Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * heapam.c
4 : * heap access method code
5 : *
6 : * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/access/heap/heapam.c
12 : *
13 : *
14 : * INTERFACE ROUTINES
15 : * heap_beginscan - begin relation scan
16 : * heap_rescan - restart a relation scan
17 : * heap_endscan - end relation scan
18 : * heap_getnext - retrieve next tuple in scan
19 : * heap_fetch - retrieve tuple with given tid
20 : * heap_insert - insert tuple into a relation
21 : * heap_multi_insert - insert multiple tuples into a relation
22 : * heap_delete - delete a tuple from a relation
23 : * heap_update - replace a tuple in a relation with another tuple
24 : *
25 : * NOTES
26 : * This file contains the heap_ routines which implement
27 : * the POSTGRES heap access method used for all POSTGRES
28 : * relations.
29 : *
30 : *-------------------------------------------------------------------------
31 : */
32 : #include "postgres.h"
33 :
34 : #include "access/heapam.h"
35 : #include "access/heaptoast.h"
36 : #include "access/hio.h"
37 : #include "access/multixact.h"
38 : #include "access/subtrans.h"
39 : #include "access/syncscan.h"
40 : #include "access/valid.h"
41 : #include "access/visibilitymap.h"
42 : #include "access/xloginsert.h"
43 : #include "catalog/pg_database.h"
44 : #include "catalog/pg_database_d.h"
45 : #include "commands/vacuum.h"
46 : #include "pgstat.h"
47 : #include "port/pg_bitutils.h"
48 : #include "storage/lmgr.h"
49 : #include "storage/predicate.h"
50 : #include "storage/procarray.h"
51 : #include "utils/datum.h"
52 : #include "utils/injection_point.h"
53 : #include "utils/inval.h"
54 : #include "utils/spccache.h"
55 : #include "utils/syscache.h"
56 :
57 :
58 : static HeapTuple heap_prepare_insert(Relation relation, HeapTuple tup,
59 : TransactionId xid, CommandId cid, int options);
60 : static XLogRecPtr log_heap_update(Relation reln, Buffer oldbuf,
61 : Buffer newbuf, HeapTuple oldtup,
62 : HeapTuple newtup, HeapTuple old_key_tuple,
63 : bool all_visible_cleared, bool new_all_visible_cleared);
64 : #ifdef USE_ASSERT_CHECKING
65 : static void check_lock_if_inplace_updateable_rel(Relation relation,
66 : ItemPointer otid,
67 : HeapTuple newtup);
68 : static void check_inplace_rel_lock(HeapTuple oldtup);
69 : #endif
70 : static Bitmapset *HeapDetermineColumnsInfo(Relation relation,
71 : Bitmapset *interesting_cols,
72 : Bitmapset *external_cols,
73 : HeapTuple oldtup, HeapTuple newtup,
74 : bool *has_external);
75 : static bool heap_acquire_tuplock(Relation relation, ItemPointer tid,
76 : LockTupleMode mode, LockWaitPolicy wait_policy,
77 : bool *have_tuple_lock);
78 : static inline BlockNumber heapgettup_advance_block(HeapScanDesc scan,
79 : BlockNumber block,
80 : ScanDirection dir);
81 : static pg_noinline BlockNumber heapgettup_initial_block(HeapScanDesc scan,
82 : ScanDirection dir);
83 : static void compute_new_xmax_infomask(TransactionId xmax, uint16 old_infomask,
84 : uint16 old_infomask2, TransactionId add_to_xmax,
85 : LockTupleMode mode, bool is_update,
86 : TransactionId *result_xmax, uint16 *result_infomask,
87 : uint16 *result_infomask2);
88 : static TM_Result heap_lock_updated_tuple(Relation rel, HeapTuple tuple,
89 : ItemPointer ctid, TransactionId xid,
90 : LockTupleMode mode);
91 : static void GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
92 : uint16 *new_infomask2);
93 : static TransactionId MultiXactIdGetUpdateXid(TransactionId xmax,
94 : uint16 t_infomask);
95 : static bool DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
96 : LockTupleMode lockmode, bool *current_is_member);
97 : static void MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
98 : Relation rel, ItemPointer ctid, XLTW_Oper oper,
99 : int *remaining);
100 : static bool ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status,
101 : uint16 infomask, Relation rel, int *remaining,
102 : bool logLockFailure);
103 : static void index_delete_sort(TM_IndexDeleteOp *delstate);
104 : static int bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate);
105 : static XLogRecPtr log_heap_new_cid(Relation relation, HeapTuple tup);
106 : static HeapTuple ExtractReplicaIdentity(Relation relation, HeapTuple tp, bool key_required,
107 : bool *copy);
108 :
109 :
110 : /*
111 : * Each tuple lock mode has a corresponding heavyweight lock, and one or two
112 : * corresponding MultiXactStatuses (one to merely lock tuples, another one to
113 : * update them). This table (and the macros below) helps us determine the
114 : * heavyweight lock mode and MultiXactStatus values to use for any particular
115 : * tuple lock strength.
116 : *
117 : * These interact with InplaceUpdateTupleLock, an alias for ExclusiveLock.
118 : *
119 : * Don't look at lockstatus/updstatus directly! Use get_mxact_status_for_lock
120 : * instead.
121 : */
122 : static const struct
123 : {
124 : LOCKMODE hwlock;
125 : int lockstatus;
126 : int updstatus;
127 : }
128 :
129 : tupleLockExtraInfo[MaxLockTupleMode + 1] =
130 : {
131 : { /* LockTupleKeyShare */
132 : AccessShareLock,
133 : MultiXactStatusForKeyShare,
134 : -1 /* KeyShare does not allow updating tuples */
135 : },
136 : { /* LockTupleShare */
137 : RowShareLock,
138 : MultiXactStatusForShare,
139 : -1 /* Share does not allow updating tuples */
140 : },
141 : { /* LockTupleNoKeyExclusive */
142 : ExclusiveLock,
143 : MultiXactStatusForNoKeyUpdate,
144 : MultiXactStatusNoKeyUpdate
145 : },
146 : { /* LockTupleExclusive */
147 : AccessExclusiveLock,
148 : MultiXactStatusForUpdate,
149 : MultiXactStatusUpdate
150 : }
151 : };
152 :
153 : /* Get the LOCKMODE for a given MultiXactStatus */
154 : #define LOCKMODE_from_mxstatus(status) \
155 : (tupleLockExtraInfo[TUPLOCK_from_mxstatus((status))].hwlock)
156 :
157 : /*
158 : * Acquire heavyweight locks on tuples, using a LockTupleMode strength value.
159 : * This is more readable than having every caller translate it to lock.h's
160 : * LOCKMODE.
161 : */
162 : #define LockTupleTuplock(rel, tup, mode) \
163 : LockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
164 : #define UnlockTupleTuplock(rel, tup, mode) \
165 : UnlockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
166 : #define ConditionalLockTupleTuplock(rel, tup, mode, log) \
167 : ConditionalLockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock, (log))
168 :
169 : #ifdef USE_PREFETCH
170 : /*
171 : * heap_index_delete_tuples and index_delete_prefetch_buffer use this
172 : * structure to coordinate prefetching activity
173 : */
174 : typedef struct
175 : {
176 : BlockNumber cur_hblkno;
177 : int next_item;
178 : int ndeltids;
179 : TM_IndexDelete *deltids;
180 : } IndexDeletePrefetchState;
181 : #endif
182 :
183 : /* heap_index_delete_tuples bottom-up index deletion costing constants */
184 : #define BOTTOMUP_MAX_NBLOCKS 6
185 : #define BOTTOMUP_TOLERANCE_NBLOCKS 3
186 :
187 : /*
188 : * heap_index_delete_tuples uses this when determining which heap blocks it
189 : * must visit to help its bottom-up index deletion caller
190 : */
191 : typedef struct IndexDeleteCounts
192 : {
193 : int16 npromisingtids; /* Number of "promising" TIDs in group */
194 : int16 ntids; /* Number of TIDs in group */
195 : int16 ifirsttid; /* Offset to group's first deltid */
196 : } IndexDeleteCounts;
197 :
198 : /*
199 : * This table maps tuple lock strength values for each particular
200 : * MultiXactStatus value.
201 : */
202 : static const int MultiXactStatusLock[MaxMultiXactStatus + 1] =
203 : {
204 : LockTupleKeyShare, /* ForKeyShare */
205 : LockTupleShare, /* ForShare */
206 : LockTupleNoKeyExclusive, /* ForNoKeyUpdate */
207 : LockTupleExclusive, /* ForUpdate */
208 : LockTupleNoKeyExclusive, /* NoKeyUpdate */
209 : LockTupleExclusive /* Update */
210 : };
211 :
212 : /* Get the LockTupleMode for a given MultiXactStatus */
213 : #define TUPLOCK_from_mxstatus(status) \
214 : (MultiXactStatusLock[(status)])
215 :
216 : /* ----------------------------------------------------------------
217 : * heap support routines
218 : * ----------------------------------------------------------------
219 : */
220 :
221 : /*
222 : * Streaming read API callback for parallel sequential scans. Returns the next
223 : * block the caller wants from the read stream or InvalidBlockNumber when done.
224 : */
225 : static BlockNumber
226 201416 : heap_scan_stream_read_next_parallel(ReadStream *stream,
227 : void *callback_private_data,
228 : void *per_buffer_data)
229 : {
230 201416 : HeapScanDesc scan = (HeapScanDesc) callback_private_data;
231 :
232 : Assert(ScanDirectionIsForward(scan->rs_dir));
233 : Assert(scan->rs_base.rs_parallel);
234 :
235 201416 : if (unlikely(!scan->rs_inited))
236 : {
237 : /* parallel scan */
238 2836 : table_block_parallelscan_startblock_init(scan->rs_base.rs_rd,
239 2836 : scan->rs_parallelworkerdata,
240 2836 : (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel);
241 :
242 : /* may return InvalidBlockNumber if there are no more blocks */
243 5672 : scan->rs_prefetch_block = table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
244 2836 : scan->rs_parallelworkerdata,
245 2836 : (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel);
246 2836 : scan->rs_inited = true;
247 : }
248 : else
249 : {
250 198580 : scan->rs_prefetch_block = table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
251 198580 : scan->rs_parallelworkerdata, (ParallelBlockTableScanDesc)
252 198580 : scan->rs_base.rs_parallel);
253 : }
254 :
255 201416 : return scan->rs_prefetch_block;
256 : }
257 :
258 : /*
259 : * Streaming read API callback for serial sequential and TID range scans.
260 : * Returns the next block the caller wants from the read stream or
261 : * InvalidBlockNumber when done.
262 : */
263 : static BlockNumber
264 7066924 : heap_scan_stream_read_next_serial(ReadStream *stream,
265 : void *callback_private_data,
266 : void *per_buffer_data)
267 : {
268 7066924 : HeapScanDesc scan = (HeapScanDesc) callback_private_data;
269 :
270 7066924 : if (unlikely(!scan->rs_inited))
271 : {
272 1839116 : scan->rs_prefetch_block = heapgettup_initial_block(scan, scan->rs_dir);
273 1839116 : scan->rs_inited = true;
274 : }
275 : else
276 5227808 : scan->rs_prefetch_block = heapgettup_advance_block(scan,
277 : scan->rs_prefetch_block,
278 : scan->rs_dir);
279 :
280 7066924 : return scan->rs_prefetch_block;
281 : }
282 :
283 : /*
284 : * Read stream API callback for bitmap heap scans.
285 : * Returns the next block the caller wants from the read stream or
286 : * InvalidBlockNumber when done.
287 : */
288 : static BlockNumber
289 397238 : bitmapheap_stream_read_next(ReadStream *pgsr, void *private_data,
290 : void *per_buffer_data)
291 : {
292 397238 : TBMIterateResult *tbmres = per_buffer_data;
293 397238 : BitmapHeapScanDesc bscan = (BitmapHeapScanDesc) private_data;
294 397238 : HeapScanDesc hscan = (HeapScanDesc) bscan;
295 397238 : TableScanDesc sscan = &hscan->rs_base;
296 :
297 : for (;;)
298 : {
299 422804 : CHECK_FOR_INTERRUPTS();
300 :
301 : /* no more entries in the bitmap */
302 422804 : if (!tbm_iterate(&sscan->st.rs_tbmiterator, tbmres))
303 25560 : return InvalidBlockNumber;
304 :
305 : /*
306 : * Ignore any claimed entries past what we think is the end of the
307 : * relation. It may have been extended after the start of our scan (we
308 : * only hold an AccessShareLock, and it could be inserts from this
309 : * backend). We don't take this optimization in SERIALIZABLE
310 : * isolation though, as we need to examine all invisible tuples
311 : * reachable by the index.
