcache entry won't go away while
1177 : * the scan has a pointer to it. Caller should be holding the rel open
1178 : * anyway, so this is redundant in all normal scenarios...
1179 : */
1180 488065 : RelationIncrementReferenceCount(relation);
1181 :
1182 : /*
1183 : * allocate and initialize scan descriptor
1184 : */
1185 488065 : if (flags & SO_TYPE_BITMAPSCAN)
1186 : {
1187 13626 : BitmapHeapScanDesc bscan = palloc_object(BitmapHeapScanDescData);
1188 :
1189 : /*
1190 : * Bitmap Heap scans do not have any fields that a normal Heap Scan
1191 : * does not have, so no special initializations required here.
1192 : */
1193 13626 : scan = (HeapScanDesc) bscan;
1194 : }
1195 : else
1196 474439 : scan = (HeapScanDesc) palloc_object(HeapScanDescData);
1197 :
1198 488065 : scan->rs_base.rs_rd = relation;
1199 488065 : scan->rs_base.rs_snapshot = snapshot;
1200 488065 : scan->rs_base.rs_nkeys = nkeys;
1201 488065 : scan->rs_base.rs_flags = flags;
1202 488065 : scan->rs_base.rs_parallel = parallel_scan;
1203 488065 : scan->rs_strategy = NULL; /* set in initscan */
1204 488065 : scan->rs_cbuf = InvalidBuffer;
1205 :
1206 : /*
1207 : * Disable page-at-a-time mode if it's not a MVCC-safe snapshot.
1208 : */
1209 488065 : if (!(snapshot && IsMVCCSnapshot(snapshot)))
1210 37849 : scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
1211 :
1212 : /* Check that a historic snapshot is not used for non-catalog tables */
1213 488065 : if (snapshot &&
1214 477134 : IsHistoricMVCCSnapshot(snapshot) &&
1215 725 : !RelationIsAccessibleInLogicalDecoding(relation))
1216 : {
1217 0 : ereport(ERROR,
1218 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
1219 : errmsg("cannot query non-catalog table \"%s\" during logical decoding",
1220 : RelationGetRelationName(relation))));
1221 : }
1222 :
1223 : /*
1224 : * For seqscan and sample scans in a serializable transaction, acquire a
1225 : * predicate lock on the entire relation. This is required not only to
1226 : * lock all the matching tuples, but also to conflict with new insertions
1227 : * into the table. In an indexscan, we take page locks on the index pages
1228 : * covering the range specified in the scan qual, but in a heap scan there
1229 : * is nothing more fine-grained to lock. A bitmap scan is a different
1230 : * story, there we have already scanned the index and locked the index
1231 : * pages covering the predicate. But in that case we still have to lock
1232 : * any matching heap tuples. For sample scan we could optimize the locking
1233 : * to be at least page-level granularity, but we'd need to add per-tuple
1234 : * locking for that.
1235 : */
1236 488065 : if (scan->rs_base.rs_flags & (SO_TYPE_SEQSCAN | SO_TYPE_SAMPLESCAN))
1237 : {
1238 : /*
1239 : * Ensure a missing snapshot is noticed reliably, even if the
1240 : * isolation mode means predicate locking isn't performed (and
1241 : * therefore the snapshot isn't used here).
1242 : */
1243 : Assert(snapshot);
1244 461720 : PredicateLockRelation(relation, snapshot);
1245 : }
1246 :
1247 : /* we only need to set this up once */
1248 488065 : scan->rs_ctup.t_tableOid = RelationGetRelid(relation);
1249 :
1250 : /*
1251 : * Allocate memory to keep track of page allocation for parallel workers
1252 : * when doing a parallel scan.
1253 : */
1254 488065 : if (parallel_scan != NULL)
1255 4520 : scan->rs_parallelworkerdata = palloc_object(ParallelBlockTableScanWorkerData);
1256 : else
1257 483545 : scan->rs_parallelworkerdata = NULL;
1258 :
1259 : /*
1260 : * we do this here instead of in initscan() because heap_rescan also calls
1261 : * initscan() and we don't want to allocate memory again
1262 : */
1263 488065 : if (nkeys > 0)
1264 267671 : scan->rs_base.rs_key = palloc_array(ScanKeyData, nkeys);
1265 : else
1266 220394 : scan->rs_base.rs_key = NULL;
1267 :
1268 488065 : initscan(scan, key, false);
1269 :
1270 488063 : scan->rs_read_stream = NULL;
1271 :
1272 : /*
1273 : * Set up a read stream for sequential scans and TID range scans. This
1274 : * should be done after initscan() because initscan() allocates the
1275 : * BufferAccessStrategy object passed to the read stream API.
1276 : */
1277 488063 : if (scan->rs_base.rs_flags & SO_TYPE_SEQSCAN ||
1278 26439 : scan->rs_base.rs_flags & SO_TYPE_TIDRANGESCAN)
1279 462938 : {
1280 : ReadStreamBlockNumberCB cb;
1281 :
1282 462938 : if (scan->rs_base.rs_parallel)
1283 4520 : cb = heap_scan_stream_read_next_parallel;
1284 : else
1285 458418 : cb = heap_scan_stream_read_next_serial;
1286 :
1287 : /* ---
1288 : * It is safe to use batchmode as the only locks taken by `cb`
1289 : * are never taken while waiting for IO:
1290 : * - SyncScanLock is used in the non-parallel case
1291 : * - in the parallel case, only spinlocks and atomics are used
1292 : * ---
1293 : */
1294 462938 : scan->rs_read_stream = read_stream_begin_relation(READ_STREAM_SEQUENTIAL |
1295 : READ_STREAM_USE_BATCHING,
1296 : scan->rs_strategy,
1297 : scan->rs_base.rs_rd,
1298 : MAIN_FORKNUM,
1299 : cb,
1300 : scan,
1301 : 0);
1302 : }
1303 25125 : else if (scan->rs_base.rs_flags & SO_TYPE_BITMAPSCAN)
1304 : {
1305 13626 : scan->rs_read_stream = read_stream_begin_relation(READ_STREAM_DEFAULT |
1306 : READ_STREAM_USE_BATCHING,
1307 : scan->rs_strategy,
1308 : scan->rs_base.rs_rd,
1309 : MAIN_FORKNUM,
1310 : bitmapheap_stream_read_next,
1311 : scan,
1312 : sizeof(TBMIterateResult));
1313 : }
1314 :
1315 488063 : scan->rs_vmbuffer = InvalidBuffer;
1316 :
1317 488063 : return (TableScanDesc) scan;
1318 : }
1319 :
1320 : void
1321 870762 : heap_rescan(TableScanDesc sscan, ScanKey key, bool set_params,
1322 : bool allow_strat, bool allow_sync, bool allow_pagemode)
1323 : {
1324 870762 : HeapScanDesc scan = (HeapScanDesc) sscan;
1325 :
1326 870762 : if (set_params)
1327 : {
1328 19 : if (allow_strat)
1329 19 : scan->rs_base.rs_flags |= SO_ALLOW_STRAT;
1330 : else
1331 0 : scan->rs_base.rs_flags &= ~SO_ALLOW_STRAT;
1332 :
1333 19 : if (allow_sync)
1334 8 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
1335 : else
1336 11 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
1337 :
1338 19 : if (allow_pagemode && scan->rs_base.rs_snapshot &&
1339 19 : IsMVCCSnapshot(scan->rs_base.rs_snapshot))
1340 19 : scan->rs_base.rs_flags |= SO_ALLOW_PAGEMODE;
1341 : else
1342 0 : scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
1343 : }
1344 :
1345 : /*
1346 : * unpin scan buffers
1347 : */
1348 870762 : if (BufferIsValid(scan->rs_cbuf))
1349 : {
1350 3015 : ReleaseBuffer(scan->rs_cbuf);
1351 3015 : scan->rs_cbuf = InvalidBuffer;
1352 : }
1353 :
1354 870762 : if (BufferIsValid(scan->rs_vmbuffer))
1355 : {
1356 24 : ReleaseBuffer(scan->rs_vmbuffer);
1357 24 : scan->rs_vmbuffer = InvalidBuffer;
1358 : }
1359 :
1360 : /*
1361 : * SO_TYPE_BITMAPSCAN would be cleaned up here, but it does not hold any
1362 : * additional data vs a normal HeapScan
1363 : */
1364 :
1365 : /*
1366 : * The read stream is reset on rescan. This must be done before
1367 : * initscan(), as some state referred to by read_stream_reset() is reset
1368 : * in initscan().
1369 : */
1370 870762 : if (scan->rs_read_stream)
1371 870739 : read_stream_reset(scan->rs_read_stream);
1372 :
1373 : /*
1374 : * reinitialize scan descriptor
1375 : */
1376 870762 : initscan(scan, key, true);
1377 870762 : }
1378 :
1379 : void
1380 484792 : heap_endscan(TableScanDesc sscan)
1381 : {
1382 484792 : HeapScanDesc scan = (HeapScanDesc) sscan;
1383 :
1384 : /* Note: no locking manipulations needed */
1385 :
1386 : /*
1387 : * unpin scan buffers
1388 : */
1389 484792 : if (BufferIsValid(scan->rs_cbuf))
1390 191247 : ReleaseBuffer(scan->rs_cbuf);
1391 :
1392 484792 : if (BufferIsValid(scan->rs_vmbuffer))
1393 2848 : ReleaseBuffer(scan->rs_vmbuffer);
1394 :
1395 : /*
1396 : * Must free the read stream before freeing the BufferAccessStrategy.
1397 : */
1398 484792 : if (scan->rs_read_stream)
1399 473360 : read_stream_end(scan->rs_read_stream);
1400 :
1401 : /*
1402 : * decrement relation reference count and free scan descriptor storage
1403 : */
1404 484792 : RelationDecrementReferenceCount(scan->rs_base.rs_rd);
1405 :
1406 484792 : if (scan->rs_base.rs_key)
1407 267633 : pfree(scan->rs_base.rs_key);
1408 :
1409 484792 : if (scan->rs_strategy != NULL)
1410 13437 : FreeAccessStrategy(scan->rs_strategy);
1411 :
1412 484792 : if (scan->rs_parallelworkerdata != NULL)
1413 4520 : pfree(scan->rs_parallelworkerdata);
1414 :
1415 484792 : if (scan->rs_base.rs_flags & SO_TEMP_SNAPSHOT)
1416 50576 : UnregisterSnapshot(scan->rs_base.rs_snapshot);
1417 :
1418 484792 : pfree(scan);
1419 484792 : }
1420 :
1421 : HeapTuple
1422 11754178 : heap_getnext(TableScanDesc sscan, ScanDirection direction)
1423 : {
1424 11754178 : HeapScanDesc scan = (HeapScanDesc) sscan;
1425 :
1426 : /*
1427 : * This is still widely used directly, without going through table AM, so
1428 : * add a safety check. It's possible we should, at a later point,
1429 : * downgrade this to an assert. The reason for checking the AM routine,
1430 : * rather than the AM oid, is that this allows to write regression tests
1431 : * that create another AM reusing the heap handler.
1432 : */
1433 11754178 : if (unlikely(sscan->rs_rd->rd_tableam != GetHeapamTableAmRoutine()))
1434 0 : ereport(ERROR,
1435 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1436 : errmsg_internal("only heap AM is supported")));
1437 :
1438 : /* Note: no locking manipulations needed */
1439 :
1440 11754178 : if (scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE)
1441 3051720 : heapgettup_pagemode(scan, direction,
1442 3051720 : scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1443 : else
1444 8702458 : heapgettup(scan, direction,
1445 8702458 : scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1446 :
1447 11754177 : if (scan->rs_ctup.t_data == NULL)
1448 81757 : return NULL;
1449 :
1450 : /*
1451 : * if we get here it means we have a new current scan tuple, so point to
1452 : * the proper return buffer and return the tuple.
1453 : */
1454 :
1455 11672420 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1456 :
1457 11672420 : return &scan->rs_ctup;
1458 : }
1459 :
1460 : bool
1461 67723108 : heap_getnextslot(TableScanDesc sscan, ScanDirection direction, TupleTableSlot *slot)
1462 : {
1463 67723108 : HeapScanDesc scan = (HeapScanDesc) sscan;
1464 :
1465 : /* Note: no locking manipulations needed */
1466 :
1467 67723108 : if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1468 67131825 : heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1469 : else
1470 591283 : heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1471 :
1472 67723080 : if (scan->rs_ctup.t_data == NULL)
1473 : {
1474 1050925 : ExecClearTuple(slot);
1475 1050925 : return false;
1476 : }
1477 :
1478 : /*
1479 : * if we get here it means we have a new current scan tuple, so point to
1480 : * the proper return buffer and return the tuple.
1481 : */
1482 :
1483 66672155 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1484 :
1485 66672155 : ExecStoreBufferHeapTuple(&scan->rs_ctup, slot,
1486 : scan->rs_cbuf);
1487 66672155 : return true;
1488 : }
1489 :
1490 : void
1491 1374 : heap_set_tidrange(TableScanDesc sscan, ItemPointer mintid,
1492 : ItemPointer maxtid)
1493 : {
1494 1374 : HeapScanDesc scan = (HeapScanDesc) sscan;
1495 : BlockNumber startBlk;
1496 : BlockNumber numBlks;
1497 : ItemPointerData highestItem;
1498 : ItemPointerData lowestItem;
1499 :
1500 : /*
1501 : * For relations without any pages, we can simply leave the TID range
1502 : * unset. There will be no tuples to scan, therefore no tuples outside
1503 : * the given TID range.
1504 : */
1505 1374 : if (scan->rs_nblocks == 0)
1506 32 : return;
1507 :
1508 : /*
1509 : * Set up some ItemPointers which point to the first and last possible
1510 : * tuples in the heap.
1511 : */
1512 1366 : ItemPointerSet(&highestItem, scan->rs_nblocks - 1, MaxOffsetNumber);
1513 1366 : ItemPointerSet(&lowestItem, 0, FirstOffsetNumber);
1514 :
1515 : /*
1516 : * If the given maximum TID is below the highest possible TID in the
1517 : * relation, then restrict the range to that, otherwise we scan to the end
1518 : * of the relation.
1519 : */
1520 1366 : if (ItemPointerCompare(maxtid, &highestItem) < 0)
1521 172 : ItemPointerCopy(maxtid, &highestItem);
1522 :
1523 : /*
1524 : * If the given minimum TID is above the lowest possible TID in the
1525 : * relation, then restrict the range to only scan for TIDs above that.
1526 : */
1527 1366 : if (ItemPointerCompare(mintid, &lowestItem) > 0)
1528 1210 : ItemPointerCopy(mintid, &lowestItem);
1529 :
1530 : /*
1531 : * Check for an empty range and protect from would be negative results
1532 : * from the numBlks calculation below.
1533 : */
1534 1366 : if (ItemPointerCompare(&highestItem, &lowestItem) < 0)
1535 : {
1536 : /* Set an empty range of blocks to scan */
1537 24 : heap_setscanlimits(sscan, 0, 0);
1538 24 : return;
1539 : }
1540 :
1541 : /*
1542 : * Calculate the first block and the number of blocks we must scan. We
1543 : * could be more aggressive here and perform some more validation to try
1544 : * and further narrow the scope of blocks to scan by checking if the
1545 : * lowestItem has an offset above MaxOffsetNumber. In this case, we could
1546 : * advance startBlk by one. Likewise, if highestItem has an offset of 0
1547 : * we could scan one fewer blocks. However, such an optimization does not
1548 : * seem worth troubling over, currently.
1549 : */
1550 1342 : startBlk = ItemPointerGetBlockNumberNoCheck(&lowestItem);
1551 :
1552 1342 : numBlks = ItemPointerGetBlockNumberNoCheck(&highestItem) -
1553 1342 : ItemPointerGetBlockNumberNoCheck(&lowestItem) + 1;
1554 :
1555 : /* Set the start block and number of blocks to scan */
1556 1342 : heap_setscanlimits(sscan, startBlk, numBlks);
1557 :
1558 : /* Finally, set the TID range in sscan */
1559 1342 : ItemPointerCopy(&lowestItem, &sscan->st.tidrange.rs_mintid);
1560 1342 : ItemPointerCopy(&highestItem, &sscan->st.tidrange.rs_maxtid);
1561 : }
1562 :
1563 : bool
1564 6512 : heap_getnextslot_tidrange(TableScanDesc sscan, ScanDirection direction,
1565 : TupleTableSlot *slot)
1566 : {
1567 6512 : HeapScanDesc scan = (HeapScanDesc) sscan;
1568 6512 : ItemPointer mintid = &sscan->st.tidrange.rs_mintid;
1569 6512 : ItemPointer maxtid = &sscan->st.tidrange.rs_maxtid;
1570 :
1571 : /* Note: no locking manipulations needed */
1572 : for (;;)
1573 : {
1574 6636 : if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1575 6636 : heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1576 : else
1577 0 : heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1578 :
1579 6626 : if (scan->rs_ctup.t_data == NULL)
1580 : {
1581 138 : ExecClearTuple(slot);
1582 138 : return false;
1583 : }
1584 :
1585 : /*
1586 : * heap_set_tidrange will have used heap_setscanlimits to limit the
1587 : * range of pages we scan to only ones that can contain the TID range
1588 : * we're scanning for. Here we must filter out any tuples from these
1589 : * pages that are outside of that range.
1590 : */
1591 6488 : if (ItemPointerCompare(&scan->rs_ctup.t_self, mintid) < 0)
1592 : {
1593 124 : ExecClearTuple(slot);
1594 :
1595 : /*
1596 : * When scanning backwards, the TIDs will be in descending order.
1597 : * Future tuples in this direction will be lower still, so we can
1598 : * just return false to indicate there will be no more tuples.
1599 : */
1600 124 : if (ScanDirectionIsBackward(direction))
1601 0 : return false;
1602 :
1603 124 : continue;
1604 : }
1605 :
1606 : /*
1607 : * Likewise for the final page, we must filter out TIDs greater than
1608 : * maxtid.
1609 : */
1610 6364 : if (ItemPointerCompare(&scan->rs_ctup.t_self, maxtid) > 0)
1611 : {
1612 74 : ExecClearTuple(slot);
1613 :
1614 : /*
1615 : * When scanning forward, the TIDs will be in ascending order.
1616 : * Future tuples in this direction will be higher still, so we can
1617 : * just return false to indicate there will be no more tuples.
1618 : */
1619 74 : if (ScanDirectionIsForward(direction))
1620 74 : return false;
1621 0 : continue;
1622 : }
1623 :
1624 6290 : break;
1625 : }
1626 :
1627 : /*
1628 : * if we get here it means we have a new current scan tuple, so point to
1629 : * the proper return buffer and return the tuple.
1630 : */
1631 6290 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1632 :
1633 6290 : ExecStoreBufferHeapTuple(&scan->rs_ctup, slot, scan->rs_cbuf);
1634 6290 : return true;
1635 : }
1636 :
1637 : /*
1638 : * heap_fetch - retrieve tuple with given tid
1639 : *
1640 : * On entry, tuple->t_self is the TID to fetch. We pin the buffer holding
1641 : * the tuple, fill in the remaining fields of *tuple, and check the tuple
1642 : * against the specified snapshot.
1643 : *
1644 : * If successful (tuple found and passes snapshot time qual), then *userbuf
1645 : * is set to the buffer holding the tuple and true is returned. The caller
1646 : * must unpin the buffer when done with the tuple.
1647 : *
1648 : * If the tuple is not found (ie, item number references a deleted slot),
1649 : * then tuple->t_data is set to NULL, *userbuf is set to InvalidBuffer,
1650 : * and false is returned.
1651 : *
1652 : * If the tuple is found but fails the time qual check, then the behavior
1653 : * depends on the keep_buf parameter. If keep_buf is false, the results
1654 : * are the same as for the tuple-not-found case. If keep_buf is true,
1655 : * then tuple->t_data and *userbuf are returned as for the success case,
1656 : * and again the caller must unpin the buffer; but false is returned.
1657 : *
1658 : * heap_fetch does not follow HOT chains: only the exact TID requested will
1659 : * be fetched.
1660 : *
1661 : * It is somewhat inconsistent that we ereport() on invalid block number but
1662 : * return false on invalid item number. There are a couple of reasons though.
1663 : * One is that the caller can relatively easily check the block number for
1664 : * validity, but cannot check the item number without reading the page
1665 : * himself. Another is that when we are following a t_ctid link, we can be
1666 : * reasonably confident that the page number is valid (since VACUUM shouldn't
1667 : * truncate off the destination page without having killed the referencing
1668 : * tuple first), but the item number might well not be good.
1669 : */
1670 : bool
1671 2839771 : heap_fetch(Relation relation,
1672 : Snapshot snapshot,
1673 : HeapTuple tuple,
1674 : Buffer *userbuf,
1675 : bool keep_buf)
1676 : {
1677 2839771 : ItemPointer tid = &(tuple->t_self);
1678 : ItemId lp;
1679 : Buffer buffer;
1680 : Page page;
1681 : OffsetNumber offnum;
1682 : bool valid;
1683 :
1684 : /*
1685 : * Fetch and pin the appropriate page of the relation.
1686 : */
1687 2839771 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
1688 :
1689 : /*
1690 : * Need share lock on buffer to examine tuple commit status.
1691 : */
1692 2839763 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
1693 2839763 : page = BufferGetPage(buffer);
1694 :
1695 : /*
1696 : * We'd better check for out-of-range offnum in case of VACUUM since the
1697 : * TID was obtained.
1698 : */
1699 2839763 : offnum = ItemPointerGetOffsetNumber(tid);
1700 2839763 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1701 : {
1702 4 : UnlockReleaseBuffer(buffer);
1703 4 : *userbuf = InvalidBuffer;
1704 4 : tuple->t_data = NULL;
1705 4 : return false;
1706 : }
1707 :
1708 : /*
1709 : * get the item line pointer corresponding to the requested tid
1710 : */
1711 2839759 : lp = PageGetItemId(page, offnum);
1712 :
1713 : /*
1714 : * Must check for deleted tuple.
1715 : */
1716 2839759 : if (!ItemIdIsNormal(lp))
1717 : {
1718 289 : UnlockReleaseBuffer(buffer);
1719 289 : *userbuf = InvalidBuffer;
1720 289 : tuple->t_data = NULL;
1721 289 : return false;
1722 : }
1723 :
1724 : /*
1725 : * fill in *tuple fields
1726 : */
1727 2839470 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
1728 2839470 : tuple->t_len = ItemIdGetLength(lp);
1729 2839470 : tuple->t_tableOid = RelationGetRelid(relation);
1730 :
1731 : /*
1732 : * check tuple visibility, then release lock
1733 : */
1734 2839470 : valid = HeapTupleSatisfiesVisibility(tuple, snapshot, buffer);
1735 :
1736 2839470 : if (valid)
1737 2839413 : PredicateLockTID(relation, &(tuple->t_self), snapshot,
1738 2839413 : HeapTupleHeaderGetXmin(tuple->t_data));
1739 :
1740 2839470 : HeapCheckForSerializableConflictOut(valid, relation, tuple, buffer, snapshot);
1741 :
1742 2839470 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1743 :
1744 2839470 : if (valid)
1745 : {
1746 : /*
1747 : * All checks passed, so return the tuple as valid. Caller is now
1748 : * responsible for releasing the buffer.
1749 : */
1750 2839413 : *userbuf = buffer;
1751 :
1752 2839413 : return true;
1753 : }
1754 :
1755 : /* Tuple failed time qual, but maybe caller wants to see it anyway. */
1756 57 : if (keep_buf)
1757 35 : *userbuf = buffer;
1758 : else
1759 : {
1760 22 : ReleaseBuffer(buffer);
1761 22 : *userbuf = InvalidBuffer;
1762 22 : tuple->t_data = NULL;
1763 : }
1764 :
1765 57 : return false;
1766 : }
1767 :
1768 : /*
1769 : * heap_get_latest_tid - get the latest tid of a specified tuple
1770 : *
1771 : * Actually, this gets the latest version that is visible according to the
1772 : * scan's snapshot. Create a scan using SnapshotDirty to get the very latest,
1773 : * possibly uncommitted version.
1774 : *
1775 : * *tid is both an input and an output parameter: it is updated to
1776 : * show the latest version of the row. Note that it will not be changed
1777 : * if no version of the row passes the snapshot test.
1778 : */
1779 : void
1780 199 : heap_get_latest_tid(TableScanDesc sscan,
1781 : ItemPointer tid)
1782 : {
1783 199 : Relation relation = sscan->rs_rd;
1784 199 : Snapshot snapshot = sscan->rs_snapshot;
1785 : ItemPointerData ctid;
1786 : TransactionId priorXmax;
1787 :
1788 : /*
1789 : * table_tuple_get_latest_tid() verified that the passed in tid is valid.
1790 : * Assume that t_ctid links are valid however - there shouldn't be invalid
1791 : * ones in the table.
1792 : */
1793 : Assert(ItemPointerIsValid(tid));
1794 :
1795 : /*
1796 : * Loop to chase down t_ctid links. At top of loop, ctid is the tuple we
1797 : * need to examine, and *tid is the TID we will return if ctid turns out
1798 : * to be bogus.
1799 : *
1800 : * Note that we will loop until we reach the end of the t_ctid chain.
1801 : * Depending on the snapshot passed, there might be at most one visible
1802 : * version of the row, but we don't try to optimize for that.
1803 : */
1804 199 : ctid = *tid;
1805 199 : priorXmax = InvalidTransactionId; /* cannot check first XMIN */
1806 : for (;;)
1807 60 : {
1808 : Buffer buffer;
1809 : Page page;
1810 : OffsetNumber offnum;
1811 : ItemId lp;
1812 : HeapTupleData tp;
1813 : bool valid;
1814 :
1815 : /*
1816 : * Read, pin, and lock the page.
1817 : */
1818 259 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(&ctid));
1819 259 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
1820 259 : page = BufferGetPage(buffer);
1821 :
1822 : /*
1823 : * Check for bogus item number. This is not treated as an error
1824 : * condition because it can happen while following a t_ctid link. We
1825 : * just assume that the prior tid is OK and return it unchanged.
1826 : */
1827 259 : offnum = ItemPointerGetOffsetNumber(&ctid);
1828 259 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1829 : {
1830 0 : UnlockReleaseBuffer(buffer);
1831 0 : break;
1832 : }
1833 259 : lp = PageGetItemId(page, offnum);
1834 259 : if (!ItemIdIsNormal(lp))
1835 : {
1836 0 : UnlockReleaseBuffer(buffer);
1837 0 : break;
1838 : }
1839 :
1840 : /* OK to access the tuple */
1841 259 : tp.t_self = ctid;
1842 259 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
1843 259 : tp.t_len = ItemIdGetLength(lp);
1844 259 : tp.t_tableOid = RelationGetRelid(relation);
1845 :
1846 : /*
1847 : * After following a t_ctid link, we might arrive at an unrelated
1848 : * tuple. Check for XMIN match.
1849 : */
1850 319 : if (TransactionIdIsValid(priorXmax) &&
1851 60 : !TransactionIdEquals(priorXmax, HeapTupleHeaderGetXmin(tp.t_data)))
1852 : {
1853 0 : UnlockReleaseBuffer(buffer);
1854 0 : break;
1855 : }
1856 :
1857 : /*
1858 : * Check tuple visibility; if visible, set it as the new result
1859 : * candidate.
1860 : */
1861 259 : valid = HeapTupleSatisfiesVisibility(&tp, snapshot, buffer);
1862 259 : HeapCheckForSerializableConflictOut(valid, relation, &tp, buffer, snapshot);
1863 259 : if (valid)
1864 183 : *tid = ctid;
1865 :
1866 : /*
1867 : * If there's a valid t_ctid link, follow it, else we're done.
1868 : */
1869 367 : if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
1870 184 : HeapTupleHeaderIsOnlyLocked(tp.t_data) ||
1871 152 : HeapTupleHeaderIndicatesMovedPartitions(tp.t_data) ||
1872 76 : ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
1873 : {
1874 199 : UnlockReleaseBuffer(buffer);
1875 199 : break;
1876 : }
1877 :
1878 60 : ctid = tp.t_data->t_ctid;
1879 60 : priorXmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
1880 60 : UnlockReleaseBuffer(buffer);
1881 : } /* end of loop */
1882 199 : }
1883 :
1884 :
1885 : /*
1886 : * UpdateXmaxHintBits - update tuple hint bits after xmax transaction ends
1887 : *
1888 : * This is called after we have waited for the XMAX transaction to terminate.
1889 : * If the transaction aborted, we guarantee the XMAX_INVALID hint bit will
1890 : * be set on exit. If the transaction committed, we set the XMAX_COMMITTED
1891 : * hint bit if possible --- but beware that that may not yet be possible,
1892 : * if the transaction committed asynchronously.
1893 : *
1894 : * Note that if the transaction was a locker only, we set HEAP_XMAX_INVALID
1895 : * even if it commits.
1896 : *
1897 : * Hence callers should look only at XMAX_INVALID.
1898 : *
1899 : * Note this is not allowed for tuples whose xmax is a multixact.
1900 : */
1901 : static void
1902 280 : UpdateXmaxHintBits(HeapTupleHeader tuple, Buffer buffer, TransactionId xid)
1903 : {
1904 : Assert(TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple), xid));
1905 : Assert(!(tuple->t_infomask & HEAP_XMAX_IS_MULTI));
1906 :
1907 280 : if (!(tuple->t_infomask & (HEAP_XMAX_COMMITTED | HEAP_XMAX_INVALID)))
1908 : {
1909 508 : if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask) &&
1910 229 : TransactionIdDidCommit(xid))
1911 202 : HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_COMMITTED,
1912 : xid);
1913 : else
1914 77 : HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_INVALID,
1915 : InvalidTransactionId);
1916 : }
1917 280 : }
1918 :
1919 :
1920 : /*
1921 : * GetBulkInsertState - prepare status object for a bulk insert
1922 : */
1923 : BulkInsertState
1924 3530 : GetBulkInsertState(void)
1925 : {
1926 : BulkInsertState bistate;
1927 :
1928 3530 : bistate = (BulkInsertState) palloc_object(BulkInsertStateData);
1929 3530 : bistate->strategy = GetAccessStrategy(BAS_BULKWRITE);
1930 3530 : bistate->current_buf = InvalidBuffer;
1931 3530 : bistate->next_free = InvalidBlockNumber;
1932 3530 : bistate->last_free = InvalidBlockNumber;
1933 3530 : bistate->already_extended_by = 0;
1934 3530 : return bistate;
1935 : }
1936 :
1937 : /*
1938 : * FreeBulkInsertState - clean up after finishing a bulk insert
1939 : */
1940 : void
1941 3288 : FreeBulkInsertState(BulkInsertState bistate)
1942 : {
1943 3288 : if (bistate->current_buf != InvalidBuffer)
1944 2587 : ReleaseBuffer(bistate->current_buf);
1945 3288 : FreeAccessStrategy(bistate->strategy);
1946 3288 : pfree(bistate);
1947 3288 : }
1948 :
1949 : /*
1950 : * ReleaseBulkInsertStatePin - release a buffer currently held in bistate
1951 : */
1952 : void
1953 90779 : ReleaseBulkInsertStatePin(BulkInsertState bistate)
1954 : {
1955 90779 : if (bistate->current_buf != InvalidBuffer)
1956 40028 : ReleaseBuffer(bistate->current_buf);
1957 90779 : bistate->current_buf = InvalidBuffer;
1958 :
1959 : /*
1960 : * Despite the name, we also reset bulk relation extension state.
1961 : * Otherwise we can end up erroring out due to looking for free space in
1962 : * ->next_free of one partition, even though ->next_free was set when
1963 : * extending another partition. It could obviously also be bad for
1964 : * efficiency to look at existing blocks at offsets from another
1965 : * partition, even if we don't error out.
1966 : */
1967 90779 : bistate->next_free = InvalidBlockNumber;
1968 90779 : bistate->last_free = InvalidBlockNumber;
1969 90779 : }
1970 :
1971 :
1972 : /*
1973 : * heap_insert - insert tuple into a heap
1974 : *
1975 : * The new tuple is stamped with current transaction ID and the specified
1976 : * command ID.
1977 : *
1978 : * See table_tuple_insert for comments about most of the input flags, except
1979 : * that this routine directly takes a tuple rather than a slot.
1980 : *
1981 : * There's corresponding HEAP_INSERT_ options to all the TABLE_INSERT_
1982 : * options, and there additionally is HEAP_INSERT_SPECULATIVE which is used to
1983 : * implement table_tuple_insert_speculative().
1984 : *
1985 : * On return the header fields of *tup are updated to match the stored tuple;
1986 : * in particular tup->t_self receives the actual TID where the tuple was
1987 : * stored. But note that any toasting of fields within the tuple data is NOT
1988 : * reflected into *tup.
1989 : */
1990 : void
1991 11442122 : heap_insert(Relation relation, HeapTuple tup, CommandId cid,
1992 : uint32 options, BulkInsertState bistate)
1993 : {
1994 11442122 : TransactionId xid = GetCurrentTransactionId();
1995 : HeapTuple heaptup;
1996 : Buffer buffer;
1997 : Page page;
1998 11442114 : Buffer vmbuffer = InvalidBuffer;
1999 11442114 : bool all_visible_cleared = false;
2000 :
2001 : /* Cheap, simplistic check that the tuple matches the rel's rowtype. */
2002 : Assert(HeapTupleHeaderGetNatts(tup->t_data) <=
2003 : RelationGetNumberOfAttributes(relation));
2004 :
2005 11442114 : AssertHasSnapshotForToast(relation);
2006 :
2007 : /*
2008 : * Fill in tuple header fields and toast the tuple if necessary.
2009 : *
2010 : * Note: below this point, heaptup is the data we actually intend to store
2011 : * into the relation; tup is the caller's original untoasted data.
2012 : */
2013 11442114 : heaptup = heap_prepare_insert(relation, tup, xid, cid, options);
2014 :
2015 : /*
2016 : * Find buffer to insert this tuple into. If the page is all visible,
2017 : * this will also pin the requisite visibility map page.
2018 : */
2019 11442114 : buffer = RelationGetBufferForTuple(relation, heaptup->t_len,
2020 : InvalidBuffer, options, bistate,
2021 : &vmbuffer, NULL,
2022 : 0);
2023 :
2024 11442114 : page = BufferGetPage(buffer);
2025 :
2026 : /*
2027 : * We're about to do the actual insert -- but check for conflict first, to
2028 : * avoid possibly having to roll back work we've just done.
2029 : *
2030 : * This is safe without a recheck as long as there is no possibility of
2031 : * another process scanning the page between this check and the insert
2032 : * being visible to the scan (i.e., an exclusive buffer content lock is
2033 : * continuously held from this point until the tuple insert is visible).
2034 : *
2035 : * For a heap insert, we only need to check for table-level SSI locks. Our
2036 : * new tuple can't possibly conflict with existing tuple locks, and heap
2037 : * page locks are only consolidated versions of tuple locks; they do not
2038 : * lock "gaps" as index page locks do. So we don't need to specify a
2039 : * buffer when making the call, which makes for a faster check.
2040 : */
2041 11442114 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2042 :
2043 : /* NO EREPORT(ERROR) from here till changes are logged */
2044 11442102 : START_CRIT_SECTION();
2045 :
2046 11442102 : RelationPutHeapTuple(relation, buffer, heaptup,
2047 11442102 : (options & HEAP_INSERT_SPECULATIVE) != 0);
2048 :
2049 11442102 : if (PageIsAllVisible(page))
2050 : {
2051 7245 : all_visible_cleared = true;
2052 7245 : PageClearAllVisible(page);
2053 7245 : visibilitymap_clear(relation,
2054 7245 : ItemPointerGetBlockNumber(&(heaptup->t_self)),
2055 : vmbuffer, VISIBILITYMAP_VALID_BITS);
2056 : }
2057 :
2058 : /*
2059 : * Set pd_prune_xid to trigger heap_page_prune_and_freeze() once the page
2060 : * is full so that we can set the page all-visible in the VM on the next
2061 : * page access.
2062 : *
2063 : * Setting pd_prune_xid is also handy if the inserting transaction
2064 : * eventually aborts making this tuple DEAD and hence available for
2065 : * pruning. If no other tuple in this page is UPDATEd/DELETEd, the aborted
2066 : * tuple would never otherwise be pruned until next vacuum is triggered.
2067 : *
2068 : * Don't set it if we are in bootstrap mode or we are inserting a frozen
2069 : * tuple, as there is no further pruning/freezing needed in those cases.
2070 : */
2071 11442102 : if (TransactionIdIsNormal(xid) && !(options & HEAP_INSERT_FROZEN))
2072 10782717 : PageSetPrunable(page, xid);
2073 :
2074 11442102 : MarkBufferDirty(buffer);
2075 :
2076 : /* XLOG stuff */
2077 11442102 : if (RelationNeedsWAL(relation))
2078 : {
2079 : xl_heap_insert xlrec;
2080 : xl_heap_header xlhdr;
2081 : XLogRecPtr recptr;
2082 9945160 : uint8 info = XLOG_HEAP_INSERT;
2083 9945160 : int bufflags = 0;
2084 :
2085 : /*
2086 : * If this is a catalog, we need to transmit combo CIDs to properly
2087 : * decode, so log that as well.
2088 : */
2089 9945160 : if (RelationIsAccessibleInLogicalDecoding(relation))
2090 5177 : log_heap_new_cid(relation, heaptup);
2091 :
2092 : /*
2093 : * If this is the single and first tuple on page, we can reinit the
2094 : * page instead of restoring the whole thing. Set flag, and hide
2095 : * buffer references from XLogInsert.
2096 : */
2097 10063099 : if (ItemPointerGetOffsetNumber(&(heaptup->t_self)) == FirstOffsetNumber &&
2098 117939 : PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
2099 : {
2100 117093 : info |= XLOG_HEAP_INIT_PAGE;
2101 117093 : bufflags |= REGBUF_WILL_INIT;
2102 : }
2103 :
2104 9945160 : xlrec.offnum = ItemPointerGetOffsetNumber(&heaptup->t_self);
2105 9945160 : xlrec.flags = 0;
2106 9945160 : if (all_visible_cleared)
2107 7241 : xlrec.flags |= XLH_INSERT_ALL_VISIBLE_CLEARED;
2108 9945160 : if (options & HEAP_INSERT_SPECULATIVE)
2109 2206 : xlrec.flags |= XLH_INSERT_IS_SPECULATIVE;
2110 : Assert(ItemPointerGetBlockNumber(&heaptup->t_self) == BufferGetBlockNumber(buffer));
2111 :
2112 : /*
2113 : * For logical decoding, we need the tuple even if we're doing a full
2114 : * page write, so make sure it's included even if we take a full-page
2115 : * image. (XXX We could alternatively store a pointer into the FPW).
2116 : */
2117 9945160 : if (RelationIsLogicallyLogged(relation) &&
2118 273231 : !(options & HEAP_INSERT_NO_LOGICAL))
2119 : {
2120 273170 : xlrec.flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
2121 273170 : bufflags |= REGBUF_KEEP_DATA;
2122 :
2123 273170 : if (IsToastRelation(relation))
2124 1805 : xlrec.flags |= XLH_INSERT_ON_TOAST_RELATION;
2125 : }
2126 :
2127 9945160 : XLogBeginInsert();
2128 9945160 : XLogRegisterData(&xlrec, SizeOfHeapInsert);
2129 :
2130 9945160 : xlhdr.t_infomask2 = heaptup->t_data->t_infomask2;
2131 9945160 : xlhdr.t_infomask = heaptup->t_data->t_infomask;
2132 9945160 : xlhdr.t_hoff = heaptup->t_data->t_hoff;
2133 :
2134 : /*
2135 : * note we mark xlhdr as belonging to buffer; if XLogInsert decides to
2136 : * write the whole page to the xlog, we don't need to store
2137 : * xl_heap_header in the xlog.
2138 : */
2139 9945160 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD | bufflags);
2140 9945160 : XLogRegisterBufData(0, &xlhdr, SizeOfHeapHeader);
2141 : /* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
2142 9945160 : XLogRegisterBufData(0,
2143 9945160 : (char *) heaptup->t_data + SizeofHeapTupleHeader,
2144 9945160 : heaptup->t_len - SizeofHeapTupleHeader);
2145 :
2146 : /* filtering by origin on a row level is much more efficient */
2147 9945160 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
2148 :
2149 9945160 : recptr = XLogInsert(RM_HEAP_ID, info);
2150 :
2151 9945160 : PageSetLSN(page, recptr);
2152 : }
2153 :
2154 11442102 : END_CRIT_SECTION();
2155 :
2156 11442102 : UnlockReleaseBuffer(buffer);
2157 11442102 : if (vmbuffer != InvalidBuffer)
2158 7587 : ReleaseBuffer(vmbuffer);
2159 :
2160 : /*
2161 : * If tuple is cacheable, mark it for invalidation from the caches in case
2162 : * we abort. Note it is OK to do this after releasing the buffer, because
2163 : * the heaptup data structure is all in local memory, not in the shared
2164 : * buffer.
2165 : */
2166 11442102 : CacheInvalidateHeapTuple(relation, heaptup, NULL);
2167 :
2168 : /* Note: speculative insertions are counted too, even if aborted later */
2169 11442102 : pgstat_count_heap_insert(relation, 1);
2170 :
2171 : /*
2172 : * If heaptup is a private copy, release it. Don't forget to copy t_self
2173 : * back to the caller's image, too.
2174 : */
2175 11442102 : if (heaptup != tup)
2176 : {
2177 22542 : tup->t_self = heaptup->t_self;
2178 22542 : heap_freetuple(heaptup);
2179 : }
2180 11442102 : }
2181 :
2182 : /*
2183 : * Subroutine for heap_insert(). Prepares a tuple for insertion. This sets the
2184 : * tuple header fields and toasts the tuple if necessary. Returns a toasted
2185 : * version of the tuple if it was toasted, or the original tuple if not. Note
2186 : * that in any case, the header fields are also set in the original tuple.
2187 : */
2188 : static HeapTuple
2189 13271536 : heap_prepare_insert(Relation relation, HeapTuple tup, TransactionId xid,
2190 : CommandId cid, uint32 options)
2191 : {
2192 : /*
2193 : * To allow parallel inserts, we need to ensure that they are safe to be
2194 : * performed in workers. We have the infrastructure to allow parallel
2195 : * inserts in general except for the cases where inserts generate a new
2196 : * CommandId (eg. inserts into a table having a foreign key column).
2197 : */
2198 13271536 : if (IsParallelWorker())
2199 0 : ereport(ERROR,
2200 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
2201 : errmsg("cannot insert tuples in a parallel worker")));
2202 :
2203 13271536 : tup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
2204 13271536 : tup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
2205 13271536 : tup->t_data->t_infomask |= HEAP_XMAX_INVALID;
2206 13271536 : HeapTupleHeaderSetXmin(tup->t_data, xid);
2207 13271536 : if (options & HEAP_INSERT_FROZEN)
2208 102651 : HeapTupleHeaderSetXminFrozen(tup->t_data);
2209 :
2210 13271536 : HeapTupleHeaderSetCmin(tup->t_data, cid);
2211 13271536 : HeapTupleHeaderSetXmax(tup->t_data, 0); /* for cleanliness */
2212 13271536 : tup->t_tableOid = RelationGetRelid(relation);
2213 :
2214 : /*
2215 : * If the new tuple is too big for storage or contains already toasted
2216 : * out-of-line attributes from some other relation, invoke the toaster.