312 : */
313 397244 : if (!IsolationIsSerializable() &&
314 397038 : tbmres->blockno >= hscan->rs_nblocks)
315 0 : continue;
316 :
317 : /*
318 : * We can skip fetching the heap page if we don't need any fields from
319 : * the heap, the bitmap entries don't need rechecking, and all tuples
320 : * on the page are visible to our transaction.
321 : */
322 397244 : if (!(sscan->rs_flags & SO_NEED_TUPLES) &&
323 100456 : !tbmres->recheck &&
324 58024 : VM_ALL_VISIBLE(sscan->rs_rd, tbmres->blockno, &bscan->rs_vmbuffer))
325 : {
326 : OffsetNumber offsets[TBM_MAX_TUPLES_PER_PAGE];
327 : int noffsets;
328 :
329 : /* can't be lossy in the skip_fetch case */
330 : Assert(!tbmres->lossy);
331 : Assert(bscan->rs_empty_tuples_pending >= 0);
332 :
333 : /*
334 : * We throw away the offsets, but this is the easiest way to get a
335 : * count of tuples.
336 : */
337 25566 : noffsets = tbm_extract_page_tuple(tbmres, offsets, TBM_MAX_TUPLES_PER_PAGE);
338 25566 : bscan->rs_empty_tuples_pending += noffsets;
339 25566 : continue;
340 : }
341 :
342 371678 : return tbmres->blockno;
343 : }
344 :
345 : /* not reachable */
346 : Assert(false);
347 : }
348 :
349 : /* ----------------
350 : * initscan - scan code common to heap_beginscan and heap_rescan
351 : * ----------------
352 : */
353 : static void
354 1886080 : initscan(HeapScanDesc scan, ScanKey key, bool keep_startblock)
355 : {
356 1886080 : ParallelBlockTableScanDesc bpscan = NULL;
357 : bool allow_strat;
358 : bool allow_sync;
359 :
360 : /*
361 : * Determine the number of blocks we have to scan.
362 : *
363 : * It is sufficient to do this once at scan start, since any tuples added
364 : * while the scan is in progress will be invisible to my snapshot anyway.
365 : * (That is not true when using a non-MVCC snapshot. However, we couldn't
366 : * guarantee to return tuples added after scan start anyway, since they
367 : * might go into pages we already scanned. To guarantee consistent
368 : * results for a non-MVCC snapshot, the caller must hold some higher-level
369 : * lock that ensures the interesting tuple(s) won't change.)
370 : */
371 1886080 : if (scan->rs_base.rs_parallel != NULL)
372 : {
373 4024 : bpscan = (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel;
374 4024 : scan->rs_nblocks = bpscan->phs_nblocks;
375 : }
376 : else
377 1882056 : scan->rs_nblocks = RelationGetNumberOfBlocks(scan->rs_base.rs_rd);
378 :
379 : /*
380 : * If the table is large relative to NBuffers, use a bulk-read access
381 : * strategy and enable synchronized scanning (see syncscan.c). Although
382 : * the thresholds for these features could be different, we make them the
383 : * same so that there are only two behaviors to tune rather than four.
384 : * (However, some callers need to be able to disable one or both of these
385 : * behaviors, independently of the size of the table; also there is a GUC
386 : * variable that can disable synchronized scanning.)
387 : *
388 : * Note that table_block_parallelscan_initialize has a very similar test;
389 : * if you change this, consider changing that one, too.
390 : */
391 1886076 : if (!RelationUsesLocalBuffers(scan->rs_base.rs_rd) &&
392 1871904 : scan->rs_nblocks > NBuffers / 4)
393 : {
394 24394 : allow_strat = (scan->rs_base.rs_flags & SO_ALLOW_STRAT) != 0;
395 24394 : allow_sync = (scan->rs_base.rs_flags & SO_ALLOW_SYNC) != 0;
396 : }
397 : else
398 1861682 : allow_strat = allow_sync = false;
399 :
400 1886076 : if (allow_strat)
401 : {
402 : /* During a rescan, keep the previous strategy object. */
403 21866 : if (scan->rs_strategy == NULL)
404 21518 : scan->rs_strategy = GetAccessStrategy(BAS_BULKREAD);
405 : }
406 : else
407 : {
408 1864210 : if (scan->rs_strategy != NULL)
409 0 : FreeAccessStrategy(scan->rs_strategy);
410 1864210 : scan->rs_strategy = NULL;
411 : }
412 :
413 1886076 : if (scan->rs_base.rs_parallel != NULL)
414 : {
415 : /* For parallel scan, believe whatever ParallelTableScanDesc says. */
416 4024 : if (scan->rs_base.rs_parallel->phs_syncscan)
417 4 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
418 : else
419 4020 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
420 : }
421 1882052 : else if (keep_startblock)
422 : {
423 : /*
424 : * When rescanning, we want to keep the previous startblock setting,
425 : * so that rewinding a cursor doesn't generate surprising results.
426 : * Reset the active syncscan setting, though.
427 : */
428 1166666 : if (allow_sync && synchronize_seqscans)
429 40 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
430 : else
431 1166626 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
432 : }
433 715386 : else if (allow_sync && synchronize_seqscans)
434 : {
435 118 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
436 118 : scan->rs_startblock = ss_get_location(scan->rs_base.rs_rd, scan->rs_nblocks);
437 : }
438 : else
439 : {
440 715268 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
441 715268 : scan->rs_startblock = 0;
442 : }
443 :
444 1886076 : scan->rs_numblocks = InvalidBlockNumber;
445 1886076 : scan->rs_inited = false;
446 1886076 : scan->rs_ctup.t_data = NULL;
447 1886076 : ItemPointerSetInvalid(&scan->rs_ctup.t_self);
448 1886076 : scan->rs_cbuf = InvalidBuffer;
449 1886076 : scan->rs_cblock = InvalidBlockNumber;
450 1886076 : scan->rs_ntuples = 0;
451 1886076 : scan->rs_cindex = 0;
452 :
453 : /*
454 : * Initialize to ForwardScanDirection because it is most common and
455 : * because heap scans go forward before going backward (e.g. CURSORs).
456 : */
457 1886076 : scan->rs_dir = ForwardScanDirection;
458 1886076 : scan->rs_prefetch_block = InvalidBlockNumber;
459 :
460 : /* page-at-a-time fields are always invalid when not rs_inited */
461 :
462 : /*
463 : * copy the scan key, if appropriate
464 : */
465 1886076 : if (key != NULL && scan->rs_base.rs_nkeys > 0)
466 405734 : memcpy(scan->rs_base.rs_key, key, scan->rs_base.rs_nkeys * sizeof(ScanKeyData));
467 :
468 : /*
469 : * Currently, we only have a stats counter for sequential heap scans (but
470 : * e.g for bitmap scans the underlying bitmap index scans will be counted,
471 : * and for sample scans we update stats for tuple fetches).
472 : */
473 1886076 : if (scan->rs_base.rs_flags & SO_TYPE_SEQSCAN)
474 1841472 : pgstat_count_heap_scan(scan->rs_base.rs_rd);
475 1886076 : }
476 :
477 : /*
478 : * heap_setscanlimits - restrict range of a heapscan
479 : *
480 : * startBlk is the page to start at
481 : * numBlks is number of pages to scan (InvalidBlockNumber means "all")
482 : */
483 : void
484 5606 : heap_setscanlimits(TableScanDesc sscan, BlockNumber startBlk, BlockNumber numBlks)
485 : {
486 5606 : HeapScanDesc scan = (HeapScanDesc) sscan;
487 :
488 : Assert(!scan->rs_inited); /* else too late to change */
489 : /* else rs_startblock is significant */
490 : Assert(!(scan->rs_base.rs_flags & SO_ALLOW_SYNC));
491 :
492 : /* Check startBlk is valid (but allow case of zero blocks...) */
493 : Assert(startBlk == 0 || startBlk < scan->rs_nblocks);
494 :
495 5606 : scan->rs_startblock = startBlk;
496 5606 : scan->rs_numblocks = numBlks;
497 5606 : }
498 :
499 : /*
500 : * Per-tuple loop for heap_prepare_pagescan(). Pulled out so it can be called
501 : * multiple times, with constant arguments for all_visible,
502 : * check_serializable.
503 : */
504 : pg_attribute_always_inline
505 : static int
506 5204386 : page_collect_tuples(HeapScanDesc scan, Snapshot snapshot,
507 : Page page, Buffer buffer,
508 : BlockNumber block, int lines,
509 : bool all_visible, bool check_serializable)
510 : {
511 5204386 : int ntup = 0;
512 : OffsetNumber lineoff;
513 :
514 261307050 : for (lineoff = FirstOffsetNumber; lineoff <= lines; lineoff++)
515 : {
516 256102680 : ItemId lpp = PageGetItemId(page, lineoff);
517 : HeapTupleData loctup;
518 : bool valid;
519 :
520 256102680 : if (!ItemIdIsNormal(lpp))
521 54793258 : continue;
522 :
523 201309422 : loctup.t_data = (HeapTupleHeader) PageGetItem(page, lpp);
524 201309422 : loctup.t_len = ItemIdGetLength(lpp);
525 201309422 : loctup.t_tableOid = RelationGetRelid(scan->rs_base.rs_rd);
526 201309422 : ItemPointerSet(&(loctup.t_self), block, lineoff);
527 :
528 201309422 : if (all_visible)
529 77700442 : valid = true;
530 : else
531 123608980 : valid = HeapTupleSatisfiesVisibility(&loctup, snapshot, buffer);
532 :
533 201309422 : if (check_serializable)
534 2810 : HeapCheckForSerializableConflictOut(valid, scan->rs_base.rs_rd,
535 : &loctup, buffer, snapshot);
536 :
537 201309406 : if (valid)
538 : {
539 185182116 : scan->rs_vistuples[ntup] = lineoff;
540 185182116 : ntup++;
541 : }
542 : }
543 :
544 : Assert(ntup <= MaxHeapTuplesPerPage);
545 :
546 5204370 : return ntup;
547 : }
548 :
549 : /*
550 : * heap_prepare_pagescan - Prepare current scan page to be scanned in pagemode
551 : *
552 : * Preparation currently consists of 1. prune the scan's rs_cbuf page, and 2.
553 : * fill the rs_vistuples[] array with the OffsetNumbers of visible tuples.
554 : */
555 : void
556 5204386 : heap_prepare_pagescan(TableScanDesc sscan)
557 : {
558 5204386 : HeapScanDesc scan = (HeapScanDesc) sscan;
559 5204386 : Buffer buffer = scan->rs_cbuf;
560 5204386 : BlockNumber block = scan->rs_cblock;
561 : Snapshot snapshot;
562 : Page page;
563 : int lines;
564 : bool all_visible;
565 : bool check_serializable;
566 :
567 : Assert(BufferGetBlockNumber(buffer) == block);
568 :
569 : /* ensure we're not accidentally being used when not in pagemode */
570 : Assert(scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE);
571 5204386 : snapshot = scan->rs_base.rs_snapshot;
572 :
573 : /*
574 : * Prune and repair fragmentation for the whole page, if possible.
575 : */
576 5204386 : heap_page_prune_opt(scan->rs_base.rs_rd, buffer);
577 :
578 : /*
579 : * We must hold share lock on the buffer content while examining tuple
580 : * visibility. Afterwards, however, the tuples we have found to be
581 : * visible are guaranteed good as long as we hold the buffer pin.
582 : */
583 5204386 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
584 :
585 5204386 : page = BufferGetPage(buffer);
586 5204386 : lines = PageGetMaxOffsetNumber(page);
587 :
588 : /*
589 : * If the all-visible flag indicates that all tuples on the page are
590 : * visible to everyone, we can skip the per-tuple visibility tests.
591 : *
592 : * Note: In hot standby, a tuple that's already visible to all
593 : * transactions on the primary might still be invisible to a read-only
594 : * transaction in the standby. We partly handle this problem by tracking
595 : * the minimum xmin of visible tuples as the cut-off XID while marking a
596 : * page all-visible on the primary and WAL log that along with the
597 : * visibility map SET operation. In hot standby, we wait for (or abort)
598 : * all transactions that can potentially may not see one or more tuples on
599 : * the page. That's how index-only scans work fine in hot standby. A
600 : * crucial difference between index-only scans and heap scans is that the
601 : * index-only scan completely relies on the visibility map where as heap
602 : * scan looks at the page-level PD_ALL_VISIBLE flag. We are not sure if
603 : * the page-level flag can be trusted in the same way, because it might
604 : * get propagated somehow without being explicitly WAL-logged, e.g. via a
605 : * full page write. Until we can prove that beyond doubt, let's check each
606 : * tuple for visibility the hard way.
607 : */
608 5204386 : all_visible = PageIsAllVisible(page) && !snapshot->takenDuringRecovery;
609 : check_serializable =
610 5204386 : CheckForSerializableConflictOutNeeded(scan->rs_base.rs_rd, snapshot);
611 :
612 : /*
613 : * We call page_collect_tuples() with constant arguments, to get the
614 : * compiler to constant fold the constant arguments. Separate calls with
615 : * constant arguments, rather than variables, are needed on several
616 : * compilers to actually perform constant folding.
617 : */
618 5204386 : if (likely(all_visible))
619 : {
620 1815780 : if (likely(!check_serializable))
621 1815780 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
622 : block, lines, true, false);
623 : else
624 0 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
625 : block, lines, true, true);
626 : }
627 : else
628 : {
629 3388606 : if (likely(!check_serializable))
630 3387368 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
631 : block, lines, false, false);
632 : else
633 1238 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
634 : block, lines, false, true);
635 : }
636 :
637 5204370 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
638 5204370 : }
639 :
640 : /*
641 : * heap_fetch_next_buffer - read and pin the next block from MAIN_FORKNUM.
642 : *
643 : * Read the next block of the scan relation from the read stream and save it
644 : * in the scan descriptor. It is already pinned.
645 : */
646 : static inline void
647 6918848 : heap_fetch_next_buffer(HeapScanDesc scan, ScanDirection dir)
648 : {
649 : Assert(scan->rs_read_stream);
650 :
651 : /* release previous scan buffer, if any */
652 6918848 : if (BufferIsValid(scan->rs_cbuf))
653 : {
654 5076892 : ReleaseBuffer(scan->rs_cbuf);
655 5076892 : scan->rs_cbuf = InvalidBuffer;
656 : }
657 :
658 : /*
659 : * Be sure to check for interrupts at least once per page. Checks at
660 : * higher code levels won't be able to stop a seqscan that encounters many
661 : * pages' worth of consecutive dead tuples.
662 : */
663 6918848 : CHECK_FOR_INTERRUPTS();
664 :
665 : /*
666 : * If the scan direction is changing, reset the prefetch block to the
667 : * current block. Otherwise, we will incorrectly prefetch the blocks
668 : * between the prefetch block and the current block again before
669 : * prefetching blocks in the new, correct scan direction.
670 : */
671 6918842 : if (unlikely(scan->rs_dir != dir))
672 : {
673 154 : scan->rs_prefetch_block = scan->rs_cblock;
674 154 : read_stream_reset(scan->rs_read_stream);
675 : }
676 :
677 6918842 : scan->rs_dir = dir;
678 :
679 6918842 : scan->rs_cbuf = read_stream_next_buffer(scan->rs_read_stream, NULL);
680 6918830 : if (BufferIsValid(scan->rs_cbuf))
681 5384878 : scan->rs_cblock = BufferGetBlockNumber(scan->rs_cbuf);
682 6918830 : }
683 :
684 : /*
685 : * heapgettup_initial_block - return the first BlockNumber to scan
686 : *
687 : * Returns InvalidBlockNumber when there are no blocks to scan. This can
688 : * occur with empty tables and in parallel scans when parallel workers get all
689 : * of the pages before we can get a chance to get our first page.
690 : */
691 : static pg_noinline BlockNumber
692 1839116 : heapgettup_initial_block(HeapScanDesc scan, ScanDirection dir)
693 : {
694 : Assert(!scan->rs_inited);
695 : Assert(scan->rs_base.rs_parallel == NULL);
696 :
697 : /* When there are no pages to scan, return InvalidBlockNumber */
698 1839116 : if (scan->rs_nblocks == 0 || scan->rs_numblocks == 0)
699 891356 : return InvalidBlockNumber;
700 :
701 947760 : if (ScanDirectionIsForward(dir))
702 : {
703 947696 : return scan->rs_startblock;
704 : }
705 : else
706 : {
707 : /*
708 : * Disable reporting to syncscan logic in a backwards scan; it's not
709 : * very likely anyone else is doing the same thing at the same time,
710 : * and much more likely that we'll just bollix things for forward
711 : * scanners.
712 : */
713 64 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
714 :
715 : /*
716 : * Start from last page of the scan. Ensure we take into account
717 : * rs_numblocks if it's been adjusted by heap_setscanlimits().
718 : */
719 64 : if (scan->rs_numblocks != InvalidBlockNumber)
720 6 : return (scan->rs_startblock + scan->rs_numblocks - 1) % scan->rs_nblocks;
721 :
722 58 : if (scan->rs_startblock > 0)
723 0 : return scan->rs_startblock - 1;
724 :
725 58 : return scan->rs_nblocks - 1;
726 : }
727 : }
728 :
729 :
730 : /*
731 : * heapgettup_start_page - helper function for heapgettup()
732 : *
733 : * Return the next page to scan based on the scan->rs_cbuf and set *linesleft
734 : * to the number of tuples on this page. Also set *lineoff to the first
735 : * offset to scan with forward scans getting the first offset and backward
736 : * getting the final offset on the page.
737 : */
738 : static Page
739 189042 : heapgettup_start_page(HeapScanDesc scan, ScanDirection dir, int *linesleft,
740 : OffsetNumber *lineoff)
741 : {
742 : Page page;
743 :
744 : Assert(scan->rs_inited);
745 : Assert(BufferIsValid(scan->rs_cbuf));
746 :
747 : /* Caller is responsible for ensuring buffer is locked if needed */
748 189042 : page = BufferGetPage(scan->rs_cbuf);
749 :
750 189042 : *linesleft = PageGetMaxOffsetNumber(page) - FirstOffsetNumber + 1;
751 :
752 189042 : if (ScanDirectionIsForward(dir))
753 189042 : *lineoff = FirstOffsetNumber;
754 : else
755 0 : *lineoff = (OffsetNumber) (*linesleft);
756 :
757 : /* lineoff now references the physically previous or next tid */
758 189042 : return page;
759 : }
760 :
761 :
762 : /*
763 : * heapgettup_continue_page - helper function for heapgettup()
764 : *
765 : * Return the next page to scan based on the scan->rs_cbuf and set *linesleft
766 : * to the number of tuples left to scan on this page. Also set *lineoff to
767 : * the next offset to scan according to the ScanDirection in 'dir'.
768 : */
769 : static inline Page
770 15655998 : heapgettup_continue_page(HeapScanDesc scan, ScanDirection dir, int *linesleft,
771 : OffsetNumber *lineoff)
772 : {
773 : Page page;
774 :
775 : Assert(scan->rs_inited);
776 : Assert(BufferIsValid(scan->rs_cbuf));
777 :
778 : /* Caller is responsible for ensuring buffer is locked if needed */
779 15655998 : page = BufferGetPage(scan->rs_cbuf);
780 :
781 15655998 : if (ScanDirectionIsForward(dir))
782 : {
783 15655998 : *lineoff = OffsetNumberNext(scan->rs_coffset);
784 15655998 : *linesleft = PageGetMaxOffsetNumber(page) - (*lineoff) + 1;
785 : }
786 : else
787 : {
788 : /*
789 : * The previous returned tuple may have been vacuumed since the
790 : * previous scan when we use a non-MVCC snapshot, so we must
791 : * re-establish the lineoff <= PageGetMaxOffsetNumber(page) invariant
792 : */
793 0 : *lineoff = Min(PageGetMaxOffsetNumber(page), OffsetNumberPrev(scan->rs_coffset));
794 0 : *linesleft = *lineoff;
795 : }
796 :
797 : /* lineoff now references the physically previous or next tid */
798 15655998 : return page;
799 : }
800 :
801 : /*
802 : * heapgettup_advance_block - helper for heap_fetch_next_buffer()
803 : *
804 : * Given the current block number, the scan direction, and various information
805 : * contained in the scan descriptor, calculate the BlockNumber to scan next
806 : * and return it. If there are no further blocks to scan, return
807 : * InvalidBlockNumber to indicate this fact to the caller.
808 : *
809 : * This should not be called to determine the initial block number -- only for
810 : * subsequent blocks.
811 : *
812 : * This also adjusts rs_numblocks when a limit has been imposed by
813 : * heap_setscanlimits().
814 : */
815 : static inline BlockNumber
816 5227808 : heapgettup_advance_block(HeapScanDesc scan, BlockNumber block, ScanDirection dir)
817 : {
818 : Assert(scan->rs_base.rs_parallel == NULL);
819 :
820 5227808 : if (likely(ScanDirectionIsForward(dir)))
821 : {
822 5227690 : block++;
823 :
824 : /* wrap back to the start of the heap */
825 5227690 : if (block >= scan->rs_nblocks)
826 762018 : block = 0;
827 :
828 : /*
829 : * Report our new scan position for synchronization purposes. We don't
830 : * do that when moving backwards, however. That would just mess up any
831 : * other forward-moving scanners.
832 : *
833 : * Note: we do this before checking for end of scan so that the final
834 : * state of the position hint is back at the start of the rel. That's
835 : * not strictly necessary, but otherwise when you run the same query
836 : * multiple times the starting position would shift a little bit
837 : * backwards on every invocation, which is confusing. We don't
838 : * guarantee any specific ordering in general, though.
839 : */
840 5227690 : if (scan->rs_base.rs_flags & SO_ALLOW_SYNC)
841 17574 : ss_report_location(scan->rs_base.rs_rd, block);
842 :
843 : /* we're done if we're back at where we started */
844 5227690 : if (block == scan->rs_startblock)
845 761936 : return InvalidBlockNumber;
846 :
847 : /* check if the limit imposed by heap_setscanlimits() is met */
848 4465754 : if (scan->rs_numblocks != InvalidBlockNumber)
849 : {
850 4924 : if (--scan->rs_numblocks == 0)
851 3056 : return InvalidBlockNumber;
852 : }
853 :
854 4462698 : return block;
855 : }
856 : else
857 : {
858 : /* we're done if the last block is the start position */
859 118 : if (block == scan->rs_startblock)
860 118 : return InvalidBlockNumber;
861 :
862 : /* check if the limit imposed by heap_setscanlimits() is met */
863 0 : if (scan->rs_numblocks != InvalidBlockNumber)
864 : {
865 0 : if (--scan->rs_numblocks == 0)
866 0 : return InvalidBlockNumber;
867 : }
868 :
869 : /* wrap to the end of the heap when the last page was page 0 */
870 0 : if (block == 0)
871 0 : block = scan->rs_nblocks;
872 :
873 0 : block--;
874 :
875 0 : return block;
876 : }
877 : }
878 :
879 : /* ----------------
880 : * heapgettup - fetch next heap tuple
881 : *
882 : * Initialize the scan if not already done; then advance to the next
883 : * tuple as indicated by "dir"; return the next tuple in scan->rs_ctup,
884 : * or set scan->rs_ctup.t_data = NULL if no more tuples.
885 : *
886 : * Note: the reason nkeys/key are passed separately, even though they are
887 : * kept in the scan descriptor, is that the caller may not want us to check
888 : * the scankeys.
889 : *
890 : * Note: when we fall off the end of the scan in either direction, we
891 : * reset rs_inited. This means that a further request with the same
892 : * scan direction will restart the scan, which is a bit odd, but a
893 : * request with the opposite scan direction will start a fresh scan
894 : * in the proper direction. The latter is required behavior for cursors,
895 : * while the former case is generally undefined behavior in Postgres
896 : * so we don't care too much.
897 : * ----------------
898 : */
899 : static void
900 15696076 : heapgettup(HeapScanDesc scan,
901 : ScanDirection dir,
902 : int nkeys,
903 : ScanKey key)
904 : {
905 15696076 : HeapTuple tuple = &(scan->rs_ctup);
906 : Page page;
907 : OffsetNumber lineoff;
908 : int linesleft;
909 :
910 15696076 : if (likely(scan->rs_inited))
911 : {
912 : /* continue from previously returned page/tuple */
913 15655998 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
914 15655998 : page = heapgettup_continue_page(scan, dir, &linesleft, &lineoff);
915 15655998 : goto continue_page;
916 : }
917 :
918 : /*
919 : * advance the scan until we find a qualifying tuple or run out of stuff
920 : * to scan
921 : */
922 : while (true)
923 : {
924 228820 : heap_fetch_next_buffer(scan, dir);
925 :
926 : /* did we run out of blocks to scan? */
927 228820 : if (!BufferIsValid(scan->rs_cbuf))
928 39778 : break;
929 :
930 : Assert(BufferGetBlockNumber(scan->rs_cbuf) == scan->rs_cblock);
931 :
932 189042 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
933 189042 : page = heapgettup_start_page(scan, dir, &linesleft, &lineoff);
934 15845040 : continue_page:
935 :
936 : /*
937 : * Only continue scanning the page while we have lines left.
938 : *
939 : * Note that this protects us from accessing line pointers past
940 : * PageGetMaxOffsetNumber(); both for forward scans when we resume the
941 : * table scan, and for when we start scanning a new page.