2217 : */
2218 13271536 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
2219 36177 : relation->rd_rel->relkind != RELKIND_MATVIEW)
2220 : {
2221 : /* toast table entries should never be recursively toasted */
2222 : Assert(!HeapTupleHasExternal(tup));
2223 36116 : return tup;
2224 : }
2225 13235420 : else if (HeapTupleHasExternal(tup) || tup->t_len > TOAST_TUPLE_THRESHOLD)
2226 22601 : return heap_toast_insert_or_update(relation, tup, NULL, options);
2227 : else
2228 13212819 : return tup;
2229 : }
2230 :
2231 : /*
2232 : * Helper for heap_multi_insert() that computes the number of entire pages
2233 : * that inserting the remaining heaptuples requires. Used to determine how
2234 : * much the relation needs to be extended by.
2235 : */
2236 : static int
2237 489679 : heap_multi_insert_pages(HeapTuple *heaptuples, int done, int ntuples, Size saveFreeSpace)
2238 : {
2239 489679 : size_t page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
2240 489679 : int npages = 1;
2241 :
2242 3000679 : for (int i = done; i < ntuples; i++)
2243 : {
2244 2511000 : size_t tup_sz = sizeof(ItemIdData) + MAXALIGN(heaptuples[i]->t_len);
2245 :
2246 2511000 : if (page_avail < tup_sz)
2247 : {
2248 17576 : npages++;
2249 17576 : page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
2250 : }
2251 2511000 : page_avail -= tup_sz;
2252 : }
2253 :
2254 489679 : return npages;
2255 : }
2256 :
2257 : /*
2258 : * heap_multi_insert - insert multiple tuples into a heap
2259 : *
2260 : * This is like heap_insert(), but inserts multiple tuples in one operation.
2261 : * That's faster than calling heap_insert() in a loop, because when multiple
2262 : * tuples can be inserted on a single page, we can write just a single WAL
2263 : * record covering all of them, and only need to lock/unlock the page once.
2264 : *
2265 : * Note: this leaks memory into the current memory context. You can create a
2266 : * temporary context before calling this, if that's a problem.
2267 : */
2268 : void
2269 481461 : heap_multi_insert(Relation relation, TupleTableSlot **slots, int ntuples,
2270 : CommandId cid, uint32 options, BulkInsertState bistate)
2271 : {
2272 481461 : TransactionId xid = GetCurrentTransactionId();
2273 : HeapTuple *heaptuples;
2274 : int i;
2275 : int ndone;
2276 : PGAlignedBlock scratch;
2277 : Page page;
2278 481461 : Buffer vmbuffer = InvalidBuffer;
2279 : bool needwal;
2280 : Size saveFreeSpace;
2281 481461 : bool need_tuple_data = RelationIsLogicallyLogged(relation);
2282 481461 : bool need_cids = RelationIsAccessibleInLogicalDecoding(relation);
2283 481461 : bool starting_with_empty_page = false;
2284 481461 : int npages = 0;
2285 481461 : int npages_used = 0;
2286 :
2287 : /* currently not needed (thus unsupported) for heap_multi_insert() */
2288 : Assert(!(options & HEAP_INSERT_NO_LOGICAL));
2289 :
2290 481461 : AssertHasSnapshotForToast(relation);
2291 :
2292 481461 : needwal = RelationNeedsWAL(relation);
2293 481461 : saveFreeSpace = RelationGetTargetPageFreeSpace(relation,
2294 : HEAP_DEFAULT_FILLFACTOR);
2295 :
2296 : /* Toast and set header data in all the slots */
2297 481461 : heaptuples = palloc(ntuples * sizeof(HeapTuple));
2298 2310883 : for (i = 0; i < ntuples; i++)
2299 : {
2300 : HeapTuple tuple;
2301 :
2302 1829422 : tuple = ExecFetchSlotHeapTuple(slots[i], true, NULL);
2303 1829422 : slots[i]->tts_tableOid = RelationGetRelid(relation);
2304 1829422 : tuple->t_tableOid = slots[i]->tts_tableOid;
2305 1829422 : heaptuples[i] = heap_prepare_insert(relation, tuple, xid, cid,
2306 : options);
2307 : }
2308 :
2309 : /*
2310 : * We're about to do the actual inserts -- but check for conflict first,
2311 : * to minimize the possibility of having to roll back work we've just
2312 : * done.
2313 : *
2314 : * A check here does not definitively prevent a serialization anomaly;
2315 : * that check MUST be done at least past the point of acquiring an
2316 : * exclusive buffer content lock on every buffer that will be affected,
2317 : * and MAY be done after all inserts are reflected in the buffers and
2318 : * those locks are released; otherwise there is a race condition. Since
2319 : * multiple buffers can be locked and unlocked in the loop below, and it
2320 : * would not be feasible to identify and lock all of those buffers before
2321 : * the loop, we must do a final check at the end.
2322 : *
2323 : * The check here could be omitted with no loss of correctness; it is
2324 : * present strictly as an optimization.
2325 : *
2326 : * For heap inserts, we only need to check for table-level SSI locks. Our
2327 : * new tuples can't possibly conflict with existing tuple locks, and heap
2328 : * page locks are only consolidated versions of tuple locks; they do not
2329 : * lock "gaps" as index page locks do. So we don't need to specify a
2330 : * buffer when making the call, which makes for a faster check.
2331 : */
2332 481461 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2333 :
2334 481461 : ndone = 0;
2335 980419 : while (ndone < ntuples)
2336 : {
2337 : Buffer buffer;
2338 498958 : bool all_visible_cleared = false;
2339 498958 : bool all_frozen_set = false;
2340 : int nthispage;
2341 :
2342 498958 : CHECK_FOR_INTERRUPTS();
2343 :
2344 : /*
2345 : * Compute number of pages needed to fit the to-be-inserted tuples in
2346 : * the worst case. This will be used to determine how much to extend
2347 : * the relation by in RelationGetBufferForTuple(), if needed. If we
2348 : * filled a prior page from scratch, we can just update our last
2349 : * computation, but if we started with a partially filled page,
2350 : * recompute from scratch, the number of potentially required pages
2351 : * can vary due to tuples needing to fit onto the page, page headers
2352 : * etc.
2353 : */
2354 498958 : if (ndone == 0 || !starting_with_empty_page)
2355 : {
2356 489679 : npages = heap_multi_insert_pages(heaptuples, ndone, ntuples,
2357 : saveFreeSpace);
2358 489679 : npages_used = 0;
2359 : }
2360 : else
2361 9279 : npages_used++;
2362 :
2363 : /*
2364 : * Find buffer where at least the next tuple will fit. If the page is
2365 : * all-visible, this will also pin the requisite visibility map page.
2366 : *
2367 : * Also pin visibility map page if COPY FREEZE inserts tuples into an
2368 : * empty page. See all_frozen_set below.
2369 : */
2370 498958 : buffer = RelationGetBufferForTuple(relation, heaptuples[ndone]->t_len,
2371 : InvalidBuffer, options, bistate,
2372 : &vmbuffer, NULL,
2373 : npages - npages_used);
2374 498958 : page = BufferGetPage(buffer);
2375 :
2376 498958 : starting_with_empty_page = PageGetMaxOffsetNumber(page) == 0;
2377 :
2378 498958 : if (starting_with_empty_page && (options & HEAP_INSERT_FROZEN))
2379 : {
2380 1665 : all_frozen_set = true;
2381 : /* Lock the vmbuffer before entering the critical section */
2382 1665 : LockBuffer(vmbuffer, BUFFER_LOCK_EXCLUSIVE);
2383 : }
2384 :
2385 : /* NO EREPORT(ERROR) from here till changes are logged */
2386 498958 : START_CRIT_SECTION();
2387 :
2388 : /*
2389 : * RelationGetBufferForTuple has ensured that the first tuple fits.
2390 : * Put that on the page, and then as many other tuples as fit.
2391 : */
2392 498958 : RelationPutHeapTuple(relation, buffer, heaptuples[ndone], false);
2393 :
2394 : /*
2395 : * For logical decoding we need combo CIDs to properly decode the
2396 : * catalog.
2397 : */
2398 498958 : if (needwal && need_cids)
2399 6959 : log_heap_new_cid(relation, heaptuples[ndone]);
2400 :
2401 1829422 : for (nthispage = 1; ndone + nthispage < ntuples; nthispage++)
2402 : {
2403 1347961 : HeapTuple heaptup = heaptuples[ndone + nthispage];
2404 :
2405 1347961 : if (PageGetHeapFreeSpace(page) < MAXALIGN(heaptup->t_len) + saveFreeSpace)
2406 17497 : break;
2407 :
2408 1330464 : RelationPutHeapTuple(relation, buffer, heaptup, false);
2409 :
2410 : /*
2411 : * For logical decoding we need combo CIDs to properly decode the
2412 : * catalog.
2413 : */
2414 1330464 : if (needwal && need_cids)
2415 5016 : log_heap_new_cid(relation, heaptup);
2416 : }
2417 :
2418 : /*
2419 : * If the page is all visible, need to clear that, unless we're only
2420 : * going to add further frozen rows to it.
2421 : *
2422 : * If we're only adding already frozen rows to a previously empty
2423 : * page, mark it as all-frozen and update the visibility map. We're
2424 : * already holding a pin on the vmbuffer.
2425 : */
2426 498958 : if (PageIsAllVisible(page) && !(options & HEAP_INSERT_FROZEN))
2427 : {
2428 3458 : all_visible_cleared = true;
2429 3458 : PageClearAllVisible(page);
2430 3458 : visibilitymap_clear(relation,
2431 : BufferGetBlockNumber(buffer),
2432 : vmbuffer, VISIBILITYMAP_VALID_BITS);
2433 : }
2434 495500 : else if (all_frozen_set)
2435 : {
2436 1665 : PageSetAllVisible(page);
2437 1665 : PageClearPrunable(page);
2438 1665 : visibilitymap_set(BufferGetBlockNumber(buffer),
2439 : vmbuffer,
2440 : VISIBILITYMAP_ALL_VISIBLE |
2441 : VISIBILITYMAP_ALL_FROZEN,
2442 : relation->rd_locator);
2443 : }
2444 :
2445 : /*
2446 : * Set pd_prune_xid. See heap_insert() for more on why we do this when
2447 : * inserting tuples. This only makes sense if we aren't already
2448 : * setting the page frozen in the VM and we're not in bootstrap mode.
2449 : */
2450 498958 : if (!all_frozen_set && TransactionIdIsNormal(xid))
2451 474721 : PageSetPrunable(page, xid);
2452 :
2453 498958 : MarkBufferDirty(buffer);
2454 :
2455 : /* XLOG stuff */
2456 498958 : if (needwal)
2457 : {
2458 : XLogRecPtr recptr;
2459 : xl_heap_multi_insert *xlrec;
2460 495058 : uint8 info = XLOG_HEAP2_MULTI_INSERT;
2461 : char *tupledata;
2462 : int totaldatalen;
2463 495058 : char *scratchptr = scratch.data;
2464 : bool init;
2465 495058 : int bufflags = 0;
2466 :
2467 : /*
2468 : * If the page was previously empty, we can reinit the page
2469 : * instead of restoring the whole thing.
2470 : */
2471 495058 : init = starting_with_empty_page;
2472 :
2473 : /* allocate xl_heap_multi_insert struct from the scratch area */
2474 495058 : xlrec = (xl_heap_multi_insert *) scratchptr;
2475 495058 : scratchptr += SizeOfHeapMultiInsert;
2476 :
2477 : /*
2478 : * Allocate offsets array. Unless we're reinitializing the page,
2479 : * in that case the tuples are stored in order starting at
2480 : * FirstOffsetNumber and we don't need to store the offsets
2481 : * explicitly.
2482 : */
2483 495058 : if (!init)
2484 478281 : scratchptr += nthispage * sizeof(OffsetNumber);
2485 :
2486 : /* the rest of the scratch space is used for tuple data */
2487 495058 : tupledata = scratchptr;
2488 :
2489 : /* check that the mutually exclusive flags are not both set */
2490 : Assert(!(all_visible_cleared && all_frozen_set));
2491 :
2492 495058 : xlrec->flags = 0;
2493 495058 : if (all_visible_cleared)
2494 3458 : xlrec->flags = XLH_INSERT_ALL_VISIBLE_CLEARED;
2495 :
2496 : /*
2497 : * We don't have to worry about including a conflict xid in the
2498 : * WAL record, as HEAP_INSERT_FROZEN intentionally violates
2499 : * visibility rules.
2500 : */
2501 495058 : if (all_frozen_set)
2502 17 : xlrec->flags = XLH_INSERT_ALL_FROZEN_SET;
2503 :
2504 495058 : xlrec->ntuples = nthispage;
2505 :
2506 : /*
2507 : * Write out an xl_multi_insert_tuple and the tuple data itself
2508 : * for each tuple.
2509 : */
2510 2117888 : for (i = 0; i < nthispage; i++)
2511 : {
2512 1622830 : HeapTuple heaptup = heaptuples[ndone + i];
2513 : xl_multi_insert_tuple *tuphdr;
2514 : int datalen;
2515 :
2516 1622830 : if (!init)
2517 982248 : xlrec->offsets[i] = ItemPointerGetOffsetNumber(&heaptup->t_self);
2518 : /* xl_multi_insert_tuple needs two-byte alignment. */
2519 1622830 : tuphdr = (xl_multi_insert_tuple *) SHORTALIGN(scratchptr);
2520 1622830 : scratchptr = ((char *) tuphdr) + SizeOfMultiInsertTuple;
2521 :
2522 1622830 : tuphdr->t_infomask2 = heaptup->t_data->t_infomask2;
2523 1622830 : tuphdr->t_infomask = heaptup->t_data->t_infomask;
2524 1622830 : tuphdr->t_hoff = heaptup->t_data->t_hoff;
2525 :
2526 : /* write bitmap [+ padding] [+ oid] + data */
2527 1622830 : datalen = heaptup->t_len - SizeofHeapTupleHeader;
2528 1622830 : memcpy(scratchptr,
2529 1622830 : (char *) heaptup->t_data + SizeofHeapTupleHeader,
2530 : datalen);
2531 1622830 : tuphdr->datalen = datalen;
2532 1622830 : scratchptr += datalen;
2533 : }
2534 495058 : totaldatalen = scratchptr - tupledata;
2535 : Assert((scratchptr - scratch.data) < BLCKSZ);
2536 :
2537 495058 : if (need_tuple_data)
2538 55 : xlrec->flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
2539 :
2540 : /*
2541 : * Signal that this is the last xl_heap_multi_insert record
2542 : * emitted by this call to heap_multi_insert(). Needed for logical
2543 : * decoding so it knows when to cleanup temporary data.
2544 : */
2545 495058 : if (ndone + nthispage == ntuples)
2546 481004 : xlrec->flags |= XLH_INSERT_LAST_IN_MULTI;
2547 :
2548 495058 : if (init)
2549 : {
2550 16777 : info |= XLOG_HEAP_INIT_PAGE;
2551 16777 : bufflags |= REGBUF_WILL_INIT;
2552 : }
2553 :
2554 : /*
2555 : * If we're doing logical decoding, include the new tuple data
2556 : * even if we take a full-page image of the page.
2557 : */
2558 495058 : if (need_tuple_data)
2559 55 : bufflags |= REGBUF_KEEP_DATA;
2560 :
2561 495058 : XLogBeginInsert();
2562 495058 : XLogRegisterData(xlrec, tupledata - scratch.data);
2563 495058 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD | bufflags);
2564 495058 : if (all_frozen_set)
2565 17 : XLogRegisterBuffer(1, vmbuffer, 0);
2566 :
2567 495058 : XLogRegisterBufData(0, tupledata, totaldatalen);
2568 :
2569 : /* filtering by origin on a row level is much more efficient */
2570 495058 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
2571 :
2572 495058 : recptr = XLogInsert(RM_HEAP2_ID, info);
2573 :
2574 495058 : PageSetLSN(page, recptr);
2575 495058 : if (all_frozen_set)
2576 : {
2577 : Assert(BufferIsDirty(vmbuffer));
2578 17 : PageSetLSN(BufferGetPage(vmbuffer), recptr);
2579 : }
2580 : }
2581 :
2582 498958 : END_CRIT_SECTION();
2583 :
2584 498958 : if (all_frozen_set)
2585 1665 : LockBuffer(vmbuffer, BUFFER_LOCK_UNLOCK);
2586 :
2587 498958 : UnlockReleaseBuffer(buffer);
2588 498958 : ndone += nthispage;
2589 :
2590 : /*
2591 : * NB: Only release vmbuffer after inserting all tuples - it's fairly
2592 : * likely that we'll insert into subsequent heap pages that are likely
2593 : * to use the same vm page.
2594 : */
2595 : }
2596 :
2597 : /* We're done with inserting all tuples, so release the last vmbuffer. */
2598 481461 : if (vmbuffer != InvalidBuffer)
2599 3561 : ReleaseBuffer(vmbuffer);
2600 :
2601 : /*
2602 : * We're done with the actual inserts. Check for conflicts again, to
2603 : * ensure that all rw-conflicts in to these inserts are detected. Without
2604 : * this final check, a sequential scan of the heap may have locked the
2605 : * table after the "before" check, missing one opportunity to detect the
2606 : * conflict, and then scanned the table before the new tuples were there,
2607 : * missing the other chance to detect the conflict.
2608 : *
2609 : * For heap inserts, we only need to check for table-level SSI locks. Our
2610 : * new tuples can't possibly conflict with existing tuple locks, and heap
2611 : * page locks are only consolidated versions of tuple locks; they do not
2612 : * lock "gaps" as index page locks do. So we don't need to specify a
2613 : * buffer when making the call.
2614 : */
2615 481461 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2616 :
2617 : /*
2618 : * If tuples are cacheable, mark them for invalidation from the caches in
2619 : * case we abort. Note it is OK to do this after releasing the buffer,
2620 : * because the heaptuples data structure is all in local memory, not in
2621 : * the shared buffer.
2622 : */
2623 481461 : if (IsCatalogRelation(relation))
2624 : {
2625 1594317 : for (i = 0; i < ntuples; i++)
2626 1114319 : CacheInvalidateHeapTuple(relation, heaptuples[i], NULL);
2627 : }
2628 :
2629 : /* copy t_self fields back to the caller's slots */
2630 2310883 : for (i = 0; i < ntuples; i++)
2631 1829422 : slots[i]->tts_tid = heaptuples[i]->t_self;
2632 :
2633 481461 : pgstat_count_heap_insert(relation, ntuples);
2634 481461 : }
2635 :
2636 : /*
2637 : * simple_heap_insert - insert a tuple
2638 : *
2639 : * Currently, this routine differs from heap_insert only in supplying
2640 : * a default command ID and not allowing access to the speedup options.
2641 : *
2642 : * This should be used rather than using heap_insert directly in most places
2643 : * where we are modifying system catalogs.
2644 : */
2645 : void
2646 1089634 : simple_heap_insert(Relation relation, HeapTuple tup)
2647 : {
2648 1089634 : heap_insert(relation, tup, GetCurrentCommandId(true), 0, NULL);
2649 1089634 : }
2650 :
2651 : /*
2652 : * Given infomask/infomask2, compute the bits that must be saved in the
2653 : * "infobits" field of xl_heap_delete, xl_heap_update, xl_heap_lock,
2654 : * xl_heap_lock_updated WAL records.
2655 : *
2656 : * See fix_infomask_from_infobits.
2657 : */
2658 : static uint8
2659 6516495 : compute_infobits(uint16 infomask, uint16 infomask2)
2660 : {
2661 : return
2662 6516495 : ((infomask & HEAP_XMAX_IS_MULTI) != 0 ? XLHL_XMAX_IS_MULTI : 0) |
2663 6516495 : ((infomask & HEAP_XMAX_LOCK_ONLY) != 0 ? XLHL_XMAX_LOCK_ONLY : 0) |
2664 6516495 : ((infomask & HEAP_XMAX_EXCL_LOCK) != 0 ? XLHL_XMAX_EXCL_LOCK : 0) |
2665 : /* note we ignore HEAP_XMAX_SHR_LOCK here */
2666 13032990 : ((infomask & HEAP_XMAX_KEYSHR_LOCK) != 0 ? XLHL_XMAX_KEYSHR_LOCK : 0) |
2667 : ((infomask2 & HEAP_KEYS_UPDATED) != 0 ?
2668 6516495 : XLHL_KEYS_UPDATED : 0);
2669 : }
2670 :
2671 : /*
2672 : * Given two versions of the same t_infomask for a tuple, compare them and
2673 : * return whether the relevant status for a tuple Xmax has changed. This is
2674 : * used after a buffer lock has been released and reacquired: we want to ensure
2675 : * that the tuple state continues to be the same it was when we previously
2676 : * examined it.
2677 : *
2678 : * Note the Xmax field itself must be compared separately.
2679 : */
2680 : static inline bool
2681 5449 : xmax_infomask_changed(uint16 new_infomask, uint16 old_infomask)
2682 : {
2683 5449 : const uint16 interesting =
2684 : HEAP_XMAX_IS_MULTI | HEAP_XMAX_LOCK_ONLY | HEAP_LOCK_MASK;
2685 :
2686 5449 : if ((new_infomask & interesting) != (old_infomask & interesting))
2687 16 : return true;
2688 :
2689 5433 : return false;
2690 : }
2691 :
2692 : /*
2693 : * heap_delete - delete a tuple
2694 : *
2695 : * See table_tuple_delete() for an explanation of the parameters, except that
2696 : * this routine directly takes a tuple rather than a slot.
2697 : *
2698 : * In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
2699 : * t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
2700 : * only for TM_SelfModified, since we cannot obtain cmax from a combo CID
2701 : * generated by another transaction).
2702 : */
2703 : TM_Result
2704 1880401 : heap_delete(Relation relation, const ItemPointerData *tid,
2705 : CommandId cid, uint32 options, Snapshot crosscheck,
2706 : bool wait, TM_FailureData *tmfd)
2707 : {
2708 : TM_Result result;
2709 1880401 : TransactionId xid = GetCurrentTransactionId();
2710 : ItemId lp;
2711 : HeapTupleData tp;
2712 : Page page;
2713 : BlockNumber block;
2714 : Buffer buffer;
2715 1880401 : Buffer vmbuffer = InvalidBuffer;
2716 : TransactionId new_xmax;
2717 : uint16 new_infomask,
2718 : new_infomask2;
2719 1880401 : bool changingPart = (options & TABLE_DELETE_CHANGING_PARTITION) != 0;
2720 1880401 : bool walLogical = (options & TABLE_DELETE_NO_LOGICAL) == 0;
2721 1880401 : bool have_tuple_lock = false;
2722 : bool iscombo;
2723 1880401 : bool all_visible_cleared = false;
2724 1880401 : HeapTuple old_key_tuple = NULL; /* replica identity of the tuple */
2725 1880401 : bool old_key_copied = false;
2726 :
2727 : Assert(ItemPointerIsValid(tid));
2728 :
2729 1880401 : AssertHasSnapshotForToast(relation);
2730 :
2731 : /*
2732 : * Forbid this during a parallel operation, lest it allocate a combo CID.
2733 : * Other workers might need that combo CID for visibility checks, and we
2734 : * have no provision for broadcasting it to them.
2735 : */
2736 1880401 : if (IsInParallelMode())
2737 0 : ereport(ERROR,
2738 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
2739 : errmsg("cannot delete tuples during a parallel operation")));
2740 :
2741 1880401 : block = ItemPointerGetBlockNumber(tid);
2742 1880401 : buffer = ReadBuffer(relation, block);
2743 1880401 : page = BufferGetPage(buffer);
2744 :
2745 : /*
2746 : * Before locking the buffer, pin the visibility map page if it appears to
2747 : * be necessary. Since we haven't got the lock yet, someone else might be
2748 : * in the middle of changing this, so we'll need to recheck after we have
2749 : * the lock.
2750 : */
2751 1880401 : if (PageIsAllVisible(page))
2752 2000 : visibilitymap_pin(relation, block, &vmbuffer);
2753 :
2754 1880401 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2755 :
2756 1880401 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
2757 : Assert(ItemIdIsNormal(lp));
2758 :
2759 1880401 : tp.t_tableOid = RelationGetRelid(relation);
2760 1880401 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
2761 1880401 : tp.t_len = ItemIdGetLength(lp);
2762 1880401 : tp.t_self = *tid;
2763 :
2764 1 : l1:
2765 :
2766 : /*
2767 : * If we didn't pin the visibility map page and the page has become all
2768 : * visible while we were busy locking the buffer, we'll have to unlock and
2769 : * re-lock, to avoid holding the buffer lock across an I/O. That's a bit
2770 : * unfortunate, but hopefully shouldn't happen often.
2771 : */
2772 1880402 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
2773 : {
2774 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2775 0 : visibilitymap_pin(relation, block, &vmbuffer);
2776 0 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2777 : }
2778 :
2779 1880402 : result = HeapTupleSatisfiesUpdate(&tp, cid, buffer);
2780 :
2781 1880402 : if (result == TM_Invisible)
2782 : {
2783 0 : UnlockReleaseBuffer(buffer);
2784 0 : ereport(ERROR,
2785 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2786 : errmsg("attempted to delete invisible tuple")));
2787 : }
2788 1880402 : else if (result == TM_BeingModified && wait)
2789 : {
2790 : TransactionId xwait;
2791 : uint16 infomask;
2792 :
2793 : /* must copy state data before unlocking buffer */
2794 40664 : xwait = HeapTupleHeaderGetRawXmax(tp.t_data);
2795 40664 : infomask = tp.t_data->t_infomask;
2796 :
2797 : /*
2798 : * Sleep until concurrent transaction ends -- except when there's a
2799 : * single locker and it's our own transaction. Note we don't care
2800 : * which lock mode the locker has, because we need the strongest one.
2801 : *
2802 : * Before sleeping, we need to acquire tuple lock to establish our
2803 : * priority for the tuple (see heap_lock_tuple). LockTuple will
2804 : * release us when we are next-in-line for the tuple.
2805 : *
2806 : * If we are forced to "start over" below, we keep the tuple lock;
2807 : * this arranges that we stay at the head of the line while rechecking
2808 : * tuple state.
2809 : */
2810 40664 : if (infomask & HEAP_XMAX_IS_MULTI)
2811 : {
2812 8 : bool current_is_member = false;
2813 :
2814 8 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
2815 : LockTupleExclusive, ¤t_is_member))
2816 : {
2817 8 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2818 :
2819 : /*
2820 : * Acquire the lock, if necessary (but skip it when we're
2821 : * requesting a lock and already have one; avoids deadlock).
2822 : */
2823 8 : if (!current_is_member)
2824 6 : heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
2825 : LockWaitBlock, &have_tuple_lock);
2826 :
2827 : /* wait for multixact */
2828 8 : MultiXactIdWait((MultiXactId) xwait, MultiXactStatusUpdate, infomask,
2829 : relation, &(tp.t_self), XLTW_Delete,
2830 : NULL);
2831 8 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2832 :
2833 : /*
2834 : * If xwait had just locked the tuple then some other xact
2835 : * could update this tuple before we get to this point. Check
2836 : * for xmax change, and start over if so.
2837 : *
2838 : * We also must start over if we didn't pin the VM page, and
2839 : * the page has become all visible.
2840 : */
2841 16 : if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
2842 16 : xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
2843 8 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
2844 : xwait))
2845 0 : goto l1;
2846 : }
2847 :
2848 : /*
2849 : * You might think the multixact is necessarily done here, but not
2850 : * so: it could have surviving members, namely our own xact or
2851 : * other subxacts of this backend. It is legal for us to delete
2852 : * the tuple in either case, however (the latter case is
2853 : * essentially a situation of upgrading our former shared lock to
2854 : * exclusive). We don't bother changing the on-disk hint bits
2855 : * since we are about to overwrite the xmax altogether.
2856 : */
2857 : }
2858 40656 : else if (!TransactionIdIsCurrentTransactionId(xwait))
2859 : {
2860 : /*
2861 : * Wait for regular transaction to end; but first, acquire tuple
2862 : * lock.
2863 : */
2864 70 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2865 70 : heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
2866 : LockWaitBlock, &have_tuple_lock);
2867 70 : XactLockTableWait(xwait, relation, &(tp.t_self), XLTW_Delete);
2868 66 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2869 :
2870 : /*
2871 : * xwait is done, but if xwait had just locked the tuple then some
2872 : * other xact could update this tuple before we get to this point.
2873 : * Check for xmax change, and start over if so.
2874 : *
2875 : * We also must start over if we didn't pin the VM page, and the
2876 : * page has become all visible.
2877 : */
2878 132 : if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
2879 131 : xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
2880 65 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
2881 : xwait))
2882 1 : goto l1;
2883 :
2884 : /* Otherwise check if it committed or aborted */
2885 65 : UpdateXmaxHintBits(tp.t_data, buffer, xwait);
2886 : }
2887 :
2888 : /*
2889 : * We may overwrite if previous xmax aborted, or if it committed but
2890 : * only locked the tuple without updating it.
2891 : */
2892 81298 : if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
2893 40688 : HEAP_XMAX_IS_LOCKED_ONLY(tp.t_data->t_infomask) ||
2894 49 : HeapTupleHeaderIsOnlyLocked(tp.t_data))
2895 40614 : result = TM_Ok;
2896 45 : else if (!ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
2897 32 : result = TM_Updated;
2898 : else
2899 13 : result = TM_Deleted;
2900 : }
2901 :
2902 : /* sanity check the result HeapTupleSatisfiesUpdate() and the logic above */
2903 : if (result != TM_Ok)
2904 : {
2905 : Assert(result == TM_SelfModified ||
2906 : result == TM_Updated ||
2907 : result == TM_Deleted ||
2908 : result == TM_BeingModified);
2909 : Assert(!(tp.t_data->t_infomask & HEAP_XMAX_INVALID));
2910 : Assert(result != TM_Updated ||
2911 : !ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid));
2912 : }
2913 :
2914 1880397 : if (crosscheck != InvalidSnapshot && result == TM_Ok)
2915 : {
2916 : /* Perform additional check for transaction-snapshot mode RI updates */
2917 1 : if (!HeapTupleSatisfiesVisibility(&tp, crosscheck, buffer))
2918 1 : result = TM_Updated;
2919 : }
2920 :
2921 1880397 : if (result != TM_Ok)
2922 : {
2923 93 : tmfd->ctid = tp.t_data->t_ctid;
2924 93 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
2925 93 : if (result == TM_SelfModified)
2926 28 : tmfd->cmax = HeapTupleHeaderGetCmax(tp.t_data);
2927 : else
2928 65 : tmfd->cmax = InvalidCommandId;
2929 93 : UnlockReleaseBuffer(buffer);
2930 93 : if (have_tuple_lock)
2931 45 : UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
2932 93 : if (vmbuffer != InvalidBuffer)
2933 0 : ReleaseBuffer(vmbuffer);
2934 93 : return result;
2935 : }
2936 :
2937 : /*
2938 : * We're about to do the actual delete -- check for conflict first, to
2939 : * avoid possibly having to roll back work we've just done.
2940 : *
2941 : * This is safe without a recheck as long as there is no possibility of
2942 : * another process scanning the page between this check and the delete
2943 : * being visible to the scan (i.e., an exclusive buffer content lock is
2944 : * continuously held from this point until the tuple delete is visible).
2945 : */
2946 1880304 : CheckForSerializableConflictIn(relation, tid, BufferGetBlockNumber(buffer));
2947 :
2948 : /* replace cid with a combo CID if necessary */
2949 1880290 : HeapTupleHeaderAdjustCmax(tp.t_data, &cid, &iscombo);
2950 :
2951 : /*
2952 : * Compute replica identity tuple before entering the critical section so
2953 : * we don't PANIC upon a memory allocation failure.
2954 : */
2955 1880290 : old_key_tuple = walLogical ?
2956 1880290 : ExtractReplicaIdentity(relation, &tp, true, &old_key_copied) : NULL;
2957 :
2958 : /*
2959 : * If this is the first possibly-multixact-able operation in the current
2960 : * transaction, set my per-backend OldestMemberMXactId setting. We can be
2961 : * certain that the transaction will never become a member of any older
2962 : * MultiXactIds than that. (We have to do this even if we end up just
2963 : * using our own TransactionId below, since some other backend could
2964 : * incorporate our XID into a MultiXact immediately afterwards.)
2965 : */
2966 1880290 : MultiXactIdSetOldestMember();
2967 :
2968 1880290 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(tp.t_data),
2969 1880290 : tp.t_data->t_infomask, tp.t_data->t_infomask2,
2970 : xid, LockTupleExclusive, true,
2971 : &new_xmax, &new_infomask, &new_infomask2);
2972 :
2973 1880290 : START_CRIT_SECTION();
2974 :
2975 : /*
2976 : * If this transaction commits, the tuple will become DEAD sooner or
2977 : * later. Set flag that this page is a candidate for pruning once our xid
2978 : * falls below the OldestXmin horizon. If the transaction finally aborts,
2979 : * the subsequent page pruning will be a no-op and the hint will be
2980 : * cleared.
2981 : */
2982 1880290 : PageSetPrunable(page, xid);
2983 :
2984 1880290 : if (PageIsAllVisible(page))
2985 : {
2986 2000 : all_visible_cleared = true;
2987 2000 : PageClearAllVisible(page);
2988 2000 : visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
2989 : vmbuffer, VISIBILITYMAP_VALID_BITS);
2990 : }
2991 :
2992 : /* store transaction information of xact deleting the tuple */
2993 1880290 : tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
2994 1880290 : tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
2995 1880290 : tp.t_data->t_infomask |= new_infomask;
2996 1880290 : tp.t_data->t_infomask2 |= new_infomask2;
2997 1880290 : HeapTupleHeaderClearHotUpdated(tp.t_data);
2998 1880290 : HeapTupleHeaderSetXmax(tp.t_data, new_xmax);
2999 1880290 : HeapTupleHeaderSetCmax(tp.t_data, cid, iscombo);
3000 : /* Make sure there is no forward chain link in t_ctid */
3001 1880290 : tp.t_data->t_ctid = tp.t_self;
3002 :
3003 : /* Signal that this is actually a move into another partition */
3004 1880290 : if (changingPart)
3005 666 : HeapTupleHeaderSetMovedPartitions(tp.t_data);
3006 :
3007 1880290 : MarkBufferDirty(buffer);
3008 :
3009 : /*
3010 : * XLOG stuff
3011 : *
3012 : * NB: heap_abort_speculative() uses the same xlog record and replay
3013 : * routines.
3014 : */
3015 1880290 : if (RelationNeedsWAL(relation))
3016 : {
3017 : xl_heap_delete xlrec;
3018 : xl_heap_header xlhdr;
3019 : XLogRecPtr recptr;
3020 :
3021 : /*
3022 : * For logical decode we need combo CIDs to properly decode the
3023 : * catalog
3024 : */
3025 1797030 : if (RelationIsAccessibleInLogicalDecoding(relation))
3026 10198 : log_heap_new_cid(relation, &tp);
3027 :
3028 1797030 : xlrec.flags = 0;
3029 1797030 : if (all_visible_cleared)
3030 2000 : xlrec.flags |= XLH_DELETE_ALL_VISIBLE_CLEARED;
3031 1797030 : if (changingPart)
3032 666 : xlrec.flags |= XLH_DELETE_IS_PARTITION_MOVE;
3033 3594060 : xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
3034 1797030 : tp.t_data->t_infomask2);
3035 1797030 : xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
3036 1797030 : xlrec.xmax = new_xmax;
3037 :
3038 1797030 : if (old_key_tuple != NULL)
3039 : {
3040 47029 : if (relation->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
3041 135 : xlrec.flags |= XLH_DELETE_CONTAINS_OLD_TUPLE;
3042 : else
3043 46894 : xlrec.flags |= XLH_DELETE_CONTAINS_OLD_KEY;
3044 : }
3045 :
3046 : /*
3047 : * Mark the change as not-for-logical-decoding if caller requested so.
3048 : *
3049 : * (This is used for changes that affect relations not visible to
3050 : * other transactions, such as the transient table during concurrent
3051 : * repack.)
3052 : */
3053 1797030 : if (!walLogical)
3054 3 : xlrec.flags |= XLH_DELETE_NO_LOGICAL;
3055 :
3056 1797030 : XLogBeginInsert();
3057 1797030 : XLogRegisterData(&xlrec, SizeOfHeapDelete);
3058 :
3059 1797030 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
3060 :
3061 : /*
3062 : * Log replica identity of the deleted tuple if there is one
3063 : */
3064 1797030 : if (old_key_tuple != NULL)
3065 : {
3066 47029 : xlhdr.t_infomask2 = old_key_tuple->t_data->t_infomask2;
3067 47029 : xlhdr.t_infomask = old_key_tuple->t_data->t_infomask;
3068 47029 : xlhdr.t_hoff = old_key_tuple->t_data->t_hoff;
3069 :
3070 47029 : XLogRegisterData(&xlhdr, SizeOfHeapHeader);
3071 47029 : XLogRegisterData((char *) old_key_tuple->t_data
3072 : + SizeofHeapTupleHeader,
3073 47029 : old_key_tuple->t_len
3074 : - SizeofHeapTupleHeader);
3075 : }
3076 :
3077 : /* filtering by origin on a row level is much more efficient */
3078 1797030 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
3079 :
3080 1797030 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
3081 :
3082 1797030 : PageSetLSN(page, recptr);
3083 : }
3084 :
3085 1880290 : END_CRIT_SECTION();
3086 :
3087 1880290 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3088 :
3089 1880290 : if (vmbuffer != InvalidBuffer)
3090 2000 : ReleaseBuffer(vmbuffer);
3091 :
3092 : /*
3093 : * If the tuple has toasted out-of-line attributes, we need to delete
3094 : * those items too. We have to do this before releasing the buffer
3095 : * because we need to look at the contents of the tuple, but it's OK to
3096 : * release the content lock on the buffer first.
3097 : */
3098 1884020 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
3099 3743 : relation->rd_rel->relkind != RELKIND_MATVIEW)
3100 : {
3101 : /* toast table entries should never be recursively toasted */
3102 : Assert(!HeapTupleHasExternal(&tp));
3103 : }
3104 1876560 : else if (HeapTupleHasExternal(&tp))
3105 522 : heap_toast_delete(relation, &tp, false);
3106 :
3107 : /*
3108 : * Mark tuple for invalidation from system caches at next command
3109 : * boundary. We have to do this before releasing the buffer because we
3110 : * need to look at the contents of the tuple.
3111 : */
3112 1880290 : CacheInvalidateHeapTuple(relation, &tp, NULL);
3113 :
3114 : /* Now we can release the buffer */
3115 1880290 : ReleaseBuffer(buffer);
3116 :
3117 : /*
3118 : * Release the lmgr tuple lock, if we had it.
3119 : */
3120 1880290 : if (have_tuple_lock)
3121 26 : UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
3122 :
3123 1880290 : pgstat_count_heap_delete(relation);
3124 :
3125 1880290 : if (old_key_tuple != NULL && old_key_copied)
3126 46895 : heap_freetuple(old_key_tuple);
3127 :
3128 1880290 : return TM_Ok;
3129 : }
3130 :
3131 : /*
3132 : * simple_heap_delete - delete a tuple
3133 : *
3134 : * This routine may be used to delete a tuple when concurrent updates of
3135 : * the target tuple are not expected (for example, because we have a lock
3136 : * on the relation associated with the tuple). Any failure is reported
3137 : * via ereport().
3138 : */
3139 : void
3140 818573 : simple_heap_delete(Relation relation, const ItemPointerData *tid)
3141 : {
3142 : TM_Result result;
3143 : TM_FailureData tmfd;
3144 :
3145 818573 : result = heap_delete(relation, tid,
3146 : GetCurrentCommandId(true),
3147 : 0,
3148 : InvalidSnapshot,
3149 : true /* wait for commit */ ,
3150 : &tmfd);
3151 818573 : switch (result)
3152 : {
3153 0 : case TM_SelfModified:
3154 : /* Tuple was already updated in current command? */
3155 0 : elog(ERROR, "tuple already updated by self");
3156 : break;
3157 :
3158 818573 : case TM_Ok:
3159 : /* done successfully */
3160 818573 : break;
3161 :
3162 0 : case TM_Updated:
3163 0 : elog(ERROR, "tuple concurrently updated");
3164 : break;
3165 :
3166 0 : case TM_Deleted:
3167 0 : elog(ERROR, "tuple concurrently deleted");
3168 : break;
3169 :
3170 0 : default:
3171 0 : elog(ERROR, "unrecognized heap_delete status: %u", result);
3172 : break;
3173 : }
3174 818573 : }
3175 :
3176 : /*
3177 : * heap_update - replace a tuple
3178 : *
3179 : * See table_tuple_update() for an explanation of the parameters, except that
3180 : * this routine directly takes a tuple rather than a slot.
3181 : *
3182 : * In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
3183 : * t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
3184 : * only for TM_SelfModified, since we cannot obtain cmax from a combo CID
3185 : * generated by another transaction).
3186 : */
3187 : TM_Result
3188 2382535 : heap_update(Relation relation, const ItemPointerData *otid, HeapTuple newtup,
3189 : CommandId cid, uint32 options pg_attribute_unused(), Snapshot crosscheck, bool wait,
3190 : TM_FailureData *tmfd, LockTupleMode *lockmode,
3191 : TU_UpdateIndexes *update_indexes)
3192 : {
3193 : TM_Result result;
3194 2382535 : TransactionId xid = GetCurrentTransactionId();
3195 : Bitmapset *hot_attrs;
3196 : Bitmapset *sum_attrs;
3197 : Bitmapset *key_attrs;
3198 : Bitmapset *id_attrs;
3199 : Bitmapset *interesting_attrs;
3200 : Bitmapset *modified_attrs;
3201 : ItemId lp;
3202 : HeapTupleData oldtup;
3203 : HeapTuple heaptup;
3204 2382535 : HeapTuple old_key_tuple = NULL;
3205 2382535 : bool old_key_copied = false;
3206 2382535 : bool walLogical = (options & TABLE_UPDATE_NO_LOGICAL) == 0;
3207 : Page page,
3208 : newpage;
3209 : BlockNumber block;
3210 : MultiXactStatus mxact_status;
3211 : Buffer buffer,
3212 : newbuf,
3213 2382535 : vmbuffer = InvalidBuffer,
3214 2382535 : vmbuffer_new = InvalidBuffer;
3215 : bool need_toast;
3216 : Size newtupsize,
3217 : pagefree;
3218 2382535 : bool have_tuple_lock = false;
3219 : bool iscombo;
3220 2382535 : bool use_hot_update = false;
3221 2382535 : bool summarized_update = false;
3222 : bool key_intact;
3223 2382535 : bool all_visible_cleared = false;
3224 2382535 : bool all_visible_cleared_new = false;
3225 : bool checked_lockers;
3226 : bool locker_remains;
3227 2382535 : bool id_has_external = false;
3228 : TransactionId xmax_new_tuple,
3229 : xmax_old_tuple;
3230 : uint16 infomask_old_tuple,
3231 : infomask2_old_tuple,
3232 : infomask_new_tuple,
3233 : infomask2_new_tuple;
3234 :
3235 : Assert(ItemPointerIsValid(otid));
3236 :
3237 : /* Cheap, simplistic check that the tuple matches the rel's rowtype. */
3238 : Assert(HeapTupleHeaderGetNatts(newtup->t_data) <=
3239 : RelationGetNumberOfAttributes(relation));
3240 :
3241 2382535 : AssertHasSnapshotForToast(relation);
3242 :
3243 : /*
3244 : * Forbid this during a parallel operation, lest it allocate a combo CID.