942 : */
943 15928798 : for (; linesleft > 0; linesleft--, lineoff += dir)
944 : {
945 : bool visible;
946 15740056 : ItemId lpp = PageGetItemId(page, lineoff);
947 :
948 15740056 : if (!ItemIdIsNormal(lpp))
949 73302 : continue;
950 :
951 15666754 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
952 15666754 : tuple->t_len = ItemIdGetLength(lpp);
953 15666754 : ItemPointerSet(&(tuple->t_self), scan->rs_cblock, lineoff);
954 :
955 15666754 : visible = HeapTupleSatisfiesVisibility(tuple,
956 : scan->rs_base.rs_snapshot,
957 : scan->rs_cbuf);
958 :
959 15666754 : HeapCheckForSerializableConflictOut(visible, scan->rs_base.rs_rd,
960 : tuple, scan->rs_cbuf,
961 : scan->rs_base.rs_snapshot);
962 :
963 : /* skip tuples not visible to this snapshot */
964 15666754 : if (!visible)
965 10456 : continue;
966 :
967 : /* skip any tuples that don't match the scan key */
968 15656298 : if (key != NULL &&
969 0 : !HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
970 : nkeys, key))
971 0 : continue;
972 :
973 15656298 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
974 15656298 : scan->rs_coffset = lineoff;
975 15656298 : return;
976 : }
977 :
978 : /*
979 : * if we get here, it means we've exhausted the items on this page and
980 : * it's time to move to the next.
981 : */
982 188742 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
983 : }
984 :
985 : /* end of scan */
986 39778 : if (BufferIsValid(scan->rs_cbuf))
987 0 : ReleaseBuffer(scan->rs_cbuf);
988 :
989 39778 : scan->rs_cbuf = InvalidBuffer;
990 39778 : scan->rs_cblock = InvalidBlockNumber;
991 39778 : scan->rs_prefetch_block = InvalidBlockNumber;
992 39778 : tuple->t_data = NULL;
993 39778 : scan->rs_inited = false;
994 : }
995 :
996 : /* ----------------
997 : * heapgettup_pagemode - fetch next heap tuple in page-at-a-time mode
998 : *
999 : * Same API as heapgettup, but used in page-at-a-time mode
1000 : *
1001 : * The internal logic is much the same as heapgettup's too, but there are some
1002 : * differences: we do not take the buffer content lock (that only needs to
1003 : * happen inside heap_prepare_pagescan), and we iterate through just the
1004 : * tuples listed in rs_vistuples[] rather than all tuples on the page. Notice
1005 : * that lineindex is 0-based, where the corresponding loop variable lineoff in
1006 : * heapgettup is 1-based.
1007 : * ----------------
1008 : */
1009 : static void
1010 93098646 : heapgettup_pagemode(HeapScanDesc scan,
1011 : ScanDirection dir,
1012 : int nkeys,
1013 : ScanKey key)
1014 : {
1015 93098646 : HeapTuple tuple = &(scan->rs_ctup);
1016 : Page page;
1017 : uint32 lineindex;
1018 : uint32 linesleft;
1019 :
1020 93098646 : if (likely(scan->rs_inited))
1021 : {
1022 : /* continue from previously returned page/tuple */
1023 91296768 : page = BufferGetPage(scan->rs_cbuf);
1024 :
1025 91296768 : lineindex = scan->rs_cindex + dir;
1026 91296768 : if (ScanDirectionIsForward(dir))
1027 91296110 : linesleft = scan->rs_ntuples - lineindex;
1028 : else
1029 658 : linesleft = scan->rs_cindex;
1030 : /* lineindex now references the next or previous visible tid */
1031 :
1032 91296768 : goto continue_page;
1033 : }
1034 :
1035 : /*
1036 : * advance the scan until we find a qualifying tuple or run out of stuff
1037 : * to scan
1038 : */
1039 : while (true)
1040 : {
1041 6690028 : heap_fetch_next_buffer(scan, dir);
1042 :
1043 : /* did we run out of blocks to scan? */
1044 6690010 : if (!BufferIsValid(scan->rs_cbuf))
1045 1494174 : break;
1046 :
1047 : Assert(BufferGetBlockNumber(scan->rs_cbuf) == scan->rs_cblock);
1048 :
1049 : /* prune the page and determine visible tuple offsets */
1050 5195836 : heap_prepare_pagescan((TableScanDesc) scan);
1051 5195820 : page = BufferGetPage(scan->rs_cbuf);
1052 5195820 : linesleft = scan->rs_ntuples;
1053 5195820 : lineindex = ScanDirectionIsForward(dir) ? 0 : linesleft - 1;
1054 :
1055 : /* block is the same for all tuples, set it once outside the loop */
1056 5195820 : ItemPointerSetBlockNumber(&tuple->t_self, scan->rs_cblock);
1057 :
1058 : /* lineindex now references the next or previous visible tid */
1059 96492588 : continue_page:
1060 :
1061 183854540 : for (; linesleft > 0; linesleft--, lineindex += dir)
1062 : {
1063 : ItemId lpp;
1064 : OffsetNumber lineoff;
1065 :
1066 : Assert(lineindex <= scan->rs_ntuples);
1067 178966390 : lineoff = scan->rs_vistuples[lineindex];
1068 178966390 : lpp = PageGetItemId(page, lineoff);
1069 : Assert(ItemIdIsNormal(lpp));
1070 :
1071 178966390 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
1072 178966390 : tuple->t_len = ItemIdGetLength(lpp);
1073 178966390 : ItemPointerSetOffsetNumber(&tuple->t_self, lineoff);
1074 :
1075 : /* skip any tuples that don't match the scan key */
1076 178966390 : if (key != NULL &&
1077 88055254 : !HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
1078 : nkeys, key))
1079 87361952 : continue;
1080 :
1081 91604438 : scan->rs_cindex = lineindex;
1082 91604438 : return;
1083 : }
1084 : }
1085 :
1086 : /* end of scan */
1087 1494174 : if (BufferIsValid(scan->rs_cbuf))
1088 0 : ReleaseBuffer(scan->rs_cbuf);
1089 1494174 : scan->rs_cbuf = InvalidBuffer;
1090 1494174 : scan->rs_cblock = InvalidBlockNumber;
1091 1494174 : scan->rs_prefetch_block = InvalidBlockNumber;
1092 1494174 : tuple->t_data = NULL;
1093 1494174 : scan->rs_inited = false;
1094 : }
1095 :
1096 :
1097 : /* ----------------------------------------------------------------
1098 : * heap access method interface
1099 : * ----------------------------------------------------------------
1100 : */
1101 :
1102 :
1103 : TableScanDesc
1104 719306 : heap_beginscan(Relation relation, Snapshot snapshot,
1105 : int nkeys, ScanKey key,
1106 : ParallelTableScanDesc parallel_scan,
1107 : uint32 flags)
1108 : {
1109 : HeapScanDesc scan;
1110 :
1111 : /*
1112 : * increment relation ref count while scanning relation
1113 : *
1114 : * This is just to make really sure the relcache entry won't go away while
1115 : * the scan has a pointer to it. Caller should be holding the rel open
1116 : * anyway, so this is redundant in all normal scenarios...
1117 : */
1118 719306 : RelationIncrementReferenceCount(relation);
1119 :
1120 : /*
1121 : * allocate and initialize scan descriptor
1122 : */
1123 719306 : if (flags & SO_TYPE_BITMAPSCAN)
1124 : {
1125 21346 : BitmapHeapScanDesc bscan = palloc(sizeof(BitmapHeapScanDescData));
1126 :
1127 21346 : bscan->rs_vmbuffer = InvalidBuffer;
1128 21346 : bscan->rs_empty_tuples_pending = 0;
1129 21346 : scan = (HeapScanDesc) bscan;
1130 : }
1131 : else
1132 697960 : scan = (HeapScanDesc) palloc(sizeof(HeapScanDescData));
1133 :
1134 719306 : scan->rs_base.rs_rd = relation;
1135 719306 : scan->rs_base.rs_snapshot = snapshot;
1136 719306 : scan->rs_base.rs_nkeys = nkeys;
1137 719306 : scan->rs_base.rs_flags = flags;
1138 719306 : scan->rs_base.rs_parallel = parallel_scan;
1139 719306 : scan->rs_strategy = NULL; /* set in initscan */
1140 719306 : scan->rs_cbuf = InvalidBuffer;
1141 :
1142 : /*
1143 : * Disable page-at-a-time mode if it's not a MVCC-safe snapshot.
1144 : */
1145 719306 : if (!(snapshot && IsMVCCSnapshot(snapshot)))
1146 56270 : scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
1147 :
1148 : /*
1149 : * For seqscan and sample scans in a serializable transaction, acquire a
1150 : * predicate lock on the entire relation. This is required not only to
1151 : * lock all the matching tuples, but also to conflict with new insertions
1152 : * into the table. In an indexscan, we take page locks on the index pages
1153 : * covering the range specified in the scan qual, but in a heap scan there
1154 : * is nothing more fine-grained to lock. A bitmap scan is a different
1155 : * story, there we have already scanned the index and locked the index
1156 : * pages covering the predicate. But in that case we still have to lock
1157 : * any matching heap tuples. For sample scan we could optimize the locking
1158 : * to be at least page-level granularity, but we'd need to add per-tuple
1159 : * locking for that.
1160 : */
1161 719306 : if (scan->rs_base.rs_flags & (SO_TYPE_SEQSCAN | SO_TYPE_SAMPLESCAN))
1162 : {
1163 : /*
1164 : * Ensure a missing snapshot is noticed reliably, even if the
1165 : * isolation mode means predicate locking isn't performed (and
1166 : * therefore the snapshot isn't used here).
1167 : */
1168 : Assert(snapshot);
1169 679196 : PredicateLockRelation(relation, snapshot);
1170 : }
1171 :
1172 : /* we only need to set this up once */
1173 719306 : scan->rs_ctup.t_tableOid = RelationGetRelid(relation);
1174 :
1175 : /*
1176 : * Allocate memory to keep track of page allocation for parallel workers
1177 : * when doing a parallel scan.
1178 : */
1179 719306 : if (parallel_scan != NULL)
1180 3916 : scan->rs_parallelworkerdata = palloc(sizeof(ParallelBlockTableScanWorkerData));
1181 : else
1182 715390 : scan->rs_parallelworkerdata = NULL;
1183 :
1184 : /*
1185 : * we do this here instead of in initscan() because heap_rescan also calls
1186 : * initscan() and we don't want to allocate memory again
1187 : */
1188 719306 : if (nkeys > 0)
1189 405734 : scan->rs_base.rs_key = (ScanKey) palloc(sizeof(ScanKeyData) * nkeys);
1190 : else
1191 313572 : scan->rs_base.rs_key = NULL;
1192 :
1193 719306 : initscan(scan, key, false);
1194 :
1195 719302 : scan->rs_read_stream = NULL;
1196 :
1197 : /*
1198 : * Set up a read stream for sequential scans and TID range scans. This
1199 : * should be done after initscan() because initscan() allocates the
1200 : * BufferAccessStrategy object passed to the read stream API.
1201 : */
1202 719302 : if (scan->rs_base.rs_flags & SO_TYPE_SEQSCAN ||
1203 40256 : scan->rs_base.rs_flags & SO_TYPE_TIDRANGESCAN)
1204 680892 : {
1205 : ReadStreamBlockNumberCB cb;
1206 :
1207 680892 : if (scan->rs_base.rs_parallel)
1208 3916 : cb = heap_scan_stream_read_next_parallel;
1209 : else
1210 676976 : cb = heap_scan_stream_read_next_serial;
1211 :
1212 : /* ---
1213 : * It is safe to use batchmode as the only locks taken by `cb`
1214 : * are never taken while waiting for IO:
1215 : * - SyncScanLock is used in the non-parallel case
1216 : * - in the parallel case, only spinlocks and atomics are used
1217 : * ---
1218 : */
1219 680892 : scan->rs_read_stream = read_stream_begin_relation(READ_STREAM_SEQUENTIAL |
1220 : READ_STREAM_USE_BATCHING,
1221 : scan->rs_strategy,
1222 : scan->rs_base.rs_rd,
1223 : MAIN_FORKNUM,
1224 : cb,
1225 : scan,
1226 : 0);
1227 : }
1228 38410 : else if (scan->rs_base.rs_flags & SO_TYPE_BITMAPSCAN)
1229 : {
1230 : /*
1231 : * Currently we can't trivially use batching, due to the
1232 : * VM_ALL_VISIBLE check in bitmapheap_stream_read_next. While that
1233 : * could be made safe, we are about to remove the all-visible logic
1234 : * from bitmap scans due to its unsoundness.