3245 : * Other workers might need that combo CID for visibility checks, and we
3246 : * have no provision for broadcasting it to them.
3247 : */
3248 2382535 : if (IsInParallelMode())
3249 0 : ereport(ERROR,
3250 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
3251 : errmsg("cannot update tuples during a parallel operation")));
3252 :
3253 : #ifdef USE_ASSERT_CHECKING
3254 : check_lock_if_inplace_updateable_rel(relation, otid, newtup);
3255 : #endif
3256 :
3257 : /*
3258 : * Fetch the list of attributes to be checked for various operations.
3259 : *
3260 : * For HOT considerations, this is wasted effort if we fail to update or
3261 : * have to put the new tuple on a different page. But we must compute the
3262 : * list before obtaining buffer lock --- in the worst case, if we are
3263 : * doing an update on one of the relevant system catalogs, we could
3264 : * deadlock if we try to fetch the list later. In any case, the relcache
3265 : * caches the data so this is usually pretty cheap.
3266 : *
3267 : * We also need columns used by the replica identity and columns that are
3268 : * considered the "key" of rows in the table.
3269 : *
3270 : * Note that we get copies of each bitmap, so we need not worry about
3271 : * relcache flush happening midway through.
3272 : */
3273 2382535 : hot_attrs = RelationGetIndexAttrBitmap(relation,
3274 : INDEX_ATTR_BITMAP_HOT_BLOCKING);
3275 2382535 : sum_attrs = RelationGetIndexAttrBitmap(relation,
3276 : INDEX_ATTR_BITMAP_SUMMARIZED);
3277 2382535 : key_attrs = RelationGetIndexAttrBitmap(relation, INDEX_ATTR_BITMAP_KEY);
3278 2382535 : id_attrs = RelationGetIndexAttrBitmap(relation,
3279 : INDEX_ATTR_BITMAP_IDENTITY_KEY);
3280 2382535 : interesting_attrs = NULL;
3281 2382535 : interesting_attrs = bms_add_members(interesting_attrs, hot_attrs);
3282 2382535 : interesting_attrs = bms_add_members(interesting_attrs, sum_attrs);
3283 2382535 : interesting_attrs = bms_add_members(interesting_attrs, key_attrs);
3284 2382535 : interesting_attrs = bms_add_members(interesting_attrs, id_attrs);
3285 :
3286 2382535 : block = ItemPointerGetBlockNumber(otid);
3287 2382535 : INJECTION_POINT("heap_update-before-pin", NULL);
3288 2382535 : buffer = ReadBuffer(relation, block);
3289 2382535 : page = BufferGetPage(buffer);
3290 :
3291 : /*
3292 : * Before locking the buffer, pin the visibility map page if it appears to
3293 : * be necessary. Since we haven't got the lock yet, someone else might be
3294 : * in the middle of changing this, so we'll need to recheck after we have
3295 : * the lock.
3296 : */
3297 2382535 : if (PageIsAllVisible(page))
3298 2404 : visibilitymap_pin(relation, block, &vmbuffer);
3299 :
3300 2382535 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3301 :
3302 2382535 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(otid));
3303 :
3304 : /*
3305 : * Usually, a buffer pin and/or snapshot blocks pruning of otid, ensuring
3306 : * we see LP_NORMAL here. When the otid origin is a syscache, we may have
3307 : * neither a pin nor a snapshot. Hence, we may see other LP_ states, each
3308 : * of which indicates concurrent pruning.
3309 : *
3310 : * Failing with TM_Updated would be most accurate. However, unlike other
3311 : * TM_Updated scenarios, we don't know the successor ctid in LP_UNUSED and
3312 : * LP_DEAD cases. While the distinction between TM_Updated and TM_Deleted
3313 : * does matter to SQL statements UPDATE and MERGE, those SQL statements
3314 : * hold a snapshot that ensures LP_NORMAL. Hence, the choice between
3315 : * TM_Updated and TM_Deleted affects only the wording of error messages.
3316 : * Settle on TM_Deleted, for two reasons. First, it avoids complicating
3317 : * the specification of when tmfd->ctid is valid. Second, it creates
3318 : * error log evidence that we took this branch.
3319 : *
3320 : * Since it's possible to see LP_UNUSED at otid, it's also possible to see
3321 : * LP_NORMAL for a tuple that replaced LP_UNUSED. If it's a tuple for an
3322 : * unrelated row, we'll fail with "duplicate key value violates unique".
3323 : * XXX if otid is the live, newer version of the newtup row, we'll discard
3324 : * changes originating in versions of this catalog row after the version
3325 : * the caller got from syscache. See syscache-update-pruned.spec.
3326 : */
3327 2382535 : if (!ItemIdIsNormal(lp))
3328 : {
3329 : Assert(RelationSupportsSysCache(RelationGetRelid(relation)));
3330 :
3331 1 : UnlockReleaseBuffer(buffer);
3332 : Assert(!have_tuple_lock);
3333 1 : if (vmbuffer != InvalidBuffer)
3334 1 : ReleaseBuffer(vmbuffer);
3335 1 : tmfd->ctid = *otid;
3336 1 : tmfd->xmax = InvalidTransactionId;
3337 1 : tmfd->cmax = InvalidCommandId;
3338 1 : *update_indexes = TU_None;
3339 :
3340 1 : bms_free(hot_attrs);
3341 1 : bms_free(sum_attrs);
3342 1 : bms_free(key_attrs);
3343 1 : bms_free(id_attrs);
3344 : /* modified_attrs not yet initialized */
3345 1 : bms_free(interesting_attrs);
3346 1 : return TM_Deleted;
3347 : }
3348 :
3349 : /*
3350 : * Fill in enough data in oldtup for HeapDetermineColumnsInfo to work
3351 : * properly.
3352 : */
3353 2382534 : oldtup.t_tableOid = RelationGetRelid(relation);
3354 2382534 : oldtup.t_data = (HeapTupleHeader) PageGetItem(page, lp);
3355 2382534 : oldtup.t_len = ItemIdGetLength(lp);
3356 2382534 : oldtup.t_self = *otid;
3357 :
3358 : /* the new tuple is ready, except for this: */
3359 2382534 : newtup->t_tableOid = RelationGetRelid(relation);
3360 :
3361 : /*
3362 : * Determine columns modified by the update. Additionally, identify
3363 : * whether any of the unmodified replica identity key attributes in the
3364 : * old tuple is externally stored or not. This is required because for
3365 : * such attributes the flattened value won't be WAL logged as part of the
3366 : * new tuple so we must include it as part of the old_key_tuple. See
3367 : * ExtractReplicaIdentity.
3368 : */
3369 2382534 : modified_attrs = HeapDetermineColumnsInfo(relation, interesting_attrs,
3370 : id_attrs, &oldtup,
3371 : newtup, &id_has_external);
3372 :
3373 : /*
3374 : * If we're not updating any "key" column, we can grab a weaker lock type.
3375 : * This allows for more concurrency when we are running simultaneously
3376 : * with foreign key checks.
3377 : *
3378 : * Note that if a column gets detoasted while executing the update, but
3379 : * the value ends up being the same, this test will fail and we will use
3380 : * the stronger lock. This is acceptable; the important case to optimize
3381 : * is updates that don't manipulate key columns, not those that
3382 : * serendipitously arrive at the same key values.
3383 : */
3384 2382534 : if (!bms_overlap(modified_attrs, key_attrs))
3385 : {
3386 2376614 : *lockmode = LockTupleNoKeyExclusive;
3387 2376614 : mxact_status = MultiXactStatusNoKeyUpdate;
3388 2376614 : key_intact = true;
3389 :
3390 : /*
3391 : * If this is the first possibly-multixact-able operation in the
3392 : * current transaction, set my per-backend OldestMemberMXactId
3393 : * setting. We can be certain that the transaction will never become a
3394 : * member of any older MultiXactIds than that. (We have to do this
3395 : * even if we end up just using our own TransactionId below, since
3396 : * some other backend could incorporate our XID into a MultiXact
3397 : * immediately afterwards.)
3398 : */
3399 2376614 : MultiXactIdSetOldestMember();
3400 : }
3401 : else
3402 : {
3403 5920 : *lockmode = LockTupleExclusive;
3404 5920 : mxact_status = MultiXactStatusUpdate;
3405 5920 : key_intact = false;
3406 : }
3407 :
3408 : /*
3409 : * Note: beyond this point, use oldtup not otid to refer to old tuple.
3410 : * otid may very well point at newtup->t_self, which we will overwrite
3411 : * with the new tuple's location, so there's great risk of confusion if we
3412 : * use otid anymore.
3413 : */
3414 :
3415 1 : l2:
3416 2382535 : checked_lockers = false;
3417 2382535 : locker_remains = false;
3418 2382535 : result = HeapTupleSatisfiesUpdate(&oldtup, cid, buffer);
3419 :
3420 : /* see below about the "no wait" case */
3421 : Assert(result != TM_BeingModified || wait);
3422 :
3423 2382535 : if (result == TM_Invisible)
3424 : {
3425 0 : UnlockReleaseBuffer(buffer);
3426 0 : ereport(ERROR,
3427 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3428 : errmsg("attempted to update invisible tuple")));
3429 : }
3430 2382535 : else if (result == TM_BeingModified && wait)
3431 : {
3432 : TransactionId xwait;
3433 : uint16 infomask;
3434 36518 : bool can_continue = false;
3435 :
3436 : /*
3437 : * XXX note that we don't consider the "no wait" case here. This
3438 : * isn't a problem currently because no caller uses that case, but it
3439 : * should be fixed if such a caller is introduced. It wasn't a
3440 : * problem previously because this code would always wait, but now
3441 : * that some tuple locks do not conflict with one of the lock modes we
3442 : * use, it is possible that this case is interesting to handle
3443 : * specially.
3444 : *
3445 : * This may cause failures with third-party code that calls
3446 : * heap_update directly.
3447 : */
3448 :
3449 : /* must copy state data before unlocking buffer */
3450 36518 : xwait = HeapTupleHeaderGetRawXmax(oldtup.t_data);
3451 36518 : infomask = oldtup.t_data->t_infomask;
3452 :
3453 : /*
3454 : * Now we have to do something about the existing locker. If it's a
3455 : * multi, sleep on it; we might be awakened before it is completely
3456 : * gone (or even not sleep at all in some cases); we need to preserve
3457 : * it as locker, unless it is gone completely.
3458 : *
3459 : * If it's not a multi, we need to check for sleeping conditions
3460 : * before actually going to sleep. If the update doesn't conflict
3461 : * with the locks, we just continue without sleeping (but making sure
3462 : * it is preserved).
3463 : *
3464 : * Before sleeping, we need to acquire tuple lock to establish our
3465 : * priority for the tuple (see heap_lock_tuple). LockTuple will
3466 : * release us when we are next-in-line for the tuple. Note we must
3467 : * not acquire the tuple lock until we're sure we're going to sleep;
3468 : * otherwise we're open for race conditions with other transactions
3469 : * holding the tuple lock which sleep on us.
3470 : *
3471 : * If we are forced to "start over" below, we keep the tuple lock;
3472 : * this arranges that we stay at the head of the line while rechecking
3473 : * tuple state.
3474 : */
3475 36518 : if (infomask & HEAP_XMAX_IS_MULTI)
3476 : {
3477 : TransactionId update_xact;
3478 : int remain;
3479 179 : bool current_is_member = false;
3480 :
3481 179 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
3482 : *lockmode, ¤t_is_member))
3483 : {
3484 8 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3485 :
3486 : /*
3487 : * Acquire the lock, if necessary (but skip it when we're
3488 : * requesting a lock and already have one; avoids deadlock).
3489 : */
3490 8 : if (!current_is_member)
3491 0 : heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3492 : LockWaitBlock, &have_tuple_lock);
3493 :
3494 : /* wait for multixact */
3495 8 : MultiXactIdWait((MultiXactId) xwait, mxact_status, infomask,
3496 : relation, &oldtup.t_self, XLTW_Update,
3497 : &remain);
3498 8 : checked_lockers = true;
3499 8 : locker_remains = remain != 0;
3500 8 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3501 :
3502 : /*
3503 : * If xwait had just locked the tuple then some other xact
3504 : * could update this tuple before we get to this point. Check
3505 : * for xmax change, and start over if so.
3506 : */
3507 8 : if (xmax_infomask_changed(oldtup.t_data->t_infomask,
3508 8 : infomask) ||
3509 8 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3510 : xwait))
3511 0 : goto l2;
3512 : }
3513 :
3514 : /*
3515 : * Note that the multixact may not be done by now. It could have
3516 : * surviving members; our own xact or other subxacts of this
3517 : * backend, and also any other concurrent transaction that locked
3518 : * the tuple with LockTupleKeyShare if we only got
3519 : * LockTupleNoKeyExclusive. If this is the case, we have to be
3520 : * careful to mark the updated tuple with the surviving members in
3521 : * Xmax.
3522 : *
3523 : * Note that there could have been another update in the
3524 : * MultiXact. In that case, we need to check whether it committed
3525 : * or aborted. If it aborted we are safe to update it again;
3526 : * otherwise there is an update conflict, and we have to return
3527 : * TableTuple{Deleted, Updated} below.
3528 : *
3529 : * In the LockTupleExclusive case, we still need to preserve the
3530 : * surviving members: those would include the tuple locks we had
3531 : * before this one, which are important to keep in case this
3532 : * subxact aborts.
3533 : */
3534 179 : if (!HEAP_XMAX_IS_LOCKED_ONLY(oldtup.t_data->t_infomask))
3535 8 : update_xact = HeapTupleGetUpdateXid(oldtup.t_data);
3536 : else
3537 171 : update_xact = InvalidTransactionId;
3538 :
3539 : /*
3540 : * There was no UPDATE in the MultiXact; or it aborted. No
3541 : * TransactionIdIsInProgress() call needed here, since we called
3542 : * MultiXactIdWait() above.
3543 : */
3544 187 : if (!TransactionIdIsValid(update_xact) ||
3545 8 : TransactionIdDidAbort(update_xact))
3546 172 : can_continue = true;
3547 : }
3548 36339 : else if (TransactionIdIsCurrentTransactionId(xwait))
3549 : {
3550 : /*
3551 : * The only locker is ourselves; we can avoid grabbing the tuple
3552 : * lock here, but must preserve our locking information.
3553 : */
3554 36209 : checked_lockers = true;
3555 36209 : locker_remains = true;
3556 36209 : can_continue = true;
3557 : }
3558 130 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) && key_intact)
3559 : {
3560 : /*
3561 : * If it's just a key-share locker, and we're not changing the key
3562 : * columns, we don't need to wait for it to end; but we need to
3563 : * preserve it as locker.
3564 : */
3565 29 : checked_lockers = true;
3566 29 : locker_remains = true;
3567 29 : can_continue = true;
3568 : }
3569 : else
3570 : {
3571 : /*
3572 : * Wait for regular transaction to end; but first, acquire tuple
3573 : * lock.
3574 : */
3575 101 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3576 101 : heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3577 : LockWaitBlock, &have_tuple_lock);
3578 101 : XactLockTableWait(xwait, relation, &oldtup.t_self,
3579 : XLTW_Update);
3580 101 : checked_lockers = true;
3581 101 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3582 :
3583 : /*
3584 : * xwait is done, but if xwait had just locked the tuple then some
3585 : * other xact could update this tuple before we get to this point.
3586 : * Check for xmax change, and start over if so.
3587 : */
3588 201 : if (xmax_infomask_changed(oldtup.t_data->t_infomask, infomask) ||
3589 100 : !TransactionIdEquals(xwait,
3590 : HeapTupleHeaderGetRawXmax(oldtup.t_data)))
3591 1 : goto l2;
3592 :
3593 : /* Otherwise check if it committed or aborted */
3594 100 : UpdateXmaxHintBits(oldtup.t_data, buffer, xwait);
3595 100 : if (oldtup.t_data->t_infomask & HEAP_XMAX_INVALID)
3596 22 : can_continue = true;
3597 : }
3598 :
3599 36517 : if (can_continue)
3600 36432 : result = TM_Ok;
3601 85 : else if (!ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid))
3602 71 : result = TM_Updated;
3603 : else
3604 14 : result = TM_Deleted;
3605 : }
3606 :
3607 : /* Sanity check the result HeapTupleSatisfiesUpdate() and the logic above */
3608 : if (result != TM_Ok)
3609 : {
3610 : Assert(result == TM_SelfModified ||
3611 : result == TM_Updated ||
3612 : result == TM_Deleted ||
3613 : result == TM_BeingModified);
3614 : Assert(!(oldtup.t_data->t_infomask & HEAP_XMAX_INVALID));
3615 : Assert(result != TM_Updated ||
3616 : !ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid));
3617 : }
3618 :
3619 2382534 : if (crosscheck != InvalidSnapshot && result == TM_Ok)
3620 : {
3621 : /* Perform additional check for transaction-snapshot mode RI updates */
3622 1 : if (!HeapTupleSatisfiesVisibility(&oldtup, crosscheck, buffer))
3623 1 : result = TM_Updated;
3624 : }
3625 :
3626 2382534 : if (result != TM_Ok)
3627 : {
3628 211 : tmfd->ctid = oldtup.t_data->t_ctid;
3629 211 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(oldtup.t_data);
3630 211 : if (result == TM_SelfModified)
3631 73 : tmfd->cmax = HeapTupleHeaderGetCmax(oldtup.t_data);
3632 : else
3633 138 : tmfd->cmax = InvalidCommandId;
3634 211 : UnlockReleaseBuffer(buffer);
3635 211 : if (have_tuple_lock)
3636 78 : UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
3637 211 : if (vmbuffer != InvalidBuffer)
3638 0 : ReleaseBuffer(vmbuffer);
3639 211 : *update_indexes = TU_None;
3640 :
3641 211 : bms_free(hot_attrs);
3642 211 : bms_free(sum_attrs);
3643 211 : bms_free(key_attrs);
3644 211 : bms_free(id_attrs);
3645 211 : bms_free(modified_attrs);
3646 211 : bms_free(interesting_attrs);
3647 211 : return result;
3648 : }
3649 :
3650 : /*
3651 : * If we didn't pin the visibility map page and the page has become all
3652 : * visible while we were busy locking the buffer, or during some
3653 : * subsequent window during which we had it unlocked, we'll have to unlock
3654 : * and re-lock, to avoid holding the buffer lock across an I/O. That's a
3655 : * bit unfortunate, especially since we'll now have to recheck whether the
3656 : * tuple has been locked or updated under us, but hopefully it won't
3657 : * happen very often.
3658 : */
3659 2382323 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
3660 : {
3661 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3662 0 : visibilitymap_pin(relation, block, &vmbuffer);
3663 0 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3664 0 : goto l2;
3665 : }
3666 :
3667 : /* Fill in transaction status data */
3668 :
3669 : /*
3670 : * If the tuple we're updating is locked, we need to preserve the locking
3671 : * info in the old tuple's Xmax. Prepare a new Xmax value for this.
3672 : */
3673 2382323 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3674 2382323 : oldtup.t_data->t_infomask,
3675 2382323 : oldtup.t_data->t_infomask2,
3676 : xid, *lockmode, true,
3677 : &xmax_old_tuple, &infomask_old_tuple,
3678 : &infomask2_old_tuple);
3679 :
3680 : /*
3681 : * And also prepare an Xmax value for the new copy of the tuple. If there
3682 : * was no xmax previously, or there was one but all lockers are now gone,
3683 : * then use InvalidTransactionId; otherwise, get the xmax from the old
3684 : * tuple. (In rare cases that might also be InvalidTransactionId and yet
3685 : * not have the HEAP_XMAX_INVALID bit set; that's fine.)
3686 : */
3687 2418733 : if ((oldtup.t_data->t_infomask & HEAP_XMAX_INVALID) ||
3688 72820 : HEAP_LOCKED_UPGRADED(oldtup.t_data->t_infomask) ||
3689 36239 : (checked_lockers && !locker_remains))
3690 2345913 : xmax_new_tuple = InvalidTransactionId;
3691 : else
3692 36410 : xmax_new_tuple = HeapTupleHeaderGetRawXmax(oldtup.t_data);
3693 :
3694 2382323 : if (!TransactionIdIsValid(xmax_new_tuple))
3695 : {
3696 2345913 : infomask_new_tuple = HEAP_XMAX_INVALID;
3697 2345913 : infomask2_new_tuple = 0;
3698 : }
3699 : else
3700 : {
3701 : /*
3702 : * If we found a valid Xmax for the new tuple, then the infomask bits
3703 : * to use on the new tuple depend on what was there on the old one.
3704 : * Note that since we're doing an update, the only possibility is that
3705 : * the lockers had FOR KEY SHARE lock.
3706 : */
3707 36410 : if (oldtup.t_data->t_infomask & HEAP_XMAX_IS_MULTI)
3708 : {
3709 172 : GetMultiXactIdHintBits(xmax_new_tuple, &infomask_new_tuple,
3710 : &infomask2_new_tuple);
3711 : }
3712 : else
3713 : {
3714 36238 : infomask_new_tuple = HEAP_XMAX_KEYSHR_LOCK | HEAP_XMAX_LOCK_ONLY;
3715 36238 : infomask2_new_tuple = 0;
3716 : }
3717 : }
3718 :
3719 : /*
3720 : * Prepare the new tuple with the appropriate initial values of Xmin and
3721 : * Xmax, as well as initial infomask bits as computed above.
3722 : */
3723 2382323 : newtup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
3724 2382323 : newtup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
3725 2382323 : HeapTupleHeaderSetXmin(newtup->t_data, xid);
3726 2382323 : HeapTupleHeaderSetCmin(newtup->t_data, cid);
3727 2382323 : newtup->t_data->t_infomask |= HEAP_UPDATED | infomask_new_tuple;
3728 2382323 : newtup->t_data->t_infomask2 |= infomask2_new_tuple;
3729 2382323 : HeapTupleHeaderSetXmax(newtup->t_data, xmax_new_tuple);
3730 :
3731 : /*
3732 : * Replace cid with a combo CID if necessary. Note that we already put
3733 : * the plain cid into the new tuple.
3734 : */
3735 2382323 : HeapTupleHeaderAdjustCmax(oldtup.t_data, &cid, &iscombo);
3736 :
3737 : /*
3738 : * If the toaster needs to be activated, OR if the new tuple will not fit
3739 : * on the same page as the old, then we need to release the content lock
3740 : * (but not the pin!) on the old tuple's buffer while we are off doing
3741 : * TOAST and/or table-file-extension work. We must mark the old tuple to
3742 : * show that it's locked, else other processes may try to update it
3743 : * themselves.
3744 : *
3745 : * We need to invoke the toaster if there are already any out-of-line
3746 : * toasted values present, or if the new tuple is over-threshold.
3747 : */
3748 2382323 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
3749 0 : relation->rd_rel->relkind != RELKIND_MATVIEW)
3750 : {
3751 : /* toast table entries should never be recursively toasted */
3752 : Assert(!HeapTupleHasExternal(&oldtup));
3753 : Assert(!HeapTupleHasExternal(newtup));
3754 0 : need_toast = false;
3755 : }
3756 : else
3757 7146468 : need_toast = (HeapTupleHasExternal(&oldtup) ||
3758 4764145 : HeapTupleHasExternal(newtup) ||
3759 2381790 : newtup->t_len > TOAST_TUPLE_THRESHOLD);
3760 :
3761 2382323 : pagefree = PageGetHeapFreeSpace(page);
3762 :
3763 2382323 : newtupsize = MAXALIGN(newtup->t_len);
3764 :
3765 2382323 : if (need_toast || newtupsize > pagefree)
3766 2197646 : {
3767 : TransactionId xmax_lock_old_tuple;
3768 : uint16 infomask_lock_old_tuple,
3769 : infomask2_lock_old_tuple;
3770 2197646 : bool cleared_all_frozen = false;
3771 :
3772 : /*
3773 : * To prevent concurrent sessions from updating the tuple, we have to
3774 : * temporarily mark it locked, while we release the page-level lock.
3775 : *
3776 : * To satisfy the rule that any xid potentially appearing in a buffer
3777 : * written out to disk, we unfortunately have to WAL log this
3778 : * temporary modification. We can reuse xl_heap_lock for this
3779 : * purpose. If we crash/error before following through with the
3780 : * actual update, xmax will be of an aborted transaction, allowing
3781 : * other sessions to proceed.
3782 : */
3783 :
3784 : /*
3785 : * Compute xmax / infomask appropriate for locking the tuple. This has
3786 : * to be done separately from the combo that's going to be used for
3787 : * updating, because the potentially created multixact would otherwise
3788 : * be wrong.
3789 : */
3790 2197646 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3791 2197646 : oldtup.t_data->t_infomask,
3792 2197646 : oldtup.t_data->t_infomask2,
3793 : xid, *lockmode, false,
3794 : &xmax_lock_old_tuple, &infomask_lock_old_tuple,
3795 : &infomask2_lock_old_tuple);
3796 :
3797 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(infomask_lock_old_tuple));
3798 :
3799 2197646 : START_CRIT_SECTION();
3800 :
3801 : /* Clear obsolete visibility flags ... */
3802 2197646 : oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
3803 2197646 : oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
3804 2197646 : HeapTupleClearHotUpdated(&oldtup);
3805 : /* ... and store info about transaction updating this tuple */
3806 : Assert(TransactionIdIsValid(xmax_lock_old_tuple));
3807 2197646 : HeapTupleHeaderSetXmax(oldtup.t_data, xmax_lock_old_tuple);
3808 2197646 : oldtup.t_data->t_infomask |= infomask_lock_old_tuple;
3809 2197646 : oldtup.t_data->t_infomask2 |= infomask2_lock_old_tuple;
3810 2197646 : HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
3811 :
3812 : /* temporarily make it look not-updated, but locked */
3813 2197646 : oldtup.t_data->t_ctid = oldtup.t_self;
3814 :
3815 : /*
3816 : * Clear all-frozen bit on visibility map if needed. We could
3817 : * immediately reset ALL_VISIBLE, but given that the WAL logging
3818 : * overhead would be unchanged, that doesn't seem necessarily
3819 : * worthwhile.
3820 : */
3821 2198915 : if (PageIsAllVisible(page) &&
3822 1269 : visibilitymap_clear(relation, block, vmbuffer,
3823 : VISIBILITYMAP_ALL_FROZEN))
3824 819 : cleared_all_frozen = true;
3825 :
3826 2197646 : MarkBufferDirty(buffer);
3827 :
3828 2197646 : if (RelationNeedsWAL(relation))
3829 : {
3830 : xl_heap_lock xlrec;
3831 : XLogRecPtr recptr;
3832 :
3833 2187511 : XLogBeginInsert();
3834 2187511 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
3835 :
3836 2187511 : xlrec.offnum = ItemPointerGetOffsetNumber(&oldtup.t_self);
3837 2187511 : xlrec.xmax = xmax_lock_old_tuple;
3838 4375022 : xlrec.infobits_set = compute_infobits(oldtup.t_data->t_infomask,
3839 2187511 : oldtup.t_data->t_infomask2);
3840 2187511 : xlrec.flags =
3841 2187511 : cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
3842 2187511 : XLogRegisterData(&xlrec, SizeOfHeapLock);
3843 2187511 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
3844 2187511 : PageSetLSN(page, recptr);
3845 : }
3846 :
3847 2197646 : END_CRIT_SECTION();
3848 :
3849 2197646 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3850 :
3851 : /*
3852 : * Let the toaster do its thing, if needed.
3853 : *
3854 : * Note: below this point, heaptup is the data we actually intend to
3855 : * store into the relation; newtup is the caller's original untoasted
3856 : * data.
3857 : */
3858 2197646 : if (need_toast)
3859 : {
3860 : /* Note we always use WAL and FSM during updates */
3861 1895 : heaptup = heap_toast_insert_or_update(relation, newtup, &oldtup, 0);
3862 1895 : newtupsize = MAXALIGN(heaptup->t_len);
3863 : }
3864 : else
3865 2195751 : heaptup = newtup;
3866 :
3867 : /*
3868 : * Now, do we need a new page for the tuple, or not? This is a bit
3869 : * tricky since someone else could have added tuples to the page while
3870 : * we weren't looking. We have to recheck the available space after
3871 : * reacquiring the buffer lock. But don't bother to do that if the
3872 : * former amount of free space is still not enough; it's unlikely
3873 : * there's more free now than before.
3874 : *
3875 : * What's more, if we need to get a new page, we will need to acquire
3876 : * buffer locks on both old and new pages. To avoid deadlock against
3877 : * some other backend trying to get the same two locks in the other
3878 : * order, we must be consistent about the order we get the locks in.
3879 : * We use the rule "lock the lower-numbered page of the relation
3880 : * first". To implement this, we must do RelationGetBufferForTuple
3881 : * while not holding the lock on the old page, and we must rely on it
3882 : * to get the locks on both pages in the correct order.
3883 : *
3884 : * Another consideration is that we need visibility map page pin(s) if
3885 : * we will have to clear the all-visible flag on either page. If we
3886 : * call RelationGetBufferForTuple, we rely on it to acquire any such
3887 : * pins; but if we don't, we have to handle that here. Hence we need
3888 : * a loop.
3889 : */
3890 : for (;;)
3891 : {
3892 2197646 : if (newtupsize > pagefree)
3893 : {
3894 : /* It doesn't fit, must use RelationGetBufferForTuple. */
3895 2196920 : newbuf = RelationGetBufferForTuple(relation, heaptup->t_len,
3896 : buffer, 0, NULL,
3897 : &vmbuffer_new, &vmbuffer,
3898 : 0);
3899 : /* We're all done. */
3900 2196920 : break;
3901 : }
3902 : /* Acquire VM page pin if needed and we don't have it. */
3903 726 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
3904 0 : visibilitymap_pin(relation, block, &vmbuffer);
3905 : /* Re-acquire the lock on the old tuple's page. */
3906 726 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3907 : /* Re-check using the up-to-date free space */
3908 726 : pagefree = PageGetHeapFreeSpace(page);
3909 726 : if (newtupsize > pagefree ||
3910 726 : (vmbuffer == InvalidBuffer && PageIsAllVisible(page)))
3911 : {
3912 : /*
3913 : * Rats, it doesn't fit anymore, or somebody just now set the
3914 : * all-visible flag. We must now unlock and loop to avoid
3915 : * deadlock. Fortunately, this path should seldom be taken.
3916 : */
3917 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3918 : }
3919 : else
3920 : {
3921 : /* We're all done. */
3922 726 : newbuf = buffer;
3923 726 : break;
3924 : }
3925 : }
3926 : }
3927 : else
3928 : {
3929 : /* No TOAST work needed, and it'll fit on same page */
3930 184677 : newbuf = buffer;
3931 184677 : heaptup = newtup;
3932 : }
3933 :
3934 2382323 : newpage = BufferGetPage(newbuf);
3935 :
3936 : /*
3937 : * We're about to do the actual update -- check for conflict first, to
3938 : * avoid possibly having to roll back work we've just done.
3939 : *
3940 : * This is safe without a recheck as long as there is no possibility of
3941 : * another process scanning the pages between this check and the update
3942 : * being visible to the scan (i.e., exclusive buffer content lock(s) are
3943 : * continuously held from this point until the tuple update is visible).
3944 : *
3945 : * For the new tuple the only check needed is at the relation level, but
3946 : * since both tuples are in the same relation and the check for oldtup
3947 : * will include checking the relation level, there is no benefit to a
3948 : * separate check for the new tuple.
3949 : */
3950 2382323 : CheckForSerializableConflictIn(relation, &oldtup.t_self,
3951 : BufferGetBlockNumber(buffer));
3952 :
3953 : /*
3954 : * At this point newbuf and buffer are both pinned and locked, and newbuf
3955 : * has enough space for the new tuple. If they are the same buffer, only
3956 : * one pin is held.
3957 : */
3958 :
3959 2382311 : if (newbuf == buffer)
3960 : {
3961 : /*
3962 : * Since the new tuple is going into the same page, we might be able
3963 : * to do a HOT update. Check if any of the index columns have been
3964 : * changed.
3965 : */
3966 185391 : if (!bms_overlap(modified_attrs, hot_attrs))
3967 : {
3968 168746 : use_hot_update = true;
3969 :
3970 : /*
3971 : * If none of the columns that are used in hot-blocking indexes
3972 : * were updated, we can apply HOT, but we do still need to check
3973 : * if we need to update the summarizing indexes, and update those
3974 : * indexes if the columns were updated, or we may fail to detect
3975 : * e.g. value bound changes in BRIN minmax indexes.
3976 : */
3977 168746 : if (bms_overlap(modified_attrs, sum_attrs))
3978 2188 : summarized_update = true;
3979 : }
3980 : }
3981 : else
3982 : {
3983 : /* Set a hint that the old page could use prune/defrag */
3984 2196920 : PageSetFull(page);
3985 : }
3986 :
3987 : /*
3988 : * Compute replica identity tuple before entering the critical section so
3989 : * we don't PANIC upon a memory allocation failure.
3990 : * ExtractReplicaIdentity() will return NULL if nothing needs to be
3991 : * logged. Pass old key required as true only if the replica identity key
3992 : * columns are modified or it has external data.
3993 : */
3994 2382311 : old_key_tuple = ExtractReplicaIdentity(relation, &oldtup,
3995 2382311 : bms_overlap(modified_attrs, id_attrs) ||
3996 : id_has_external,
3997 2382311 : &old_key_copied);
3998 :
3999 : /* NO EREPORT(ERROR) from here till changes are logged */
4000 2382311 : START_CRIT_SECTION();
4001 :
4002 : /*
4003 : * If this transaction commits, the old tuple will become DEAD sooner or
4004 : * later. Set flag that this page is a candidate for pruning once our xid
4005 : * falls below the OldestXmin horizon. If the transaction finally aborts,
4006 : * the subsequent page pruning will be a no-op and the hint will be
4007 : * cleared.
4008 : *
4009 : * We set the new page prunable as well. See heap_insert() for more on why
4010 : * we do this when inserting tuples.
4011 : */
4012 2382311 : PageSetPrunable(page, xid);
4013 2382311 : if (newbuf != buffer)
4014 2196920 : PageSetPrunable(newpage, xid);
4015 :
4016 2382311 : if (use_hot_update)
4017 : {
4018 : /* Mark the old tuple as HOT-updated */
4019 168746 : HeapTupleSetHotUpdated(&oldtup);
4020 : /* And mark the new tuple as heap-only */
4021 168746 : HeapTupleSetHeapOnly(heaptup);
4022 : /* Mark the caller's copy too, in case different from heaptup */
4023 168746 : HeapTupleSetHeapOnly(newtup);
4024 : }
4025 : else
4026 : {
4027 : /* Make sure tuples are correctly marked as not-HOT */
4028 2213565 : HeapTupleClearHotUpdated(&oldtup);
4029 2213565 : HeapTupleClearHeapOnly(heaptup);
4030 2213565 : HeapTupleClearHeapOnly(newtup);
4031 : }
4032 :
4033 2382311 : RelationPutHeapTuple(relation, newbuf, heaptup, false); /* insert new tuple */
4034 :
4035 :
4036 : /* Clear obsolete visibility flags, possibly set by ourselves above... */
4037 2382311 : oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
4038 2382311 : oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
4039 : /* ... and store info about transaction updating this tuple */
4040 : Assert(TransactionIdIsValid(xmax_old_tuple));
4041 2382311 : HeapTupleHeaderSetXmax(oldtup.t_data, xmax_old_tuple);
4042 2382311 : oldtup.t_data->t_infomask |= infomask_old_tuple;
4043 2382311 : oldtup.t_data->t_infomask2 |= infomask2_old_tuple;
4044 2382311 : HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
4045 :
4046 : /* record address of new tuple in t_ctid of old one */
4047 2382311 : oldtup.t_data->t_ctid = heaptup->t_self;
4048 :
4049 : /* clear PD_ALL_VISIBLE flags, reset all visibilitymap bits */
4050 2382311 : if (PageIsAllVisible(page))
4051 : {
4052 2403 : all_visible_cleared = true;
4053 2403 : PageClearAllVisible(page);
4054 2403 : visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
4055 : vmbuffer, VISIBILITYMAP_VALID_BITS);
4056 : }
4057 2382311 : if (newbuf != buffer && PageIsAllVisible(newpage))
4058 : {
4059 815 : all_visible_cleared_new = true;
4060 815 : PageClearAllVisible(newpage);
4061 815 : visibilitymap_clear(relation, BufferGetBlockNumber(newbuf),
4062 : vmbuffer_new, VISIBILITYMAP_VALID_BITS);
4063 : }
4064 :
4065 2382311 : if (newbuf != buffer)
4066 2196920 : MarkBufferDirty(newbuf);
4067 2382311 : MarkBufferDirty(buffer);
4068 :
4069 : /* XLOG stuff */
4070 2382311 : if (RelationNeedsWAL(relation))
4071 : {
4072 : XLogRecPtr recptr;
4073 :
4074 : /*
4075 : * For logical decoding we need combo CIDs to properly decode the
4076 : * catalog.
4077 : */
4078 2370519 : if (RelationIsAccessibleInLogicalDecoding(relation))
4079 : {
4080 2652 : log_heap_new_cid(relation, &oldtup);
4081 2652 : log_heap_new_cid(relation, heaptup);
4082 : }
4083 :
4084 2370519 : recptr = log_heap_update(relation, buffer,
4085 : newbuf, &oldtup, heaptup,
4086 : old_key_tuple,
4087 : all_visible_cleared,
4088 : all_visible_cleared_new,
4089 : walLogical);
4090 2370519 : if (newbuf != buffer)
4091 : {
4092 2186793 : PageSetLSN(newpage, recptr);
4093 : }
4094 2370519 : PageSetLSN(page, recptr);
4095 : }
4096 :
4097 2382311 : END_CRIT_SECTION();
4098 :
4099 2382311 : if (newbuf != buffer)
4100 2196920 : LockBuffer(newbuf, BUFFER_LOCK_UNLOCK);
4101 2382311 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
4102 :
4103 : /*
4104 : * Mark old tuple for invalidation from system caches at next command
4105 : * boundary, and mark the new tuple for invalidation in case we abort. We
4106 : * have to do this before releasing the buffer because oldtup is in the
4107 : * buffer. (heaptup is all in local memory, but it's necessary to process
4108 : * both tuple versions in one call to inval.c so we can avoid redundant
4109 : * sinval messages.)
4110 : */
4111 2382311 : CacheInvalidateHeapTuple(relation, &oldtup, heaptup);
4112 :
4113 : /* Now we can release the buffer(s) */
4114 2382311 : if (newbuf != buffer)
4115 2196920 : ReleaseBuffer(newbuf);
4116 2382311 : ReleaseBuffer(buffer);
4117 2382311 : if (BufferIsValid(vmbuffer_new))
4118 816 : ReleaseBuffer(vmbuffer_new);
4119 2382311 : if (BufferIsValid(vmbuffer))
4120 2403 : ReleaseBuffer(vmbuffer);
4121 :
4122 : /*
4123 : * Release the lmgr tuple lock, if we had it.
4124 : */
4125 2382311 : if (have_tuple_lock)
4126 22 : UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
4127 :
4128 2382311 : pgstat_count_heap_update(relation, use_hot_update, newbuf != buffer);
4129 :
4130 : /*
4131 : * If heaptup is a private copy, release it. Don't forget to copy t_self
4132 : * back to the caller's image, too.
4133 : */
4134 2382311 : if (heaptup != newtup)
4135 : {
4136 1839 : newtup->t_self = heaptup->t_self;
4137 1839 : heap_freetuple(heaptup);
4138 : }
4139 :
4140 : /*
4141 : * If it is a HOT update, the update may still need to update summarized
4142 : * indexes, lest we fail to update those summaries and get incorrect
4143 : * results (for example, minmax bounds of the block may change with this
4144 : * update).
4145 : */
4146 2382311 : if (use_hot_update)
4147 : {
4148 168746 : if (summarized_update)
4149 2188 : *update_indexes = TU_Summarizing;
4150 : else
4151 166558 : *update_indexes = TU_None;
4152 : }
4153 : else
4154 2213565 : *update_indexes = TU_All;
4155 :
4156 2382311 : if (old_key_tuple != NULL && old_key_copied)
4157 99 : heap_freetuple(old_key_tuple);
4158 :
4159 2382311 : bms_free(hot_attrs);
4160 2382311 : bms_free(sum_attrs);
4161 2382311 : bms_free(key_attrs);
4162 2382311 : bms_free(id_attrs);
4163 2382311 : bms_free(modified_attrs);
4164 2382311 : bms_free(interesting_attrs);
4165 :
4166 2382311 : return TM_Ok;
4167 : }
4168 :
4169 : #ifdef USE_ASSERT_CHECKING
4170 : /*
4171 : * Confirm adequate lock held during heap_update(), per rules from
4172 : * README.tuplock section "Locking to write inplace-updated tables".
4173 : */
4174 : static void
4175 : check_lock_if_inplace_updateable_rel(Relation relation,
4176 : const ItemPointerData *otid,
4177 : HeapTuple newtup)
4178 : {
4179 : /* LOCKTAG_TUPLE acceptable for any catalog */
4180 : switch (RelationGetRelid(relation))
4181 : {
4182 : case RelationRelationId:
4183 : case DatabaseRelationId:
4184 : {
4185 : LOCKTAG tuptag;
4186 :
4187 : SET_LOCKTAG_TUPLE(tuptag,
4188 : relation->rd_lockInfo.lockRelId.dbId,
4189 : relation->rd_lockInfo.lockRelId.relId,
4190 : ItemPointerGetBlockNumber(otid),
4191 : ItemPointerGetOffsetNumber(otid));
4192 : if (LockHeldByMe(&tuptag, InplaceUpdateTupleLock, false))
4193 : return;
4194 : }
4195 : break;
4196 : default:
4197 : Assert(!IsInplaceUpdateRelation(relation));
4198 : return;
4199 : }
4200 :
4201 : switch (RelationGetRelid(relation))
4202 : {
4203 : case RelationRelationId:
4204 : {
4205 : /* LOCKTAG_TUPLE or LOCKTAG_RELATION ok */
4206 : Form_pg_class classForm = (Form_pg_class) GETSTRUCT(newtup);
4207 : Oid relid = classForm->oid;
4208 : Oid dbid;
4209 : LOCKTAG tag;
4210 :
4211 : if (IsSharedRelation(relid))
4212 : dbid = InvalidOid;
4213 : else
4214 : dbid = MyDatabaseId;
4215 :
4216 : if (classForm->relkind == RELKIND_INDEX)
4217 : {
4218 : Relation irel = index_open(relid, AccessShareLock);
4219 :
4220 : SET_LOCKTAG_RELATION(tag, dbid, irel->rd_index->indrelid);
4221 : index_close(irel, AccessShareLock);
4222 : }
4223 : else
4224 : SET_LOCKTAG_RELATION(tag, dbid, relid);
4225 :
4226 : if (!LockHeldByMe(&tag, ShareUpdateExclusiveLock, false) &&
4227 : !LockHeldByMe(&tag, ShareRowExclusiveLock, true))
4228 : elog(WARNING,
4229 : "missing lock for relation \"%s\" (OID %u, relkind %c) @ TID (%u,%u)",
4230 : NameStr(classForm->relname),
4231 : relid,
4232 : classForm->relkind,
4233 : ItemPointerGetBlockNumber(otid),
4234 : ItemPointerGetOffsetNumber(otid));
4235 : }
4236 : break;
4237 : case DatabaseRelationId:
4238 : {
4239 : /* LOCKTAG_TUPLE required */
4240 : Form_pg_database dbForm = (Form_pg_database) GETSTRUCT(newtup);
4241 :
4242 : elog(WARNING,
4243 : "missing lock on database \"%s\" (OID %u) @ TID (%u,%u)",
4244 : NameStr(dbForm->datname),
4245 : dbForm->oid,
4246 : ItemPointerGetBlockNumber(otid),
4247 : ItemPointerGetOffsetNumber(otid));
4248 : }
4249 : break;
4250 : }
4251 : }
4252 :
4253 : /*
4254 : * Confirm adequate relation lock held, per rules from README.tuplock section
4255 : * "Locking to write inplace-updated tables".
4256 : */
4257 : static void
4258 : check_inplace_rel_lock(HeapTuple oldtup)
4259 : {
4260 : Form_pg_class classForm = (Form_pg_class) GETSTRUCT(oldtup);
4261 : Oid relid = classForm->oid;
4262 : Oid dbid;
4263 : LOCKTAG tag;
4264 :
4265 : if (IsSharedRelation(relid))
4266 : dbid = InvalidOid;
4267 : else
4268 : dbid = MyDatabaseId;
4269 :
4270 : if (classForm->relkind == RELKIND_INDEX)
4271 : {
4272 : Relation irel = index_open(relid, AccessShareLock);
4273 :
4274 : SET_LOCKTAG_RELATION(tag, dbid, irel->rd_index->indrelid);
4275 : index_close(irel, AccessShareLock);
4276 : }
4277 : else
4278 : SET_LOCKTAG_RELATION(tag, dbid, relid);
4279 :
4280 : if (!LockHeldByMe(&tag, ShareUpdateExclusiveLock, true))
4281 : elog(WARNING,
4282 : "missing lock for relation \"%s\" (OID %u, relkind %c) @ TID (%u,%u)",
4283 : NameStr(classForm->relname),
4284 : relid,
4285 : classForm->relkind,
4286 : ItemPointerGetBlockNumber(&oldtup->t_self),
4287 : ItemPointerGetOffsetNumber(&oldtup->t_self));
4288 : }
4289 : #endif
4290 :
4291 : /*
4292 : * Check if the specified attribute's values are the same. Subroutine for
4293 : * HeapDetermineColumnsInfo.
4294 : */
4295 : static bool
4296 978443 : heap_attr_equals(TupleDesc tupdesc, int attrnum, Datum value1, Datum value2,
4297 : bool isnull1, bool isnull2)
4298 : {
4299 : /*
4300 : * If one value is NULL and other is not, then they are certainly not
4301 : * equal
4302 : */
4303 978443 : if (isnull1 != isnull2)
4304 60 : return false;
4305 :
4306 : /*
4307 : * If both are NULL, they can be considered equal.
4308 : */
4309 978383 : if (isnull1)
4310 6641 : return true;
4311 :
4312 : /*
4313 : * We do simple binary comparison of the two datums. This may be overly
4314 : * strict because there can be multiple binary representations for the
4315 : * same logical value. But we should be OK as long as there are no false
4316 : * positives. Using a type-specific equality operator is messy because
4317 : * there could be multiple notions of equality in different operator
4318 : * classes; furthermore, we cannot safely invoke user-defined functions
4319 : * while holding exclusive buffer lock.
4320 : */
4321 971742 : if (attrnum <= 0)
4322 : {
4323 : /* The only allowed system columns are OIDs, so do this */
4324 0 : return (DatumGetObjectId(value1) == DatumGetObjectId(value2));
4325 : }
4326 : else
4327 : {
4328 : CompactAttribute *att;
4329 :
4330 : Assert(attrnum <= tupdesc->natts);
4331 971742 : att = TupleDescCompactAttr(tupdesc, attrnum - 1);
4332 971742 : return datumIsEqual(value1, value2, att->attbyval, att->attlen);
4333 : }
4334 : }
4335 :
4336 : /*
4337 : * Check which columns are being updated.
4338 : *
4339 : * Given an updated tuple, determine (and return into the output bitmapset),
4340 : * from those listed as interesting, the set of columns that changed.
4341 : *
4342 : * has_external indicates if any of the unmodified attributes (from those
4343 : * listed as interesting) of the old tuple is a member of external_cols and is
4344 : * stored externally.
4345 : */
4346 : static Bitmapset *
4347 2382534 : HeapDetermineColumnsInfo(Relation relation,
4348 : Bitmapset *interesting_cols,
4349 : Bitmapset *external_cols,
4350 : HeapTuple oldtup, HeapTuple newtup,
4351 : bool *has_external)
4352 : {
4353 : int attidx;
4354 2382534 : Bitmapset *modified = NULL;
4355 2382534 : TupleDesc tupdesc = RelationGetDescr(relation);
4356 :
4357 2382534 : attidx = -1;
4358 3360977 : while ((attidx = bms_next_member(interesting_cols, attidx)) >= 0)
4359 : {
4360 : /* attidx is zero-based, attrnum is the normal attribute number */
4361 978443 : AttrNumber attrnum = attidx + FirstLowInvalidHeapAttributeNumber;
4362 : Datum value1,
4363 : value2;
4364 : bool isnull1,
4365 : isnull2;
4366 :
4367 : /*
4368 : * If it's a whole-tuple reference, say "not equal". It's not really
4369 : * worth supporting this case, since it could only succeed after a
4370 : * no-op update, which is hardly a case worth optimizing for.
4371 : */
4372 978443 : if (attrnum == 0)
4373 : {
4374 0 : modified = bms_add_member(modified, attidx);
4375 946074 : continue;
4376 : }
4377 :
4378 : /*
4379 : * Likewise, automatically say "not equal" for any system attribute
4380 : * other than tableOID; we cannot expect these to be consistent in a
4381 : * HOT chain, or even to be set correctly yet in the new tuple.
4382 : */
4383 978443 : if (attrnum < 0)
4384 : {
4385 0 : if (attrnum != TableOidAttributeNumber)
4386 : {
4387 0 : modified = bms_add_member(modified, attidx);
4388 0 : continue;
4389 : }
4390 : }
4391 :
4392 : /*
4393 : * Extract the corresponding values. XXX this is pretty inefficient
4394 : * if there are many indexed columns. Should we do a single
4395 : * heap_deform_tuple call on each tuple, instead? But that doesn't
4396 : * work for system columns ...
4397 : */
4398 978443 : value1 = heap_getattr(oldtup, attrnum, tupdesc, &isnull1);
4399 978443 : value2 = heap_getattr(newtup, attrnum, tupdesc, &isnull2);
4400 :
4401 978443 : if (!heap_attr_equals(tupdesc, attrnum, value1,
4402 : value2, isnull1, isnull2))
4403 : {
4404 54518 : modified = bms_add_member(modified, attidx);
4405 54518 : continue;
4406 : }
4407 :
4408 : /*
4409 : * No need to check attributes that can't be stored externally. Note
4410 : * that system attributes can't be stored externally.
4411 : */
4412 923925 : if (attrnum < 0 || isnull1 ||
4413 917284 : TupleDescCompactAttr(tupdesc, attrnum - 1)->attlen != -1)
4414 891556 : continue;
4415 :
4416 : /*
4417 : * Check if the old tuple's attribute is stored externally and is a
4418 : * member of external_cols.
4419 : */
4420 32374 : if (VARATT_IS_EXTERNAL((varlena *) DatumGetPointer(value1)) &&
4421 5 : bms_is_member(attidx, external_cols))
4422 2 : *has_external = true;
4423 : }
4424 :
4425 2382534 : return modified;
4426 : }
4427 :
4428 : /*
4429 : * simple_heap_update - replace a tuple
4430 : *
4431 : * This routine may be used to update a tuple when concurrent updates of
4432 : * the target tuple are not expected (for example, because we have a lock
4433 : * on the relation associated with the tuple). Any failure is reported
4434 : * via ereport().
4435 : */
4436 : void
4437 138213 : simple_heap_update(Relation relation, const ItemPointerData *otid, HeapTuple tup,
4438 : TU_UpdateIndexes *update_indexes)
4439 : {
4440 : TM_Result result;
4441 : TM_FailureData tmfd;
4442 : LockTupleMode lockmode;
4443 :
4444 138213 : result = heap_update(relation, otid, tup,
4445 : GetCurrentCommandId(true), 0,
4446 : InvalidSnapshot,
4447 : true /* wait for commit */ ,
4448 : &tmfd, &lockmode, update_indexes);
4449 138213 : switch (result)
4450 : {
4451 0 : case TM_SelfModified:
4452 : /* Tuple was already updated in current command? */
4453 0 : elog(ERROR, "tuple already updated by self");
4454 : break;
4455 :
4456 138212 : case TM_Ok:
4457 : /* done successfully */
4458 138212 : break;
4459 :
4460 0 : case TM_Updated:
4461 0 : elog(ERROR, "tuple concurrently updated");
4462 : break;
4463 :
4464 1 : case TM_Deleted:
4465 1 : elog(ERROR, "tuple concurrently deleted");
4466 : break;
4467 :
4468 0 : default:
4469 0 : elog(ERROR, "unrecognized heap_update status: %u", result);
4470 : break;
4471 : }
4472 138212 : }
4473 :
4474 :
4475 : /*
4476 : * Return the MultiXactStatus corresponding to the given tuple lock mode.
4477 : */
4478 : static MultiXactStatus
4479 115517 : get_mxact_status_for_lock(LockTupleMode mode, bool is_update)
4480 : {
4481 : int retval;
4482 :
4483 115517 : if (is_update)
4484 216 : retval = tupleLockExtraInfo[mode].updstatus;
4485 : else
4486 115301 : retval = tupleLockExtraInfo[mode].lockstatus;
4487 :
4488 115517 : if (retval == -1)
4489 0 : elog(ERROR, "invalid lock tuple mode %d/%s", mode,
4490 : is_update ? "true" : "false");
4491 :
4492 115517 : return (MultiXactStatus) retval;
4493 : }
4494 :
4495 : /*
4496 : * heap_lock_tuple - lock a tuple in shared or exclusive mode
4497 : *
4498 : * Note that this acquires a buffer pin, which the caller must release.
4499 : *
4500 : * Input parameters:
4501 : * relation: relation containing tuple (caller must hold suitable lock)
4502 : * cid: current command ID (used for visibility test, and stored into
4503 : * tuple's cmax if lock is successful)
4504 : * mode: indicates if shared or exclusive tuple lock is desired
4505 : * wait_policy: what to do if tuple lock is not available
4506 : * follow_updates: if true, follow the update chain to also lock descendant
4507 : * tuples.
4508 : *
4509 : * Output parameters:
4510 : * *tuple: all fields filled in
4511 : * *buffer: set to buffer holding tuple (pinned but not locked at exit)
4512 : * *tmfd: filled in failure cases (see below)
4513 : *
4514 : * Function results are the same as the ones for table_tuple_lock().
4515 : *
4516 : * In the failure cases other than TM_Invisible, the routine fills
4517 : * *tmfd with the tuple's t_ctid, t_xmax (resolving a possible MultiXact,
4518 : * if necessary), and t_cmax (the last only for TM_SelfModified,
4519 : * since we cannot obtain cmax from a combo CID generated by another
4520 : * transaction).
4521 : * See comments for struct TM_FailureData for additional info.
4522 : *
4523 : * See README.tuplock for a thorough explanation of this mechanism.
4524 : */
4525 : TM_Result
4526 570477 : heap_lock_tuple(Relation relation, HeapTuple tuple,
4527 : CommandId cid, LockTupleMode mode, LockWaitPolicy wait_policy,
4528 : bool follow_updates,
4529 : Buffer *buffer, TM_FailureData *tmfd)
4530 : {
4531 : TM_Result result;
4532 570477 : ItemPointer tid = &(tuple->t_self);
4533 : ItemId lp;
4534 : Page page;
4535 570477 : Buffer vmbuffer = InvalidBuffer;
4536 : BlockNumber block;
4537 : TransactionId xid,
4538 : xmax;
4539 : uint16 old_infomask,
4540 : new_infomask,
4541 : new_infomask2;
4542 570477 : bool first_time = true;
4543 570477 : bool skip_tuple_lock = false;
4544 570477 : bool have_tuple_lock = false;
4545 570477 : bool cleared_all_frozen = false;
4546 :
4547 570477 : *buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
4548 570477 : block = ItemPointerGetBlockNumber(tid);
4549 :
4550 : /*
4551 : * Before locking the buffer, pin the visibility map page if it appears to
4552 : * be necessary. Since we haven't got the lock yet, someone else might be
4553 : * in the middle of changing this, so we'll need to recheck after we have
4554 : * the lock.
4555 : */
4556 570477 : if (PageIsAllVisible(BufferGetPage(*buffer)))
4557 413522 : visibilitymap_pin(relation, block, &vmbuffer);
4558 :
4559 570477 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4560 :
4561 570477 : page = BufferGetPage(*buffer);
4562 570477 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
4563 : Assert(ItemIdIsNormal(lp));
4564 :
4565 570477 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
4566 570477 : tuple->t_len = ItemIdGetLength(lp);
4567 570477 : tuple->t_tableOid = RelationGetRelid(relation);
4568 :
4569 15 : l3:
4570 570492 : result = HeapTupleSatisfiesUpdate(tuple, cid, *buffer);
4571 :
4572 570492 : if (result == TM_Invisible)
4573 : {
4574 : /*
4575 : * This is possible, but only when locking a tuple for ON CONFLICT DO
4576 : * SELECT/UPDATE. We return this value here rather than throwing an
4577 : * error in order to give that case the opportunity to throw a more
4578 : * specific error.
4579 : */
4580 28 : result = TM_Invisible;
4581 28 : goto out_locked;
4582 : }
4583 570464 : else if (result == TM_BeingModified ||
4584 80195 : result == TM_Updated ||
4585 : result == TM_Deleted)
4586 : {
4587 : TransactionId xwait;
4588 : uint16 infomask;
4589 : uint16 infomask2;
4590 : bool require_sleep;
4591 : ItemPointerData t_ctid;
4592 :
4593 : /* must copy state data before unlocking buffer */
4594 490271 : xwait = HeapTupleHeaderGetRawXmax(tuple->t_data);
4595 490271 : infomask = tuple->t_data->t_infomask;
4596 490271 : infomask2 = tuple->t_data->t_infomask2;
4597 490271 : ItemPointerCopy(&tuple->t_data->t_ctid, &t_ctid);
4598 :
4599 490271 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
4600 :
4601 : /*
4602 : * If any subtransaction of the current top transaction already holds
4603 : * a lock as strong as or stronger than what we're requesting, we
4604 : * effectively hold the desired lock already. We *must* succeed
4605 : * without trying to take the tuple lock, else we will deadlock
4606 : * against anyone wanting to acquire a stronger lock.
4607 : *
4608 : * Note we only do this the first time we loop on the HTSU result;
4609 : * there is no point in testing in subsequent passes, because
4610 : * evidently our own transaction cannot have acquired a new lock after
4611 : * the first time we checked.
4612 : */
4613 490271 : if (first_time)
4614 : {
4615 490259 : first_time = false;
4616 :
4617 490259 : if (infomask & HEAP_XMAX_IS_MULTI)
4618 : {
4619 : int i;
4620 : int nmembers;
4621 : MultiXactMember *members;
4622 :
4623 : /*
4624 : * We don't need to allow old multixacts here; if that had
4625 : * been the case, HeapTupleSatisfiesUpdate would have returned
4626 : * MayBeUpdated and we wouldn't be here.
4627 : */
4628 : nmembers =
4629 73297 : GetMultiXactIdMembers(xwait, &members, false,
4630 73297 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
4631 :
4632 1422660 : for (i = 0; i < nmembers; i++)
4633 : {
4634 : /* only consider members of our own transaction */
4635 1349377 : if (!TransactionIdIsCurrentTransactionId(members[i].xid))
4636 1349327 : continue;
4637 :
4638 50 : if (TUPLOCK_from_mxstatus(members[i].status) >= mode)
4639 : {
4640 14 : pfree(members);
4641 14 : result = TM_Ok;
4642 14 : goto out_unlocked;
4643 : }
4644 : else
4645 : {
4646 : /*
4647 : * Disable acquisition of the heavyweight tuple lock.
4648 : * Otherwise, when promoting a weaker lock, we might
4649 : * deadlock with another locker that has acquired the
4650 : * heavyweight tuple lock and is waiting for our
4651 : * transaction to finish.
4652 : *
4653 : * Note that in this case we still need to wait for
4654 : * the multixact if required, to avoid acquiring
4655 : * conflicting locks.
4656 : */
4657 36 : skip_tuple_lock = true;
4658 : }
4659 : }
4660 :
4661 73283 : if (members)
4662 73283 : pfree(members);
4663 : }
4664 416962 : else if (TransactionIdIsCurrentTransactionId(xwait))
4665 : {
4666 415543 : switch (mode)
4667 : {
4668 409099 : case LockTupleKeyShare:
4669 : Assert(HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) ||
4670 : HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4671 : HEAP_XMAX_IS_EXCL_LOCKED(infomask));
4672 409099 : result = TM_Ok;
4673 409099 : goto out_unlocked;
4674 41 : case LockTupleShare:
4675 47 : if (HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4676 6 : HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4677 : {
4678 35 : result = TM_Ok;
4679 35 : goto out_unlocked;
4680 : }
4681 6 : break;
4682 82 : case LockTupleNoKeyExclusive:
4683 82 : if (HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4684 : {
4685 70 : result = TM_Ok;
4686 70 : goto out_unlocked;
4687 : }
4688 12 : break;
4689 6321 : case LockTupleExclusive:
4690 6321 : if (HEAP_XMAX_IS_EXCL_LOCKED(infomask) &&
4691 1279 : infomask2 & HEAP_KEYS_UPDATED)
4692 : {
4693 1251 : result = TM_Ok;
4694 1251 : goto out_unlocked;
4695 : }
4696 5070 : break;
4697 : }
4698 : }
4699 : }
4700 :
4701 : /*
4702 : * Initially assume that we will have to wait for the locking
4703 : * transaction(s) to finish. We check various cases below in which
4704 : * this can be turned off.
4705 : */
4706 79802 : require_sleep = true;
4707 79802 : if (mode == LockTupleKeyShare)
4708 : {
4709 : /*
4710 : * If we're requesting KeyShare, and there's no update present, we
4711 : * don't need to wait. Even if there is an update, we can still
4712 : * continue if the key hasn't been modified.
4713 : *
4714 : * However, if there are updates, we need to walk the update chain
4715 : * to mark future versions of the row as locked, too. That way,
4716 : * if somebody deletes that future version, we're protected
4717 : * against the key going away. This locking of future versions
4718 : * could block momentarily, if a concurrent transaction is
4719 : * deleting a key; or it could return a value to the effect that
4720 : * the transaction deleting the key has already committed. So we
4721 : * do this before re-locking the buffer; otherwise this would be
4722 : * prone to deadlocks.
4723 : *
4724 : * Note that the TID we're locking was grabbed before we unlocked
4725 : * the buffer. For it to change while we're not looking, the
4726 : * other properties we're testing for below after re-locking the
4727 : * buffer would also change, in which case we would restart this
4728 : * loop above.
4729 : */
4730 73892 : if (!(infomask2 & HEAP_KEYS_UPDATED))
4731 : {
4732 : bool updated;
4733 :
4734 73846 : updated = !HEAP_XMAX_IS_LOCKED_ONLY(infomask);
4735 :
4736 : /*
4737 : * If there are updates, follow the update chain; bail out if
4738 : * that cannot be done.
4739 : */
4740 73846 : if (follow_updates && updated &&
4741 2171 : !ItemPointerEquals(&tuple->t_self, &t_ctid))
4742 : {
4743 : TM_Result res;
4744 :
4745 2171 : res = heap_lock_updated_tuple(relation,
4746 : infomask, xwait, &t_ctid,
4747 : GetCurrentTransactionId(),
4748 : mode);
4749 2171 : if (res != TM_Ok)
4750 : {
4751 6 : result = res;
4752 : /* recovery code expects to have buffer lock held */
4753 6 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4754 209 : goto failed;
4755 : }
4756 : }
4757 :
4758 73840 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4759 :
4760 : /*
4761 : * Make sure it's still an appropriate lock, else start over.
4762 : * Also, if it wasn't updated before we released the lock, but
4763 : * is updated now, we start over too; the reason is that we
4764 : * now need to follow the update chain to lock the new
4765 : * versions.
4766 : */
4767 73840 : if (!HeapTupleHeaderIsOnlyLocked(tuple->t_data) &&
4768 2153 : ((tuple->t_data->t_infomask2 & HEAP_KEYS_UPDATED) ||
4769 2153 : !updated))
4770 15 : goto l3;
4771 :
4772 : /* Things look okay, so we can skip sleeping */
4773 73840 : require_sleep = false;
4774 :
4775 : /*
4776 : * Note we allow Xmax to change here; other updaters/lockers
4777 : * could have modified it before we grabbed the buffer lock.
4778 : * However, this is not a problem, because with the recheck we
4779 : * just did we ensure that they still don't conflict with the
4780 : * lock we want.
4781 : */
4782 : }
4783 : }
4784 5910 : else if (mode == LockTupleShare)
4785 : {
4786 : /*
4787 : * If we're requesting Share, we can similarly avoid sleeping if
4788 : * there's no update and no exclusive lock present.
4789 : */
4790 480 : if (HEAP_XMAX_IS_LOCKED_ONLY(infomask) &&
4791 480 : !HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4792 : {
4793 474 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4794 :
4795 : /*
4796 : * Make sure it's still an appropriate lock, else start over.
4797 : * See above about allowing xmax to change.
4798 : */
4799 948 : if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
4800 474 : HEAP_XMAX_IS_EXCL_LOCKED(tuple->t_data->t_infomask))
4801 0 : goto l3;
4802 474 : require_sleep = false;
4803 : }
4804 : }
4805 5430 : else if (mode == LockTupleNoKeyExclusive)
4806 : {
4807 : /*
4808 : * If we're requesting NoKeyExclusive, we might also be able to
4809 : * avoid sleeping; just ensure that there no conflicting lock
4810 : * already acquired.
4811 : */
4812 175 : if (infomask & HEAP_XMAX_IS_MULTI)
4813 : {
4814 26 : if (!DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
4815 : mode, NULL))
4816 : {
4817 : /*
4818 : * No conflict, but if the xmax changed under us in the
4819 : * meantime, start over.
4820 : */
4821 13 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4822 26 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4823 13 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4824 : xwait))
4825 0 : goto l3;
4826 :
4827 : /* otherwise, we're good */
4828 13 : require_sleep = false;
4829 : }
4830 : }
4831 149 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask))
4832 : {
4833 18 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4834 :
4835 : /* if the xmax changed in the meantime, start over */
4836 36 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4837 18 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4838 : xwait))
4839 0 : goto l3;
4840 : /* otherwise, we're good */
4841 18 : require_sleep = false;
4842 : }
4843 : }
4844 :
4845 : /*
4846 : * As a check independent from those above, we can also avoid sleeping
4847 : * if the current transaction is the sole locker of the tuple. Note
4848 : * that the strength of the lock already held is irrelevant; this is
4849 : * not about recording the lock in Xmax (which will be done regardless
4850 : * of this optimization, below). Also, note that the cases where we
4851 : * hold a lock stronger than we are requesting are already handled
4852 : * above by not doing anything.
4853 : *
4854 : * Note we only deal with the non-multixact case here; MultiXactIdWait
4855 : * is well equipped to deal with this situation on its own.
4856 : */
4857 85203 : if (require_sleep && !(infomask & HEAP_XMAX_IS_MULTI) &&
4858 5407 : TransactionIdIsCurrentTransactionId(xwait))
4859 : {
4860 : /* ... but if the xmax changed in the meantime, start over */
4861 5070 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4862 10140 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4863 5070 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4864 : xwait))
4865 0 : goto l3;
4866 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask));
4867 5070 : require_sleep = false;
4868 : }
4869 :
4870 : /*
4871 : * Time to sleep on the other transaction/multixact, if necessary.
4872 : *
4873 : * If the other transaction is an update/delete that's already
4874 : * committed, then sleeping cannot possibly do any good: if we're
4875 : * required to sleep, get out to raise an error instead.
4876 : *
4877 : * By here, we either have already acquired the buffer exclusive lock,
4878 : * or we must wait for the locking transaction or multixact; so below
4879 : * we ensure that we grab buffer lock after the sleep.
4880 : */
4881 79796 : if (require_sleep && (result == TM_Updated || result == TM_Deleted))
4882 : {
4883 164 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4884 164 : goto failed;
4885 : }
4886 79632 : else if (require_sleep)
4887 : {
4888 : /*
4889 : * Acquire tuple lock to establish our priority for the tuple, or
4890 : * die trying. LockTuple will release us when we are next-in-line
4891 : * for the tuple. We must do this even if we are share-locking,
4892 : * but not if we already have a weaker lock on the tuple.
4893 : *
4894 : * If we are forced to "start over" below, we keep the tuple lock;
4895 : * this arranges that we stay at the head of the line while
4896 : * rechecking tuple state.
4897 : */
4898 217 : if (!skip_tuple_lock &&
4899 200 : !heap_acquire_tuplock(relation, tid, mode, wait_policy,
4900 : &have_tuple_lock))
4901 : {
4902 : /*
4903 : * This can only happen if wait_policy is Skip and the lock
4904 : * couldn't be obtained.
4905 : */
4906 1 : result = TM_WouldBlock;
4907 : /* recovery code expects to have buffer lock held */
4908 1 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4909 1 : goto failed;
4910 : }
4911 :
4912 215 : if (infomask & HEAP_XMAX_IS_MULTI)
4913 : {
4914 43 : MultiXactStatus status = get_mxact_status_for_lock(mode, false);
4915 :
4916 : /* We only ever lock tuples, never update them */
4917 43 : if (status >= MultiXactStatusNoKeyUpdate)
4918 0 : elog(ERROR, "invalid lock mode in heap_lock_tuple");
4919 :
4920 : /* wait for multixact to end, or die trying */
4921 43 : switch (wait_policy)
4922 : {
4923 37 : case LockWaitBlock:
4924 37 : MultiXactIdWait((MultiXactId) xwait, status, infomask,
4925 37 : relation, &tuple->t_self, XLTW_Lock, NULL);
4926 37 : break;
4927 2 : case LockWaitSkip:
4928 2 : if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
4929 : status, infomask, relation,
4930 : NULL, false))
4931 : {
4932 2 : result = TM_WouldBlock;
4933 : /* recovery code expects to have buffer lock held */
4934 2 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4935 2 : goto failed;
4936 : }
4937 0 : break;
4938 4 : case LockWaitError:
4939 4 : if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
4940 : status, infomask, relation,
4941 : NULL, log_lock_failures))
4942 4 : ereport(ERROR,
4943 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
4944 : errmsg("could not obtain lock on row in relation \"%s\"",
4945 : RelationGetRelationName(relation))));
4946 :
4947 0 : break;
4948 : }
4949 :
4950 : /*
4951 : * Of course, the multixact might not be done here: if we're
4952 : * requesting a light lock mode, other transactions with light
4953 : * locks could still be alive, as well as locks owned by our
4954 : * own xact or other subxacts of this backend. We need to
4955 : * preserve the surviving MultiXact members. Note that it
4956 : * isn't absolutely necessary in the latter case, but doing so
4957 : * is simpler.
4958 : */
4959 : }
4960 : else
4961 : {
4962 : /* wait for regular transaction to end, or die trying */
4963 172 : switch (wait_policy)
4964 : {
4965 131 : case LockWaitBlock:
4966 131 : XactLockTableWait(xwait, relation, &tuple->t_self,
4967 : XLTW_Lock);
4968 131 : break;
4969 33 : case LockWaitSkip:
4970 33 : if (!ConditionalXactLockTableWait(xwait, false))
4971 : {
4972 33 : result = TM_WouldBlock;
4973 : /* recovery code expects to have buffer lock held */
4974 33 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4975 33 : goto failed;
4976 : }
4977 0 : break;
4978 8 : case LockWaitError:
4979 8 : if (!ConditionalXactLockTableWait(xwait, log_lock_failures))
4980 8 : ereport(ERROR,
4981 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
4982 : errmsg("could not obtain lock on row in relation \"%s\"",
4983 : RelationGetRelationName(relation))));
4984 0 : break;
4985 : }
4986 : }
4987 :
4988 : /* if there are updates, follow the update chain */
4989 168 : if (follow_updates && !HEAP_XMAX_IS_LOCKED_ONLY(infomask) &&
4990 75 : !ItemPointerEquals(&tuple->t_self, &t_ctid))
4991 : {
4992 : TM_Result res;
4993 :
4994 55 : res = heap_lock_updated_tuple(relation,
4995 : infomask, xwait, &t_ctid,
4996 : GetCurrentTransactionId(),
4997 : mode);
4998 55 : if (res != TM_Ok)
4999 : {
5000 3 : result = res;
5001 : /* recovery code expects to have buffer lock held */
5002 3 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5003 3 : goto failed;
5004 : }
5005 : }
5006 :
5007 165 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5008 :
5009 : /*
5010 : * xwait is done, but if xwait had just locked the tuple then some
5011 : * other xact could update this tuple before we get to this point.
5012 : * Check for xmax change, and start over if so.
5013 : */
5014 316 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
5015 151 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
5016 : xwait))
5017 15 : goto l3;
5018 :
5019 150 : if (!(infomask & HEAP_XMAX_IS_MULTI))
5020 : {
5021 : /*
5022 : * Otherwise check if it committed or aborted. Note we cannot
5023 : * be here if the tuple was only locked by somebody who didn't
5024 : * conflict with us; that would have been handled above. So
5025 : * that transaction must necessarily be gone by now. But
5026 : * don't check for this in the multixact case, because some
5027 : * locker transactions might still be running.
5028 : */
5029 115 : UpdateXmaxHintBits(tuple->t_data, *buffer, xwait);
5030 : }
5031 : }
5032 :
5033 : /* By here, we're certain that we hold buffer exclusive lock again */
5034 :
5035 : /*
5036 : * We may lock if previous xmax aborted, or if it committed but only
5037 : * locked the tuple without updating it; or if we didn't have to wait
5038 : * at all for whatever reason.
5039 : */
5040 79565 : if (!require_sleep ||
5041 264 : (tuple->t_data->t_infomask & HEAP_XMAX_INVALID) ||
5042 212 : HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
5043 98 : HeapTupleHeaderIsOnlyLocked(tuple->t_data))
5044 79476 : result = TM_Ok;
5045 89 : else if (!ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid))
5046 65 : result = TM_Updated;
5047 : else
5048 24 : result = TM_Deleted;
5049 : }
5050 :
5051 80193 : failed:
5052 159967 : if (result != TM_Ok)
5053 : {
5054 : Assert(result == TM_SelfModified || result == TM_Updated ||
5055 : result == TM_Deleted || result == TM_WouldBlock);
5056 :
5057 : /*
5058 : * When locking a tuple under LockWaitSkip semantics and we fail with
5059 : * TM_WouldBlock above, it's possible for concurrent transactions to
5060 : * release the lock and set HEAP_XMAX_INVALID in the meantime. So
5061 : * this assert is slightly different from the equivalent one in
5062 : * heap_delete and heap_update.
5063 : */
5064 : Assert((result == TM_WouldBlock) ||
5065 : !(tuple->t_data->t_infomask & HEAP_XMAX_INVALID));
5066 : Assert(result != TM_Updated ||
5067 : !ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid));
5068 306 : tmfd->ctid = tuple->t_data->t_ctid;
5069 306 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(tuple->t_data);
5070 306 : if (result == TM_SelfModified)
5071 8 : tmfd->cmax = HeapTupleHeaderGetCmax(tuple->t_data);
5072 : else
5073 298 : tmfd->cmax = InvalidCommandId;
5074 306 : goto out_locked;
5075 : }
5076 :
5077 : /*
5078 : * If we didn't pin the visibility map page and the page has become all
5079 : * visible while we were busy locking the buffer, or during some
5080 : * subsequent window during which we had it unlocked, we'll have to unlock
5081 : * and re-lock, to avoid holding the buffer lock across I/O. That's a bit
5082 : * unfortunate, especially since we'll now have to recheck whether the
5083 : * tuple has been locked or updated under us, but hopefully it won't
5084 : * happen very often.
5085 : */
5086 159661 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
5087 : {
5088 0 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
5089 0 : visibilitymap_pin(relation, block, &vmbuffer);
5090 0 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5091 0 : goto l3;
5092 : }
5093 :
5094 159661 : xmax = HeapTupleHeaderGetRawXmax(tuple->t_data);
5095 159661 : old_infomask = tuple->t_data->t_infomask;
5096 :
5097 : /*
5098 : * If this is the first possibly-multixact-able operation in the current
5099 : * transaction, set my per-backend OldestMemberMXactId setting. We can be
5100 : * certain that the transaction will never become a member of any older
5101 : * MultiXactIds than that. (We have to do this even if we end up just
5102 : * using our own TransactionId below, since some other backend could
5103 : * incorporate our XID into a MultiXact immediately afterwards.)
5104 : */
5105 159661 : MultiXactIdSetOldestMember();
5106 :
5107 : /*
5108 : * Compute the new xmax and infomask to store into the tuple. Note we do
5109 : * not modify the tuple just yet, because that would leave it in the wrong
5110 : * state if multixact.c elogs.
5111 : */
5112 159661 : compute_new_xmax_infomask(xmax, old_infomask, tuple->t_data->t_infomask2,
5113 : GetCurrentTransactionId(), mode, false,
5114 : &xid, &new_infomask, &new_infomask2);
5115 :
5116 159661 : START_CRIT_SECTION();
5117 :
5118 : /*
5119 : * Store transaction information of xact locking the tuple.
5120 : *
5121 : * Note: Cmax is meaningless in this context, so don't set it; this avoids
5122 : * possibly generating a useless combo CID. Moreover, if we're locking a
5123 : * previously updated tuple, it's important to preserve the Cmax.
5124 : *
5125 : * Also reset the HOT UPDATE bit, but only if there's no update; otherwise
5126 : * we would break the HOT chain.
5127 : */
5128 159661 : tuple->t_data->t_infomask &= ~HEAP_XMAX_BITS;
5129 159661 : tuple->t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
5130 159661 : tuple->t_data->t_infomask |= new_infomask;
5131 159661 : tuple->t_data->t_infomask2 |= new_infomask2;
5132 159661 : if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
5133 157512 : HeapTupleHeaderClearHotUpdated(tuple->t_data);
5134 159661 : HeapTupleHeaderSetXmax(tuple->t_data, xid);
5135 :
5136 : /*
5137 : * Make sure there is no forward chain link in t_ctid. Note that in the
5138 : * cases where the tuple has been updated, we must not overwrite t_ctid,
5139 : * because it was set by the updater. Moreover, if the tuple has been
5140 : * updated, we need to follow the update chain to lock the new versions of
5141 : * the tuple as well.
5142 : */
5143 159661 : if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
5144 157512 : tuple->t_data->t_ctid = *tid;
5145 :
5146 : /* Clear only the all-frozen bit on visibility map if needed */
5147 163286 : if (PageIsAllVisible(page) &&
5148 3625 : visibilitymap_clear(relation, block, vmbuffer,
5149 : VISIBILITYMAP_ALL_FROZEN))
5150 16 : cleared_all_frozen = true;
5151 :
5152 :
5153 159661 : MarkBufferDirty(*buffer);
5154 :
5155 : /*
5156 : * XLOG stuff. You might think that we don't need an XLOG record because
5157 : * there is no state change worth restoring after a crash. You would be
5158 : * wrong however: we have just written either a TransactionId or a
5159 : * MultiXactId that may never have been seen on disk before, and we need
5160 : * to make sure that there are XLOG entries covering those ID numbers.
5161 : * Else the same IDs might be re-used after a crash, which would be
5162 : * disastrous if this page made it to disk before the crash. Essentially
5163 : * we have to enforce the WAL log-before-data rule even in this case.
5164 : * (Also, in a PITR log-shipping or 2PC environment, we have to have XLOG
5165 : * entries for everything anyway.)
5166 : */
5167 159661 : if (RelationNeedsWAL(relation))
5168 : {
5169 : xl_heap_lock xlrec;
5170 : XLogRecPtr recptr;
5171 :
5172 159232 : XLogBeginInsert();
5173 159232 : XLogRegisterBuffer(0, *buffer, REGBUF_STANDARD);
5174 :
5175 159232 : xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
5176 159232 : xlrec.xmax = xid;
5177 318464 : xlrec.infobits_set = compute_infobits(new_infomask,
5178 159232 : tuple->t_data->t_infomask2);
5179 159232 : xlrec.flags = cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
5180 159232 : XLogRegisterData(&xlrec, SizeOfHeapLock);
5181 :
5182 : /* we don't decode row locks atm, so no need to log the origin */
5183 :
5184 159232 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
5185 :
5186 159232 : PageSetLSN(page, recptr);
5187 : }
5188 :
5189 159661 : END_CRIT_SECTION();
5190 :
5191 159661 : result = TM_Ok;
5192 :
5193 159995 : out_locked:
5194 159995 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
5195 :
5196 570464 : out_unlocked:
5197 570464 : if (BufferIsValid(vmbuffer))
5198 413522 : ReleaseBuffer(vmbuffer);
5199 :
5200 : /*
5201 : * Don't update the visibility map here. Locking a tuple doesn't change
5202 : * visibility info.
5203 : */
5204 :
5205 : /*
5206 : * Now that we have successfully marked the tuple as locked, we can
5207 : * release the lmgr tuple lock, if we had it.
5208 : */
5209 570464 : if (have_tuple_lock)
5210 181 : UnlockTupleTuplock(relation, tid, mode);
5211 :
5212 570464 : return result;
5213 : }
5214 :
5215 : /*
5216 : * Acquire heavyweight lock on the given tuple, in preparation for acquiring
5217 : * its normal, Xmax-based tuple lock.
5218 : *
5219 : * have_tuple_lock is an input and output parameter: on input, it indicates
5220 : * whether the lock has previously been acquired (and this function does
5221 : * nothing in that case). If this function returns success, have_tuple_lock
5222 : * has been flipped to true.
5223 : *
5224 : * Returns false if it was unable to obtain the lock; this can only happen if
5225 : * wait_policy is Skip.
5226 : */
5227 : static bool
5228 377 : heap_acquire_tuplock(Relation relation, const ItemPointerData *tid, LockTupleMode mode,
5229 : LockWaitPolicy wait_policy, bool *have_tuple_lock)
5230 : {
5231 377 : if (*have_tuple_lock)
5232 9 : return true;
5233 :
5234 368 : switch (wait_policy)
5235 : {
5236 323 : case LockWaitBlock:
5237 323 : LockTupleTuplock(relation, tid, mode);
5238 323 : break;
5239 :
5240 34 : case LockWaitSkip:
5241 34 : if (!ConditionalLockTupleTuplock(relation, tid, mode, false))
5242 1 : return false;
5243 33 : break;
5244 :
5245 11 : case LockWaitError:
5246 11 : if (!ConditionalLockTupleTuplock(relation, tid, mode, log_lock_failures))
5247 1 : ereport(ERROR,
5248 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5249 : errmsg("could not obtain lock on row in relation \"%s\"",
5250 : RelationGetRelationName(relation))));
5251 10 : break;
5252 : }
5253 366 : *have_tuple_lock = true;
5254 :
5255 366 : return true;
5256 : }
5257 :
5258 : /*
5259 : * Given an original set of Xmax and infomask, and a transaction (identified by
5260 : * add_to_xmax) acquiring a new lock of some mode, compute the new Xmax and
5261 : * corresponding infomasks to use on the tuple.
5262 : *
5263 : * Note that this might have side effects such as creating a new MultiXactId.
5264 : *
5265 : * Most callers will have called HeapTupleSatisfiesUpdate before this function;
5266 : * that will have set the HEAP_XMAX_INVALID bit if the xmax was a MultiXactId
5267 : * but it was not running anymore. There is a race condition, which is that the
5268 : * MultiXactId may have finished since then, but that uncommon case is handled
5269 : * either here, or within MultiXactIdExpand.
5270 : *
5271 : * There is a similar race condition possible when the old xmax was a regular
5272 : * TransactionId. We test TransactionIdIsInProgress again just to narrow the
5273 : * window, but it's still possible to end up creating an unnecessary
5274 : * MultiXactId. Fortunately this is harmless.
5275 : */
5276 : static void
5277 6622111 : compute_new_xmax_infomask(TransactionId xmax, uint16 old_infomask,
5278 : uint16 old_infomask2, TransactionId add_to_xmax,
5279 : LockTupleMode mode, bool is_update,
5280 : TransactionId *result_xmax, uint16 *result_infomask,
5281 : uint16 *result_infomask2)
5282 : {
5283 : TransactionId new_xmax;
5284 : uint16 new_infomask,
5285 : new_infomask2;
5286 :
5287 : Assert(TransactionIdIsCurrentTransactionId(add_to_xmax));
5288 :
5289 104487 : l5:
5290 6726598 : new_infomask = 0;
5291 6726598 : new_infomask2 = 0;
5292 6726598 : if (old_infomask & HEAP_XMAX_INVALID)
5293 : {
5294 : /*
5295 : * No previous locker; we just insert our own TransactionId.
5296 : *
5297 : * Note that it's critical that this case be the first one checked,
5298 : * because there are several blocks below that come back to this one
5299 : * to implement certain optimizations; old_infomask might contain
5300 : * other dirty bits in those cases, but we don't really care.