1235 : */
1236 21346 : scan->rs_read_stream = read_stream_begin_relation(READ_STREAM_DEFAULT,
1237 : scan->rs_strategy,
1238 : scan->rs_base.rs_rd,
1239 : MAIN_FORKNUM,
1240 : bitmapheap_stream_read_next,
1241 : scan,
1242 : sizeof(TBMIterateResult));
1243 : }
1244 :
1245 :
1246 719302 : return (TableScanDesc) scan;
1247 : }
1248 :
1249 : void
1250 1166774 : heap_rescan(TableScanDesc sscan, ScanKey key, bool set_params,
1251 : bool allow_strat, bool allow_sync, bool allow_pagemode)
1252 : {
1253 1166774 : HeapScanDesc scan = (HeapScanDesc) sscan;
1254 :
1255 1166774 : if (set_params)
1256 : {
1257 30 : if (allow_strat)
1258 30 : scan->rs_base.rs_flags |= SO_ALLOW_STRAT;
1259 : else
1260 0 : scan->rs_base.rs_flags &= ~SO_ALLOW_STRAT;
1261 :
1262 30 : if (allow_sync)
1263 12 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
1264 : else
1265 18 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
1266 :
1267 30 : if (allow_pagemode && scan->rs_base.rs_snapshot &&
1268 30 : IsMVCCSnapshot(scan->rs_base.rs_snapshot))
1269 30 : scan->rs_base.rs_flags |= SO_ALLOW_PAGEMODE;
1270 : else
1271 0 : scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
1272 : }
1273 :
1274 : /*
1275 : * unpin scan buffers
1276 : */
1277 1166774 : if (BufferIsValid(scan->rs_cbuf))
1278 : {
1279 7272 : ReleaseBuffer(scan->rs_cbuf);
1280 7272 : scan->rs_cbuf = InvalidBuffer;
1281 : }
1282 :
1283 1166774 : if (scan->rs_base.rs_flags & SO_TYPE_BITMAPSCAN)
1284 : {
1285 4246 : BitmapHeapScanDesc bscan = (BitmapHeapScanDesc) scan;
1286 :
1287 : /*
1288 : * Reset empty_tuples_pending, a field only used by bitmap heap scan,
1289 : * to avoid incorrectly emitting NULL-filled tuples from a previous
1290 : * scan on rescan.
1291 : */
1292 4246 : bscan->rs_empty_tuples_pending = 0;
1293 :
1294 4246 : if (BufferIsValid(bscan->rs_vmbuffer))
1295 : {
1296 0 : ReleaseBuffer(bscan->rs_vmbuffer);
1297 0 : bscan->rs_vmbuffer = InvalidBuffer;
1298 : }
1299 : }
1300 :
1301 : /*
1302 : * The read stream is reset on rescan. This must be done before
1303 : * initscan(), as some state referred to by read_stream_reset() is reset
1304 : * in initscan().
1305 : */
1306 1166774 : if (scan->rs_read_stream)
1307 1166738 : read_stream_reset(scan->rs_read_stream);
1308 :
1309 : /*
1310 : * reinitialize scan descriptor
1311 : */
1312 1166774 : initscan(scan, key, true);
1313 1166774 : }
1314 :
1315 : void
1316 714764 : heap_endscan(TableScanDesc sscan)
1317 : {
1318 714764 : HeapScanDesc scan = (HeapScanDesc) sscan;
1319 :
1320 : /* Note: no locking manipulations needed */
1321 :
1322 : /*
1323 : * unpin scan buffers
1324 : */
1325 714764 : if (BufferIsValid(scan->rs_cbuf))
1326 297556 : ReleaseBuffer(scan->rs_cbuf);
1327 :
1328 714764 : if (scan->rs_base.rs_flags & SO_TYPE_BITMAPSCAN)
1329 : {
1330 21244 : BitmapHeapScanDesc bscan = (BitmapHeapScanDesc) sscan;
1331 :
1332 21244 : bscan->rs_empty_tuples_pending = 0;
1333 21244 : if (BufferIsValid(bscan->rs_vmbuffer))
1334 0 : ReleaseBuffer(bscan->rs_vmbuffer);
1335 : }
1336 :
1337 : /*
1338 : * Must free the read stream before freeing the BufferAccessStrategy.
1339 : */
1340 714764 : if (scan->rs_read_stream)
1341 697790 : read_stream_end(scan->rs_read_stream);
1342 :
1343 : /*
1344 : * decrement relation reference count and free scan descriptor storage
1345 : */
1346 714764 : RelationDecrementReferenceCount(scan->rs_base.rs_rd);
1347 :
1348 714764 : if (scan->rs_base.rs_key)
1349 405674 : pfree(scan->rs_base.rs_key);
1350 :
1351 714764 : if (scan->rs_strategy != NULL)
1352 21498 : FreeAccessStrategy(scan->rs_strategy);
1353 :
1354 714764 : if (scan->rs_parallelworkerdata != NULL)
1355 3916 : pfree(scan->rs_parallelworkerdata);
1356 :
1357 714764 : if (scan->rs_base.rs_flags & SO_TEMP_SNAPSHOT)
1358 70756 : UnregisterSnapshot(scan->rs_base.rs_snapshot);
1359 :
1360 714764 : pfree(scan);
1361 714764 : }
1362 :
1363 : HeapTuple
1364 18925336 : heap_getnext(TableScanDesc sscan, ScanDirection direction)
1365 : {
1366 18925336 : HeapScanDesc scan = (HeapScanDesc) sscan;
1367 :
1368 : /*
1369 : * This is still widely used directly, without going through table AM, so
1370 : * add a safety check. It's possible we should, at a later point,
1371 : * downgrade this to an assert. The reason for checking the AM routine,
1372 : * rather than the AM oid, is that this allows to write regression tests
1373 : * that create another AM reusing the heap handler.
1374 : */
1375 18925336 : if (unlikely(sscan->rs_rd->rd_tableam != GetHeapamTableAmRoutine()))
1376 0 : ereport(ERROR,
1377 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1378 : errmsg_internal("only heap AM is supported")));
1379 :
1380 : /*
1381 : * We don't expect direct calls to heap_getnext with valid CheckXidAlive
1382 : * for catalog or regular tables. See detailed comments in xact.c where
1383 : * these variables are declared. Normally we have such a check at tableam
1384 : * level API but this is called from many places so we need to ensure it
1385 : * here.
1386 : */
1387 18925336 : if (unlikely(TransactionIdIsValid(CheckXidAlive) && !bsysscan))
1388 0 : elog(ERROR, "unexpected heap_getnext call during logical decoding");
1389 :
1390 : /* Note: no locking manipulations needed */
1391 :
1392 18925336 : if (scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE)
1393 4220838 : heapgettup_pagemode(scan, direction,
1394 4220838 : scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1395 : else
1396 14704498 : heapgettup(scan, direction,
1397 14704498 : scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1398 :
1399 18925336 : if (scan->rs_ctup.t_data == NULL)
1400 121112 : return NULL;
1401 :
1402 : /*
1403 : * if we get here it means we have a new current scan tuple, so point to
1404 : * the proper return buffer and return the tuple.
1405 : */
1406 :
1407 18804224 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1408 :
1409 18804224 : return &scan->rs_ctup;
1410 : }
1411 :
1412 : bool
1413 89861514 : heap_getnextslot(TableScanDesc sscan, ScanDirection direction, TupleTableSlot *slot)
1414 : {
1415 89861514 : HeapScanDesc scan = (HeapScanDesc) sscan;
1416 :
1417 : /* Note: no locking manipulations needed */
1418 :
1419 89861514 : if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1420 88869936 : heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1421 : else
1422 991578 : heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1423 :
1424 89861492 : if (scan->rs_ctup.t_data == NULL)
1425 : {
1426 1412746 : ExecClearTuple(slot);
1427 1412746 : return false;
1428 : }
1429 :
1430 : /*
1431 : * if we get here it means we have a new current scan tuple, so point to
1432 : * the proper return buffer and return the tuple.
1433 : */
1434 :
1435 88448746 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1436 :
1437 88448746 : ExecStoreBufferHeapTuple(&scan->rs_ctup, slot,
1438 : scan->rs_cbuf);
1439 88448746 : return true;
1440 : }
1441 :
1442 : void
1443 1912 : heap_set_tidrange(TableScanDesc sscan, ItemPointer mintid,
1444 : ItemPointer maxtid)
1445 : {
1446 1912 : HeapScanDesc scan = (HeapScanDesc) sscan;
1447 : BlockNumber startBlk;
1448 : BlockNumber numBlks;
1449 : ItemPointerData highestItem;
1450 : ItemPointerData lowestItem;
1451 :
1452 : /*
1453 : * For relations without any pages, we can simply leave the TID range
1454 : * unset. There will be no tuples to scan, therefore no tuples outside
1455 : * the given TID range.
1456 : */
1457 1912 : if (scan->rs_nblocks == 0)
1458 48 : return;
1459 :
1460 : /*
1461 : * Set up some ItemPointers which point to the first and last possible
1462 : * tuples in the heap.
1463 : */
1464 1900 : ItemPointerSet(&highestItem, scan->rs_nblocks - 1, MaxOffsetNumber);
1465 1900 : ItemPointerSet(&lowestItem, 0, FirstOffsetNumber);
1466 :
1467 : /*
1468 : * If the given maximum TID is below the highest possible TID in the
1469 : * relation, then restrict the range to that, otherwise we scan to the end
1470 : * of the relation.
1471 : */
1472 1900 : if (ItemPointerCompare(maxtid, &highestItem) < 0)
1473 132 : ItemPointerCopy(maxtid, &highestItem);
1474 :
1475 : /*
1476 : * If the given minimum TID is above the lowest possible TID in the
1477 : * relation, then restrict the range to only scan for TIDs above that.
1478 : */
1479 1900 : if (ItemPointerCompare(mintid, &lowestItem) > 0)
1480 1762 : ItemPointerCopy(mintid, &lowestItem);
1481 :
1482 : /*
1483 : * Check for an empty range and protect from would be negative results
1484 : * from the numBlks calculation below.
1485 : */
1486 1900 : if (ItemPointerCompare(&highestItem, &lowestItem) < 0)
1487 : {
1488 : /* Set an empty range of blocks to scan */
1489 36 : heap_setscanlimits(sscan, 0, 0);
1490 36 : return;
1491 : }
1492 :
1493 : /*
1494 : * Calculate the first block and the number of blocks we must scan. We
1495 : * could be more aggressive here and perform some more validation to try
1496 : * and further narrow the scope of blocks to scan by checking if the
1497 : * lowestItem has an offset above MaxOffsetNumber. In this case, we could
1498 : * advance startBlk by one. Likewise, if highestItem has an offset of 0
1499 : * we could scan one fewer blocks. However, such an optimization does not
1500 : * seem worth troubling over, currently.
1501 : */
1502 1864 : startBlk = ItemPointerGetBlockNumberNoCheck(&lowestItem);
1503 :
1504 1864 : numBlks = ItemPointerGetBlockNumberNoCheck(&highestItem) -
1505 1864 : ItemPointerGetBlockNumberNoCheck(&lowestItem) + 1;
1506 :
1507 : /* Set the start block and number of blocks to scan */
1508 1864 : heap_setscanlimits(sscan, startBlk, numBlks);
1509 :
1510 : /* Finally, set the TID range in sscan */
1511 1864 : ItemPointerCopy(&lowestItem, &sscan->st.tidrange.rs_mintid);
1512 1864 : ItemPointerCopy(&highestItem, &sscan->st.tidrange.rs_maxtid);
1513 : }
1514 :
1515 : bool
1516 7686 : heap_getnextslot_tidrange(TableScanDesc sscan, ScanDirection direction,
1517 : TupleTableSlot *slot)
1518 : {
1519 7686 : HeapScanDesc scan = (HeapScanDesc) sscan;
1520 7686 : ItemPointer mintid = &sscan->st.tidrange.rs_mintid;
1521 7686 : ItemPointer maxtid = &sscan->st.tidrange.rs_maxtid;
1522 :
1523 : /* Note: no locking manipulations needed */
1524 : for (;;)
1525 : {
1526 7872 : if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1527 7872 : heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1528 : else
1529 0 : heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1530 :
1531 7860 : if (scan->rs_ctup.t_data == NULL)
1532 : {
1533 94 : ExecClearTuple(slot);
1534 94 : return false;
1535 : }
1536 :
1537 : /*
1538 : * heap_set_tidrange will have used heap_setscanlimits to limit the
1539 : * range of pages we scan to only ones that can contain the TID range
1540 : * we're scanning for. Here we must filter out any tuples from these
1541 : * pages that are outside of that range.
1542 : */
1543 7766 : if (ItemPointerCompare(&scan->rs_ctup.t_self, mintid) < 0)
1544 : {
1545 186 : ExecClearTuple(slot);
1546 :
1547 : /*
1548 : * When scanning backwards, the TIDs will be in descending order.
1549 : * Future tuples in this direction will be lower still, so we can
1550 : * just return false to indicate there will be no more tuples.
1551 : */
1552 186 : if (ScanDirectionIsBackward(direction))
1553 0 : return false;
1554 :
1555 186 : continue;
1556 : }
1557 :
1558 : /*
1559 : * Likewise for the final page, we must filter out TIDs greater than
1560 : * maxtid.