5301 : */
5302 6545418 : if (is_update)
5303 : {
5304 4262397 : new_xmax = add_to_xmax;
5305 4262397 : if (mode == LockTupleExclusive)
5306 1918262 : new_infomask2 |= HEAP_KEYS_UPDATED;
5307 : }
5308 : else
5309 : {
5310 2283021 : new_infomask |= HEAP_XMAX_LOCK_ONLY;
5311 2283021 : switch (mode)
5312 : {
5313 4987 : case LockTupleKeyShare:
5314 4987 : new_xmax = add_to_xmax;
5315 4987 : new_infomask |= HEAP_XMAX_KEYSHR_LOCK;
5316 4987 : break;
5317 814 : case LockTupleShare:
5318 814 : new_xmax = add_to_xmax;
5319 814 : new_infomask |= HEAP_XMAX_SHR_LOCK;
5320 814 : break;
5321 2181220 : case LockTupleNoKeyExclusive:
5322 2181220 : new_xmax = add_to_xmax;
5323 2181220 : new_infomask |= HEAP_XMAX_EXCL_LOCK;
5324 2181220 : break;
5325 96000 : case LockTupleExclusive:
5326 96000 : new_xmax = add_to_xmax;
5327 96000 : new_infomask |= HEAP_XMAX_EXCL_LOCK;
5328 96000 : new_infomask2 |= HEAP_KEYS_UPDATED;
5329 96000 : break;
5330 0 : default:
5331 0 : new_xmax = InvalidTransactionId; /* silence compiler */
5332 0 : elog(ERROR, "invalid lock mode");
5333 : }
5334 : }
5335 : }
5336 181180 : else if (old_infomask & HEAP_XMAX_IS_MULTI)
5337 : {
5338 : MultiXactStatus new_status;
5339 :
5340 : /*
5341 : * Currently we don't allow XMAX_COMMITTED to be set for multis, so
5342 : * cross-check.
5343 : */
5344 : Assert(!(old_infomask & HEAP_XMAX_COMMITTED));
5345 :
5346 : /*
5347 : * A multixact together with LOCK_ONLY set but neither lock bit set
5348 : * (i.e. a pg_upgraded share locked tuple) cannot possibly be running
5349 : * anymore. This check is critical for databases upgraded by
5350 : * pg_upgrade; both MultiXactIdIsRunning and MultiXactIdExpand assume
5351 : * that such multis are never passed.
5352 : */
5353 75561 : if (HEAP_LOCKED_UPGRADED(old_infomask))
5354 : {
5355 0 : old_infomask &= ~HEAP_XMAX_IS_MULTI;
5356 0 : old_infomask |= HEAP_XMAX_INVALID;
5357 0 : goto l5;
5358 : }
5359 :
5360 : /*
5361 : * If the XMAX is already a MultiXactId, then we need to expand it to
5362 : * include add_to_xmax; but if all the members were lockers and are
5363 : * all gone, we can do away with the IS_MULTI bit and just set
5364 : * add_to_xmax as the only locker/updater. If all lockers are gone
5365 : * and we have an updater that aborted, we can also do without a
5366 : * multi.
5367 : *
5368 : * The cost of doing GetMultiXactIdMembers would be paid by
5369 : * MultiXactIdExpand if we weren't to do this, so this check is not
5370 : * incurring extra work anyhow.
5371 : */
5372 75561 : if (!MultiXactIdIsRunning(xmax, HEAP_XMAX_IS_LOCKED_ONLY(old_infomask)))
5373 : {
5374 26 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) ||
5375 10 : !TransactionIdDidCommit(MultiXactIdGetUpdateXid(xmax,
5376 : old_infomask)))
5377 : {
5378 : /*
5379 : * Reset these bits and restart; otherwise fall through to
5380 : * create a new multi below.
5381 : */
5382 26 : old_infomask &= ~HEAP_XMAX_IS_MULTI;
5383 26 : old_infomask |= HEAP_XMAX_INVALID;
5384 26 : goto l5;
5385 : }
5386 : }
5387 :
5388 75535 : new_status = get_mxact_status_for_lock(mode, is_update);
5389 :
5390 75535 : new_xmax = MultiXactIdExpand((MultiXactId) xmax, add_to_xmax,
5391 : new_status);
5392 75535 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5393 : }
5394 105619 : else if (old_infomask & HEAP_XMAX_COMMITTED)
5395 : {
5396 : /*
5397 : * It's a committed update, so we need to preserve him as updater of
5398 : * the tuple.
5399 : */
5400 : MultiXactStatus status;
5401 : MultiXactStatus new_status;
5402 :
5403 13 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5404 0 : status = MultiXactStatusUpdate;
5405 : else
5406 13 : status = MultiXactStatusNoKeyUpdate;
5407 :
5408 13 : new_status = get_mxact_status_for_lock(mode, is_update);
5409 :
5410 : /*
5411 : * since it's not running, it's obviously impossible for the old
5412 : * updater to be identical to the current one, so we need not check
5413 : * for that case as we do in the block above.
5414 : */
5415 13 : new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5416 13 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5417 : }
5418 105606 : else if (TransactionIdIsInProgress(xmax))
5419 : {
5420 : /*
5421 : * If the XMAX is a valid, in-progress TransactionId, then we need to
5422 : * create a new MultiXactId that includes both the old locker or
5423 : * updater and our own TransactionId.
5424 : */
5425 : MultiXactStatus new_status;
5426 : MultiXactStatus old_status;
5427 : LockTupleMode old_mode;
5428 :
5429 105591 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
5430 : {
5431 105563 : if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
5432 5715 : old_status = MultiXactStatusForKeyShare;
5433 99848 : else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
5434 471 : old_status = MultiXactStatusForShare;
5435 99377 : else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
5436 : {
5437 99377 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5438 92905 : old_status = MultiXactStatusForUpdate;
5439 : else
5440 6472 : old_status = MultiXactStatusForNoKeyUpdate;
5441 : }
5442 : else
5443 : {
5444 : /*
5445 : * LOCK_ONLY can be present alone only when a page has been
5446 : * upgraded by pg_upgrade. But in that case,
5447 : * TransactionIdIsInProgress() should have returned false. We
5448 : * assume it's no longer locked in this case.
5449 : */
5450 0 : elog(WARNING, "LOCK_ONLY found for Xid in progress %u", xmax);
5451 0 : old_infomask |= HEAP_XMAX_INVALID;
5452 0 : old_infomask &= ~HEAP_XMAX_LOCK_ONLY;
5453 0 : goto l5;
5454 : }
5455 : }
5456 : else
5457 : {
5458 : /* it's an update, but which kind? */
5459 28 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5460 0 : old_status = MultiXactStatusUpdate;
5461 : else
5462 28 : old_status = MultiXactStatusNoKeyUpdate;
5463 : }
5464 :
5465 105591 : old_mode = TUPLOCK_from_mxstatus(old_status);
5466 :
5467 : /*
5468 : * If the lock to be acquired is for the same TransactionId as the
5469 : * existing lock, there's an optimization possible: consider only the
5470 : * strongest of both locks as the only one present, and restart.
5471 : */
5472 105591 : if (xmax == add_to_xmax)
5473 : {
5474 : /*
5475 : * Note that it's not possible for the original tuple to be
5476 : * updated: we wouldn't be here because the tuple would have been
5477 : * invisible and we wouldn't try to update it. As a subtlety,
5478 : * this code can also run when traversing an update chain to lock
5479 : * future versions of a tuple. But we wouldn't be here either,
5480 : * because the add_to_xmax would be different from the original
5481 : * updater.
5482 : */
5483 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
5484 :
5485 : /* acquire the strongest of both */
5486 104447 : if (mode < old_mode)
5487 52213 : mode = old_mode;
5488 : /* mustn't touch is_update */
5489 :
5490 104447 : old_infomask |= HEAP_XMAX_INVALID;
5491 104447 : goto l5;
5492 : }
5493 :
5494 : /* otherwise, just fall back to creating a new multixact */
5495 1144 : new_status = get_mxact_status_for_lock(mode, is_update);
5496 1144 : new_xmax = MultiXactIdCreate(xmax, old_status,
5497 : add_to_xmax, new_status);
5498 1144 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5499 : }
5500 20 : else if (!HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) &&
5501 5 : TransactionIdDidCommit(xmax))
5502 1 : {
5503 : /*
5504 : * It's a committed update, so we gotta preserve him as updater of the
5505 : * tuple.
5506 : */
5507 : MultiXactStatus status;
5508 : MultiXactStatus new_status;
5509 :
5510 1 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5511 0 : status = MultiXactStatusUpdate;
5512 : else
5513 1 : status = MultiXactStatusNoKeyUpdate;
5514 :
5515 1 : new_status = get_mxact_status_for_lock(mode, is_update);
5516 :
5517 : /*
5518 : * since it's not running, it's obviously impossible for the old
5519 : * updater to be identical to the current one, so we need not check
5520 : * for that case as we do in the block above.
5521 : */
5522 1 : new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5523 1 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5524 : }
5525 : else
5526 : {
5527 : /*
5528 : * Can get here iff the locking/updating transaction was running when
5529 : * the infomask was extracted from the tuple, but finished before
5530 : * TransactionIdIsInProgress got to run. Deal with it as if there was
5531 : * no locker at all in the first place.
5532 : */
5533 14 : old_infomask |= HEAP_XMAX_INVALID;
5534 14 : goto l5;
5535 : }
5536 :
5537 6622111 : *result_infomask = new_infomask;
5538 6622111 : *result_infomask2 = new_infomask2;
5539 6622111 : *result_xmax = new_xmax;
5540 6622111 : }
5541 :
5542 : /*
5543 : * Subroutine for heap_lock_updated_tuple_rec.
5544 : *
5545 : * Given a hypothetical multixact status held by the transaction identified
5546 : * with the given xid, does the current transaction need to wait, fail, or can
5547 : * it continue if it wanted to acquire a lock of the given mode? "needwait"
5548 : * is set to true if waiting is necessary; if it can continue, then TM_Ok is
5549 : * returned. If the lock is already held by the current transaction, return
5550 : * TM_SelfModified. In case of a conflict with another transaction, a
5551 : * different HeapTupleSatisfiesUpdate return code is returned.
5552 : *
5553 : * The held status is said to be hypothetical because it might correspond to a
5554 : * lock held by a single Xid, i.e. not a real MultiXactId; we express it this
5555 : * way for simplicity of API.
5556 : */
5557 : static TM_Result
5558 38781 : test_lockmode_for_conflict(MultiXactStatus status, TransactionId xid,
5559 : LockTupleMode mode, HeapTuple tup,
5560 : bool *needwait)
5561 : {
5562 : MultiXactStatus wantedstatus;
5563 :
5564 38781 : *needwait = false;
5565 38781 : wantedstatus = get_mxact_status_for_lock(mode, false);
5566 :
5567 : /*
5568 : * Note: we *must* check TransactionIdIsInProgress before
5569 : * TransactionIdDidAbort/Commit; see comment at top of heapam_visibility.c
5570 : * for an explanation.
5571 : */
5572 38781 : if (TransactionIdIsCurrentTransactionId(xid))
5573 : {
5574 : /*
5575 : * The tuple has already been locked by our own transaction. This is
5576 : * very rare but can happen if multiple transactions are trying to
5577 : * lock an ancient version of the same tuple.
5578 : */
5579 0 : return TM_SelfModified;
5580 : }
5581 38781 : else if (TransactionIdIsInProgress(xid))
5582 : {
5583 : /*
5584 : * If the locking transaction is running, what we do depends on
5585 : * whether the lock modes conflict: if they do, then we must wait for
5586 : * it to finish; otherwise we can fall through to lock this tuple
5587 : * version without waiting.
5588 : */
5589 36539 : if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
5590 36539 : LOCKMODE_from_mxstatus(wantedstatus)))
5591 : {
5592 8 : *needwait = true;
5593 : }
5594 :
5595 : /*
5596 : * If we set needwait above, then this value doesn't matter;
5597 : * otherwise, this value signals to caller that it's okay to proceed.
5598 : */
5599 36539 : return TM_Ok;
5600 : }
5601 2242 : else if (TransactionIdDidAbort(xid))
5602 206 : return TM_Ok;
5603 2036 : else if (TransactionIdDidCommit(xid))
5604 : {
5605 : /*
5606 : * The other transaction committed. If it was only a locker, then the
5607 : * lock is completely gone now and we can return success; but if it
5608 : * was an update, then what we do depends on whether the two lock
5609 : * modes conflict. If they conflict, then we must report error to
5610 : * caller. But if they don't, we can fall through to allow the current
5611 : * transaction to lock the tuple.
5612 : *
5613 : * Note: the reason we worry about ISUPDATE here is because as soon as
5614 : * a transaction ends, all its locks are gone and meaningless, and
5615 : * thus we can ignore them; whereas its updates persist. In the
5616 : * TransactionIdIsInProgress case, above, we don't need to check
5617 : * because we know the lock is still "alive" and thus a conflict needs
5618 : * always be checked.
5619 : */
5620 2036 : if (!ISUPDATE_from_mxstatus(status))
5621 2026 : return TM_Ok;
5622 :
5623 10 : if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
5624 10 : LOCKMODE_from_mxstatus(wantedstatus)))
5625 : {
5626 : /* bummer */
5627 9 : if (!ItemPointerEquals(&tup->t_self, &tup->t_data->t_ctid))
5628 7 : return TM_Updated;
5629 : else
5630 2 : return TM_Deleted;
5631 : }
5632 :
5633 1 : return TM_Ok;
5634 : }
5635 :
5636 : /* Not in progress, not aborted, not committed -- must have crashed */
5637 0 : return TM_Ok;
5638 : }
5639 :
5640 :
5641 : /*
5642 : * Recursive part of heap_lock_updated_tuple
5643 : *
5644 : * Fetch the tuple pointed to by tid in rel, and mark it as locked by the given
5645 : * xid with the given mode; if this tuple is updated, recurse to lock the new
5646 : * version as well.
5647 : */
5648 : static TM_Result
5649 2224 : heap_lock_updated_tuple_rec(Relation rel, TransactionId priorXmax,
5650 : const ItemPointerData *tid, TransactionId xid,
5651 : LockTupleMode mode)
5652 : {
5653 : TM_Result result;
5654 : ItemPointerData tupid;
5655 : HeapTupleData mytup;
5656 : Buffer buf;
5657 : uint16 new_infomask,
5658 : new_infomask2,
5659 : old_infomask,
5660 : old_infomask2;
5661 : TransactionId xmax,
5662 : new_xmax;
5663 2224 : bool cleared_all_frozen = false;
5664 : bool pinned_desired_page;
5665 2224 : Buffer vmbuffer = InvalidBuffer;
5666 : BlockNumber block;
5667 :
5668 2224 : ItemPointerCopy(tid, &tupid);
5669 :
5670 : for (;;)
5671 : {
5672 2227 : new_infomask = 0;
5673 2227 : new_xmax = InvalidTransactionId;
5674 2227 : block = ItemPointerGetBlockNumber(&tupid);
5675 2227 : ItemPointerCopy(&tupid, &(mytup.t_self));
5676 :
5677 2227 : if (!heap_fetch(rel, SnapshotAny, &mytup, &buf, false))
5678 : {
5679 : /*
5680 : * if we fail to find the updated version of the tuple, it's
5681 : * because it was vacuumed/pruned away after its creator
5682 : * transaction aborted. So behave as if we got to the end of the
5683 : * chain, and there's no further tuple to lock: return success to
5684 : * caller.
5685 : */
5686 0 : result = TM_Ok;
5687 0 : goto out_unlocked;
5688 : }
5689 :
5690 2227 : l4:
5691 2235 : CHECK_FOR_INTERRUPTS();
5692 :
5693 : /*
5694 : * Before locking the buffer, pin the visibility map page if it
5695 : * appears to be necessary. Since we haven't got the lock yet,
5696 : * someone else might be in the middle of changing this, so we'll need
5697 : * to recheck after we have the lock.
5698 : */
5699 2235 : if (PageIsAllVisible(BufferGetPage(buf)))
5700 : {
5701 0 : visibilitymap_pin(rel, block, &vmbuffer);
5702 0 : pinned_desired_page = true;
5703 : }
5704 : else
5705 2235 : pinned_desired_page = false;
5706 :
5707 2235 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
5708 :
5709 : /*
5710 : * If we didn't pin the visibility map page and the page has become
5711 : * all visible while we were busy locking the buffer, we'll have to
5712 : * unlock and re-lock, to avoid holding the buffer lock across I/O.
5713 : * That's a bit unfortunate, but hopefully shouldn't happen often.
5714 : *
5715 : * Note: in some paths through this function, we will reach here
5716 : * holding a pin on a vm page that may or may not be the one matching
5717 : * this page. If this page isn't all-visible, we won't use the vm
5718 : * page, but we hold onto such a pin till the end of the function.
5719 : */
5720 2235 : if (!pinned_desired_page && PageIsAllVisible(BufferGetPage(buf)))
5721 : {
5722 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5723 0 : visibilitymap_pin(rel, block, &vmbuffer);
5724 0 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
5725 : }
5726 :
5727 : /*
5728 : * Check the tuple XMIN against prior XMAX, if any. If we reached the
5729 : * end of the chain, we're done, so return success.
5730 : */
5731 4470 : if (TransactionIdIsValid(priorXmax) &&
5732 2235 : !TransactionIdEquals(HeapTupleHeaderGetXmin(mytup.t_data),
5733 : priorXmax))
5734 : {
5735 2 : result = TM_Ok;
5736 2 : goto out_locked;
5737 : }
5738 :
5739 : /*
5740 : * Also check Xmin: if this tuple was created by an aborted
5741 : * (sub)transaction, then we already locked the last live one in the
5742 : * chain, thus we're done, so return success.
5743 : */
5744 2233 : if (TransactionIdDidAbort(HeapTupleHeaderGetXmin(mytup.t_data)))
5745 : {
5746 25 : result = TM_Ok;
5747 25 : goto out_locked;
5748 : }
5749 :
5750 2208 : old_infomask = mytup.t_data->t_infomask;
5751 2208 : old_infomask2 = mytup.t_data->t_infomask2;
5752 2208 : xmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
5753 :
5754 : /*
5755 : * If this tuple version has been updated or locked by some concurrent
5756 : * transaction(s), what we do depends on whether our lock mode
5757 : * conflicts with what those other transactions hold, and also on the
5758 : * status of them.
5759 : */
5760 2208 : if (!(old_infomask & HEAP_XMAX_INVALID))
5761 : {
5762 : TransactionId rawxmax;
5763 : bool needwait;
5764 :
5765 2145 : rawxmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
5766 2145 : if (old_infomask & HEAP_XMAX_IS_MULTI)
5767 : {
5768 : int nmembers;
5769 : int i;
5770 : MultiXactMember *members;
5771 :
5772 : /*
5773 : * We don't need a test for pg_upgrade'd tuples: this is only
5774 : * applied to tuples after the first in an update chain. Said
5775 : * first tuple in the chain may well be locked-in-9.2-and-
5776 : * pg_upgraded, but that one was already locked by our caller,
5777 : * not us; and any subsequent ones cannot be because our
5778 : * caller must necessarily have obtained a snapshot later than
5779 : * the pg_upgrade itself.
5780 : */
5781 : Assert(!HEAP_LOCKED_UPGRADED(mytup.t_data->t_infomask));
5782 :
5783 2109 : nmembers = GetMultiXactIdMembers(rawxmax, &members, false,
5784 2109 : HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
5785 40854 : for (i = 0; i < nmembers; i++)
5786 : {
5787 38745 : result = test_lockmode_for_conflict(members[i].status,
5788 38745 : members[i].xid,
5789 : mode,
5790 : &mytup,
5791 : &needwait);
5792 :
5793 : /*
5794 : * If the tuple was already locked by ourselves in a
5795 : * previous iteration of this (say heap_lock_tuple was
5796 : * forced to restart the locking loop because of a change
5797 : * in xmax), then we hold the lock already on this tuple
5798 : * version and we don't need to do anything; and this is
5799 : * not an error condition either. We just need to skip
5800 : * this tuple and continue locking the next version in the
5801 : * update chain.
5802 : */
5803 38745 : if (result == TM_SelfModified)
5804 : {
5805 0 : pfree(members);
5806 0 : goto next;
5807 : }
5808 :
5809 38745 : if (needwait)
5810 : {
5811 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5812 0 : XactLockTableWait(members[i].xid, rel,
5813 : &mytup.t_self,
5814 : XLTW_LockUpdated);
5815 0 : pfree(members);
5816 0 : goto l4;
5817 : }
5818 38745 : if (result != TM_Ok)
5819 : {
5820 0 : pfree(members);
5821 0 : goto out_locked;
5822 : }
5823 : }
5824 2109 : if (members)
5825 2109 : pfree(members);
5826 : }
5827 : else
5828 : {
5829 : MultiXactStatus status;
5830 :
5831 : /*
5832 : * For a non-multi Xmax, we first need to compute the
5833 : * corresponding MultiXactStatus by using the infomask bits.
5834 : */
5835 36 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
5836 : {
5837 16 : if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
5838 16 : status = MultiXactStatusForKeyShare;
5839 0 : else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
5840 0 : status = MultiXactStatusForShare;
5841 0 : else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
5842 : {
5843 0 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5844 0 : status = MultiXactStatusForUpdate;
5845 : else
5846 0 : status = MultiXactStatusForNoKeyUpdate;
5847 : }
5848 : else
5849 : {
5850 : /*
5851 : * LOCK_ONLY present alone (a pg_upgraded tuple marked
5852 : * as share-locked in the old cluster) shouldn't be
5853 : * seen in the middle of an update chain.
5854 : */
5855 0 : elog(ERROR, "invalid lock status in tuple");
5856 : }
5857 : }
5858 : else
5859 : {
5860 : /* it's an update, but which kind? */
5861 20 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5862 15 : status = MultiXactStatusUpdate;
5863 : else
5864 5 : status = MultiXactStatusNoKeyUpdate;
5865 : }
5866 :
5867 36 : result = test_lockmode_for_conflict(status, rawxmax, mode,
5868 : &mytup, &needwait);
5869 :
5870 : /*
5871 : * If the tuple was already locked by ourselves in a previous
5872 : * iteration of this (say heap_lock_tuple was forced to
5873 : * restart the locking loop because of a change in xmax), then
5874 : * we hold the lock already on this tuple version and we don't
5875 : * need to do anything; and this is not an error condition
5876 : * either. We just need to skip this tuple and continue
5877 : * locking the next version in the update chain.
5878 : */
5879 36 : if (result == TM_SelfModified)
5880 0 : goto next;
5881 :
5882 36 : if (needwait)
5883 : {
5884 8 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5885 8 : XactLockTableWait(rawxmax, rel, &mytup.t_self,
5886 : XLTW_LockUpdated);
5887 8 : goto l4;
5888 : }
5889 28 : if (result != TM_Ok)
5890 : {
5891 9 : goto out_locked;
5892 : }
5893 : }
5894 : }
5895 :
5896 : /* compute the new Xmax and infomask values for the tuple ... */
5897 2191 : compute_new_xmax_infomask(xmax, old_infomask, mytup.t_data->t_infomask2,
5898 : xid, mode, false,
5899 : &new_xmax, &new_infomask, &new_infomask2);
5900 :
5901 2191 : if (PageIsAllVisible(BufferGetPage(buf)) &&
5902 0 : visibilitymap_clear(rel, block, vmbuffer,
5903 : VISIBILITYMAP_ALL_FROZEN))
5904 0 : cleared_all_frozen = true;
5905 :
5906 2191 : START_CRIT_SECTION();
5907 :
5908 : /* ... and set them */
5909 2191 : HeapTupleHeaderSetXmax(mytup.t_data, new_xmax);
5910 2191 : mytup.t_data->t_infomask &= ~HEAP_XMAX_BITS;
5911 2191 : mytup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
5912 2191 : mytup.t_data->t_infomask |= new_infomask;
5913 2191 : mytup.t_data->t_infomask2 |= new_infomask2;
5914 :
5915 2191 : MarkBufferDirty(buf);
5916 :
5917 : /* XLOG stuff */
5918 2191 : if (RelationNeedsWAL(rel))
5919 : {
5920 : xl_heap_lock_updated xlrec;
5921 : XLogRecPtr recptr;
5922 2191 : Page page = BufferGetPage(buf);
5923 :
5924 2191 : XLogBeginInsert();
5925 2191 : XLogRegisterBuffer(0, buf, REGBUF_STANDARD);
5926 :
5927 2191 : xlrec.offnum = ItemPointerGetOffsetNumber(&mytup.t_self);
5928 2191 : xlrec.xmax = new_xmax;
5929 2191 : xlrec.infobits_set = compute_infobits(new_infomask, new_infomask2);
5930 2191 : xlrec.flags =
5931 2191 : cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
5932 :
5933 2191 : XLogRegisterData(&xlrec, SizeOfHeapLockUpdated);
5934 :
5935 2191 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_LOCK_UPDATED);
5936 :
5937 2191 : PageSetLSN(page, recptr);
5938 : }
5939 :
5940 2191 : END_CRIT_SECTION();
5941 :
5942 2191 : next:
5943 : /* if we find the end of update chain, we're done. */
5944 4382 : if (mytup.t_data->t_infomask & HEAP_XMAX_INVALID ||
5945 4382 : HeapTupleHeaderIndicatesMovedPartitions(mytup.t_data) ||
5946 2195 : ItemPointerEquals(&mytup.t_self, &mytup.t_data->t_ctid) ||
5947 4 : HeapTupleHeaderIsOnlyLocked(mytup.t_data))
5948 : {
5949 2188 : result = TM_Ok;
5950 2188 : goto out_locked;
5951 : }
5952 :
5953 : /* tail recursion */
5954 3 : priorXmax = HeapTupleHeaderGetUpdateXid(mytup.t_data);
5955 3 : ItemPointerCopy(&(mytup.t_data->t_ctid), &tupid);
5956 3 : UnlockReleaseBuffer(buf);
5957 : }
5958 :
5959 : result = TM_Ok;
5960 :
5961 2224 : out_locked:
5962 2224 : UnlockReleaseBuffer(buf);
5963 :
5964 2224 : out_unlocked:
5965 2224 : if (vmbuffer != InvalidBuffer)
5966 0 : ReleaseBuffer(vmbuffer);
5967 :
5968 2224 : return result;
5969 : }
5970 :
5971 : /*
5972 : * heap_lock_updated_tuple
5973 : * Follow update chain when locking an updated tuple, acquiring locks (row
5974 : * marks) on the updated versions.
5975 : *
5976 : * 'prior_infomask', 'prior_raw_xmax' and 'prior_ctid' are the corresponding
5977 : * fields from the initial tuple. We will lock the tuples starting from the
5978 : * one that 'prior_ctid' points to. Note: This function does not lock the
5979 : * initial tuple itself.
5980 : *
5981 : * This function doesn't check visibility, it just unconditionally marks the
5982 : * tuple(s) as locked. If any tuple in the updated chain is being deleted
5983 : * concurrently (or updated with the key being modified), sleep until the
5984 : * transaction doing it is finished.
5985 : *
5986 : * Note that we don't acquire heavyweight tuple locks on the tuples we walk
5987 : * when we have to wait for other transactions to release them, as opposed to
5988 : * what heap_lock_tuple does. The reason is that having more than one
5989 : * transaction walking the chain is probably uncommon enough that risk of
5990 : * starvation is not likely: one of the preconditions for being here is that
5991 : * the snapshot in use predates the update that created this tuple (because we
5992 : * started at an earlier version of the tuple), but at the same time such a
5993 : * transaction cannot be using repeatable read or serializable isolation
5994 : * levels, because that would lead to a serializability failure.
5995 : */
5996 : static TM_Result
5997 2226 : heap_lock_updated_tuple(Relation rel,
5998 : uint16 prior_infomask,
5999 : TransactionId prior_raw_xmax,
6000 : const ItemPointerData *prior_ctid,
6001 : TransactionId xid, LockTupleMode mode)
6002 : {
6003 2226 : INJECTION_POINT("heap_lock_updated_tuple", NULL);
6004 :
6005 : /*
6006 : * If the tuple has moved into another partition (effectively a delete)
6007 : * stop here.
6008 : */
6009 2226 : if (!ItemPointerIndicatesMovedPartitions(prior_ctid))
6010 : {
6011 : TransactionId prior_xmax;
6012 :
6013 : /*
6014 : * If this is the first possibly-multixact-able operation in the
6015 : * current transaction, set my per-backend OldestMemberMXactId
6016 : * setting. We can be certain that the transaction will never become a
6017 : * member of any older MultiXactIds than that. (We have to do this
6018 : * even if we end up just using our own TransactionId below, since
6019 : * some other backend could incorporate our XID into a MultiXact
6020 : * immediately afterwards.)
6021 : */
6022 2224 : MultiXactIdSetOldestMember();
6023 :
6024 4448 : prior_xmax = (prior_infomask & HEAP_XMAX_IS_MULTI) ?
6025 2224 : MultiXactIdGetUpdateXid(prior_raw_xmax, prior_infomask) : prior_raw_xmax;
6026 2224 : return heap_lock_updated_tuple_rec(rel, prior_xmax, prior_ctid, xid, mode);
6027 : }
6028 :
6029 : /* nothing to lock */
6030 2 : return TM_Ok;
6031 : }
6032 :
6033 : /*
6034 : * heap_finish_speculative - mark speculative insertion as successful
6035 : *
6036 : * To successfully finish a speculative insertion we have to clear speculative
6037 : * token from tuple. To do so the t_ctid field, which will contain a
6038 : * speculative token value, is modified in place to point to the tuple itself,
6039 : * which is characteristic of a newly inserted ordinary tuple.
6040 : *
6041 : * NB: It is not ok to commit without either finishing or aborting a
6042 : * speculative insertion. We could treat speculative tuples of committed
6043 : * transactions implicitly as completed, but then we would have to be prepared
6044 : * to deal with speculative tokens on committed tuples. That wouldn't be
6045 : * difficult - no-one looks at the ctid field of a tuple with invalid xmax -
6046 : * but clearing the token at completion isn't very expensive either.
6047 : * An explicit confirmation WAL record also makes logical decoding simpler.
6048 : */
6049 : void
6050 2215 : heap_finish_speculative(Relation relation, const ItemPointerData *tid)
6051 : {
6052 : Buffer buffer;
6053 : Page page;
6054 : OffsetNumber offnum;
6055 : ItemId lp;
6056 : HeapTupleHeader htup;
6057 :
6058 2215 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
6059 2215 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6060 2215 : page = BufferGetPage(buffer);
6061 :
6062 2215 : offnum = ItemPointerGetOffsetNumber(tid);
6063 2215 : if (offnum < 1 || offnum > PageGetMaxOffsetNumber(page))
6064 0 : elog(ERROR, "offnum out of range");
6065 2215 : lp = PageGetItemId(page, offnum);
6066 2215 : if (!ItemIdIsNormal(lp))
6067 0 : elog(ERROR, "invalid lp");
6068 :
6069 2215 : htup = (HeapTupleHeader) PageGetItem(page, lp);
6070 :
6071 : /* NO EREPORT(ERROR) from here till changes are logged */
6072 2215 : START_CRIT_SECTION();
6073 :
6074 : Assert(HeapTupleHeaderIsSpeculative(htup));
6075 :
6076 2215 : MarkBufferDirty(buffer);
6077 :
6078 : /*
6079 : * Replace the speculative insertion token with a real t_ctid, pointing to
6080 : * itself like it does on regular tuples.
6081 : */
6082 2215 : htup->t_ctid = *tid;
6083 :
6084 : /* XLOG stuff */
6085 2215 : if (RelationNeedsWAL(relation))
6086 : {
6087 : xl_heap_confirm xlrec;
6088 : XLogRecPtr recptr;
6089 :
6090 2195 : xlrec.offnum = ItemPointerGetOffsetNumber(tid);
6091 :
6092 2195 : XLogBeginInsert();
6093 :
6094 : /* We want the same filtering on this as on a plain insert */
6095 2195 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
6096 :
6097 2195 : XLogRegisterData(&xlrec, SizeOfHeapConfirm);
6098 2195 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
6099 :
6100 2195 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_CONFIRM);
6101 :
6102 2195 : PageSetLSN(page, recptr);
6103 : }
6104 :
6105 2215 : END_CRIT_SECTION();
6106 :
6107 2215 : UnlockReleaseBuffer(buffer);
6108 2215 : }
6109 :
6110 : /*
6111 : * heap_abort_speculative - kill a speculatively inserted tuple
6112 : *
6113 : * Marks a tuple that was speculatively inserted in the same command as dead,
6114 : * by setting its xmin as invalid. That makes it immediately appear as dead
6115 : * to all transactions, including our own. In particular, it makes
6116 : * HeapTupleSatisfiesDirty() regard the tuple as dead, so that another backend
6117 : * inserting a duplicate key value won't unnecessarily wait for our whole
6118 : * transaction to finish (it'll just wait for our speculative insertion to
6119 : * finish).
6120 : *
6121 : * Killing the tuple prevents "unprincipled deadlocks", which are deadlocks
6122 : * that arise due to a mutual dependency that is not user visible. By
6123 : * definition, unprincipled deadlocks cannot be prevented by the user
6124 : * reordering lock acquisition in client code, because the implementation level
6125 : * lock acquisitions are not under the user's direct control. If speculative
6126 : * inserters did not take this precaution, then under high concurrency they
6127 : * could deadlock with each other, which would not be acceptable.
6128 : *
6129 : * This is somewhat redundant with heap_delete, but we prefer to have a
6130 : * dedicated routine with stripped down requirements. Note that this is also
6131 : * used to delete the TOAST tuples created during speculative insertion.
6132 : *
6133 : * This routine does not affect logical decoding as it only looks at
6134 : * confirmation records.
6135 : */
6136 : void
6137 16 : heap_abort_speculative(Relation relation, const ItemPointerData *tid)
6138 : {
6139 16 : TransactionId xid = GetCurrentTransactionId();
6140 : ItemId lp;
6141 : HeapTupleData tp;
6142 : Page page;
6143 : BlockNumber block;
6144 : Buffer buffer;
6145 :
6146 : Assert(ItemPointerIsValid(tid));
6147 :
6148 16 : block = ItemPointerGetBlockNumber(tid);
6149 16 : buffer = ReadBuffer(relation, block);
6150 16 : page = BufferGetPage(buffer);
6151 :
6152 16 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6153 :
6154 : /*
6155 : * Page can't be all visible, we just inserted into it, and are still
6156 : * running.
6157 : */
6158 : Assert(!PageIsAllVisible(page));
6159 :
6160 16 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
6161 : Assert(ItemIdIsNormal(lp));
6162 :
6163 16 : tp.t_tableOid = RelationGetRelid(relation);
6164 16 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
6165 16 : tp.t_len = ItemIdGetLength(lp);
6166 16 : tp.t_self = *tid;
6167 :
6168 : /*
6169 : * Sanity check that the tuple really is a speculatively inserted tuple,
6170 : * inserted by us.
6171 : */
6172 16 : if (tp.t_data->t_choice.t_heap.t_xmin != xid)
6173 0 : elog(ERROR, "attempted to kill a tuple inserted by another transaction");
6174 16 : if (!(IsToastRelation(relation) || HeapTupleHeaderIsSpeculative(tp.t_data)))
6175 0 : elog(ERROR, "attempted to kill a non-speculative tuple");
6176 : Assert(!HeapTupleHeaderIsHeapOnly(tp.t_data));
6177 :
6178 : /*
6179 : * No need to check for serializable conflicts here. There is never a
6180 : * need for a combo CID, either. No need to extract replica identity, or
6181 : * do anything special with infomask bits.
6182 : */
6183 :
6184 16 : START_CRIT_SECTION();
6185 :
6186 : /*
6187 : * The tuple will become DEAD immediately. Flag that this page is a
6188 : * candidate for pruning by setting xmin to TransactionXmin. While not
6189 : * immediately prunable, it is the oldest xid we can cheaply determine
6190 : * that's safe against wraparound / being older than the table's
6191 : * relfrozenxid. To defend against the unlikely case of a new relation
6192 : * having a newer relfrozenxid than our TransactionXmin, use relfrozenxid
6193 : * if so (vacuum can't subsequently move relfrozenxid to beyond
6194 : * TransactionXmin, so there's no race here).
6195 : */
6196 : Assert(TransactionIdIsValid(TransactionXmin));
6197 : {
6198 16 : TransactionId relfrozenxid = relation->rd_rel->relfrozenxid;
6199 : TransactionId prune_xid;
6200 :
6201 16 : if (TransactionIdPrecedes(TransactionXmin, relfrozenxid))
6202 0 : prune_xid = relfrozenxid;
6203 : else
6204 16 : prune_xid = TransactionXmin;
6205 16 : PageSetPrunable(page, prune_xid);
6206 : }
6207 :
6208 : /* store transaction information of xact deleting the tuple */
6209 16 : tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
6210 16 : tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
6211 :
6212 : /*
6213 : * Set the tuple header xmin to InvalidTransactionId. This makes the
6214 : * tuple immediately invisible everyone. (In particular, to any
6215 : * transactions waiting on the speculative token, woken up later.)
6216 : */
6217 16 : HeapTupleHeaderSetXmin(tp.t_data, InvalidTransactionId);
6218 :
6219 : /* Clear the speculative insertion token too */
6220 16 : tp.t_data->t_ctid = tp.t_self;
6221 :
6222 16 : MarkBufferDirty(buffer);
6223 :
6224 : /*
6225 : * XLOG stuff
6226 : *
6227 : * The WAL records generated here match heap_delete(). The same recovery
6228 : * routines are used.
6229 : */
6230 16 : if (RelationNeedsWAL(relation))
6231 : {
6232 : xl_heap_delete xlrec;
6233 : XLogRecPtr recptr;
6234 :
6235 12 : xlrec.flags = XLH_DELETE_IS_SUPER;
6236 24 : xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
6237 12 : tp.t_data->t_infomask2);
6238 12 : xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
6239 12 : xlrec.xmax = xid;
6240 :
6241 12 : XLogBeginInsert();
6242 12 : XLogRegisterData(&xlrec, SizeOfHeapDelete);
6243 12 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
6244 :
6245 : /* No replica identity & replication origin logged */
6246 :
6247 12 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
6248 :
6249 12 : PageSetLSN(page, recptr);
6250 : }
6251 :
6252 16 : END_CRIT_SECTION();
6253 :
6254 16 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6255 :
6256 16 : if (HeapTupleHasExternal(&tp))
6257 : {
6258 : Assert(!IsToastRelation(relation));
6259 1 : heap_toast_delete(relation, &tp, true);
6260 : }
6261 :
6262 : /*
6263 : * Never need to mark tuple for invalidation, since catalogs don't support
6264 : * speculative insertion
6265 : */
6266 :
6267 : /* Now we can release the buffer */
6268 16 : ReleaseBuffer(buffer);
6269 :
6270 : /* count deletion, as we counted the insertion too */
6271 16 : pgstat_count_heap_delete(relation);
6272 16 : }
6273 :
6274 : /*
6275 : * heap_inplace_lock - protect inplace update from concurrent heap_update()
6276 : *
6277 : * Evaluate whether the tuple's state is compatible with a no-key update.
6278 : * Current transaction rowmarks are fine, as is KEY SHARE from any
6279 : * transaction. If compatible, return true with the buffer exclusive-locked,
6280 : * and the caller must release that by calling
6281 : * heap_inplace_update_and_unlock(), calling heap_inplace_unlock(), or raising
6282 : * an error. Otherwise, call release_callback(arg), wait for blocking
6283 : * transactions to end, and return false.
6284 : *
6285 : * Since this is intended for system catalogs and SERIALIZABLE doesn't cover
6286 : * DDL, this doesn't guarantee any particular predicate locking.
6287 : *
6288 : * heap_delete() is a rarer source of blocking transactions (xwait). We'll
6289 : * wait for such a transaction just like for the normal heap_update() case.
6290 : * Normal concurrent DROP commands won't cause that, because all inplace
6291 : * updaters take some lock that conflicts with DROP. An explicit SQL "DELETE
6292 : * FROM pg_class" can cause it. By waiting, if the concurrent transaction
6293 : * executed both "DELETE FROM pg_class" and "INSERT INTO pg_class", our caller
6294 : * can find the successor tuple.
6295 : *
6296 : * Readers of inplace-updated fields expect changes to those fields are
6297 : * durable. For example, vac_truncate_clog() reads datfrozenxid from
6298 : * pg_database tuples via catalog snapshots. A future snapshot must not
6299 : * return a lower datfrozenxid for the same database OID (lower in the
6300 : * FullTransactionIdPrecedes() sense). We achieve that since no update of a
6301 : * tuple can start while we hold a lock on its buffer. In cases like
6302 : * BEGIN;GRANT;CREATE INDEX;COMMIT we're inplace-updating a tuple visible only
6303 : * to this transaction. ROLLBACK then is one case where it's okay to lose
6304 : * inplace updates. (Restoring relhasindex=false on ROLLBACK is fine, since
6305 : * any concurrent CREATE INDEX would have blocked, then inplace-updated the
6306 : * committed tuple.)
6307 : *
6308 : * In principle, we could avoid waiting by overwriting every tuple in the
6309 : * updated tuple chain. Reader expectations permit updating a tuple only if
6310 : * it's aborted, is the tail of the chain, or we already updated the tuple
6311 : * referenced in its t_ctid. Hence, we would need to overwrite the tuples in
6312 : * order from tail to head. That would imply either (a) mutating all tuples
6313 : * in one critical section or (b) accepting a chance of partial completion.
6314 : * Partial completion of a relfrozenxid update would have the weird
6315 : * consequence that the table's next VACUUM could see the table's relfrozenxid
6316 : * move forward between vacuum_get_cutoffs() and finishing.
6317 : */
6318 : bool
6319 217441 : heap_inplace_lock(Relation relation,
6320 : HeapTuple oldtup_ptr, Buffer buffer,
6321 : void (*release_callback) (void *), void *arg)
6322 : {
6323 217441 : HeapTupleData oldtup = *oldtup_ptr; /* minimize diff vs. heap_update() */
6324 : TM_Result result;
6325 : bool ret;
6326 :
6327 : #ifdef USE_ASSERT_CHECKING
6328 : if (RelationGetRelid(relation) == RelationRelationId)
6329 : check_inplace_rel_lock(oldtup_ptr);
6330 : #endif
6331 :
6332 : Assert(BufferIsValid(buffer));
6333 :
6334 : /*
6335 : * Register shared cache invals if necessary. Other sessions may finish
6336 : * inplace updates of this tuple between this step and LockTuple(). Since
6337 : * inplace updates don't change cache keys, that's harmless.
6338 : *
6339 : * While it's tempting to register invals only after confirming we can
6340 : * return true, the following obstacle precludes reordering steps that
6341 : * way. Registering invals might reach a CatalogCacheInitializeCache()
6342 : * that locks "buffer". That would hang indefinitely if running after our
6343 : * own LockBuffer(). Hence, we must register invals before LockBuffer().
6344 : */
6345 217441 : CacheInvalidateHeapTupleInplace(relation, oldtup_ptr);
6346 :
6347 217441 : LockTuple(relation, &oldtup.t_self, InplaceUpdateTupleLock);
6348 217441 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6349 :
6350 : /*----------
6351 : * Interpret HeapTupleSatisfiesUpdate() like heap_update() does, except:
6352 : *
6353 : * - wait unconditionally
6354 : * - already locked tuple above, since inplace needs that unconditionally
6355 : * - don't recheck header after wait: simpler to defer to next iteration
6356 : * - don't try to continue even if the updater aborts: likewise
6357 : * - no crosscheck
6358 : */
6359 217441 : result = HeapTupleSatisfiesUpdate(&oldtup, GetCurrentCommandId(false),
6360 : buffer);
6361 :
6362 217441 : if (result == TM_Invisible)
6363 : {
6364 : /* no known way this can happen */
6365 0 : ereport(ERROR,
6366 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6367 : errmsg_internal("attempted to overwrite invisible tuple")));
6368 : }
6369 217441 : else if (result == TM_SelfModified)
6370 : {
6371 : /*
6372 : * CREATE INDEX might reach this if an expression is silly enough to
6373 : * call e.g. SELECT ... FROM pg_class FOR SHARE. C code of other SQL
6374 : * statements might get here after a heap_update() of the same row, in
6375 : * the absence of an intervening CommandCounterIncrement().
6376 : */
6377 0 : ereport(ERROR,
6378 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6379 : errmsg("tuple to be updated was already modified by an operation triggered by the current command")));
6380 : }
6381 217441 : else if (result == TM_BeingModified)
6382 : {
6383 : TransactionId xwait;
6384 : uint16 infomask;
6385 :
6386 29 : xwait = HeapTupleHeaderGetRawXmax(oldtup.t_data);
6387 29 : infomask = oldtup.t_data->t_infomask;
6388 :
6389 29 : if (infomask & HEAP_XMAX_IS_MULTI)
6390 : {
6391 5 : LockTupleMode lockmode = LockTupleNoKeyExclusive;
6392 5 : MultiXactStatus mxact_status = MultiXactStatusNoKeyUpdate;
6393 : int remain;
6394 :
6395 5 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
6396 : lockmode, NULL))
6397 : {
6398 2 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6399 2 : release_callback(arg);
6400 2 : ret = false;
6401 2 : MultiXactIdWait((MultiXactId) xwait, mxact_status, infomask,
6402 : relation, &oldtup.t_self, XLTW_Update,
6403 : &remain);
6404 : }
6405 : else
6406 3 : ret = true;
6407 : }
6408 24 : else if (TransactionIdIsCurrentTransactionId(xwait))
6409 1 : ret = true;
6410 23 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask))
6411 1 : ret = true;
6412 : else
6413 : {
6414 22 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6415 22 : release_callback(arg);
6416 22 : ret = false;
6417 22 : XactLockTableWait(xwait, relation, &oldtup.t_self,
6418 : XLTW_Update);
6419 : }
6420 : }
6421 : else
6422 : {
6423 217412 : ret = (result == TM_Ok);
6424 217412 : if (!ret)
6425 : {
6426 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6427 0 : release_callback(arg);
6428 : }
6429 : }
6430 :
6431 : /*
6432 : * GetCatalogSnapshot() relies on invalidation messages to know when to
6433 : * take a new snapshot. COMMIT of xwait is responsible for sending the
6434 : * invalidation. We're not acquiring heavyweight locks sufficient to
6435 : * block if not yet sent, so we must take a new snapshot to ensure a later
6436 : * attempt has a fair chance. While we don't need this if xwait aborted,
6437 : * don't bother optimizing that.
6438 : */
6439 217441 : if (!ret)
6440 : {
6441 24 : UnlockTuple(relation, &oldtup.t_self, InplaceUpdateTupleLock);
6442 24 : ForgetInplace_Inval();
6443 24 : InvalidateCatalogSnapshot();
6444 : }
6445 217441 : return ret;
6446 : }
6447 :
6448 : /*
6449 : * heap_inplace_update_and_unlock - core of systable_inplace_update_finish
6450 : *
6451 : * The tuple cannot change size, and therefore its header fields and null
6452 : * bitmap (if any) don't change either.
6453 : *
6454 : * Since we hold LOCKTAG_TUPLE, no updater has a local copy of this tuple.
6455 : */
6456 : void
6457 100019 : heap_inplace_update_and_unlock(Relation relation,
6458 : HeapTuple oldtup, HeapTuple tuple,
6459 : Buffer buffer)
6460 : {
6461 100019 : HeapTupleHeader htup = oldtup->t_data;
6462 : uint32 oldlen;
6463 : uint32 newlen;
6464 : char *dst;
6465 : char *src;
6466 100019 : int nmsgs = 0;
6467 100019 : SharedInvalidationMessage *invalMessages = NULL;
6468 100019 : bool RelcacheInitFileInval = false;
6469 :
6470 : Assert(ItemPointerEquals(&oldtup->t_self, &tuple->t_self));
6471 100019 : oldlen = oldtup->t_len - htup->t_hoff;
6472 100019 : newlen = tuple->t_len - tuple->t_data->t_hoff;
6473 100019 : if (oldlen != newlen || htup->t_hoff != tuple->t_data->t_hoff)
6474 0 : elog(ERROR, "wrong tuple length");
6475 :
6476 100019 : dst = (char *) htup + htup->t_hoff;
6477 100019 : src = (char *) tuple->t_data + tuple->t_data->t_hoff;
6478 :
6479 : /* Like RecordTransactionCommit(), log only if needed */
6480 100019 : if (XLogStandbyInfoActive())
6481 64998 : nmsgs = inplaceGetInvalidationMessages(&invalMessages,
6482 : &RelcacheInitFileInval);
6483 :
6484 : /*
6485 : * Unlink relcache init files as needed. If unlinking, acquire
6486 : * RelCacheInitLock until after associated invalidations. By doing this
6487 : * in advance, if we checkpoint and then crash between inplace
6488 : * XLogInsert() and inval, we don't rely on StartupXLOG() ->
6489 : * RelationCacheInitFileRemove(). That uses elevel==LOG, so replay would
6490 : * neglect to PANIC on EIO.
6491 : */
6492 100019 : PreInplace_Inval();
6493 :
6494 : /*----------
6495 : * NO EREPORT(ERROR) from here till changes are complete
6496 : *
6497 : * Our exclusive buffer lock won't stop a reader having already pinned and
6498 : * checked visibility for this tuple. With the usual order of changes
6499 : * (i.e. updating the buffer contents before WAL logging), a reader could
6500 : * observe our not-yet-persistent update to relfrozenxid and update
6501 : * datfrozenxid based on that. A crash in that moment could allow
6502 : * datfrozenxid to overtake relfrozenxid:
6503 : *
6504 : * ["D" is a VACUUM (ONLY_DATABASE_STATS)]
6505 : * ["R" is a VACUUM tbl]
6506 : * D: vac_update_datfrozenxid() -> systable_beginscan(pg_class)
6507 : * D: systable_getnext() returns pg_class tuple of tbl
6508 : * R: memcpy() into pg_class tuple of tbl
6509 : * D: raise pg_database.datfrozenxid, XLogInsert(), finish
6510 : * [crash]
6511 : * [recovery restores datfrozenxid w/o relfrozenxid]
6512 : *
6513 : * We avoid that by using a temporary copy of the buffer to hide our
6514 : * change from other backends until the change has been WAL-logged. We
6515 : * apply our change to the temporary copy and WAL-log it, before modifying
6516 : * the real page. That way any action a reader of the in-place-updated
6517 : * value takes will be WAL logged after this change.
6518 : */
6519 100019 : START_CRIT_SECTION();
6520 :
6521 100019 : MarkBufferDirty(buffer);
6522 :
6523 : /* XLOG stuff */
6524 100019 : if (RelationNeedsWAL(relation))
6525 : {
6526 : xl_heap_inplace xlrec;
6527 : PGAlignedBlock copied_buffer;
6528 100011 : char *origdata = (char *) BufferGetBlock(buffer);
6529 100011 : Page page = BufferGetPage(buffer);
6530 100011 : uint16 lower = ((PageHeader) page)->pd_lower;
6531 100011 : uint16 upper = ((PageHeader) page)->pd_upper;
6532 : uintptr_t dst_offset_in_block;
6533 : RelFileLocator rlocator;
6534 : ForkNumber forkno;
6535 : BlockNumber blkno;
6536 : XLogRecPtr recptr;
6537 :
6538 100011 : xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
6539 100011 : xlrec.dbId = MyDatabaseId;
6540 100011 : xlrec.tsId = MyDatabaseTableSpace;
6541 100011 : xlrec.relcacheInitFileInval = RelcacheInitFileInval;
6542 100011 : xlrec.nmsgs = nmsgs;
6543 :
6544 100011 : XLogBeginInsert();
6545 100011 : XLogRegisterData(&xlrec, MinSizeOfHeapInplace);
6546 100011 : if (nmsgs != 0)
6547 47043 : XLogRegisterData(invalMessages,
6548 : nmsgs * sizeof(SharedInvalidationMessage));
6549 :
6550 : /* register block matching what buffer will look like after changes */
6551 100011 : memcpy(copied_buffer.data, origdata, lower);
6552 100011 : memcpy(copied_buffer.data + upper, origdata + upper, BLCKSZ - upper);
6553 100011 : dst_offset_in_block = dst - origdata;
6554 100011 : memcpy(copied_buffer.data + dst_offset_in_block, src, newlen);
6555 100011 : BufferGetTag(buffer, &rlocator, &forkno, &blkno);
6556 : Assert(forkno == MAIN_FORKNUM);
6557 100011 : XLogRegisterBlock(0, &rlocator, forkno, blkno, copied_buffer.data,
6558 : REGBUF_STANDARD);
6559 100011 : XLogRegisterBufData(0, src, newlen);
6560 :
6561 : /* inplace updates aren't decoded atm, don't log the origin */
6562 :
6563 100011 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_INPLACE);
6564 :
6565 100011 : PageSetLSN(page, recptr);
6566 : }
6567 :
6568 100019 : memcpy(dst, src, newlen);
6569 :
6570 100019 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6571 :
6572 : /*
6573 : * Send invalidations to shared queue. SearchSysCacheLocked1() assumes we
6574 : * do this before UnlockTuple().
6575 : */
6576 100019 : AtInplace_Inval();
6577 :
6578 100019 : END_CRIT_SECTION();
6579 100019 : UnlockTuple(relation, &tuple->t_self, InplaceUpdateTupleLock);
6580 :
6581 100019 : AcceptInvalidationMessages(); /* local processing of just-sent inval */
6582 :
6583 : /*
6584 : * Queue a transactional inval, for logical decoding and for third-party
6585 : * code that might have been relying on it since long before inplace
6586 : * update adopted immediate invalidation. See README.tuplock section
6587 : * "Reading inplace-updated columns" for logical decoding details.
6588 : */
6589 100019 : if (!IsBootstrapProcessingMode())
6590 82064 : CacheInvalidateHeapTuple(relation, tuple, NULL);
6591 100019 : }
6592 :
6593 : /*
6594 : * heap_inplace_unlock - reverse of heap_inplace_lock
6595 : */
6596 : void
6597 117398 : heap_inplace_unlock(Relation relation,
6598 : HeapTuple oldtup, Buffer buffer)
6599 : {
6600 117398 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6601 117398 : UnlockTuple(relation, &oldtup->t_self, InplaceUpdateTupleLock);
6602 117398 : ForgetInplace_Inval();
6603 117398 : }
6604 :
6605 : #define FRM_NOOP 0x0001
6606 : #define FRM_INVALIDATE_XMAX 0x0002
6607 : #define FRM_RETURN_IS_XID 0x0004
6608 : #define FRM_RETURN_IS_MULTI 0x0008
6609 : #define FRM_MARK_COMMITTED 0x0010
6610 :
6611 : /*
6612 : * FreezeMultiXactId
6613 : * Determine what to do during freezing when a tuple is marked by a
6614 : * MultiXactId.
6615 : *
6616 : * "flags" is an output value; it's used to tell caller what to do on return.
6617 : * "pagefrz" is an input/output value, used to manage page level freezing.
6618 : *
6619 : * Possible values that we can set in "flags":
6620 : * FRM_NOOP
6621 : * don't do anything -- keep existing Xmax
6622 : * FRM_INVALIDATE_XMAX
6623 : * mark Xmax as InvalidTransactionId and set XMAX_INVALID flag.
6624 : * FRM_RETURN_IS_XID
6625 : * The Xid return value is a single update Xid to set as xmax.
6626 : * FRM_MARK_COMMITTED
6627 : * Xmax can be marked as HEAP_XMAX_COMMITTED
6628 : * FRM_RETURN_IS_MULTI
6629 : * The return value is a new MultiXactId to set as new Xmax.
6630 : * (caller must obtain proper infomask bits using GetMultiXactIdHintBits)
6631 : *
6632 : * Caller delegates control of page freezing to us. In practice we always
6633 : * force freezing of caller's page unless FRM_NOOP processing is indicated.
6634 : * We help caller ensure that XIDs < FreezeLimit and MXIDs < MultiXactCutoff
6635 : * can never be left behind. We freely choose when and how to process each
6636 : * Multi, without ever violating the cutoff postconditions for freezing.
6637 : *
6638 : * It's useful to remove Multis on a proactive timeline (relative to freezing
6639 : * XIDs) to keep MultiXact member SLRU buffer misses to a minimum. It can also
6640 : * be cheaper in the short run, for us, since we too can avoid SLRU buffer
6641 : * misses through eager processing.
6642 : *
6643 : * NB: Creates a _new_ MultiXactId when FRM_RETURN_IS_MULTI is set, though only
6644 : * when FreezeLimit and/or MultiXactCutoff cutoffs leave us with no choice.
6645 : * This can usually be put off, which is usually enough to avoid it altogether.
6646 : * Allocating new multis during VACUUM should be avoided on general principle;
6647 : * only VACUUM can advance relminmxid, so allocating new Multis here comes with
6648 : * its own special risks.
6649 : *
6650 : * NB: Caller must maintain "no freeze" NewRelfrozenXid/NewRelminMxid trackers
6651 : * using heap_tuple_should_freeze when we haven't forced page-level freezing.
6652 : *
6653 : * NB: Caller should avoid needlessly calling heap_tuple_should_freeze when we
6654 : * have already forced page-level freezing, since that might incur the same
6655 : * SLRU buffer misses that we specifically intended to avoid by freezing.
6656 : */
6657 : static TransactionId
6658 7 : FreezeMultiXactId(MultiXactId multi, uint16 t_infomask,
6659 : const struct VacuumCutoffs *cutoffs, uint16 *flags,
6660 : HeapPageFreeze *pagefrz)
6661 : {
6662 : TransactionId newxmax;
6663 : MultiXactMember *members;
6664 : int nmembers;
6665 : bool need_replace;
6666 : int nnewmembers;
6667 : MultiXactMember *newmembers;
6668 : bool has_lockers;
6669 : TransactionId update_xid;
6670 : bool update_committed;
6671 : TransactionId FreezePageRelfrozenXid;
6672 :
6673 7 : *flags = 0;
6674 :
6675 : /* We should only be called in Multis */
6676 : Assert(t_infomask & HEAP_XMAX_IS_MULTI);
6677 :
6678 14 : if (!MultiXactIdIsValid(multi) ||
6679 7 : HEAP_LOCKED_UPGRADED(t_infomask))
6680 : {
6681 0 : *flags |= FRM_INVALIDATE_XMAX;
6682 0 : pagefrz->freeze_required = true;
6683 0 : return InvalidTransactionId;
6684 : }
6685 7 : else if (MultiXactIdPrecedes(multi, cutoffs->relminmxid))
6686 0 : ereport(ERROR,
6687 : (errcode(ERRCODE_DATA_CORRUPTED),
6688 : errmsg_internal("found multixact %u from before relminmxid %u",
6689 : multi, cutoffs->relminmxid)));
6690 7 : else if (MultiXactIdPrecedes(multi, cutoffs->OldestMxact))
6691 : {
6692 : TransactionId update_xact;
6693 :
6694 : /*
6695 : * This old multi cannot possibly have members still running, but
6696 : * verify just in case. If it was a locker only, it can be removed
6697 : * without any further consideration; but if it contained an update,
6698 : * we might need to preserve it.
6699 : */
6700 5 : if (MultiXactIdIsRunning(multi,
6701 5 : HEAP_XMAX_IS_LOCKED_ONLY(t_infomask)))
6702 0 : ereport(ERROR,
6703 : (errcode(ERRCODE_DATA_CORRUPTED),
6704 : errmsg_internal("multixact %u from before multi freeze cutoff %u found to be still running",
6705 : multi, cutoffs->OldestMxact)));
6706 :
6707 5 : if (HEAP_XMAX_IS_LOCKED_ONLY(t_infomask))
6708 : {
6709 5 : *flags |= FRM_INVALIDATE_XMAX;
6710 5 : pagefrz->freeze_required = true;
6711 5 : return InvalidTransactionId;
6712 : }
6713 :
6714 : /* replace multi with single XID for its updater? */
6715 0 : update_xact = MultiXactIdGetUpdateXid(multi, t_infomask);
6716 0 : if (TransactionIdPrecedes(update_xact, cutoffs->relfrozenxid))
6717 0 : ereport(ERROR,
6718 : (errcode(ERRCODE_DATA_CORRUPTED),
6719 : errmsg_internal("multixact %u contains update XID %u from before relfrozenxid %u",
6720 : multi, update_xact,
6721 : cutoffs->relfrozenxid)));
6722 0 : else if (TransactionIdPrecedes(update_xact, cutoffs->OldestXmin))
6723 : {
6724 : /*
6725 : * Updater XID has to have aborted (otherwise the tuple would have
6726 : * been pruned away instead, since updater XID is < OldestXmin).
6727 : * Just remove xmax.
6728 : */
6729 0 : if (TransactionIdDidCommit(update_xact))
6730 0 : ereport(ERROR,
6731 : (errcode(ERRCODE_DATA_CORRUPTED),
6732 : errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
6733 : multi, update_xact,
6734 : cutoffs->OldestXmin)));
6735 0 : *flags |= FRM_INVALIDATE_XMAX;
6736 0 : pagefrz->freeze_required = true;
6737 0 : return InvalidTransactionId;
6738 : }
6739 :
6740 : /* Have to keep updater XID as new xmax */
6741 0 : *flags |= FRM_RETURN_IS_XID;
6742 0 : pagefrz->freeze_required = true;
6743 0 : return update_xact;
6744 : }
6745 :
6746 : /*
6747 : * Some member(s) of this Multi may be below FreezeLimit xid cutoff, so we
6748 : * need to walk the whole members array to figure out what to do, if
6749 : * anything.
6750 : */
6751 : nmembers =
6752 2 : GetMultiXactIdMembers(multi, &members, false,
6753 2 : HEAP_XMAX_IS_LOCKED_ONLY(t_infomask));
6754 2 : if (nmembers <= 0)
6755 : {
6756 : /* Nothing worth keeping */
6757 0 : *flags |= FRM_INVALIDATE_XMAX;
6758 0 : pagefrz->freeze_required = true;
6759 0 : return InvalidTransactionId;
6760 : }
6761 :
6762 : /*
6763 : * The FRM_NOOP case is the only case where we might need to ratchet back
6764 : * FreezePageRelfrozenXid or FreezePageRelminMxid. It is also the only
6765 : * case where our caller might ratchet back its NoFreezePageRelfrozenXid
6766 : * or NoFreezePageRelminMxid "no freeze" trackers to deal with a multi.
6767 : * FRM_NOOP handling should result in the NewRelfrozenXid/NewRelminMxid
6768 : * trackers managed by VACUUM being ratcheting back by xmax to the degree
6769 : * required to make it safe to leave xmax undisturbed, independent of
6770 : * whether or not page freezing is triggered somewhere else.
6771 : *
6772 : * Our policy is to force freezing in every case other than FRM_NOOP,
6773 : * which obviates the need to maintain either set of trackers, anywhere.
6774 : * Every other case will reliably execute a freeze plan for xmax that
6775 : * either replaces xmax with an XID/MXID >= OldestXmin/OldestMxact, or
6776 : * sets xmax to an InvalidTransactionId XID, rendering xmax fully frozen.
6777 : * (VACUUM's NewRelfrozenXid/NewRelminMxid trackers are initialized with
6778 : * OldestXmin/OldestMxact, so later values never need to be tracked here.)
6779 : */
6780 2 : need_replace = false;
6781 2 : FreezePageRelfrozenXid = pagefrz->FreezePageRelfrozenXid;
6782 4 : for (int i = 0; i < nmembers; i++)
6783 : {
6784 3 : TransactionId xid = members[i].xid;
6785 :
6786 : Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
6787 :
6788 3 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
6789 : {
6790 : /* Can't violate the FreezeLimit postcondition */
6791 1 : need_replace = true;
6792 1 : break;
6793 : }
6794 2 : if (TransactionIdPrecedes(xid, FreezePageRelfrozenXid))
6795 0 : FreezePageRelfrozenXid = xid;
6796 : }
6797 :
6798 : /* Can't violate the MultiXactCutoff postcondition, either */
6799 2 : if (!need_replace)
6800 1 : need_replace = MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff);
6801 :
6802 2 : if (!need_replace)
6803 : {
6804 : /*
6805 : * vacuumlazy.c might ratchet back NewRelminMxid, NewRelfrozenXid, or
6806 : * both together to make it safe to retain this particular multi after
6807 : * freezing its page
6808 : */
6809 1 : *flags |= FRM_NOOP;
6810 1 : pagefrz->FreezePageRelfrozenXid = FreezePageRelfrozenXid;
6811 1 : if (MultiXactIdPrecedes(multi, pagefrz->FreezePageRelminMxid))
6812 0 : pagefrz->FreezePageRelminMxid = multi;
6813 1 : pfree(members);
6814 1 : return multi;
6815 : }
6816 :
6817 : /*
6818 : * Do a more thorough second pass over the multi to figure out which
6819 : * member XIDs actually need to be kept. Checking the precise status of
6820 : * individual members might even show that we don't need to keep anything.
6821 : * That is quite possible even though the Multi must be >= OldestMxact,
6822 : * since our second pass only keeps member XIDs when it's truly necessary;
6823 : * even member XIDs >= OldestXmin often won't be kept by second pass.
6824 : */
6825 1 : nnewmembers = 0;
6826 1 : newmembers = palloc_array(MultiXactMember, nmembers);
6827 1 : has_lockers = false;
6828 1 : update_xid = InvalidTransactionId;
6829 1 : update_committed = false;
6830 :
6831 : /*
6832 : * Determine whether to keep each member xid, or to ignore it instead
6833 : */
6834 3 : for (int i = 0; i < nmembers; i++)
6835 : {
6836 2 : TransactionId xid = members[i].xid;
6837 2 : MultiXactStatus mstatus = members[i].status;
6838 :
6839 : Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
6840 :
6841 2 : if (!ISUPDATE_from_mxstatus(mstatus))
6842 : {
6843 : /*
6844 : * Locker XID (not updater XID). We only keep lockers that are
6845 : * still running.
6846 : */
6847 4 : if (TransactionIdIsCurrentTransactionId(xid) ||
6848 2 : TransactionIdIsInProgress(xid))
6849 : {
6850 1 : if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
6851 0 : ereport(ERROR,
6852 : (errcode(ERRCODE_DATA_CORRUPTED),
6853 : errmsg_internal("multixact %u contains running locker XID %u from before removable cutoff %u",
6854 : multi, xid,
6855 : cutoffs->OldestXmin)));
6856 1 : newmembers[nnewmembers++] = members[i];
6857 1 : has_lockers = true;
6858 : }
6859 :
6860 2 : continue;
6861 : }
6862 :
6863 : /*
6864 : * Updater XID (not locker XID). Should we keep it?
6865 : *
6866 : * Since the tuple wasn't totally removed when vacuum pruned, the
6867 : * update Xid cannot possibly be older than OldestXmin cutoff unless
6868 : * the updater XID aborted. If the updater transaction is known
6869 : * aborted or crashed then it's okay to ignore it, otherwise not.
6870 : *
6871 : * In any case the Multi should never contain two updaters, whatever
6872 : * their individual commit status. Check for that first, in passing.
6873 : */
6874 0 : if (TransactionIdIsValid(update_xid))
6875 0 : ereport(ERROR,
6876 : (errcode(ERRCODE_DATA_CORRUPTED),
6877 : errmsg_internal("multixact %u has two or more updating members",
6878 : multi),
6879 : errdetail_internal("First updater XID=%u second updater XID=%u.",
6880 : update_xid, xid)));
6881 :
6882 : /*
6883 : * As with all tuple visibility routines, it's critical to test
6884 : * TransactionIdIsInProgress before TransactionIdDidCommit, because of
6885 : * race conditions explained in detail in heapam_visibility.c.
6886 : */
6887 0 : if (TransactionIdIsCurrentTransactionId(xid) ||
6888 0 : TransactionIdIsInProgress(xid))
6889 0 : update_xid = xid;
6890 0 : else if (TransactionIdDidCommit(xid))
6891 : {
6892 : /*
6893 : * The transaction committed, so we can tell caller to set
6894 : * HEAP_XMAX_COMMITTED. (We can only do this because we know the
6895 : * transaction is not running.)
6896 : */
6897 0 : update_committed = true;
6898 0 : update_xid = xid;
6899 : }
6900 : else
6901 : {
6902 : /*
6903 : * Not in progress, not committed -- must be aborted or crashed;
6904 : * we can ignore it.
6905 : */
6906 0 : continue;
6907 : }
6908 :
6909 : /*
6910 : * We determined that updater must be kept -- add it to pending new
6911 : * members list
6912 : */
6913 0 : if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
6914 0 : ereport(ERROR,
6915 : (errcode(ERRCODE_DATA_CORRUPTED),
6916 : errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
6917 : multi, xid, cutoffs->OldestXmin)));
6918 0 : newmembers[nnewmembers++] = members[i];
6919 : }
6920 :
6921 1 : pfree(members);
6922 :
6923 : /*
6924 : * Determine what to do with caller's multi based on information gathered
6925 : * during our second pass
6926 : */
6927 1 : if (nnewmembers == 0)
6928 : {
6929 : /* Nothing worth keeping */
6930 0 : *flags |= FRM_INVALIDATE_XMAX;
6931 0 : newxmax = InvalidTransactionId;
6932 : }
6933 1 : else if (TransactionIdIsValid(update_xid) && !has_lockers)
6934 : {
6935 : /*
6936 : * If there's a single member and it's an update, pass it back alone
6937 : * without creating a new Multi. (XXX we could do this when there's a
6938 : * single remaining locker, too, but that would complicate the API too
6939 : * much; moreover, the case with the single updater is more
6940 : * interesting, because those are longer-lived.)
6941 : */
6942 : Assert(nnewmembers == 1);
6943 0 : *flags |= FRM_RETURN_IS_XID;
6944 0 : if (update_committed)
6945 0 : *flags |= FRM_MARK_COMMITTED;
6946 0 : newxmax = update_xid;
6947 : }
6948 : else
6949 : {
6950 : /*
6951 : * Create a new multixact with the surviving members of the previous
6952 : * one, to set as new Xmax in the tuple
6953 : */
6954 1 : newxmax = MultiXactIdCreateFromMembers(nnewmembers, newmembers);
6955 1 : *flags |= FRM_RETURN_IS_MULTI;
6956 : }
6957 :
6958 1 : pfree(newmembers);
6959 :
6960 1 : pagefrz->freeze_required = true;
6961 1 : return newxmax;
6962 : }
6963 :
6964 : /*
6965 : * heap_prepare_freeze_tuple
6966 : *
6967 : * Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
6968 : * are older than the OldestXmin and/or OldestMxact freeze cutoffs. If so,
6969 : * setup enough state (in the *frz output argument) to enable caller to
6970 : * process this tuple as part of freezing its page, and return true. Return
6971 : * false if nothing can be changed about the tuple right now.
6972 : *
6973 : * FreezePageConflictXid is advanced only for xmin/xvac freezing, not for xmax
6974 : * changes. We only remove xmax state here when it is lock-only, or when the
6975 : * updater XID (including an updater member of a MultiXact) must be aborted;
6976 : * otherwise, the tuple would already be removable. Neither case affects
6977 : * visibility on a standby.
6978 : *
6979 : * Also sets *totally_frozen to true if the tuple will be totally frozen once
6980 : * caller executes returned freeze plan (or if the tuple was already totally
6981 : * frozen by an earlier VACUUM). This indicates that there are no remaining
6982 : * XIDs or MultiXactIds that will need to be processed by a future VACUUM.
6983 : *
6984 : * VACUUM caller must assemble HeapTupleFreeze freeze plan entries for every
6985 : * tuple that we returned true for, and then execute freezing. Caller must
6986 : * initialize pagefrz fields for page as a whole before first call here for
6987 : * each heap page.
6988 : *
6989 : * VACUUM caller decides on whether or not to freeze the page as a whole.
6990 : * We'll often prepare freeze plans for a page that caller just discards.
6991 : * However, VACUUM doesn't always get to make a choice; it must freeze when
6992 : * pagefrz.freeze_required is set, to ensure that any XIDs < FreezeLimit (and
6993 : * MXIDs < MultiXactCutoff) can never be left behind. We help to make sure
6994 : * that VACUUM always follows that rule.
6995 : *
6996 : * We sometimes force freezing of xmax MultiXactId values long before it is
6997 : * strictly necessary to do so just to ensure the FreezeLimit postcondition.
6998 : * It's worth processing MultiXactIds proactively when it is cheap to do so,
6999 : * and it's convenient to make that happen by piggy-backing it on the "force
7000 : * freezing" mechanism. Conversely, we sometimes delay freezing MultiXactIds
7001 : * because it is expensive right now (though only when it's still possible to
7002 : * do so without violating the FreezeLimit/MultiXactCutoff postcondition).
7003 : *
7004 : * It is assumed that the caller has checked the tuple with
7005 : * HeapTupleSatisfiesVacuum() and determined that it is not HEAPTUPLE_DEAD
7006 : * (else we should be removing the tuple, not freezing it).
7007 : *
7008 : * NB: This function has side effects: it might allocate a new MultiXactId.
7009 : * It will be set as tuple's new xmax when our *frz output is processed within
7010 : * heap_execute_freeze_tuple later on. If the tuple is in a shared buffer
7011 : * then caller had better have an exclusive lock on it already.
7012 : */
7013 : bool
7014 10259186 : heap_prepare_freeze_tuple(HeapTupleHeader tuple,
7015 : const struct VacuumCutoffs *cutoffs,
7016 : HeapPageFreeze *pagefrz,
7017 : HeapTupleFreeze *frz, bool *totally_frozen)
7018 : {
7019 10259186 : bool xmin_already_frozen = false,
7020 10259186 : xmax_already_frozen = false;
7021 10259186 : bool freeze_xmin = false,
7022 10259186 : replace_xvac = false,
7023 10259186 : replace_xmax = false,
7024 10259186 : freeze_xmax = false;
7025 : TransactionId xid;
7026 :
7027 10259186 : frz->xmax = HeapTupleHeaderGetRawXmax(tuple);
7028 10259186 : frz->t_infomask2 = tuple->t_infomask2;
7029 10259186 : frz->t_infomask = tuple->t_infomask;
7030 10259186 : frz->frzflags = 0;
7031 10259186 : frz->checkflags = 0;
7032 :
7033 : /*
7034 : * Process xmin, while keeping track of whether it's already frozen, or
7035 : * will become frozen iff our freeze plan is executed by caller (could be
7036 : * neither).
7037 : */
7038 10259186 : xid = HeapTupleHeaderGetXmin(tuple);
7039 10259186 : if (!TransactionIdIsNormal(xid))
7040 6451180 : xmin_already_frozen = true;
7041 : else
7042 : {
7043 3808006 : if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
7044 0 : ereport(ERROR,
7045 : (errcode(ERRCODE_DATA_CORRUPTED),
7046 : errmsg_internal("found xmin %u from before relfrozenxid %u",
7047 : xid, cutoffs->relfrozenxid)));
7048 :
7049 : /* Will set freeze_xmin flags in freeze plan below */
7050 3808006 : freeze_xmin = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
7051 :
7052 : /* Verify that xmin committed if and when freeze plan is executed */
7053 3808006 : if (freeze_xmin)
7054 : {
7055 3075306 : frz->checkflags |= HEAP_FREEZE_CHECK_XMIN_COMMITTED;
7056 3075306 : if (TransactionIdFollows(xid, pagefrz->FreezePageConflictXid))
7057 487204 : pagefrz->FreezePageConflictXid = xid;
7058 : }
7059 : }
7060 :
7061 : /*
7062 : * Old-style VACUUM FULL is gone, but we have to process xvac for as long
7063 : * as we support having MOVED_OFF/MOVED_IN tuples in the database
7064 : */
7065 10259186 : xid = HeapTupleHeaderGetXvac(tuple);
7066 10259186 : if (TransactionIdIsNormal(xid))
7067 : {
7068 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7069 : Assert(TransactionIdPrecedes(xid, cutoffs->OldestXmin));
7070 :
7071 : /*
7072 : * For Xvac, we always freeze proactively. This allows totally_frozen
7073 : * tracking to ignore xvac.
7074 : */
7075 0 : replace_xvac = pagefrz->freeze_required = true;
7076 :
7077 0 : if (TransactionIdFollows(xid, pagefrz->FreezePageConflictXid))
7078 0 : pagefrz->FreezePageConflictXid = xid;
7079 :
7080 : /* Will set replace_xvac flags in freeze plan below */
7081 : }
7082 :
7083 : /* Now process xmax */
7084 10259186 : xid = frz->xmax;
7085 10259186 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7086 : {
7087 : /* Raw xmax is a MultiXactId */
7088 : TransactionId newxmax;
7089 : uint16 flags;
7090 :
7091 : /*
7092 : * We will either remove xmax completely (in the "freeze_xmax" path),
7093 : * process xmax by replacing it (in the "replace_xmax" path), or
7094 : * perform no-op xmax processing. The only constraint is that the
7095 : * FreezeLimit/MultiXactCutoff postcondition must never be violated.
7096 : */
7097 7 : newxmax = FreezeMultiXactId(xid, tuple->t_infomask, cutoffs,
7098 : &flags, pagefrz);
7099 :
7100 7 : if (flags & FRM_NOOP)
7101 : {
7102 : /*
7103 : * xmax is a MultiXactId, and nothing about it changes for now.
7104 : * This is the only case where 'freeze_required' won't have been
7105 : * set for us by FreezeMultiXactId, as well as the only case where
7106 : * neither freeze_xmax nor replace_xmax are set (given a multi).
7107 : *
7108 : * This is a no-op, but the call to FreezeMultiXactId might have
7109 : * ratcheted back NewRelfrozenXid and/or NewRelminMxid trackers
7110 : * for us (the "freeze page" variants, specifically). That'll
7111 : * make it safe for our caller to freeze the page later on, while
7112 : * leaving this particular xmax undisturbed.
7113 : *
7114 : * FreezeMultiXactId is _not_ responsible for the "no freeze"
7115 : * NewRelfrozenXid/NewRelminMxid trackers, though -- that's our
7116 : * job. A call to heap_tuple_should_freeze for this same tuple
7117 : * will take place below if 'freeze_required' isn't set already.
7118 : * (This repeats work from FreezeMultiXactId, but allows "no
7119 : * freeze" tracker maintenance to happen in only one place.)
7120 : */
7121 : Assert(!MultiXactIdPrecedes(newxmax, cutoffs->MultiXactCutoff));
7122 : Assert(MultiXactIdIsValid(newxmax) && xid == newxmax);
7123 : }
7124 6 : else if (flags & FRM_RETURN_IS_XID)
7125 : {
7126 : /*
7127 : * xmax will become an updater Xid (original MultiXact's updater
7128 : * member Xid will be carried forward as a simple Xid in Xmax).
7129 : */
7130 : Assert(!TransactionIdPrecedes(newxmax, cutoffs->OldestXmin));
7131 :
7132 : /*
7133 : * NB -- some of these transformations are only valid because we
7134 : * know the return Xid is a tuple updater (i.e. not merely a
7135 : * locker.) Also note that the only reason we don't explicitly
7136 : * worry about HEAP_KEYS_UPDATED is because it lives in
7137 : * t_infomask2 rather than t_infomask.
7138 : */
7139 0 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7140 0 : frz->xmax = newxmax;
7141 0 : if (flags & FRM_MARK_COMMITTED)
7142 0 : frz->t_infomask |= HEAP_XMAX_COMMITTED;
7143 0 : replace_xmax = true;
7144 : }
7145 6 : else if (flags & FRM_RETURN_IS_MULTI)
7146 : {
7147 : uint16 newbits;
7148 : uint16 newbits2;
7149 :
7150 : /*
7151 : * xmax is an old MultiXactId that we have to replace with a new
7152 : * MultiXactId, to carry forward two or more original member XIDs.
7153 : */
7154 : Assert(!MultiXactIdPrecedes(newxmax, cutoffs->OldestMxact));
7155 :
7156 : /*
7157 : * We can't use GetMultiXactIdHintBits directly on the new multi
7158 : * here; that routine initializes the masks to all zeroes, which
7159 : * would lose other bits we need. Doing it this way ensures all
7160 : * unrelated bits remain untouched.
7161 : */
7162 1 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7163 1 : frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
7164 1 : GetMultiXactIdHintBits(newxmax, &newbits, &newbits2);
7165 1 : frz->t_infomask |= newbits;
7166 1 : frz->t_infomask2 |= newbits2;
7167 1 : frz->xmax = newxmax;
7168 1 : replace_xmax = true;
7169 : }
7170 : else
7171 : {
7172 : /*
7173 : * Freeze plan for tuple "freezes xmax" in the strictest sense:
7174 : * it'll leave nothing in xmax (neither an Xid nor a MultiXactId).
7175 : */
7176 : Assert(flags & FRM_INVALIDATE_XMAX);
7177 : Assert(!TransactionIdIsValid(newxmax));
7178 :
7179 : /* Will set freeze_xmax flags in freeze plan below */
7180 5 : freeze_xmax = true;
7181 : }
7182 :
7183 : /* MultiXactId processing forces freezing (barring FRM_NOOP case) */
7184 : Assert(pagefrz->freeze_required || (!freeze_xmax && !replace_xmax));
7185 : }
7186 10259179 : else if (TransactionIdIsNormal(xid))
7187 : {
7188 : /* Raw xmax is normal XID */
7189 4359867 : if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
7190 0 : ereport(ERROR,
7191 : (errcode(ERRCODE_DATA_CORRUPTED),
7192 : errmsg_internal("found xmax %u from before relfrozenxid %u",
7193 : xid, cutoffs->relfrozenxid)));
7194 :
7195 : /* Will set freeze_xmax flags in freeze plan below */
7196 4359867 : freeze_xmax = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
7197 :
7198 : /*
7199 : * Verify that xmax aborted if and when freeze plan is executed,
7200 : * provided it's from an update. (A lock-only xmax can be removed
7201 : * independent of this, since the lock is released at xact end.)
7202 : */
7203 4359867 : if (freeze_xmax && !HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask))
7204 978 : frz->checkflags |= HEAP_FREEZE_CHECK_XMAX_ABORTED;
7205 : }
7206 5899312 : else if (!TransactionIdIsValid(xid))
7207 : {
7208 : /* Raw xmax is InvalidTransactionId XID */
7209 : Assert((tuple->t_infomask & HEAP_XMAX_IS_MULTI) == 0);
7210 5899312 : xmax_already_frozen = true;
7211 : }
7212 : else
7213 0 : ereport(ERROR,
7214 : (errcode(ERRCODE_DATA_CORRUPTED),
7215 : errmsg_internal("found raw xmax %u (infomask 0x%04x) not invalid and not multi",
7216 : xid, tuple->t_infomask)));
7217 :
7218 10259186 : if (freeze_xmin)
7219 : {
7220 : Assert(!xmin_already_frozen);
7221 :
7222 3075306 : frz->t_infomask |= HEAP_XMIN_FROZEN;
7223 : }
7224 10259186 : if (replace_xvac)
7225 : {
7226 : /*
7227 : * If a MOVED_OFF tuple is not dead, the xvac transaction must have
7228 : * failed; whereas a non-dead MOVED_IN tuple must mean the xvac
7229 : * transaction succeeded.
7230 : */
7231 : Assert(pagefrz->freeze_required);
7232 0 : if (tuple->t_infomask & HEAP_MOVED_OFF)
7233 0 : frz->frzflags |= XLH_INVALID_XVAC;
7234 : else
7235 0 : frz->frzflags |= XLH_FREEZE_XVAC;
7236 : }
7237 : if (replace_xmax)
7238 : {
7239 : Assert(!xmax_already_frozen && !freeze_xmax);
7240 : Assert(pagefrz->freeze_required);
7241 :
7242 : /* Already set replace_xmax flags in freeze plan earlier */
7243 : }
7244 10259186 : if (freeze_xmax)
7245 : {
7246 : Assert(!xmax_already_frozen && !replace_xmax);
7247 :
7248 1868 : frz->xmax = InvalidTransactionId;
7249 :
7250 : /*
7251 : * The tuple might be marked either XMAX_INVALID or XMAX_COMMITTED +
7252 : * LOCKED. Normalize to INVALID just to be sure no one gets confused.
7253 : * Also get rid of the HEAP_KEYS_UPDATED bit.
7254 : */
7255 1868 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7256 1868 : frz->t_infomask |= HEAP_XMAX_INVALID;
7257 1868 : frz->t_infomask2 &= ~HEAP_HOT_UPDATED;
7258 1868 : frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
7259 : }
7260 :
7261 : /*
7262 : * Determine if this tuple is already totally frozen, or will become
7263 : * totally frozen (provided caller executes freeze plans for the page)
7264 : */
7265 19783804 : *totally_frozen = ((freeze_xmin || xmin_already_frozen) &&
7266 9524618 : (freeze_xmax || xmax_already_frozen));
7267 :
7268 10259186 : if (!pagefrz->freeze_required && !(xmin_already_frozen &&
7269 : xmax_already_frozen))
7270 : {
7271 : /*
7272 : * So far no previous tuple from the page made freezing mandatory.
7273 : * Does this tuple force caller to freeze the entire page?
7274 : */
7275 6361181 : pagefrz->freeze_required =
7276 6361181 : heap_tuple_should_freeze(tuple, cutoffs,
7277 : &pagefrz->NoFreezePageRelfrozenXid,
7278 : &pagefrz->NoFreezePageRelminMxid);
7279 : }
7280 :
7281 : /* Tell caller if this tuple has a usable freeze plan set in *frz */
7282 10259186 : return freeze_xmin || replace_xvac || replace_xmax || freeze_xmax;
7283 : }
7284 :
7285 : /*
7286 : * Perform xmin/xmax XID status sanity checks before actually executing freeze
7287 : * plans.
7288 : *
7289 : * heap_prepare_freeze_tuple doesn't perform these checks directly because
7290 : * pg_xact lookups are relatively expensive. They shouldn't be repeated by
7291 : * successive VACUUMs that each decide against freezing the same page.