1561 : */
1562 7580 : if (ItemPointerCompare(&scan->rs_ctup.t_self, maxtid) > 0)
1563 : {
1564 72 : ExecClearTuple(slot);
1565 :
1566 : /*
1567 : * When scanning forward, the TIDs will be in ascending order.
1568 : * Future tuples in this direction will be higher still, so we can
1569 : * just return false to indicate there will be no more tuples.
1570 : */
1571 72 : if (ScanDirectionIsForward(direction))
1572 72 : return false;
1573 0 : continue;
1574 : }
1575 :
1576 7508 : break;
1577 : }
1578 :
1579 : /*
1580 : * if we get here it means we have a new current scan tuple, so point to
1581 : * the proper return buffer and return the tuple.
1582 : */
1583 7508 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1584 :
1585 7508 : ExecStoreBufferHeapTuple(&scan->rs_ctup, slot, scan->rs_cbuf);
1586 7508 : return true;
1587 : }
1588 :
1589 : /*
1590 : * heap_fetch - retrieve tuple with given tid
1591 : *
1592 : * On entry, tuple->t_self is the TID to fetch. We pin the buffer holding
1593 : * the tuple, fill in the remaining fields of *tuple, and check the tuple
1594 : * against the specified snapshot.
1595 : *
1596 : * If successful (tuple found and passes snapshot time qual), then *userbuf
1597 : * is set to the buffer holding the tuple and true is returned. The caller
1598 : * must unpin the buffer when done with the tuple.
1599 : *
1600 : * If the tuple is not found (ie, item number references a deleted slot),
1601 : * then tuple->t_data is set to NULL, *userbuf is set to InvalidBuffer,
1602 : * and false is returned.
1603 : *
1604 : * If the tuple is found but fails the time qual check, then the behavior
1605 : * depends on the keep_buf parameter. If keep_buf is false, the results
1606 : * are the same as for the tuple-not-found case. If keep_buf is true,
1607 : * then tuple->t_data and *userbuf are returned as for the success case,
1608 : * and again the caller must unpin the buffer; but false is returned.
1609 : *
1610 : * heap_fetch does not follow HOT chains: only the exact TID requested will
1611 : * be fetched.
1612 : *
1613 : * It is somewhat inconsistent that we ereport() on invalid block number but
1614 : * return false on invalid item number. There are a couple of reasons though.
1615 : * One is that the caller can relatively easily check the block number for
1616 : * validity, but cannot check the item number without reading the page
1617 : * himself. Another is that when we are following a t_ctid link, we can be
1618 : * reasonably confident that the page number is valid (since VACUUM shouldn't
1619 : * truncate off the destination page without having killed the referencing
1620 : * tuple first), but the item number might well not be good.
1621 : */
1622 : bool
1623 353378 : heap_fetch(Relation relation,
1624 : Snapshot snapshot,
1625 : HeapTuple tuple,
1626 : Buffer *userbuf,
1627 : bool keep_buf)
1628 : {
1629 353378 : ItemPointer tid = &(tuple->t_self);
1630 : ItemId lp;
1631 : Buffer buffer;
1632 : Page page;
1633 : OffsetNumber offnum;
1634 : bool valid;
1635 :
1636 : /*
1637 : * Fetch and pin the appropriate page of the relation.
1638 : */
1639 353378 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
1640 :
1641 : /*
1642 : * Need share lock on buffer to examine tuple commit status.
1643 : */
1644 353378 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
1645 353378 : page = BufferGetPage(buffer);
1646 :
1647 : /*
1648 : * We'd better check for out-of-range offnum in case of VACUUM since the
1649 : * TID was obtained.
1650 : */
1651 353378 : offnum = ItemPointerGetOffsetNumber(tid);
1652 353378 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1653 : {
1654 6 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1655 6 : ReleaseBuffer(buffer);
1656 6 : *userbuf = InvalidBuffer;
1657 6 : tuple->t_data = NULL;
1658 6 : return false;
1659 : }
1660 :
1661 : /*
1662 : * get the item line pointer corresponding to the requested tid
1663 : */
1664 353372 : lp = PageGetItemId(page, offnum);
1665 :
1666 : /*
1667 : * Must check for deleted tuple.
1668 : */
1669 353372 : if (!ItemIdIsNormal(lp))
1670 : {
1671 650 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1672 650 : ReleaseBuffer(buffer);
1673 650 : *userbuf = InvalidBuffer;
1674 650 : tuple->t_data = NULL;
1675 650 : return false;
1676 : }
1677 :
1678 : /*
1679 : * fill in *tuple fields
1680 : */
1681 352722 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
1682 352722 : tuple->t_len = ItemIdGetLength(lp);
1683 352722 : tuple->t_tableOid = RelationGetRelid(relation);
1684 :
1685 : /*
1686 : * check tuple visibility, then release lock
1687 : */
1688 352722 : valid = HeapTupleSatisfiesVisibility(tuple, snapshot, buffer);
1689 :
1690 352722 : if (valid)
1691 352632 : PredicateLockTID(relation, &(tuple->t_self), snapshot,
1692 352632 : HeapTupleHeaderGetXmin(tuple->t_data));
1693 :
1694 352722 : HeapCheckForSerializableConflictOut(valid, relation, tuple, buffer, snapshot);
1695 :
1696 352722 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1697 :
1698 352722 : if (valid)
1699 : {
1700 : /*
1701 : * All checks passed, so return the tuple as valid. Caller is now
1702 : * responsible for releasing the buffer.
1703 : */
1704 352632 : *userbuf = buffer;
1705 :
1706 352632 : return true;
1707 : }
1708 :
1709 : /* Tuple failed time qual, but maybe caller wants to see it anyway. */
1710 90 : if (keep_buf)
1711 58 : *userbuf = buffer;
1712 : else
1713 : {
1714 32 : ReleaseBuffer(buffer);
1715 32 : *userbuf = InvalidBuffer;
1716 32 : tuple->t_data = NULL;
1717 : }
1718 :
1719 90 : return false;
1720 : }
1721 :
1722 : /*
1723 : * heap_hot_search_buffer - search HOT chain for tuple satisfying snapshot
1724 : *
1725 : * On entry, *tid is the TID of a tuple (either a simple tuple, or the root
1726 : * of a HOT chain), and buffer is the buffer holding this tuple. We search
1727 : * for the first chain member satisfying the given snapshot. If one is
1728 : * found, we update *tid to reference that tuple's offset number, and
1729 : * return true. If no match, return false without modifying *tid.
1730 : *
1731 : * heapTuple is a caller-supplied buffer. When a match is found, we return
1732 : * the tuple here, in addition to updating *tid. If no match is found, the
1733 : * contents of this buffer on return are undefined.
1734 : *
1735 : * If all_dead is not NULL, we check non-visible tuples to see if they are
1736 : * globally dead; *all_dead is set true if all members of the HOT chain
1737 : * are vacuumable, false if not.
1738 : *
1739 : * Unlike heap_fetch, the caller must already have pin and (at least) share
1740 : * lock on the buffer; it is still pinned/locked at exit.
1741 : */
1742 : bool
1743 41689684 : heap_hot_search_buffer(ItemPointer tid, Relation relation, Buffer buffer,
1744 : Snapshot snapshot, HeapTuple heapTuple,
1745 : bool *all_dead, bool first_call)
1746 : {
1747 41689684 : Page page = BufferGetPage(buffer);
1748 41689684 : TransactionId prev_xmax = InvalidTransactionId;
1749 : BlockNumber blkno;
1750 : OffsetNumber offnum;
1751 : bool at_chain_start;
1752 : bool valid;
1753 : bool skip;
1754 41689684 : GlobalVisState *vistest = NULL;
1755 :
1756 : /* If this is not the first call, previous call returned a (live!) tuple */
1757 41689684 : if (all_dead)
1758 35839132 : *all_dead = first_call;
1759 :
1760 41689684 : blkno = ItemPointerGetBlockNumber(tid);
1761 41689684 : offnum = ItemPointerGetOffsetNumber(tid);
1762 41689684 : at_chain_start = first_call;
1763 41689684 : skip = !first_call;
1764 :
1765 : /* XXX: we should assert that a snapshot is pushed or registered */
1766 : Assert(TransactionIdIsValid(RecentXmin));
1767 : Assert(BufferGetBlockNumber(buffer) == blkno);
1768 :
1769 : /* Scan through possible multiple members of HOT-chain */
1770 : for (;;)
1771 2976110 : {
1772 : ItemId lp;
1773 :
1774 : /* check for bogus TID */
1775 44665794 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1776 : break;
1777 :
1778 44665794 : lp = PageGetItemId(page, offnum);
1779 :
1780 : /* check for unused, dead, or redirected items */
1781 44665794 : if (!ItemIdIsNormal(lp))
1782 : {
1783 : /* We should only see a redirect at start of chain */
1784 1655842 : if (ItemIdIsRedirected(lp) && at_chain_start)
1785 : {
1786 : /* Follow the redirect */
1787 877780 : offnum = ItemIdGetRedirect(lp);
1788 877780 : at_chain_start = false;
1789 877780 : continue;
1790 : }
1791 : /* else must be end of chain */
1792 778062 : break;
1793 : }
1794 :
1795 : /*
1796 : * Update heapTuple to point to the element of the HOT chain we're
1797 : * currently investigating. Having t_self set correctly is important
1798 : * because the SSI checks and the *Satisfies routine for historical
1799 : * MVCC snapshots need the correct tid to decide about the visibility.
1800 : */
1801 43009952 : heapTuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
1802 43009952 : heapTuple->t_len = ItemIdGetLength(lp);
1803 43009952 : heapTuple->t_tableOid = RelationGetRelid(relation);
1804 43009952 : ItemPointerSet(&heapTuple->t_self, blkno, offnum);
1805 :
1806 : /*
1807 : * Shouldn't see a HEAP_ONLY tuple at chain start.
1808 : */
1809 43009952 : if (at_chain_start && HeapTupleIsHeapOnly(heapTuple))
1810 0 : break;
1811 :
1812 : /*
1813 : * The xmin should match the previous xmax value, else chain is
1814 : * broken.
1815 : */
1816 45108282 : if (TransactionIdIsValid(prev_xmax) &&
1817 2098330 : !TransactionIdEquals(prev_xmax,
1818 : HeapTupleHeaderGetXmin(heapTuple->t_data)))
1819 0 : break;
1820 :
1821 : /*
1822 : * When first_call is true (and thus, skip is initially false) we'll
1823 : * return the first tuple we find. But on later passes, heapTuple
1824 : * will initially be pointing to the tuple we returned last time.
1825 : * Returning it again would be incorrect (and would loop forever), so
1826 : * we skip it and return the next match we find.
1827 : */
1828 43009952 : if (!skip)
1829 : {
1830 : /* If it's visible per the snapshot, we must return it */
1831 42851814 : valid = HeapTupleSatisfiesVisibility(heapTuple, snapshot, buffer);
1832 42851814 : HeapCheckForSerializableConflictOut(valid, relation, heapTuple,
1833 : buffer, snapshot);
1834 :
1835 42851804 : if (valid)
1836 : {
1837 28564690 : ItemPointerSetOffsetNumber(tid, offnum);
1838 28564690 : PredicateLockTID(relation, &heapTuple->t_self, snapshot,
1839 28564690 : HeapTupleHeaderGetXmin(heapTuple->t_data));
1840 28564690 : if (all_dead)
1841 23279812 : *all_dead = false;
1842 28564690 : return true;
1843 : }
1844 : }
1845 14445252 : skip = false;
1846 :
1847 : /*
1848 : * If we can't see it, maybe no one else can either. At caller
1849 : * request, check whether all chain members are dead to all
1850 : * transactions.
1851 : *
1852 : * Note: if you change the criterion here for what is "dead", fix the
1853 : * planner's get_actual_variable_range() function to match.
1854 : */
1855 14445252 : if (all_dead && *all_dead)
1856 : {
1857 12779070 : if (!vistest)
1858 12553490 : vistest = GlobalVisTestFor(relation);
1859 :
1860 12779070 : if (!HeapTupleIsSurelyDead(heapTuple, vistest))
1861 12082684 : *all_dead = false;
1862 : }
1863 :
1864 : /*
1865 : * Check to see if HOT chain continues past this tuple; if so fetch
1866 : * the next offnum and loop around.
1867 : */
1868 14445252 : if (HeapTupleIsHotUpdated(heapTuple))
1869 : {
1870 : Assert(ItemPointerGetBlockNumber(&heapTuple->t_data->t_ctid) ==
1871 : blkno);
1872 2098330 : offnum = ItemPointerGetOffsetNumber(&heapTuple->t_data->t_ctid);
1873 2098330 : at_chain_start = false;
1874 2098330 : prev_xmax = HeapTupleHeaderGetUpdateXid(heapTuple->t_data);
1875 : }
1876 : else
1877 12346922 : break; /* end of chain */
1878 : }
1879 :
1880 13124984 : return false;
1881 : }
1882 :
1883 : /*
1884 : * heap_get_latest_tid - get the latest tid of a specified tuple
1885 : *
1886 : * Actually, this gets the latest version that is visible according to the
1887 : * scan's snapshot. Create a scan using SnapshotDirty to get the very latest,
1888 : * possibly uncommitted version.