7292 : */
7293 : void
7294 26653 : heap_pre_freeze_checks(Buffer buffer,
7295 : HeapTupleFreeze *tuples, int ntuples)
7296 : {
7297 26653 : Page page = BufferGetPage(buffer);
7298 :
7299 1258589 : for (int i = 0; i < ntuples; i++)
7300 : {
7301 1231936 : HeapTupleFreeze *frz = tuples + i;
7302 1231936 : ItemId itemid = PageGetItemId(page, frz->offset);
7303 : HeapTupleHeader htup;
7304 :
7305 1231936 : htup = (HeapTupleHeader) PageGetItem(page, itemid);
7306 :
7307 : /* Deliberately avoid relying on tuple hint bits here */
7308 1231936 : if (frz->checkflags & HEAP_FREEZE_CHECK_XMIN_COMMITTED)
7309 : {
7310 1231935 : TransactionId xmin = HeapTupleHeaderGetRawXmin(htup);
7311 :
7312 : Assert(!HeapTupleHeaderXminFrozen(htup));
7313 1231935 : if (unlikely(!TransactionIdDidCommit(xmin)))
7314 0 : ereport(ERROR,
7315 : (errcode(ERRCODE_DATA_CORRUPTED),
7316 : errmsg_internal("uncommitted xmin %u needs to be frozen",
7317 : xmin)));
7318 : }
7319 :
7320 : /*
7321 : * TransactionIdDidAbort won't work reliably in the presence of XIDs
7322 : * left behind by transactions that were in progress during a crash,
7323 : * so we can only check that xmax didn't commit
7324 : */
7325 1231936 : if (frz->checkflags & HEAP_FREEZE_CHECK_XMAX_ABORTED)
7326 : {
7327 354 : TransactionId xmax = HeapTupleHeaderGetRawXmax(htup);
7328 :
7329 : Assert(TransactionIdIsNormal(xmax));
7330 354 : if (unlikely(TransactionIdDidCommit(xmax)))
7331 0 : ereport(ERROR,
7332 : (errcode(ERRCODE_DATA_CORRUPTED),
7333 : errmsg_internal("cannot freeze committed xmax %u",
7334 : xmax)));
7335 : }
7336 : }
7337 26653 : }
7338 :
7339 : /*
7340 : * Helper which executes freezing of one or more heap tuples on a page on
7341 : * behalf of caller. Caller passes an array of tuple plans from
7342 : * heap_prepare_freeze_tuple. Caller must set 'offset' in each plan for us.
7343 : * Must be called in a critical section that also marks the buffer dirty and,
7344 : * if needed, emits WAL.
7345 : */
7346 : void
7347 26653 : heap_freeze_prepared_tuples(Buffer buffer, HeapTupleFreeze *tuples, int ntuples)
7348 : {
7349 26653 : Page page = BufferGetPage(buffer);
7350 :
7351 1258589 : for (int i = 0; i < ntuples; i++)
7352 : {
7353 1231936 : HeapTupleFreeze *frz = tuples + i;
7354 1231936 : ItemId itemid = PageGetItemId(page, frz->offset);
7355 : HeapTupleHeader htup;
7356 :
7357 1231936 : htup = (HeapTupleHeader) PageGetItem(page, itemid);
7358 1231936 : heap_execute_freeze_tuple(htup, frz);
7359 : }
7360 26653 : }
7361 :
7362 : /*
7363 : * heap_freeze_tuple
7364 : * Freeze tuple in place, without WAL logging.
7365 : *
7366 : * Useful for callers like CLUSTER that perform their own WAL logging.
7367 : */
7368 : bool
7369 465229 : heap_freeze_tuple(HeapTupleHeader tuple,
7370 : TransactionId relfrozenxid, TransactionId relminmxid,
7371 : TransactionId FreezeLimit, TransactionId MultiXactCutoff)
7372 : {
7373 : HeapTupleFreeze frz;
7374 : bool do_freeze;
7375 : bool totally_frozen;
7376 : struct VacuumCutoffs cutoffs;
7377 : HeapPageFreeze pagefrz;
7378 :
7379 465229 : cutoffs.relfrozenxid = relfrozenxid;
7380 465229 : cutoffs.relminmxid = relminmxid;
7381 465229 : cutoffs.OldestXmin = FreezeLimit;
7382 465229 : cutoffs.OldestMxact = MultiXactCutoff;
7383 465229 : cutoffs.FreezeLimit = FreezeLimit;
7384 465229 : cutoffs.MultiXactCutoff = MultiXactCutoff;
7385 :
7386 465229 : pagefrz.freeze_required = true;
7387 465229 : pagefrz.FreezePageRelfrozenXid = FreezeLimit;
7388 465229 : pagefrz.FreezePageRelminMxid = MultiXactCutoff;
7389 465229 : pagefrz.FreezePageConflictXid = InvalidTransactionId;
7390 465229 : pagefrz.NoFreezePageRelfrozenXid = FreezeLimit;
7391 465229 : pagefrz.NoFreezePageRelminMxid = MultiXactCutoff;
7392 :
7393 465229 : do_freeze = heap_prepare_freeze_tuple(tuple, &cutoffs,
7394 : &pagefrz, &frz, &totally_frozen);
7395 :
7396 : /*
7397 : * Note that because this is not a WAL-logged operation, we don't need to
7398 : * fill in the offset in the freeze record.
7399 : */
7400 :
7401 465229 : if (do_freeze)
7402 357718 : heap_execute_freeze_tuple(tuple, &frz);
7403 465229 : return do_freeze;
7404 : }
7405 :
7406 : /*
7407 : * For a given MultiXactId, return the hint bits that should be set in the
7408 : * tuple's infomask.
7409 : *
7410 : * Normally this should be called for a multixact that was just created, and
7411 : * so is on our local cache, so the GetMembers call is fast.
7412 : */
7413 : static void
7414 76866 : GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
7415 : uint16 *new_infomask2)
7416 : {
7417 : int nmembers;
7418 : MultiXactMember *members;
7419 : int i;
7420 76866 : uint16 bits = HEAP_XMAX_IS_MULTI;
7421 76866 : uint16 bits2 = 0;
7422 76866 : bool has_update = false;
7423 76866 : LockTupleMode strongest = LockTupleKeyShare;
7424 :
7425 : /*
7426 : * We only use this in multis we just created, so they cannot be values
7427 : * pre-pg_upgrade.
7428 : */
7429 76866 : nmembers = GetMultiXactIdMembers(multi, &members, false, false);
7430 :
7431 1472801 : for (i = 0; i < nmembers; i++)
7432 : {
7433 : LockTupleMode mode;
7434 :
7435 : /*
7436 : * Remember the strongest lock mode held by any member of the
7437 : * multixact.
7438 : */
7439 1395935 : mode = TUPLOCK_from_mxstatus(members[i].status);
7440 1395935 : if (mode > strongest)
7441 2929 : strongest = mode;
7442 :
7443 : /* See what other bits we need */
7444 1395935 : switch (members[i].status)
7445 : {
7446 1393513 : case MultiXactStatusForKeyShare:
7447 : case MultiXactStatusForShare:
7448 : case MultiXactStatusForNoKeyUpdate:
7449 1393513 : break;
7450 :
7451 53 : case MultiXactStatusForUpdate:
7452 53 : bits2 |= HEAP_KEYS_UPDATED;
7453 53 : break;
7454 :
7455 2359 : case MultiXactStatusNoKeyUpdate:
7456 2359 : has_update = true;
7457 2359 : break;
7458 :
7459 10 : case MultiXactStatusUpdate:
7460 10 : bits2 |= HEAP_KEYS_UPDATED;
7461 10 : has_update = true;
7462 10 : break;
7463 : }
7464 : }
7465 :
7466 76866 : if (strongest == LockTupleExclusive ||
7467 : strongest == LockTupleNoKeyExclusive)
7468 2450 : bits |= HEAP_XMAX_EXCL_LOCK;
7469 74416 : else if (strongest == LockTupleShare)
7470 476 : bits |= HEAP_XMAX_SHR_LOCK;
7471 73940 : else if (strongest == LockTupleKeyShare)
7472 73940 : bits |= HEAP_XMAX_KEYSHR_LOCK;
7473 :
7474 76866 : if (!has_update)
7475 74497 : bits |= HEAP_XMAX_LOCK_ONLY;
7476 :
7477 76866 : if (nmembers > 0)
7478 76866 : pfree(members);
7479 :
7480 76866 : *new_infomask = bits;
7481 76866 : *new_infomask2 = bits2;
7482 76866 : }
7483 :
7484 : /*
7485 : * MultiXactIdGetUpdateXid
7486 : *
7487 : * Given a multixact Xmax and corresponding infomask, which does not have the
7488 : * HEAP_XMAX_LOCK_ONLY bit set, obtain and return the Xid of the updating
7489 : * transaction.
7490 : *
7491 : * Caller is expected to check the status of the updating transaction, if
7492 : * necessary.
7493 : */
7494 : static TransactionId
7495 162054 : MultiXactIdGetUpdateXid(TransactionId xmax, uint16 t_infomask)
7496 : {
7497 162054 : TransactionId update_xact = InvalidTransactionId;
7498 : MultiXactMember *members;
7499 : int nmembers;
7500 :
7501 : Assert(!(t_infomask & HEAP_XMAX_LOCK_ONLY));
7502 : Assert(t_infomask & HEAP_XMAX_IS_MULTI);
7503 :
7504 : /*
7505 : * Since we know the LOCK_ONLY bit is not set, this cannot be a multi from
7506 : * pre-pg_upgrade.
7507 : */
7508 162054 : nmembers = GetMultiXactIdMembers(xmax, &members, false, false);
7509 :
7510 162054 : if (nmembers > 0)
7511 : {
7512 : int i;
7513 :
7514 245875 : for (i = 0; i < nmembers; i++)
7515 : {
7516 : /* Ignore lockers */
7517 245875 : if (!ISUPDATE_from_mxstatus(members[i].status))
7518 83821 : continue;
7519 :
7520 : /* there can be at most one updater */
7521 : Assert(update_xact == InvalidTransactionId);
7522 162054 : update_xact = members[i].xid;
7523 : #ifndef USE_ASSERT_CHECKING
7524 :
7525 : /*
7526 : * in an assert-enabled build, walk the whole array to ensure
7527 : * there's no other updater.
7528 : */
7529 162054 : break;
7530 : #endif
7531 : }
7532 :
7533 162054 : pfree(members);
7534 : }
7535 :
7536 162054 : return update_xact;
7537 : }
7538 :
7539 : /*
7540 : * HeapTupleGetUpdateXid
7541 : * As above, but use a HeapTupleHeader
7542 : *
7543 : * See also HeapTupleHeaderGetUpdateXid, which can be used without previously
7544 : * checking the hint bits.
7545 : */
7546 : TransactionId
7547 159913 : HeapTupleGetUpdateXid(const HeapTupleHeaderData *tup)
7548 : {
7549 159913 : return MultiXactIdGetUpdateXid(HeapTupleHeaderGetRawXmax(tup),
7550 159913 : tup->t_infomask);
7551 : }
7552 :
7553 : /*
7554 : * Does the given multixact conflict with the current transaction grabbing a
7555 : * tuple lock of the given strength?
7556 : *
7557 : * The passed infomask pairs up with the given multixact in the tuple header.
7558 : *
7559 : * If current_is_member is not NULL, it is set to 'true' if the current
7560 : * transaction is a member of the given multixact.
7561 : */
7562 : static bool
7563 218 : DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
7564 : LockTupleMode lockmode, bool *current_is_member)
7565 : {
7566 : int nmembers;
7567 : MultiXactMember *members;
7568 218 : bool result = false;
7569 218 : LOCKMODE wanted = tupleLockExtraInfo[lockmode].hwlock;
7570 :
7571 218 : if (HEAP_LOCKED_UPGRADED(infomask))
7572 0 : return false;
7573 :
7574 218 : nmembers = GetMultiXactIdMembers(multi, &members, false,
7575 218 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
7576 218 : if (nmembers >= 0)
7577 : {
7578 : int i;
7579 :
7580 2682 : for (i = 0; i < nmembers; i++)
7581 : {
7582 : TransactionId memxid;
7583 : LOCKMODE memlockmode;
7584 :
7585 2471 : if (result && (current_is_member == NULL || *current_is_member))
7586 : break;
7587 :
7588 2464 : memlockmode = LOCKMODE_from_mxstatus(members[i].status);
7589 :
7590 : /* ignore members from current xact (but track their presence) */
7591 2464 : memxid = members[i].xid;
7592 2464 : if (TransactionIdIsCurrentTransactionId(memxid))
7593 : {
7594 92 : if (current_is_member != NULL)
7595 78 : *current_is_member = true;
7596 92 : continue;
7597 : }
7598 2372 : else if (result)
7599 8 : continue;
7600 :
7601 : /* ignore members that don't conflict with the lock we want */
7602 2364 : if (!DoLockModesConflict(memlockmode, wanted))
7603 2325 : continue;
7604 :
7605 39 : if (ISUPDATE_from_mxstatus(members[i].status))
7606 : {
7607 : /* ignore aborted updaters */
7608 17 : if (TransactionIdDidAbort(memxid))
7609 1 : continue;
7610 : }
7611 : else
7612 : {
7613 : /* ignore lockers-only that are no longer in progress */
7614 22 : if (!TransactionIdIsInProgress(memxid))
7615 7 : continue;
7616 : }
7617 :
7618 : /*
7619 : * Whatever remains are either live lockers that conflict with our
7620 : * wanted lock, and updaters that are not aborted. Those conflict
7621 : * with what we want. Set up to return true, but keep going to
7622 : * look for the current transaction among the multixact members,
7623 : * if needed.
7624 : */
7625 31 : result = true;
7626 : }
7627 218 : pfree(members);
7628 : }
7629 :
7630 218 : return result;
7631 : }
7632 :
7633 : /*
7634 : * Do_MultiXactIdWait
7635 : * Actual implementation for the two functions below.
7636 : *
7637 : * 'multi', 'status' and 'infomask' indicate what to sleep on (the status is
7638 : * needed to ensure we only sleep on conflicting members, and the infomask is
7639 : * used to optimize multixact access in case it's a lock-only multi); 'nowait'
7640 : * indicates whether to use conditional lock acquisition, to allow callers to
7641 : * fail if lock is unavailable. 'rel', 'ctid' and 'oper' are used to set up
7642 : * context information for error messages. 'remaining', if not NULL, receives
7643 : * the number of members that are still running, including any (non-aborted)
7644 : * subtransactions of our own transaction. 'logLockFailure' indicates whether
7645 : * to log details when a lock acquisition fails with 'nowait' enabled.
7646 : *
7647 : * We do this by sleeping on each member using XactLockTableWait. Any
7648 : * members that belong to the current backend are *not* waited for, however;
7649 : * this would not merely be useless but would lead to Assert failure inside
7650 : * XactLockTableWait. By the time this returns, it is certain that all
7651 : * transactions *of other backends* that were members of the MultiXactId
7652 : * that conflict with the requested status are dead (and no new ones can have
7653 : * been added, since it is not legal to add members to an existing
7654 : * MultiXactId).
7655 : *
7656 : * But by the time we finish sleeping, someone else may have changed the Xmax
7657 : * of the containing tuple, so the caller needs to iterate on us somehow.
7658 : *
7659 : * Note that in case we return false, the number of remaining members is
7660 : * not to be trusted.
7661 : */
7662 : static bool
7663 61 : Do_MultiXactIdWait(MultiXactId multi, MultiXactStatus status,
7664 : uint16 infomask, bool nowait,
7665 : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
7666 : int *remaining, bool logLockFailure)
7667 : {
7668 61 : bool result = true;
7669 : MultiXactMember *members;
7670 : int nmembers;
7671 61 : int remain = 0;
7672 :
7673 : /* for pre-pg_upgrade tuples, no need to sleep at all */
7674 61 : nmembers = HEAP_LOCKED_UPGRADED(infomask) ? -1 :
7675 61 : GetMultiXactIdMembers(multi, &members, false,
7676 61 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
7677 :
7678 61 : if (nmembers >= 0)
7679 : {
7680 : int i;
7681 :
7682 191 : for (i = 0; i < nmembers; i++)
7683 : {
7684 136 : TransactionId memxid = members[i].xid;
7685 136 : MultiXactStatus memstatus = members[i].status;
7686 :
7687 136 : if (TransactionIdIsCurrentTransactionId(memxid))
7688 : {
7689 25 : remain++;
7690 25 : continue;
7691 : }
7692 :
7693 111 : if (!DoLockModesConflict(LOCKMODE_from_mxstatus(memstatus),
7694 111 : LOCKMODE_from_mxstatus(status)))
7695 : {
7696 22 : if (remaining && TransactionIdIsInProgress(memxid))
7697 8 : remain++;
7698 22 : continue;
7699 : }
7700 :
7701 : /*
7702 : * This member conflicts with our multi, so we have to sleep (or
7703 : * return failure, if asked to avoid waiting.)
7704 : *
7705 : * Note that we don't set up an error context callback ourselves,
7706 : * but instead we pass the info down to XactLockTableWait. This
7707 : * might seem a bit wasteful because the context is set up and
7708 : * tore down for each member of the multixact, but in reality it
7709 : * should be barely noticeable, and it avoids duplicate code.
7710 : */
7711 89 : if (nowait)
7712 : {
7713 6 : result = ConditionalXactLockTableWait(memxid, logLockFailure);
7714 6 : if (!result)
7715 6 : break;
7716 : }
7717 : else
7718 83 : XactLockTableWait(memxid, rel, ctid, oper);
7719 : }
7720 :
7721 61 : pfree(members);
7722 : }
7723 :
7724 61 : if (remaining)
7725 10 : *remaining = remain;
7726 :
7727 61 : return result;
7728 : }
7729 :
7730 : /*
7731 : * MultiXactIdWait
7732 : * Sleep on a MultiXactId.
7733 : *
7734 : * By the time we finish sleeping, someone else may have changed the Xmax
7735 : * of the containing tuple, so the caller needs to iterate on us somehow.
7736 : *
7737 : * We return (in *remaining, if not NULL) the number of members that are still
7738 : * running, including any (non-aborted) subtransactions of our own transaction.
7739 : */
7740 : static void
7741 55 : MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
7742 : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
7743 : int *remaining)
7744 : {
7745 55 : (void) Do_MultiXactIdWait(multi, status, infomask, false,
7746 : rel, ctid, oper, remaining, false);
7747 55 : }
7748 :
7749 : /*
7750 : * ConditionalMultiXactIdWait
7751 : * As above, but only lock if we can get the lock without blocking.
7752 : *
7753 : * By the time we finish sleeping, someone else may have changed the Xmax
7754 : * of the containing tuple, so the caller needs to iterate on us somehow.
7755 : *
7756 : * If the multixact is now all gone, return true. Returns false if some
7757 : * transactions might still be running.
7758 : *
7759 : * We return (in *remaining, if not NULL) the number of members that are still
7760 : * running, including any (non-aborted) subtransactions of our own transaction.
7761 : */
7762 : static bool
7763 6 : ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status,
7764 : uint16 infomask, Relation rel, int *remaining,
7765 : bool logLockFailure)
7766 : {
7767 6 : return Do_MultiXactIdWait(multi, status, infomask, true,
7768 : rel, NULL, XLTW_None, remaining, logLockFailure);
7769 : }
7770 :
7771 : /*
7772 : * heap_tuple_needs_eventual_freeze
7773 : *
7774 : * Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
7775 : * will eventually require freezing (if tuple isn't removed by pruning first).
7776 : */
7777 : bool
7778 117727 : heap_tuple_needs_eventual_freeze(HeapTupleHeader tuple)
7779 : {
7780 : TransactionId xid;
7781 :
7782 : /*
7783 : * If xmin is a normal transaction ID, this tuple is definitely not
7784 : * frozen.
7785 : */
7786 117727 : xid = HeapTupleHeaderGetXmin(tuple);
7787 117727 : if (TransactionIdIsNormal(xid))
7788 2969 : return true;
7789 :
7790 : /*
7791 : * If xmax is a valid xact or multixact, this tuple is also not frozen.
7792 : */
7793 114758 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7794 : {
7795 : MultiXactId multi;
7796 :
7797 0 : multi = HeapTupleHeaderGetRawXmax(tuple);
7798 0 : if (MultiXactIdIsValid(multi))
7799 0 : return true;
7800 : }
7801 : else
7802 : {
7803 114758 : xid = HeapTupleHeaderGetRawXmax(tuple);
7804 114758 : if (TransactionIdIsNormal(xid))
7805 8 : return true;
7806 : }
7807 :
7808 114750 : if (tuple->t_infomask & HEAP_MOVED)
7809 : {
7810 0 : xid = HeapTupleHeaderGetXvac(tuple);
7811 0 : if (TransactionIdIsNormal(xid))
7812 0 : return true;
7813 : }
7814 :
7815 114750 : return false;
7816 : }
7817 :
7818 : /*
7819 : * heap_tuple_should_freeze
7820 : *
7821 : * Return value indicates if heap_prepare_freeze_tuple sibling function would
7822 : * (or should) force freezing of the heap page that contains caller's tuple.
7823 : * Tuple header XIDs/MXIDs < FreezeLimit/MultiXactCutoff trigger freezing.
7824 : * This includes (xmin, xmax, xvac) fields, as well as MultiXact member XIDs.
7825 : *
7826 : * The *NoFreezePageRelfrozenXid and *NoFreezePageRelminMxid input/output
7827 : * arguments help VACUUM track the oldest extant XID/MXID remaining in rel.
7828 : * Our working assumption is that caller won't decide to freeze this tuple.
7829 : * It's up to caller to only ratchet back its own top-level trackers after the
7830 : * point that it fully commits to not freezing the tuple/page in question.
7831 : */
7832 : bool
7833 6363291 : heap_tuple_should_freeze(HeapTupleHeader tuple,
7834 : const struct VacuumCutoffs *cutoffs,
7835 : TransactionId *NoFreezePageRelfrozenXid,
7836 : MultiXactId *NoFreezePageRelminMxid)
7837 : {
7838 : TransactionId xid;
7839 : MultiXactId multi;
7840 6363291 : bool freeze = false;
7841 :
7842 : /* First deal with xmin */
7843 6363291 : xid = HeapTupleHeaderGetXmin(tuple);
7844 6363291 : if (TransactionIdIsNormal(xid))
7845 : {
7846 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7847 2365074 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7848 26548 : *NoFreezePageRelfrozenXid = xid;
7849 2365074 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7850 23745 : freeze = true;
7851 : }
7852 :
7853 : /* Now deal with xmax */
7854 6363291 : xid = InvalidTransactionId;
7855 6363291 : multi = InvalidMultiXactId;
7856 6363291 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7857 2 : multi = HeapTupleHeaderGetRawXmax(tuple);
7858 : else
7859 6363289 : xid = HeapTupleHeaderGetRawXmax(tuple);
7860 :
7861 6363291 : if (TransactionIdIsNormal(xid))
7862 : {
7863 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7864 : /* xmax is a non-permanent XID */
7865 4285301 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7866 1 : *NoFreezePageRelfrozenXid = xid;
7867 4285301 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7868 8 : freeze = true;
7869 : }
7870 2077990 : else if (!MultiXactIdIsValid(multi))
7871 : {
7872 : /* xmax is a permanent XID or invalid MultiXactId/XID */
7873 : }
7874 2 : else if (HEAP_LOCKED_UPGRADED(tuple->t_infomask))
7875 : {
7876 : /* xmax is a pg_upgrade'd MultiXact, which can't have updater XID */
7877 0 : if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
7878 0 : *NoFreezePageRelminMxid = multi;
7879 : /* heap_prepare_freeze_tuple always freezes pg_upgrade'd xmax */
7880 0 : freeze = true;
7881 : }
7882 : else
7883 : {
7884 : /* xmax is a MultiXactId that may have an updater XID */
7885 : MultiXactMember *members;
7886 : int nmembers;
7887 :
7888 : Assert(MultiXactIdPrecedesOrEquals(cutoffs->relminmxid, multi));
7889 2 : if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
7890 2 : *NoFreezePageRelminMxid = multi;
7891 2 : if (MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff))
7892 2 : freeze = true;
7893 :
7894 : /* need to check whether any member of the mxact is old */
7895 2 : nmembers = GetMultiXactIdMembers(multi, &members, false,
7896 2 : HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask));
7897 :
7898 5 : for (int i = 0; i < nmembers; i++)
7899 : {
7900 3 : xid = members[i].xid;
7901 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7902 3 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7903 0 : *NoFreezePageRelfrozenXid = xid;
7904 3 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7905 0 : freeze = true;
7906 : }
7907 2 : if (nmembers > 0)
7908 1 : pfree(members);
7909 : }
7910 :
7911 6363291 : if (tuple->t_infomask & HEAP_MOVED)
7912 : {
7913 0 : xid = HeapTupleHeaderGetXvac(tuple);
7914 0 : if (TransactionIdIsNormal(xid))
7915 : {
7916 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7917 0 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7918 0 : *NoFreezePageRelfrozenXid = xid;
7919 : /* heap_prepare_freeze_tuple forces xvac freezing */
7920 0 : freeze = true;
7921 : }
7922 : }
7923 :
7924 6363291 : return freeze;
7925 : }
7926 :
7927 : /*
7928 : * Maintain snapshotConflictHorizon for caller by ratcheting forward its value
7929 : * using any committed XIDs contained in 'tuple', an obsolescent heap tuple
7930 : * that caller is in the process of physically removing, e.g. via HOT pruning
7931 : * or index deletion.
7932 : *
7933 : * Caller must initialize its value to InvalidTransactionId, which is
7934 : * generally interpreted as "definitely no need for a recovery conflict".
7935 : * Final value must reflect all heap tuples that caller will physically remove
7936 : * (or remove TID references to) via its ongoing pruning/deletion operation.
7937 : * ResolveRecoveryConflictWithSnapshot() is passed the final value (taken from
7938 : * caller's WAL record) by REDO routine when it replays caller's operation.
7939 : */
7940 : void
7941 3981350 : HeapTupleHeaderAdvanceConflictHorizon(HeapTupleHeader tuple,
7942 : TransactionId *snapshotConflictHorizon)
7943 : {
7944 3981350 : TransactionId xmin = HeapTupleHeaderGetXmin(tuple);
7945 3981350 : TransactionId xmax = HeapTupleHeaderGetUpdateXid(tuple);
7946 3981350 : TransactionId xvac = HeapTupleHeaderGetXvac(tuple);
7947 :
7948 3981350 : if (tuple->t_infomask & HEAP_MOVED)
7949 : {
7950 0 : if (TransactionIdPrecedes(*snapshotConflictHorizon, xvac))
7951 0 : *snapshotConflictHorizon = xvac;
7952 : }
7953 :
7954 : /*
7955 : * Ignore tuples inserted by an aborted transaction or if the tuple was
7956 : * updated/deleted by the inserting transaction.
7957 : *
7958 : * Look for a committed hint bit, or if no xmin bit is set, check clog.
7959 : */
7960 3981350 : if (HeapTupleHeaderXminCommitted(tuple) ||
7961 125631 : (!HeapTupleHeaderXminInvalid(tuple) && TransactionIdDidCommit(xmin)))
7962 : {
7963 7522041 : if (xmax != xmin &&
7964 3661520 : TransactionIdFollows(xmax, *snapshotConflictHorizon))
7965 130140 : *snapshotConflictHorizon = xmax;
7966 : }
7967 3981350 : }
7968 :
7969 : #ifdef USE_PREFETCH
7970 : /*
7971 : * Helper function for heap_index_delete_tuples. Issues prefetch requests for
7972 : * prefetch_count buffers. The prefetch_state keeps track of all the buffers
7973 : * we can prefetch, and which have already been prefetched; each call to this
7974 : * function picks up where the previous call left off.
7975 : *
7976 : * Note: we expect the deltids array to be sorted in an order that groups TIDs
7977 : * by heap block, with all TIDs for each block appearing together in exactly
7978 : * one group.
7979 : */
7980 : static void
7981 27184 : index_delete_prefetch_buffer(Relation rel,
7982 : IndexDeletePrefetchState *prefetch_state,
7983 : int prefetch_count)
7984 : {
7985 27184 : BlockNumber cur_hblkno = prefetch_state->cur_hblkno;
7986 27184 : int count = 0;
7987 : int i;
7988 27184 : int ndeltids = prefetch_state->ndeltids;
7989 27184 : TM_IndexDelete *deltids = prefetch_state->deltids;
7990 :
7991 27184 : for (i = prefetch_state->next_item;
7992 932371 : i < ndeltids && count < prefetch_count;
7993 905187 : i++)
7994 : {
7995 905187 : ItemPointer htid = &deltids[i].tid;
7996 :
7997 1802044 : if (cur_hblkno == InvalidBlockNumber ||
7998 896857 : ItemPointerGetBlockNumber(htid) != cur_hblkno)
7999 : {
8000 24370 : cur_hblkno = ItemPointerGetBlockNumber(htid);
8001 24370 : PrefetchBuffer(rel, MAIN_FORKNUM, cur_hblkno);
8002 24370 : count++;
8003 : }
8004 : }
8005 :
8006 : /*
8007 : * Save the prefetch position so that next time we can continue from that
8008 : * position.
8009 : */
8010 27184 : prefetch_state->next_item = i;
8011 27184 : prefetch_state->cur_hblkno = cur_hblkno;
8012 27184 : }
8013 : #endif
8014 :
8015 : /*
8016 : * Helper function for heap_index_delete_tuples. Checks for index corruption
8017 : * involving an invalid TID in index AM caller's index page.
8018 : *
8019 : * This is an ideal place for these checks. The index AM must hold a buffer
8020 : * lock on the index page containing the TIDs we examine here, so we don't
8021 : * have to worry about concurrent VACUUMs at all. We can be sure that the
8022 : * index is corrupt when htid points directly to an LP_UNUSED item or
8023 : * heap-only tuple, which is not the case during standard index scans.
8024 : */
8025 : static inline void
8026 750457 : index_delete_check_htid(TM_IndexDeleteOp *delstate,
8027 : Page page, OffsetNumber maxoff,
8028 : const ItemPointerData *htid, TM_IndexStatus *istatus)
8029 : {
8030 750457 : OffsetNumber indexpagehoffnum = ItemPointerGetOffsetNumber(htid);
8031 : ItemId iid;
8032 :
8033 : Assert(OffsetNumberIsValid(istatus->idxoffnum));
8034 :
8035 750457 : if (unlikely(indexpagehoffnum > maxoff))
8036 0 : ereport(ERROR,
8037 : (errcode(ERRCODE_INDEX_CORRUPTED),
8038 : 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\"",
8039 : ItemPointerGetBlockNumber(htid),
8040 : indexpagehoffnum,
8041 : istatus->idxoffnum, delstate->iblknum,
8042 : RelationGetRelationName(delstate->irel))));
8043 :
8044 750457 : iid = PageGetItemId(page, indexpagehoffnum);
8045 750457 : if (unlikely(!ItemIdIsUsed(iid)))
8046 0 : ereport(ERROR,
8047 : (errcode(ERRCODE_INDEX_CORRUPTED),
8048 : errmsg_internal("heap tid from index tuple (%u,%u) points to unused heap page item at offset %u of block %u in index \"%s\"",
8049 : ItemPointerGetBlockNumber(htid),
8050 : indexpagehoffnum,
8051 : istatus->idxoffnum, delstate->iblknum,
8052 : RelationGetRelationName(delstate->irel))));
8053 :
8054 750457 : if (ItemIdHasStorage(iid))
8055 : {
8056 : HeapTupleHeader htup;
8057 :
8058 : Assert(ItemIdIsNormal(iid));
8059 463493 : htup = (HeapTupleHeader) PageGetItem(page, iid);
8060 :
8061 463493 : if (unlikely(HeapTupleHeaderIsHeapOnly(htup)))
8062 0 : ereport(ERROR,
8063 : (errcode(ERRCODE_INDEX_CORRUPTED),
8064 : errmsg_internal("heap tid from index tuple (%u,%u) points to heap-only tuple at offset %u of block %u in index \"%s\"",
8065 : ItemPointerGetBlockNumber(htid),
8066 : indexpagehoffnum,
8067 : istatus->idxoffnum, delstate->iblknum,
8068 : RelationGetRelationName(delstate->irel))));
8069 : }
8070 750457 : }
8071 :
8072 : /*
8073 : * heapam implementation of tableam's index_delete_tuples interface.
8074 : *
8075 : * This helper function is called by index AMs during index tuple deletion.
8076 : * See tableam header comments for an explanation of the interface implemented
8077 : * here and a general theory of operation. Note that each call here is either
8078 : * a simple index deletion call, or a bottom-up index deletion call.
8079 : *
8080 : * It's possible for this to generate a fair amount of I/O, since we may be
8081 : * deleting hundreds of tuples from a single index block. To amortize that
8082 : * cost to some degree, this uses prefetching and combines repeat accesses to
8083 : * the same heap block.
8084 : */
8085 : TransactionId
8086 8330 : heap_index_delete_tuples(Relation rel, TM_IndexDeleteOp *delstate)
8087 : {
8088 : /* Initial assumption is that earlier pruning took care of conflict */
8089 8330 : TransactionId snapshotConflictHorizon = InvalidTransactionId;
8090 8330 : BlockNumber blkno = InvalidBlockNumber;
8091 8330 : Buffer buf = InvalidBuffer;
8092 8330 : Page page = NULL;
8093 8330 : OffsetNumber maxoff = InvalidOffsetNumber;
8094 : TransactionId priorXmax;
8095 : #ifdef USE_PREFETCH
8096 : IndexDeletePrefetchState prefetch_state;
8097 : int prefetch_distance;
8098 : #endif
8099 : SnapshotData SnapshotNonVacuumable;
8100 8330 : int finalndeltids = 0,
8101 8330 : nblocksaccessed = 0;
8102 :
8103 : /* State that's only used in bottom-up index deletion case */
8104 8330 : int nblocksfavorable = 0;
8105 8330 : int curtargetfreespace = delstate->bottomupfreespace,
8106 8330 : lastfreespace = 0,
8107 8330 : actualfreespace = 0;
8108 8330 : bool bottomup_final_block = false;
8109 :
8110 8330 : InitNonVacuumableSnapshot(SnapshotNonVacuumable, GlobalVisTestFor(rel));
8111 :
8112 : /* Sort caller's deltids array by TID for further processing */
8113 8330 : index_delete_sort(delstate);
8114 :
8115 : /*
8116 : * Bottom-up case: resort deltids array in an order attuned to where the
8117 : * greatest number of promising TIDs are to be found, and determine how
8118 : * many blocks from the start of sorted array should be considered
8119 : * favorable. This will also shrink the deltids array in order to
8120 : * eliminate completely unfavorable blocks up front.
8121 : */
8122 8330 : if (delstate->bottomup)
8123 2851 : nblocksfavorable = bottomup_sort_and_shrink(delstate);
8124 :
8125 : #ifdef USE_PREFETCH
8126 : /* Initialize prefetch state. */
8127 8330 : prefetch_state.cur_hblkno = InvalidBlockNumber;
8128 8330 : prefetch_state.next_item = 0;
8129 8330 : prefetch_state.ndeltids = delstate->ndeltids;
8130 8330 : prefetch_state.deltids = delstate->deltids;
8131 :
8132 : /*
8133 : * Determine the prefetch distance that we will attempt to maintain.
8134 : *
8135 : * Since the caller holds a buffer lock somewhere in rel, we'd better make
8136 : * sure that isn't a catalog relation before we call code that does
8137 : * syscache lookups, to avoid risk of deadlock.
8138 : */
8139 8330 : if (IsCatalogRelation(rel))
8140 6146 : prefetch_distance = maintenance_io_concurrency;
8141 : else
8142 : prefetch_distance =
8143 2184 : get_tablespace_maintenance_io_concurrency(rel->rd_rel->reltablespace);
8144 :
8145 : /* Cap initial prefetch distance for bottom-up deletion caller */
8146 8330 : if (delstate->bottomup)
8147 : {
8148 : Assert(nblocksfavorable >= 1);
8149 : Assert(nblocksfavorable <= BOTTOMUP_MAX_NBLOCKS);
8150 2851 : prefetch_distance = Min(prefetch_distance, nblocksfavorable);
8151 : }
8152 :
8153 : /* Start prefetching. */
8154 8330 : index_delete_prefetch_buffer(rel, &prefetch_state, prefetch_distance);
8155 : #endif
8156 :
8157 : /* Iterate over deltids, determine which to delete, check their horizon */
8158 : Assert(delstate->ndeltids > 0);
8159 758787 : for (int i = 0; i < delstate->ndeltids; i++)
8160 : {
8161 753308 : TM_IndexDelete *ideltid = &delstate->deltids[i];
8162 753308 : TM_IndexStatus *istatus = delstate->status + ideltid->id;
8163 753308 : ItemPointer htid = &ideltid->tid;
8164 : OffsetNumber offnum;
8165 :
8166 : /*
8167 : * Read buffer, and perform required extra steps each time a new block
8168 : * is encountered. Avoid refetching if it's the same block as the one
8169 : * from the last htid.
8170 : */
8171 1498286 : if (blkno == InvalidBlockNumber ||
8172 744978 : ItemPointerGetBlockNumber(htid) != blkno)
8173 : {
8174 : /*
8175 : * Consider giving up early for bottom-up index deletion caller
8176 : * first. (Only prefetch next-next block afterwards, when it
8177 : * becomes clear that we're at least going to access the next
8178 : * block in line.)
8179 : *
8180 : * Sometimes the first block frees so much space for bottom-up
8181 : * caller that the deletion process can end without accessing any
8182 : * more blocks. It is usually necessary to access 2 or 3 blocks
8183 : * per bottom-up deletion operation, though.
8184 : */
8185 21705 : if (delstate->bottomup)
8186 : {
8187 : /*
8188 : * We often allow caller to delete a few additional items
8189 : * whose entries we reached after the point that space target
8190 : * from caller was satisfied. The cost of accessing the page
8191 : * was already paid at that point, so it made sense to finish
8192 : * it off. When that happened, we finalize everything here
8193 : * (by finishing off the whole bottom-up deletion operation
8194 : * without needlessly paying the cost of accessing any more
8195 : * blocks).
8196 : */
8197 6163 : if (bottomup_final_block)
8198 165 : break;
8199 :
8200 : /*
8201 : * Give up when we didn't enable our caller to free any
8202 : * additional space as a result of processing the page that we
8203 : * just finished up with. This rule is the main way in which
8204 : * we keep the cost of bottom-up deletion under control.
8205 : */
8206 5998 : if (nblocksaccessed >= 1 && actualfreespace == lastfreespace)
8207 2686 : break;
8208 3312 : lastfreespace = actualfreespace; /* for next time */
8209 :
8210 : /*
8211 : * Deletion operation (which is bottom-up) will definitely
8212 : * access the next block in line. Prepare for that now.
8213 : *
8214 : * Decay target free space so that we don't hang on for too
8215 : * long with a marginal case. (Space target is only truly
8216 : * helpful when it allows us to recognize that we don't need
8217 : * to access more than 1 or 2 blocks to satisfy caller due to
8218 : * agreeable workload characteristics.)
8219 : *
8220 : * We are a bit more patient when we encounter contiguous
8221 : * blocks, though: these are treated as favorable blocks. The
8222 : * decay process is only applied when the next block in line
8223 : * is not a favorable/contiguous block. This is not an
8224 : * exception to the general rule; we still insist on finding
8225 : * at least one deletable item per block accessed. See
8226 : * bottomup_nblocksfavorable() for full details of the theory
8227 : * behind favorable blocks and heap block locality in general.
8228 : *
8229 : * Note: The first block in line is always treated as a
8230 : * favorable block, so the earliest possible point that the
8231 : * decay can be applied is just before we access the second
8232 : * block in line. The Assert() verifies this for us.
8233 : */
8234 : Assert(nblocksaccessed > 0 || nblocksfavorable > 0);
8235 3312 : if (nblocksfavorable > 0)
8236 3035 : nblocksfavorable--;
8237 : else
8238 277 : curtargetfreespace /= 2;
8239 : }
8240 :
8241 : /* release old buffer */
8242 18854 : if (BufferIsValid(buf))
8243 10524 : UnlockReleaseBuffer(buf);
8244 :
8245 18854 : blkno = ItemPointerGetBlockNumber(htid);
8246 18854 : buf = ReadBuffer(rel, blkno);
8247 18854 : nblocksaccessed++;
8248 : Assert(!delstate->bottomup ||
8249 : nblocksaccessed <= BOTTOMUP_MAX_NBLOCKS);
8250 :
8251 : #ifdef USE_PREFETCH
8252 :
8253 : /*
8254 : * To maintain the prefetch distance, prefetch one more page for
8255 : * each page we read.
8256 : */
8257 18854 : index_delete_prefetch_buffer(rel, &prefetch_state, 1);
8258 : #endif
8259 :
8260 18854 : LockBuffer(buf, BUFFER_LOCK_SHARE);
8261 :
8262 18854 : page = BufferGetPage(buf);
8263 18854 : maxoff = PageGetMaxOffsetNumber(page);
8264 : }
8265 :
8266 : /*
8267 : * In passing, detect index corruption involving an index page with a
8268 : * TID that points to a location in the heap that couldn't possibly be
8269 : * correct. We only do this with actual TIDs from caller's index page
8270 : * (not items reached by traversing through a HOT chain).
8271 : */
8272 750457 : index_delete_check_htid(delstate, page, maxoff, htid, istatus);
8273 :
8274 750457 : if (istatus->knowndeletable)
8275 : Assert(!delstate->bottomup && !istatus->promising);
8276 : else
8277 : {
8278 584624 : ItemPointerData tmp = *htid;
8279 : HeapTupleData heapTuple;
8280 :
8281 : /* Are any tuples from this HOT chain non-vacuumable? */
8282 584624 : if (heap_hot_search_buffer(&tmp, rel, buf, &SnapshotNonVacuumable,
8283 : &heapTuple, NULL, true))
8284 351085 : continue; /* can't delete entry */
8285 :
8286 : /* Caller will delete, since whole HOT chain is vacuumable */
8287 233539 : istatus->knowndeletable = true;
8288 :
8289 : /* Maintain index free space info for bottom-up deletion case */
8290 233539 : if (delstate->bottomup)
8291 : {
8292 : Assert(istatus->freespace > 0);
8293 11254 : actualfreespace += istatus->freespace;
8294 11254 : if (actualfreespace >= curtargetfreespace)
8295 2565 : bottomup_final_block = true;
8296 : }
8297 : }
8298 :
8299 : /*
8300 : * Maintain snapshotConflictHorizon value for deletion operation as a
8301 : * whole by advancing current value using heap tuple headers. This is
8302 : * loosely based on the logic for pruning a HOT chain.