1889 : *
1890 : * *tid is both an input and an output parameter: it is updated to
1891 : * show the latest version of the row. Note that it will not be changed
1892 : * if no version of the row passes the snapshot test.
1893 : */
1894 : void
1895 300 : heap_get_latest_tid(TableScanDesc sscan,
1896 : ItemPointer tid)
1897 : {
1898 300 : Relation relation = sscan->rs_rd;
1899 300 : Snapshot snapshot = sscan->rs_snapshot;
1900 : ItemPointerData ctid;
1901 : TransactionId priorXmax;
1902 :
1903 : /*
1904 : * table_tuple_get_latest_tid() verified that the passed in tid is valid.
1905 : * Assume that t_ctid links are valid however - there shouldn't be invalid
1906 : * ones in the table.
1907 : */
1908 : Assert(ItemPointerIsValid(tid));
1909 :
1910 : /*
1911 : * Loop to chase down t_ctid links. At top of loop, ctid is the tuple we
1912 : * need to examine, and *tid is the TID we will return if ctid turns out
1913 : * to be bogus.
1914 : *
1915 : * Note that we will loop until we reach the end of the t_ctid chain.
1916 : * Depending on the snapshot passed, there might be at most one visible
1917 : * version of the row, but we don't try to optimize for that.
1918 : */
1919 300 : ctid = *tid;
1920 300 : priorXmax = InvalidTransactionId; /* cannot check first XMIN */
1921 : for (;;)
1922 90 : {
1923 : Buffer buffer;
1924 : Page page;
1925 : OffsetNumber offnum;
1926 : ItemId lp;
1927 : HeapTupleData tp;
1928 : bool valid;
1929 :
1930 : /*
1931 : * Read, pin, and lock the page.
1932 : */
1933 390 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(&ctid));
1934 390 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
1935 390 : page = BufferGetPage(buffer);
1936 :
1937 : /*
1938 : * Check for bogus item number. This is not treated as an error
1939 : * condition because it can happen while following a t_ctid link. We
1940 : * just assume that the prior tid is OK and return it unchanged.
1941 : */
1942 390 : offnum = ItemPointerGetOffsetNumber(&ctid);
1943 390 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1944 : {
1945 0 : UnlockReleaseBuffer(buffer);
1946 0 : break;
1947 : }
1948 390 : lp = PageGetItemId(page, offnum);
1949 390 : if (!ItemIdIsNormal(lp))
1950 : {
1951 0 : UnlockReleaseBuffer(buffer);
1952 0 : break;
1953 : }
1954 :
1955 : /* OK to access the tuple */
1956 390 : tp.t_self = ctid;
1957 390 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
1958 390 : tp.t_len = ItemIdGetLength(lp);
1959 390 : tp.t_tableOid = RelationGetRelid(relation);
1960 :
1961 : /*
1962 : * After following a t_ctid link, we might arrive at an unrelated
1963 : * tuple. Check for XMIN match.
1964 : */
1965 480 : if (TransactionIdIsValid(priorXmax) &&
1966 90 : !TransactionIdEquals(priorXmax, HeapTupleHeaderGetXmin(tp.t_data)))
1967 : {
1968 0 : UnlockReleaseBuffer(buffer);
1969 0 : break;
1970 : }
1971 :
1972 : /*
1973 : * Check tuple visibility; if visible, set it as the new result
1974 : * candidate.
1975 : */
1976 390 : valid = HeapTupleSatisfiesVisibility(&tp, snapshot, buffer);
1977 390 : HeapCheckForSerializableConflictOut(valid, relation, &tp, buffer, snapshot);
1978 390 : if (valid)
1979 276 : *tid = ctid;
1980 :
1981 : /*
1982 : * If there's a valid t_ctid link, follow it, else we're done.
1983 : */
1984 552 : if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
1985 276 : HeapTupleHeaderIsOnlyLocked(tp.t_data) ||
1986 228 : HeapTupleHeaderIndicatesMovedPartitions(tp.t_data) ||
1987 114 : ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
1988 : {
1989 300 : UnlockReleaseBuffer(buffer);
1990 300 : break;
1991 : }
1992 :
1993 90 : ctid = tp.t_data->t_ctid;
1994 90 : priorXmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
1995 90 : UnlockReleaseBuffer(buffer);
1996 : } /* end of loop */
1997 300 : }
1998 :
1999 :
2000 : /*
2001 : * UpdateXmaxHintBits - update tuple hint bits after xmax transaction ends
2002 : *
2003 : * This is called after we have waited for the XMAX transaction to terminate.
2004 : * If the transaction aborted, we guarantee the XMAX_INVALID hint bit will
2005 : * be set on exit. If the transaction committed, we set the XMAX_COMMITTED
2006 : * hint bit if possible --- but beware that that may not yet be possible,
2007 : * if the transaction committed asynchronously.
2008 : *
2009 : * Note that if the transaction was a locker only, we set HEAP_XMAX_INVALID
2010 : * even if it commits.
2011 : *
2012 : * Hence callers should look only at XMAX_INVALID.
2013 : *
2014 : * Note this is not allowed for tuples whose xmax is a multixact.
2015 : */
2016 : static void
2017 350 : UpdateXmaxHintBits(HeapTupleHeader tuple, Buffer buffer, TransactionId xid)
2018 : {
2019 : Assert(TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple), xid));
2020 : Assert(!(tuple->t_infomask & HEAP_XMAX_IS_MULTI));
2021 :
2022 350 : if (!(tuple->t_infomask & (HEAP_XMAX_COMMITTED | HEAP_XMAX_INVALID)))
2023 : {
2024 634 : if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask) &&
2025 284 : TransactionIdDidCommit(xid))
2026 232 : HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_COMMITTED,
2027 : xid);
2028 : else
2029 118 : HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_INVALID,
2030 : InvalidTransactionId);
2031 : }
2032 350 : }
2033 :
2034 :
2035 : /*
2036 : * GetBulkInsertState - prepare status object for a bulk insert
2037 : */
2038 : BulkInsertState
2039 4468 : GetBulkInsertState(void)
2040 : {
2041 : BulkInsertState bistate;
2042 :
2043 4468 : bistate = (BulkInsertState) palloc(sizeof(BulkInsertStateData));
2044 4468 : bistate->strategy = GetAccessStrategy(BAS_BULKWRITE);
2045 4468 : bistate->current_buf = InvalidBuffer;
2046 4468 : bistate->next_free = InvalidBlockNumber;
2047 4468 : bistate->last_free = InvalidBlockNumber;
2048 4468 : bistate->already_extended_by = 0;
2049 4468 : return bistate;
2050 : }
2051 :
2052 : /*
2053 : * FreeBulkInsertState - clean up after finishing a bulk insert
2054 : */
2055 : void
2056 4190 : FreeBulkInsertState(BulkInsertState bistate)
2057 : {
2058 4190 : if (bistate->current_buf != InvalidBuffer)
2059 3450 : ReleaseBuffer(bistate->current_buf);
2060 4190 : FreeAccessStrategy(bistate->strategy);
2061 4190 : pfree(bistate);
2062 4190 : }
2063 :
2064 : /*
2065 : * ReleaseBulkInsertStatePin - release a buffer currently held in bistate
2066 : */
2067 : void
2068 161512 : ReleaseBulkInsertStatePin(BulkInsertState bistate)
2069 : {
2070 161512 : if (bistate->current_buf != InvalidBuffer)
2071 60042 : ReleaseBuffer(bistate->current_buf);
2072 161512 : bistate->current_buf = InvalidBuffer;
2073 :
2074 : /*
2075 : * Despite the name, we also reset bulk relation extension state.
2076 : * Otherwise we can end up erroring out due to looking for free space in
2077 : * ->next_free of one partition, even though ->next_free was set when
2078 : * extending another partition. It could obviously also be bad for
2079 : * efficiency to look at existing blocks at offsets from another
2080 : * partition, even if we don't error out.
2081 : */
2082 161512 : bistate->next_free = InvalidBlockNumber;
2083 161512 : bistate->last_free = InvalidBlockNumber;
2084 161512 : }
2085 :
2086 :
2087 : /*
2088 : * heap_insert - insert tuple into a heap
2089 : *
2090 : * The new tuple is stamped with current transaction ID and the specified
2091 : * command ID.
2092 : *
2093 : * See table_tuple_insert for comments about most of the input flags, except
2094 : * that this routine directly takes a tuple rather than a slot.
2095 : *
2096 : * There's corresponding HEAP_INSERT_ options to all the TABLE_INSERT_
2097 : * options, and there additionally is HEAP_INSERT_SPECULATIVE which is used to
2098 : * implement table_tuple_insert_speculative().
2099 : *
2100 : * On return the header fields of *tup are updated to match the stored tuple;
2101 : * in particular tup->t_self receives the actual TID where the tuple was
2102 : * stored. But note that any toasting of fields within the tuple data is NOT
2103 : * reflected into *tup.
2104 : */
2105 : void
2106 15846670 : heap_insert(Relation relation, HeapTuple tup, CommandId cid,
2107 : int options, BulkInsertState bistate)
2108 : {
2109 15846670 : TransactionId xid = GetCurrentTransactionId();
2110 : HeapTuple heaptup;
2111 : Buffer buffer;
2112 15846660 : Buffer vmbuffer = InvalidBuffer;
2113 15846660 : bool all_visible_cleared = false;
2114 :
2115 : /* Cheap, simplistic check that the tuple matches the rel's rowtype. */
2116 : Assert(HeapTupleHeaderGetNatts(tup->t_data) <=
2117 : RelationGetNumberOfAttributes(relation));
2118 :
2119 : /*
2120 : * Fill in tuple header fields and toast the tuple if necessary.
2121 : *
2122 : * Note: below this point, heaptup is the data we actually intend to store
2123 : * into the relation; tup is the caller's original untoasted data.
2124 : */
2125 15846660 : heaptup = heap_prepare_insert(relation, tup, xid, cid, options);
2126 :
2127 : /*
2128 : * Find buffer to insert this tuple into. If the page is all visible,
2129 : * this will also pin the requisite visibility map page.
2130 : */
2131 15846660 : buffer = RelationGetBufferForTuple(relation, heaptup->t_len,
2132 : InvalidBuffer, options, bistate,
2133 : &vmbuffer, NULL,
2134 : 0);
2135 :
2136 : /*
2137 : * We're about to do the actual insert -- but check for conflict first, to
2138 : * avoid possibly having to roll back work we've just done.
2139 : *
2140 : * This is safe without a recheck as long as there is no possibility of
2141 : * another process scanning the page between this check and the insert
2142 : * being visible to the scan (i.e., an exclusive buffer content lock is
2143 : * continuously held from this point until the tuple insert is visible).
2144 : *
2145 : * For a heap insert, we only need to check for table-level SSI locks. Our
2146 : * new tuple can't possibly conflict with existing tuple locks, and heap
2147 : * page locks are only consolidated versions of tuple locks; they do not
2148 : * lock "gaps" as index page locks do. So we don't need to specify a
2149 : * buffer when making the call, which makes for a faster check.
2150 : */
2151 15846660 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2152 :
2153 : /* NO EREPORT(ERROR) from here till changes are logged */
2154 15846636 : START_CRIT_SECTION();
2155 :
2156 15846636 : RelationPutHeapTuple(relation, buffer, heaptup,
2157 15846636 : (options & HEAP_INSERT_SPECULATIVE) != 0);
2158 :
2159 15846636 : if (PageIsAllVisible(BufferGetPage(buffer)))
2160 : {
2161 13358 : all_visible_cleared = true;
2162 13358 : PageClearAllVisible(BufferGetPage(buffer));
2163 13358 : visibilitymap_clear(relation,
2164 13358 : ItemPointerGetBlockNumber(&(heaptup->t_self)),
2165 : vmbuffer, VISIBILITYMAP_VALID_BITS);
2166 : }
2167 :
2168 : /*
2169 : * XXX Should we set PageSetPrunable on this page ?
2170 : *
2171 : * The inserting transaction may eventually abort thus making this tuple
2172 : * DEAD and hence available for pruning. Though we don't want to optimize
2173 : * for aborts, if no other tuple in this page is UPDATEd/DELETEd, the
2174 : * aborted tuple will never be pruned until next vacuum is triggered.
2175 : *
2176 : * If you do add PageSetPrunable here, add it in heap_xlog_insert too.