8303 : */
8304 399372 : offnum = ItemPointerGetOffsetNumber(htid);
8305 399372 : priorXmax = InvalidTransactionId; /* cannot check first XMIN */
8306 : for (;;)
8307 23855 : {
8308 : ItemId lp;
8309 : HeapTupleHeader htup;
8310 :
8311 : /* Sanity check (pure paranoia) */
8312 423227 : if (offnum < FirstOffsetNumber)
8313 0 : break;
8314 :
8315 : /*
8316 : * An offset past the end of page's line pointer array is possible
8317 : * when the array was truncated
8318 : */
8319 423227 : if (offnum > maxoff)
8320 0 : break;
8321 :
8322 423227 : lp = PageGetItemId(page, offnum);
8323 423227 : if (ItemIdIsRedirected(lp))
8324 : {
8325 11523 : offnum = ItemIdGetRedirect(lp);
8326 11523 : continue;
8327 : }
8328 :
8329 : /*
8330 : * We'll often encounter LP_DEAD line pointers (especially with an
8331 : * entry marked knowndeletable by our caller up front). No heap
8332 : * tuple headers get examined for an htid that leads us to an
8333 : * LP_DEAD item. This is okay because the earlier pruning
8334 : * operation that made the line pointer LP_DEAD in the first place
8335 : * must have considered the original tuple header as part of
8336 : * generating its own snapshotConflictHorizon value.
8337 : *
8338 : * Relying on XLOG_HEAP2_PRUNE_VACUUM_SCAN records like this is
8339 : * the same strategy that index vacuuming uses in all cases. Index
8340 : * VACUUM WAL records don't even have a snapshotConflictHorizon
8341 : * field of their own for this reason.
8342 : */
8343 411704 : if (!ItemIdIsNormal(lp))
8344 260171 : break;
8345 :
8346 151533 : htup = (HeapTupleHeader) PageGetItem(page, lp);
8347 :
8348 : /*
8349 : * Check the tuple XMIN against prior XMAX, if any
8350 : */
8351 163865 : if (TransactionIdIsValid(priorXmax) &&
8352 12332 : !TransactionIdEquals(HeapTupleHeaderGetXmin(htup), priorXmax))
8353 0 : break;
8354 :
8355 151533 : HeapTupleHeaderAdvanceConflictHorizon(htup,
8356 : &snapshotConflictHorizon);
8357 :
8358 : /*
8359 : * If the tuple is not HOT-updated, then we are at the end of this
8360 : * HOT-chain. No need to visit later tuples from the same update
8361 : * chain (they get their own index entries) -- just move on to
8362 : * next htid from index AM caller.
8363 : */
8364 151533 : if (!HeapTupleHeaderIsHotUpdated(htup))
8365 139201 : break;
8366 :
8367 : /* Advance to next HOT chain member */
8368 : Assert(ItemPointerGetBlockNumber(&htup->t_ctid) == blkno);
8369 12332 : offnum = ItemPointerGetOffsetNumber(&htup->t_ctid);
8370 12332 : priorXmax = HeapTupleHeaderGetUpdateXid(htup);
8371 : }
8372 :
8373 : /* Enable further/final shrinking of deltids for caller */
8374 399372 : finalndeltids = i + 1;
8375 : }
8376 :
8377 8330 : UnlockReleaseBuffer(buf);
8378 :
8379 : /*
8380 : * Shrink deltids array to exclude non-deletable entries at the end. This
8381 : * is not just a minor optimization. Final deltids array size might be
8382 : * zero for a bottom-up caller. Index AM is explicitly allowed to rely on
8383 : * ndeltids being zero in all cases with zero total deletable entries.
8384 : */
8385 : Assert(finalndeltids > 0 || delstate->bottomup);
8386 8330 : delstate->ndeltids = finalndeltids;
8387 :
8388 8330 : return snapshotConflictHorizon;
8389 : }
8390 :
8391 : /*
8392 : * Specialized inlineable comparison function for index_delete_sort()
8393 : */
8394 : static inline int
8395 17907956 : index_delete_sort_cmp(TM_IndexDelete *deltid1, TM_IndexDelete *deltid2)
8396 : {
8397 17907956 : ItemPointer tid1 = &deltid1->tid;
8398 17907956 : ItemPointer tid2 = &deltid2->tid;
8399 :
8400 : {
8401 17907956 : BlockNumber blk1 = ItemPointerGetBlockNumber(tid1);
8402 17907956 : BlockNumber blk2 = ItemPointerGetBlockNumber(tid2);
8403 :
8404 17907956 : if (blk1 != blk2)
8405 7273735 : return (blk1 < blk2) ? -1 : 1;
8406 : }
8407 : {
8408 10634221 : OffsetNumber pos1 = ItemPointerGetOffsetNumber(tid1);
8409 10634221 : OffsetNumber pos2 = ItemPointerGetOffsetNumber(tid2);
8410 :
8411 10634221 : if (pos1 != pos2)
8412 10634221 : return (pos1 < pos2) ? -1 : 1;
8413 : }
8414 :
8415 : Assert(false);
8416 :
8417 0 : return 0;
8418 : }
8419 :
8420 : /*
8421 : * Sort deltids array from delstate by TID. This prepares it for further
8422 : * processing by heap_index_delete_tuples().
8423 : *
8424 : * This operation becomes a noticeable consumer of CPU cycles with some
8425 : * workloads, so we go to the trouble of specialization/micro optimization.
8426 : * We use shellsort for this because it's easy to specialize, compiles to
8427 : * relatively few instructions, and is adaptive to presorted inputs/subsets
8428 : * (which are typical here).
8429 : */
8430 : static void
8431 8330 : index_delete_sort(TM_IndexDeleteOp *delstate)
8432 : {
8433 8330 : TM_IndexDelete *deltids = delstate->deltids;
8434 8330 : int ndeltids = delstate->ndeltids;
8435 :
8436 : /*
8437 : * Shellsort gap sequence (taken from Sedgewick-Incerpi paper).
8438 : *
8439 : * This implementation is fast with array sizes up to ~4500. This covers
8440 : * all supported BLCKSZ values.
8441 : */
8442 8330 : const int gaps[9] = {1968, 861, 336, 112, 48, 21, 7, 3, 1};
8443 :
8444 : /* Think carefully before changing anything here -- keep swaps cheap */
8445 : StaticAssertDecl(sizeof(TM_IndexDelete) <= 8,
8446 : "element size exceeds 8 bytes");
8447 :
8448 83300 : for (int g = 0; g < lengthof(gaps); g++)
8449 : {
8450 10814386 : for (int hi = gaps[g], i = hi; i < ndeltids; i++)
8451 : {
8452 10739416 : TM_IndexDelete d = deltids[i];
8453 10739416 : int j = i;
8454 :
8455 18428850 : while (j >= hi && index_delete_sort_cmp(&deltids[j - hi], &d) >= 0)
8456 : {
8457 7689434 : deltids[j] = deltids[j - hi];
8458 7689434 : j -= hi;
8459 : }
8460 10739416 : deltids[j] = d;
8461 : }
8462 : }
8463 8330 : }
8464 :
8465 : /*
8466 : * Returns how many blocks should be considered favorable/contiguous for a
8467 : * bottom-up index deletion pass. This is a number of heap blocks that starts
8468 : * from and includes the first block in line.
8469 : *
8470 : * There is always at least one favorable block during bottom-up index
8471 : * deletion. In the worst case (i.e. with totally random heap blocks) the
8472 : * first block in line (the only favorable block) can be thought of as a
8473 : * degenerate array of contiguous blocks that consists of a single block.
8474 : * heap_index_delete_tuples() will expect this.
8475 : *
8476 : * Caller passes blockgroups, a description of the final order that deltids
8477 : * will be sorted in for heap_index_delete_tuples() bottom-up index deletion
8478 : * processing. Note that deltids need not actually be sorted just yet (caller
8479 : * only passes deltids to us so that we can interpret blockgroups).
8480 : *
8481 : * You might guess that the existence of contiguous blocks cannot matter much,
8482 : * since in general the main factor that determines which blocks we visit is
8483 : * the number of promising TIDs, which is a fixed hint from the index AM.
8484 : * We're not really targeting the general case, though -- the actual goal is
8485 : * to adapt our behavior to a wide variety of naturally occurring conditions.
8486 : * The effects of most of the heuristics we apply are only noticeable in the
8487 : * aggregate, over time and across many _related_ bottom-up index deletion
8488 : * passes.
8489 : *
8490 : * Deeming certain blocks favorable allows heapam to recognize and adapt to
8491 : * workloads where heap blocks visited during bottom-up index deletion can be
8492 : * accessed contiguously, in the sense that each newly visited block is the
8493 : * neighbor of the block that bottom-up deletion just finished processing (or
8494 : * close enough to it). It will likely be cheaper to access more favorable
8495 : * blocks sooner rather than later (e.g. in this pass, not across a series of
8496 : * related bottom-up passes). Either way it is probably only a matter of time
8497 : * (or a matter of further correlated version churn) before all blocks that
8498 : * appear together as a single large batch of favorable blocks get accessed by
8499 : * _some_ bottom-up pass. Large batches of favorable blocks tend to either
8500 : * appear almost constantly or not even once (it all depends on per-index
8501 : * workload characteristics).
8502 : *
8503 : * Note that the blockgroups sort order applies a power-of-two bucketing
8504 : * scheme that creates opportunities for contiguous groups of blocks to get
8505 : * batched together, at least with workloads that are naturally amenable to
8506 : * being driven by heap block locality. This doesn't just enhance the spatial
8507 : * locality of bottom-up heap block processing in the obvious way. It also
8508 : * enables temporal locality of access, since sorting by heap block number
8509 : * naturally tends to make the bottom-up processing order deterministic.
8510 : *
8511 : * Consider the following example to get a sense of how temporal locality
8512 : * might matter: There is a heap relation with several indexes, each of which
8513 : * is low to medium cardinality. It is subject to constant non-HOT updates.
8514 : * The updates are skewed (in one part of the primary key, perhaps). None of
8515 : * the indexes are logically modified by the UPDATE statements (if they were
8516 : * then bottom-up index deletion would not be triggered in the first place).
8517 : * Naturally, each new round of index tuples (for each heap tuple that gets a
8518 : * heap_update() call) will have the same heap TID in each and every index.
8519 : * Since these indexes are low cardinality and never get logically modified,
8520 : * heapam processing during bottom-up deletion passes will access heap blocks
8521 : * in approximately sequential order. Temporal locality of access occurs due
8522 : * to bottom-up deletion passes behaving very similarly across each of the
8523 : * indexes at any given moment. This keeps the number of buffer misses needed
8524 : * to visit heap blocks to a minimum.
8525 : */
8526 : static int
8527 2851 : bottomup_nblocksfavorable(IndexDeleteCounts *blockgroups, int nblockgroups,
8528 : TM_IndexDelete *deltids)
8529 : {
8530 2851 : int64 lastblock = -1;
8531 2851 : int nblocksfavorable = 0;
8532 :
8533 : Assert(nblockgroups >= 1);
8534 : Assert(nblockgroups <= BOTTOMUP_MAX_NBLOCKS);
8535 :
8536 : /*
8537 : * We tolerate heap blocks that will be accessed only slightly out of
8538 : * physical order. Small blips occur when a pair of almost-contiguous
8539 : * blocks happen to fall into different buckets (perhaps due only to a
8540 : * small difference in npromisingtids that the bucketing scheme didn't
8541 : * quite manage to ignore). We effectively ignore these blips by applying
8542 : * a small tolerance. The precise tolerance we use is a little arbitrary,
8543 : * but it works well enough in practice.
8544 : */
8545 8878 : for (int b = 0; b < nblockgroups; b++)
8546 : {
8547 8465 : IndexDeleteCounts *group = blockgroups + b;
8548 8465 : TM_IndexDelete *firstdtid = deltids + group->ifirsttid;
8549 8465 : BlockNumber block = ItemPointerGetBlockNumber(&firstdtid->tid);
8550 :
8551 8465 : if (lastblock != -1 &&
8552 5614 : ((int64) block < lastblock - BOTTOMUP_TOLERANCE_NBLOCKS ||
8553 4933 : (int64) block > lastblock + BOTTOMUP_TOLERANCE_NBLOCKS))
8554 : break;
8555 :
8556 6027 : nblocksfavorable++;
8557 6027 : lastblock = block;
8558 : }
8559 :
8560 : /* Always indicate that there is at least 1 favorable block */
8561 : Assert(nblocksfavorable >= 1);
8562 :
8563 2851 : return nblocksfavorable;
8564 : }
8565 :
8566 : /*
8567 : * qsort comparison function for bottomup_sort_and_shrink()
8568 : */
8569 : static int
8570 272569 : bottomup_sort_and_shrink_cmp(const void *arg1, const void *arg2)
8571 : {
8572 272569 : const IndexDeleteCounts *group1 = (const IndexDeleteCounts *) arg1;
8573 272569 : const IndexDeleteCounts *group2 = (const IndexDeleteCounts *) arg2;
8574 :
8575 : /*
8576 : * Most significant field is npromisingtids (which we invert the order of
8577 : * so as to sort in desc order).
8578 : *
8579 : * Caller should have already normalized npromisingtids fields into
8580 : * power-of-two values (buckets).
8581 : */
8582 272569 : if (group1->npromisingtids > group2->npromisingtids)
8583 12719 : return -1;
8584 259850 : if (group1->npromisingtids < group2->npromisingtids)
8585 16327 : return 1;
8586 :
8587 : /*
8588 : * Tiebreak: desc ntids sort order.
8589 : *
8590 : * We cannot expect power-of-two values for ntids fields. We should
8591 : * behave as if they were already rounded up for us instead.
8592 : */
8593 243523 : if (group1->ntids != group2->ntids)
8594 : {
8595 172977 : uint32 ntids1 = pg_nextpower2_32((uint32) group1->ntids);
8596 172977 : uint32 ntids2 = pg_nextpower2_32((uint32) group2->ntids);
8597 :
8598 172977 : if (ntids1 > ntids2)
8599 27185 : return -1;
8600 145792 : if (ntids1 < ntids2)
8601 31873 : return 1;
8602 : }
8603 :
8604 : /*
8605 : * Tiebreak: asc offset-into-deltids-for-block (offset to first TID for
8606 : * block in deltids array) order.
8607 : *
8608 : * This is equivalent to sorting in ascending heap block number order
8609 : * (among otherwise equal subsets of the array). This approach allows us
8610 : * to avoid accessing the out-of-line TID. (We rely on the assumption
8611 : * that the deltids array was sorted in ascending heap TID order when
8612 : * these offsets to the first TID from each heap block group were formed.)
8613 : */
8614 184465 : if (group1->ifirsttid > group2->ifirsttid)
8615 90050 : return 1;
8616 94415 : if (group1->ifirsttid < group2->ifirsttid)
8617 94415 : return -1;
8618 :
8619 0 : pg_unreachable();
8620 :
8621 : return 0;
8622 : }
8623 :
8624 : /*
8625 : * heap_index_delete_tuples() helper function for bottom-up deletion callers.
8626 : *
8627 : * Sorts deltids array in the order needed for useful processing by bottom-up
8628 : * deletion. The array should already be sorted in TID order when we're
8629 : * called. The sort process groups heap TIDs from deltids into heap block
8630 : * groupings. Earlier/more-promising groups/blocks are usually those that are
8631 : * known to have the most "promising" TIDs.
8632 : *
8633 : * Sets new size of deltids array (ndeltids) in state. deltids will only have
8634 : * TIDs from the BOTTOMUP_MAX_NBLOCKS most promising heap blocks when we
8635 : * return. This often means that deltids will be shrunk to a small fraction
8636 : * of its original size (we eliminate many heap blocks from consideration for
8637 : * caller up front).
8638 : *
8639 : * Returns the number of "favorable" blocks. See bottomup_nblocksfavorable()
8640 : * for a definition and full details.
8641 : */
8642 : static int
8643 2851 : bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate)
8644 : {
8645 : IndexDeleteCounts *blockgroups;
8646 : TM_IndexDelete *reordereddeltids;
8647 2851 : BlockNumber curblock = InvalidBlockNumber;
8648 2851 : int nblockgroups = 0;
8649 2851 : int ncopied = 0;
8650 2851 : int nblocksfavorable = 0;
8651 :
8652 : Assert(delstate->bottomup);
8653 : Assert(delstate->ndeltids > 0);
8654 :
8655 : /* Calculate per-heap-block count of TIDs */
8656 2851 : blockgroups = palloc_array(IndexDeleteCounts, delstate->ndeltids);
8657 1373390 : for (int i = 0; i < delstate->ndeltids; i++)
8658 : {
8659 1370539 : TM_IndexDelete *ideltid = &delstate->deltids[i];
8660 1370539 : TM_IndexStatus *istatus = delstate->status + ideltid->id;
8661 1370539 : ItemPointer htid = &ideltid->tid;
8662 1370539 : bool promising = istatus->promising;
8663 :
8664 1370539 : if (curblock != ItemPointerGetBlockNumber(htid))
8665 : {
8666 : /* New block group */
8667 53159 : nblockgroups++;
8668 :
8669 : Assert(curblock < ItemPointerGetBlockNumber(htid) ||
8670 : !BlockNumberIsValid(curblock));
8671 :
8672 53159 : curblock = ItemPointerGetBlockNumber(htid);
8673 53159 : blockgroups[nblockgroups - 1].ifirsttid = i;
8674 53159 : blockgroups[nblockgroups - 1].ntids = 1;
8675 53159 : blockgroups[nblockgroups - 1].npromisingtids = 0;
8676 : }
8677 : else
8678 : {
8679 1317380 : blockgroups[nblockgroups - 1].ntids++;
8680 : }
8681 :
8682 1370539 : if (promising)
8683 221387 : blockgroups[nblockgroups - 1].npromisingtids++;
8684 : }
8685 :
8686 : /*
8687 : * We're about ready to sort block groups to determine the optimal order
8688 : * for visiting heap blocks. But before we do, round the number of
8689 : * promising tuples for each block group up to the next power-of-two,
8690 : * unless it is very low (less than 4), in which case we round up to 4.
8691 : * npromisingtids is far too noisy to trust when choosing between a pair
8692 : * of block groups that both have very low values.
8693 : *
8694 : * This scheme divides heap blocks/block groups into buckets. Each bucket
8695 : * contains blocks that have _approximately_ the same number of promising
8696 : * TIDs as each other. The goal is to ignore relatively small differences
8697 : * in the total number of promising entries, so that the whole process can
8698 : * give a little weight to heapam factors (like heap block locality)
8699 : * instead. This isn't a trade-off, really -- we have nothing to lose. It
8700 : * would be foolish to interpret small differences in npromisingtids
8701 : * values as anything more than noise.
8702 : *
8703 : * We tiebreak on nhtids when sorting block group subsets that have the
8704 : * same npromisingtids, but this has the same issues as npromisingtids,
8705 : * and so nhtids is subject to the same power-of-two bucketing scheme. The
8706 : * only reason that we don't fix nhtids in the same way here too is that
8707 : * we'll need accurate nhtids values after the sort. We handle nhtids
8708 : * bucketization dynamically instead (in the sort comparator).
8709 : *
8710 : * See bottomup_nblocksfavorable() for a full explanation of when and how
8711 : * heap locality/favorable blocks can significantly influence when and how
8712 : * heap blocks are accessed.
8713 : */
8714 56010 : for (int b = 0; b < nblockgroups; b++)
8715 : {
8716 53159 : IndexDeleteCounts *group = blockgroups + b;
8717 :
8718 : /* Better off falling back on nhtids with low npromisingtids */
8719 53159 : if (group->npromisingtids <= 4)
8720 44850 : group->npromisingtids = 4;
8721 : else
8722 8309 : group->npromisingtids =
8723 8309 : pg_nextpower2_32((uint32) group->npromisingtids);
8724 : }
8725 :
8726 : /* Sort groups and rearrange caller's deltids array */
8727 2851 : qsort(blockgroups, nblockgroups, sizeof(IndexDeleteCounts),
8728 : bottomup_sort_and_shrink_cmp);
8729 2851 : reordereddeltids = palloc(delstate->ndeltids * sizeof(TM_IndexDelete));
8730 :
8731 2851 : nblockgroups = Min(BOTTOMUP_MAX_NBLOCKS, nblockgroups);
8732 : /* Determine number of favorable blocks at the start of final deltids */
8733 2851 : nblocksfavorable = bottomup_nblocksfavorable(blockgroups, nblockgroups,
8734 : delstate->deltids);
8735 :
8736 18620 : for (int b = 0; b < nblockgroups; b++)
8737 : {
8738 15769 : IndexDeleteCounts *group = blockgroups + b;
8739 15769 : TM_IndexDelete *firstdtid = delstate->deltids + group->ifirsttid;
8740 :
8741 15769 : memcpy(reordereddeltids + ncopied, firstdtid,
8742 15769 : sizeof(TM_IndexDelete) * group->ntids);
8743 15769 : ncopied += group->ntids;
8744 : }
8745 :
8746 : /* Copy final grouped and sorted TIDs back into start of caller's array */
8747 2851 : memcpy(delstate->deltids, reordereddeltids,
8748 : sizeof(TM_IndexDelete) * ncopied);
8749 2851 : delstate->ndeltids = ncopied;
8750 :
8751 2851 : pfree(reordereddeltids);
8752 2851 : pfree(blockgroups);
8753 :
8754 2851 : return nblocksfavorable;
8755 : }
8756 :
8757 : /*
8758 : * Perform XLogInsert for a heap-update operation. Caller must already
8759 : * have modified the buffer(s) and marked them dirty.
8760 : */
8761 : static XLogRecPtr
8762 2370519 : log_heap_update(Relation reln, Buffer oldbuf,
8763 : Buffer newbuf, HeapTuple oldtup, HeapTuple newtup,
8764 : HeapTuple old_key_tuple,
8765 : bool all_visible_cleared, bool new_all_visible_cleared,
8766 : bool walLogical)
8767 : {
8768 : xl_heap_update xlrec;
8769 : xl_heap_header xlhdr;
8770 : xl_heap_header xlhdr_idx;
8771 : uint8 info;
8772 : uint16 prefix_suffix[2];
8773 2370519 : uint16 prefixlen = 0,
8774 2370519 : suffixlen = 0;
8775 : XLogRecPtr recptr;
8776 2370519 : Page page = BufferGetPage(newbuf);
8777 2370519 : bool need_tuple_data = walLogical && RelationIsLogicallyLogged(reln);
8778 : bool init;
8779 : int bufflags;
8780 :
8781 : /* Caller should not call me on a non-WAL-logged relation */
8782 : Assert(RelationNeedsWAL(reln));
8783 :
8784 2370519 : XLogBeginInsert();
8785 :
8786 2370519 : if (HeapTupleIsHeapOnly(newtup))
8787 167158 : info = XLOG_HEAP_HOT_UPDATE;
8788 : else
8789 2203361 : info = XLOG_HEAP_UPDATE;
8790 :
8791 : /*
8792 : * If the old and new tuple are on the same page, we only need to log the
8793 : * parts of the new tuple that were changed. That saves on the amount of
8794 : * WAL we need to write. Currently, we just count any unchanged bytes in
8795 : * the beginning and end of the tuple. That's quick to check, and
8796 : * perfectly covers the common case that only one field is updated.
8797 : *
8798 : * We could do this even if the old and new tuple are on different pages,
8799 : * but only if we don't make a full-page image of the old page, which is
8800 : * difficult to know in advance. Also, if the old tuple is corrupt for
8801 : * some reason, it would allow the corruption to propagate the new page,
8802 : * so it seems best to avoid. Under the general assumption that most
8803 : * updates tend to create the new tuple version on the same page, there
8804 : * isn't much to be gained by doing this across pages anyway.
8805 : *
8806 : * Skip this if we're taking a full-page image of the new page, as we
8807 : * don't include the new tuple in the WAL record in that case. Also
8808 : * disable if effective_wal_level='logical', as logical decoding needs to
8809 : * be able to read the new tuple in whole from the WAL record alone.
8810 : */
8811 2370519 : if (oldbuf == newbuf && !need_tuple_data &&
8812 171763 : !XLogCheckBufferNeedsBackup(newbuf))
8813 : {
8814 171235 : char *oldp = (char *) oldtup->t_data + oldtup->t_data->t_hoff;
8815 171235 : char *newp = (char *) newtup->t_data + newtup->t_data->t_hoff;
8816 171235 : int oldlen = oldtup->t_len - oldtup->t_data->t_hoff;
8817 171235 : int newlen = newtup->t_len - newtup->t_data->t_hoff;
8818 :
8819 : /* Check for common prefix between old and new tuple */
8820 14582957 : for (prefixlen = 0; prefixlen < Min(oldlen, newlen); prefixlen++)
8821 : {
8822 14555140 : if (newp[prefixlen] != oldp[prefixlen])
8823 143418 : break;
8824 : }
8825 :
8826 : /*
8827 : * Storing the length of the prefix takes 2 bytes, so we need to save
8828 : * at least 3 bytes or there's no point.
8829 : */
8830 171235 : if (prefixlen < 3)
8831 22809 : prefixlen = 0;
8832 :
8833 : /* Same for suffix */
8834 6272855 : for (suffixlen = 0; suffixlen < Min(oldlen, newlen) - prefixlen; suffixlen++)
8835 : {
8836 6244708 : if (newp[newlen - suffixlen - 1] != oldp[oldlen - suffixlen - 1])
8837 143088 : break;
8838 : }
8839 171235 : if (suffixlen < 3)
8840 41765 : suffixlen = 0;
8841 : }
8842 :
8843 : /* Prepare main WAL data chain */
8844 2370519 : xlrec.flags = 0;
8845 2370519 : if (all_visible_cleared)
8846 2402 : xlrec.flags |= XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED;
8847 2370519 : if (new_all_visible_cleared)
8848 815 : xlrec.flags |= XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED;
8849 2370519 : if (prefixlen > 0)
8850 148426 : xlrec.flags |= XLH_UPDATE_PREFIX_FROM_OLD;
8851 2370519 : if (suffixlen > 0)
8852 129470 : xlrec.flags |= XLH_UPDATE_SUFFIX_FROM_OLD;
8853 2370519 : if (need_tuple_data)
8854 : {
8855 47045 : xlrec.flags |= XLH_UPDATE_CONTAINS_NEW_TUPLE;
8856 47045 : if (old_key_tuple)
8857 : {
8858 164 : if (reln->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
8859 68 : xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_TUPLE;
8860 : else
8861 96 : xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_KEY;
8862 : }
8863 : }
8864 :
8865 : /* If new tuple is the single and first tuple on page... */
8866 2383599 : if (ItemPointerGetOffsetNumber(&(newtup->t_self)) == FirstOffsetNumber &&
8867 13080 : PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
8868 : {
8869 12908 : info |= XLOG_HEAP_INIT_PAGE;
8870 12908 : init = true;
8871 : }
8872 : else
8873 2357611 : init = false;
8874 :
8875 : /* Prepare WAL data for the old page */
8876 2370519 : xlrec.old_offnum = ItemPointerGetOffsetNumber(&oldtup->t_self);
8877 2370519 : xlrec.old_xmax = HeapTupleHeaderGetRawXmax(oldtup->t_data);
8878 4741038 : xlrec.old_infobits_set = compute_infobits(oldtup->t_data->t_infomask,
8879 2370519 : oldtup->t_data->t_infomask2);
8880 :
8881 : /* Prepare WAL data for the new page */
8882 2370519 : xlrec.new_offnum = ItemPointerGetOffsetNumber(&newtup->t_self);
8883 2370519 : xlrec.new_xmax = HeapTupleHeaderGetRawXmax(newtup->t_data);
8884 :
8885 2370519 : bufflags = REGBUF_STANDARD;
8886 2370519 : if (init)
8887 12908 : bufflags |= REGBUF_WILL_INIT;
8888 2370519 : if (need_tuple_data)
8889 47045 : bufflags |= REGBUF_KEEP_DATA;
8890 :
8891 2370519 : XLogRegisterBuffer(0, newbuf, bufflags);
8892 2370519 : if (oldbuf != newbuf)
8893 2186793 : XLogRegisterBuffer(1, oldbuf, REGBUF_STANDARD);
8894 :
8895 2370519 : XLogRegisterData(&xlrec, SizeOfHeapUpdate);
8896 :
8897 : /*
8898 : * Prepare WAL data for the new tuple.
8899 : */
8900 2370519 : if (prefixlen > 0 || suffixlen > 0)
8901 : {
8902 170666 : if (prefixlen > 0 && suffixlen > 0)
8903 : {
8904 107230 : prefix_suffix[0] = prefixlen;
8905 107230 : prefix_suffix[1] = suffixlen;
8906 107230 : XLogRegisterBufData(0, &prefix_suffix, sizeof(uint16) * 2);
8907 : }
8908 63436 : else if (prefixlen > 0)
8909 : {
8910 41196 : XLogRegisterBufData(0, &prefixlen, sizeof(uint16));
8911 : }
8912 : else
8913 : {
8914 22240 : XLogRegisterBufData(0, &suffixlen, sizeof(uint16));
8915 : }
8916 : }
8917 :
8918 2370519 : xlhdr.t_infomask2 = newtup->t_data->t_infomask2;
8919 2370519 : xlhdr.t_infomask = newtup->t_data->t_infomask;
8920 2370519 : xlhdr.t_hoff = newtup->t_data->t_hoff;
8921 : Assert(SizeofHeapTupleHeader + prefixlen + suffixlen <= newtup->t_len);
8922 :
8923 : /*
8924 : * PG73FORMAT: write bitmap [+ padding] [+ oid] + data
8925 : *
8926 : * The 'data' doesn't include the common prefix or suffix.
8927 : */
8928 2370519 : XLogRegisterBufData(0, &xlhdr, SizeOfHeapHeader);
8929 2370519 : if (prefixlen == 0)
8930 : {
8931 2222093 : XLogRegisterBufData(0,
8932 2222093 : (char *) newtup->t_data + SizeofHeapTupleHeader,
8933 2222093 : newtup->t_len - SizeofHeapTupleHeader - suffixlen);
8934 : }
8935 : else
8936 : {
8937 : /*
8938 : * Have to write the null bitmap and data after the common prefix as
8939 : * two separate rdata entries.
8940 : */
8941 : /* bitmap [+ padding] [+ oid] */
8942 148426 : if (newtup->t_data->t_hoff - SizeofHeapTupleHeader > 0)
8943 : {
8944 148426 : XLogRegisterBufData(0,
8945 148426 : (char *) newtup->t_data + SizeofHeapTupleHeader,
8946 148426 : newtup->t_data->t_hoff - SizeofHeapTupleHeader);
8947 : }
8948 :
8949 : /* data after common prefix */
8950 148426 : XLogRegisterBufData(0,
8951 148426 : (char *) newtup->t_data + newtup->t_data->t_hoff + prefixlen,
8952 148426 : newtup->t_len - newtup->t_data->t_hoff - prefixlen - suffixlen);
8953 : }
8954 :
8955 : /* We need to log a tuple identity */
8956 2370519 : if (need_tuple_data && old_key_tuple)
8957 : {
8958 : /* don't really need this, but its more comfy to decode */
8959 164 : xlhdr_idx.t_infomask2 = old_key_tuple->t_data->t_infomask2;
8960 164 : xlhdr_idx.t_infomask = old_key_tuple->t_data->t_infomask;
8961 164 : xlhdr_idx.t_hoff = old_key_tuple->t_data->t_hoff;
8962 :
8963 164 : XLogRegisterData(&xlhdr_idx, SizeOfHeapHeader);
8964 :
8965 : /* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
8966 164 : XLogRegisterData((char *) old_key_tuple->t_data + SizeofHeapTupleHeader,
8967 164 : old_key_tuple->t_len - SizeofHeapTupleHeader);
8968 : }
8969 :
8970 : /* filtering by origin on a row level is much more efficient */
8971 2370519 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
8972 :
8973 2370519 : recptr = XLogInsert(RM_HEAP_ID, info);
8974 :
8975 2370519 : return recptr;
8976 : }
8977 :
8978 : /*
8979 : * Perform XLogInsert of an XLOG_HEAP2_NEW_CID record
8980 : *
8981 : * This is only used when effective_wal_level is logical, and only for
8982 : * catalog tuples.
8983 : */
8984 : static XLogRecPtr
8985 32654 : log_heap_new_cid(Relation relation, HeapTuple tup)
8986 : {
8987 : xl_heap_new_cid xlrec;
8988 :
8989 : XLogRecPtr recptr;
8990 32654 : HeapTupleHeader hdr = tup->t_data;
8991 :
8992 : Assert(ItemPointerIsValid(&tup->t_self));
8993 : Assert(tup->t_tableOid != InvalidOid);
8994 :
8995 32654 : xlrec.top_xid = GetTopTransactionId();
8996 32654 : xlrec.target_locator = relation->rd_locator;
8997 32654 : xlrec.target_tid = tup->t_self;
8998 :
8999 : /*
9000 : * If the tuple got inserted & deleted in the same TX we definitely have a
9001 : * combo CID, set cmin and cmax.
9002 : */
9003 32654 : if (hdr->t_infomask & HEAP_COMBOCID)
9004 : {
9005 : Assert(!(hdr->t_infomask & HEAP_XMAX_INVALID));
9006 : Assert(!HeapTupleHeaderXminInvalid(hdr));
9007 2219 : xlrec.cmin = HeapTupleHeaderGetCmin(hdr);
9008 2219 : xlrec.cmax = HeapTupleHeaderGetCmax(hdr);
9009 2219 : xlrec.combocid = HeapTupleHeaderGetRawCommandId(hdr);
9010 : }
9011 : /* No combo CID, so only cmin or cmax can be set by this TX */
9012 : else
9013 : {
9014 : /*
9015 : * Tuple inserted.
9016 : *
9017 : * We need to check for LOCK ONLY because multixacts might be
9018 : * transferred to the new tuple in case of FOR KEY SHARE updates in
9019 : * which case there will be an xmax, although the tuple just got
9020 : * inserted.
9021 : */
9022 41067 : if (hdr->t_infomask & HEAP_XMAX_INVALID ||
9023 10632 : HEAP_XMAX_IS_LOCKED_ONLY(hdr->t_infomask))
9024 : {
9025 19804 : xlrec.cmin = HeapTupleHeaderGetRawCommandId(hdr);
9026 19804 : xlrec.cmax = InvalidCommandId;
9027 : }
9028 : /* Tuple from a different tx updated or deleted. */
9029 : else
9030 : {
9031 10631 : xlrec.cmin = InvalidCommandId;
9032 10631 : xlrec.cmax = HeapTupleHeaderGetRawCommandId(hdr);
9033 : }
9034 30435 : xlrec.combocid = InvalidCommandId;
9035 : }
9036 :
9037 : /*
9038 : * Note that we don't need to register the buffer here, because this
9039 : * operation does not modify the page. The insert/update/delete that
9040 : * called us certainly did, but that's WAL-logged separately.
9041 : */
9042 32654 : XLogBeginInsert();
9043 32654 : XLogRegisterData(&xlrec, SizeOfHeapNewCid);
9044 :
9045 : /* will be looked at irrespective of origin */
9046 :
9047 32654 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_NEW_CID);
9048 :
9049 32654 : return recptr;
9050 : }
9051 :
9052 : /*
9053 : * Build a heap tuple representing the configured REPLICA IDENTITY to represent
9054 : * the old tuple in an UPDATE or DELETE.
9055 : *
9056 : * Returns NULL if there's no need to log an identity or if there's no suitable
9057 : * key defined.
9058 : *
9059 : * Pass key_required true if any replica identity columns changed value, or if
9060 : * any of them have any external data. Delete must always pass true.
9061 : *
9062 : * *copy is set to true if the returned tuple is a modified copy rather than
9063 : * the same tuple that was passed in.
9064 : */
9065 : static HeapTuple
9066 4262598 : ExtractReplicaIdentity(Relation relation, HeapTuple tp, bool key_required,
9067 : bool *copy)
9068 : {
9069 4262598 : TupleDesc desc = RelationGetDescr(relation);
9070 4262598 : char replident = relation->rd_rel->relreplident;
9071 : Bitmapset *idattrs;
9072 : HeapTuple key_tuple;
9073 : bool nulls[MaxHeapAttributeNumber];
9074 : Datum values[MaxHeapAttributeNumber];
9075 :
9076 4262598 : *copy = false;
9077 :
9078 4262598 : if (!RelationIsLogicallyLogged(relation))
9079 4162257 : return NULL;
9080 :
9081 100341 : if (replident == REPLICA_IDENTITY_NOTHING)
9082 236 : return NULL;
9083 :
9084 100105 : if (replident == REPLICA_IDENTITY_FULL)
9085 : {
9086 : /*
9087 : * When logging the entire old tuple, it very well could contain
9088 : * toasted columns. If so, force them to be inlined.
9089 : */
9090 203 : if (HeapTupleHasExternal(tp))
9091 : {
9092 4 : *copy = true;
9093 4 : tp = toast_flatten_tuple(tp, desc);
9094 : }
9095 203 : return tp;
9096 : }
9097 :
9098 : /* if the key isn't required and we're only logging the key, we're done */
9099 99902 : if (!key_required)
9100 46891 : return NULL;
9101 :
9102 : /* find out the replica identity columns */
9103 53011 : idattrs = RelationGetIndexAttrBitmap(relation,
9104 : INDEX_ATTR_BITMAP_IDENTITY_KEY);
9105 :
9106 : /*
9107 : * If there's no defined replica identity columns, treat as !key_required.
9108 : * (This case should not be reachable from heap_update, since that should
9109 : * calculate key_required accurately. But heap_delete just passes
9110 : * constant true for key_required, so we can hit this case in deletes.)
9111 : */
9112 53011 : if (bms_is_empty(idattrs))
9113 6021 : return NULL;
9114 :
9115 : /*
9116 : * Construct a new tuple containing only the replica identity columns,
9117 : * with nulls elsewhere. While we're at it, assert that the replica
9118 : * identity columns aren't null.
9119 : */
9120 46990 : heap_deform_tuple(tp, desc, values, nulls);
9121 :
9122 150970 : for (int i = 0; i < desc->natts; i++)
9123 : {
9124 103980 : if (bms_is_member(i + 1 - FirstLowInvalidHeapAttributeNumber,
9125 : idattrs))
9126 : Assert(!nulls[i]);
9127 : else
9128 56970 : nulls[i] = true;
9129 : }
9130 :
9131 46990 : key_tuple = heap_form_tuple(desc, values, nulls);
9132 46990 : *copy = true;
9133 :
9134 46990 : bms_free(idattrs);
9135 :
9136 : /*
9137 : * If the tuple, which by here only contains indexed columns, still has
9138 : * toasted columns, force them to be inlined. This is somewhat unlikely
9139 : * since there's limits on the size of indexed columns, so we don't
9140 : * duplicate toast_flatten_tuple()s functionality in the above loop over
9141 : * the indexed columns, even if it would be more efficient.
9142 : */
9143 46990 : if (HeapTupleHasExternal(key_tuple))
9144 : {
9145 4 : HeapTuple oldtup = key_tuple;
9146 :
9147 4 : key_tuple = toast_flatten_tuple(oldtup, desc);
9148 4 : heap_freetuple(oldtup);
9149 : }
9150 :
9151 46990 : return key_tuple;
9152 : }
9153 :
9154 : /*
9155 : * HeapCheckForSerializableConflictOut
9156 : * We are reading a tuple. If it's not visible, there may be a
9157 : * rw-conflict out with the inserter. Otherwise, if it is visible to us
9158 : * but has been deleted, there may be a rw-conflict out with the deleter.
9159 : *
9160 : * We will determine the top level xid of the writing transaction with which
9161 : * we may be in conflict, and ask CheckForSerializableConflictOut() to check
9162 : * for overlap with our own transaction.
9163 : *
9164 : * This function should be called just about anywhere in heapam.c where a
9165 : * tuple has been read. The caller must hold at least a shared lock on the
9166 : * buffer, because this function might set hint bits on the tuple. There is
9167 : * currently no known reason to call this function from an index AM.
9168 : */
9169 : void
9170 42082912 : HeapCheckForSerializableConflictOut(bool visible, Relation relation,
9171 : HeapTuple tuple, Buffer buffer,
9172 : Snapshot snapshot)
9173 : {
9174 : TransactionId xid;
9175 : HTSV_Result htsvResult;
9176 :
9177 42082912 : if (!CheckForSerializableConflictOutNeeded(relation, snapshot))
9178 42074967 : return;
9179 :
9180 : /*
9181 : * Check to see whether the tuple has been written to by a concurrent
9182 : * transaction, either to create it not visible to us, or to delete it
9183 : * while it is visible to us. The "visible" bool indicates whether the
9184 : * tuple is visible to us, while HeapTupleSatisfiesVacuum checks what else
9185 : * is going on with it.
9186 : *
9187 : * In the event of a concurrently inserted tuple that also happens to have
9188 : * been concurrently updated (by a separate transaction), the xmin of the
9189 : * tuple will be used -- not the updater's xid.
9190 : */
9191 7945 : htsvResult = HeapTupleSatisfiesVacuum(tuple, TransactionXmin, buffer);
9192 7945 : switch (htsvResult)
9193 : {
9194 7137 : case HEAPTUPLE_LIVE:
9195 7137 : if (visible)
9196 7124 : return;
9197 13 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9198 13 : break;
9199 361 : case HEAPTUPLE_RECENTLY_DEAD:
9200 : case HEAPTUPLE_DELETE_IN_PROGRESS:
9201 361 : if (visible)
9202 286 : xid = HeapTupleHeaderGetUpdateXid(tuple->t_data);
9203 : else
9204 75 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9205 :
9206 361 : if (TransactionIdPrecedes(xid, TransactionXmin))
9207 : {
9208 : /* This is like the HEAPTUPLE_DEAD case */
9209 : Assert(!visible);
9210 71 : return;
9211 : }
9212 290 : break;
9213 327 : case HEAPTUPLE_INSERT_IN_PROGRESS:
9214 327 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9215 327 : break;
9216 120 : case HEAPTUPLE_DEAD:
9217 : Assert(!visible);
9218 120 : return;
9219 0 : default:
9220 :
9221 : /*
9222 : * The only way to get to this default clause is if a new value is
9223 : * added to the enum type without adding it to this switch
9224 : * statement. That's a bug, so elog.
9225 : */
9226 0 : elog(ERROR, "unrecognized return value from HeapTupleSatisfiesVacuum: %u", htsvResult);
9227 :
9228 : /*
9229 : * In spite of having all enum values covered and calling elog on
9230 : * this default, some compilers think this is a code path which
9231 : * allows xid to be used below without initialization. Silence
9232 : * that warning.
9233 : */
9234 : xid = InvalidTransactionId;
9235 : }
9236 :
9237 : Assert(TransactionIdIsValid(xid));
9238 : Assert(TransactionIdFollowsOrEquals(xid, TransactionXmin));
9239 :
9240 : /*
9241 : * Find top level xid. Bail out if xid is too early to be a conflict, or
9242 : * if it's our own xid.
9243 : */
9244 630 : if (TransactionIdEquals(xid, GetTopTransactionIdIfAny()))
9245 64 : return;
9246 566 : xid = SubTransGetTopmostTransaction(xid);
9247 566 : if (TransactionIdPrecedes(xid, TransactionXmin))
9248 0 : return;
9249 :
9250 566 : CheckForSerializableConflictOut(relation, xid, snapshot);
9251 : }
|