2177 : */
2178 :
2179 15846636 : MarkBufferDirty(buffer);
2180 :
2181 : /* XLOG stuff */
2182 15846636 : if (RelationNeedsWAL(relation))
2183 : {
2184 : xl_heap_insert xlrec;
2185 : xl_heap_header xlhdr;
2186 : XLogRecPtr recptr;
2187 13726868 : Page page = BufferGetPage(buffer);
2188 13726868 : uint8 info = XLOG_HEAP_INSERT;
2189 13726868 : int bufflags = 0;
2190 :
2191 : /*
2192 : * If this is a catalog, we need to transmit combo CIDs to properly
2193 : * decode, so log that as well.
2194 : */
2195 13726868 : if (RelationIsAccessibleInLogicalDecoding(relation))
2196 6476 : log_heap_new_cid(relation, heaptup);
2197 :
2198 : /*
2199 : * If this is the single and first tuple on page, we can reinit the
2200 : * page instead of restoring the whole thing. Set flag, and hide
2201 : * buffer references from XLogInsert.
2202 : */
2203 13902832 : if (ItemPointerGetOffsetNumber(&(heaptup->t_self)) == FirstOffsetNumber &&
2204 175964 : PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
2205 : {
2206 174566 : info |= XLOG_HEAP_INIT_PAGE;
2207 174566 : bufflags |= REGBUF_WILL_INIT;
2208 : }
2209 :
2210 13726868 : xlrec.offnum = ItemPointerGetOffsetNumber(&heaptup->t_self);
2211 13726868 : xlrec.flags = 0;
2212 13726868 : if (all_visible_cleared)
2213 13352 : xlrec.flags |= XLH_INSERT_ALL_VISIBLE_CLEARED;
2214 13726868 : if (options & HEAP_INSERT_SPECULATIVE)
2215 4110 : xlrec.flags |= XLH_INSERT_IS_SPECULATIVE;
2216 : Assert(ItemPointerGetBlockNumber(&heaptup->t_self) == BufferGetBlockNumber(buffer));
2217 :
2218 : /*
2219 : * For logical decoding, we need the tuple even if we're doing a full
2220 : * page write, so make sure it's included even if we take a full-page
2221 : * image. (XXX We could alternatively store a pointer into the FPW).
2222 : */
2223 13726868 : if (RelationIsLogicallyLogged(relation) &&
2224 499750 : !(options & HEAP_INSERT_NO_LOGICAL))
2225 : {
2226 499696 : xlrec.flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
2227 499696 : bufflags |= REGBUF_KEEP_DATA;
2228 :
2229 499696 : if (IsToastRelation(relation))
2230 3572 : xlrec.flags |= XLH_INSERT_ON_TOAST_RELATION;
2231 : }
2232 :
2233 13726868 : XLogBeginInsert();
2234 13726868 : XLogRegisterData(&xlrec, SizeOfHeapInsert);
2235 :
2236 13726868 : xlhdr.t_infomask2 = heaptup->t_data->t_infomask2;
2237 13726868 : xlhdr.t_infomask = heaptup->t_data->t_infomask;
2238 13726868 : xlhdr.t_hoff = heaptup->t_data->t_hoff;
2239 :
2240 : /*
2241 : * note we mark xlhdr as belonging to buffer; if XLogInsert decides to
2242 : * write the whole page to the xlog, we don't need to store
2243 : * xl_heap_header in the xlog.
2244 : */
2245 13726868 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD | bufflags);
2246 13726868 : XLogRegisterBufData(0, &xlhdr, SizeOfHeapHeader);
2247 : /* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
2248 13726868 : XLogRegisterBufData(0,
2249 13726868 : (char *) heaptup->t_data + SizeofHeapTupleHeader,
2250 13726868 : heaptup->t_len - SizeofHeapTupleHeader);
2251 :
2252 : /* filtering by origin on a row level is much more efficient */
2253 13726868 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
2254 :
2255 13726868 : recptr = XLogInsert(RM_HEAP_ID, info);
2256 :
2257 13726868 : PageSetLSN(page, recptr);
2258 : }
2259 :
2260 15846636 : END_CRIT_SECTION();
2261 :
2262 15846636 : UnlockReleaseBuffer(buffer);
2263 15846636 : if (vmbuffer != InvalidBuffer)
2264 13934 : ReleaseBuffer(vmbuffer);
2265 :
2266 : /*
2267 : * If tuple is cachable, mark it for invalidation from the caches in case
2268 : * we abort. Note it is OK to do this after releasing the buffer, because
2269 : * the heaptup data structure is all in local memory, not in the shared
2270 : * buffer.
2271 : */
2272 15846636 : CacheInvalidateHeapTuple(relation, heaptup, NULL);
2273 :
2274 : /* Note: speculative insertions are counted too, even if aborted later */
2275 15846636 : pgstat_count_heap_insert(relation, 1);
2276 :
2277 : /*
2278 : * If heaptup is a private copy, release it. Don't forget to copy t_self
2279 : * back to the caller's image, too.
2280 : */
2281 15846636 : if (heaptup != tup)
2282 : {
2283 34268 : tup->t_self = heaptup->t_self;
2284 34268 : heap_freetuple(heaptup);
2285 : }
2286 15846636 : }
2287 :
2288 : /*
2289 : * Subroutine for heap_insert(). Prepares a tuple for insertion. This sets the
2290 : * tuple header fields and toasts the tuple if necessary. Returns a toasted
2291 : * version of the tuple if it was toasted, or the original tuple if not. Note
2292 : * that in any case, the header fields are also set in the original tuple.
2293 : */
2294 : static HeapTuple
2295 18691594 : heap_prepare_insert(Relation relation, HeapTuple tup, TransactionId xid,
2296 : CommandId cid, int options)
2297 : {
2298 : /*
2299 : * To allow parallel inserts, we need to ensure that they are safe to be
2300 : * performed in workers. We have the infrastructure to allow parallel
2301 : * inserts in general except for the cases where inserts generate a new
2302 : * CommandId (eg. inserts into a table having a foreign key column).
2303 : */
2304 18691594 : if (IsParallelWorker())
2305 0 : ereport(ERROR,
2306 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
2307 : errmsg("cannot insert tuples in a parallel worker")));
2308 :
2309 18691594 : tup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
2310 18691594 : tup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
2311 18691594 : tup->t_data->t_infomask |= HEAP_XMAX_INVALID;
2312 18691594 : HeapTupleHeaderSetXmin(tup->t_data, xid);
2313 18691594 : if (options & HEAP_INSERT_FROZEN)
2314 204222 : HeapTupleHeaderSetXminFrozen(tup->t_data);
2315 :
2316 18691594 : HeapTupleHeaderSetCmin(tup->t_data, cid);
2317 18691594 : HeapTupleHeaderSetXmax(tup->t_data, 0); /* for cleanliness */
2318 18691594 : tup->t_tableOid = RelationGetRelid(relation);
2319 :
2320 : /*
2321 : * If the new tuple is too big for storage or contains already toasted
2322 : * out-of-line attributes from some other relation, invoke the toaster.
2323 : */
2324 18691594 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
2325 58212 : relation->rd_rel->relkind != RELKIND_MATVIEW)
2326 : {
2327 : /* toast table entries should never be recursively toasted */
2328 : Assert(!HeapTupleHasExternal(tup));
2329 58116 : return tup;
2330 : }
2331 18633478 : else if (HeapTupleHasExternal(tup) || tup->t_len > TOAST_TUPLE_THRESHOLD)
2332 34350 : return heap_toast_insert_or_update(relation, tup, NULL, options);
2333 : else
2334 18599128 : return tup;
2335 : }
2336 :
2337 : /*
2338 : * Helper for heap_multi_insert() that computes the number of entire pages
2339 : * that inserting the remaining heaptuples requires. Used to determine how
2340 : * much the relation needs to be extended by.
2341 : */
2342 : static int
2343 684574 : heap_multi_insert_pages(HeapTuple *heaptuples, int done, int ntuples, Size saveFreeSpace)
2344 : {
2345 684574 : size_t page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
2346 684574 : int npages = 1;
2347 :
2348 4718388 : for (int i = done; i < ntuples; i++)
2349 : {
2350 4033814 : size_t tup_sz = sizeof(ItemIdData) + MAXALIGN(heaptuples[i]->t_len);
2351 :
2352 4033814 : if (page_avail < tup_sz)
2353 : {
2354 31214 : npages++;
2355 31214 : page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
2356 : }
2357 4033814 : page_avail -= tup_sz;
2358 : }
2359 :
2360 684574 : return npages;
2361 : }
2362 :
2363 : /*
2364 : * heap_multi_insert - insert multiple tuples into a heap
2365 : *
2366 : * This is like heap_insert(), but inserts multiple tuples in one operation.
2367 : * That's faster than calling heap_insert() in a loop, because when multiple
2368 : * tuples can be inserted on a single page, we can write just a single WAL
2369 : * record covering all of them, and only need to lock/unlock the page once.
2370 : *
2371 : * Note: this leaks memory into the current memory context. You can create a
2372 : * temporary context before calling this, if that's a problem.
2373 : */
2374 : void
2375 672182 : heap_multi_insert(Relation relation, TupleTableSlot **slots, int ntuples,
2376 : CommandId cid, int options, BulkInsertState bistate)
2377 : {
2378 672182 : TransactionId xid = GetCurrentTransactionId();
2379 : HeapTuple *heaptuples;
2380 : int i;
2381 : int ndone;
2382 : PGAlignedBlock scratch;
2383 : Page page;
2384 672182 : Buffer vmbuffer = InvalidBuffer;
2385 : bool needwal;
2386 : Size saveFreeSpace;
2387 672182 : bool need_tuple_data = RelationIsLogicallyLogged(relation);
2388 672182 : bool need_cids = RelationIsAccessibleInLogicalDecoding(relation);
2389 672182 : bool starting_with_empty_page = false;
2390 672182 : int npages = 0;
2391 672182 : int npages_used = 0;
2392 :
2393 : /* currently not needed (thus unsupported) for heap_multi_insert() */
2394 : Assert(!(options & HEAP_INSERT_NO_LOGICAL));
2395 :
2396 672182 : needwal = RelationNeedsWAL(relation);
2397 672182 : saveFreeSpace = RelationGetTargetPageFreeSpace(relation,
2398 : HEAP_DEFAULT_FILLFACTOR);
2399 :
2400 : /* Toast and set header data in all the slots */
2401 672182 : heaptuples = palloc(ntuples * sizeof(HeapTuple));
2402 3517116 : for (i = 0; i < ntuples; i++)
2403 : {
2404 : HeapTuple tuple;
2405 :
2406 2844934 : tuple = ExecFetchSlotHeapTuple(slots[i], true, NULL);
2407 2844934 : slots[i]->tts_tableOid = RelationGetRelid(relation);
2408 2844934 : tuple->t_tableOid = slots[i]->tts_tableOid;
2409 2844934 : heaptuples[i] = heap_prepare_insert(relation, tuple, xid, cid,
2410 : options);
2411 : }
2412 :
2413 : /*
2414 : * We're about to do the actual inserts -- but check for conflict first,
2415 : * to minimize the possibility of having to roll back work we've just
2416 : * done.
2417 : *
2418 : * A check here does not definitively prevent a serialization anomaly;
2419 : * that check MUST be done at least past the point of acquiring an
2420 : * exclusive buffer content lock on every buffer that will be affected,
2421 : * and MAY be done after all inserts are reflected in the buffers and
2422 : * those locks are released; otherwise there is a race condition. Since
2423 : * multiple buffers can be locked and unlocked in the loop below, and it
2424 : * would not be feasible to identify and lock all of those buffers before
2425 : * the loop, we must do a final check at the end.
2426 : *
2427 : * The check here could be omitted with no loss of correctness; it is
2428 : * present strictly as an optimization.
2429 : *
2430 : * For heap inserts, we only need to check for table-level SSI locks. Our
2431 : * new tuples can't possibly conflict with existing tuple locks, and heap
2432 : * page locks are only consolidated versions of tuple locks; they do not
2433 : * lock "gaps" as index page locks do. So we don't need to specify a
2434 : * buffer when making the call, which makes for a faster check.
2435 : */
2436 672182 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2437 :
2438 672182 : ndone = 0;
2439 1373074 : while (ndone < ntuples)
2440 : {
2441 : Buffer buffer;
2442 700892 : bool all_visible_cleared = false;
2443 700892 : bool all_frozen_set = false;
2444 : int nthispage;
2445 :
2446 700892 : CHECK_FOR_INTERRUPTS();
2447 :
2448 : /*
2449 : * Compute number of pages needed to fit the to-be-inserted tuples in
2450 : * the worst case. This will be used to determine how much to extend
2451 : * the relation by in RelationGetBufferForTuple(), if needed. If we
2452 : * filled a prior page from scratch, we can just update our last
2453 : * computation, but if we started with a partially filled page,
2454 : * recompute from scratch, the number of potentially required pages
2455 : * can vary due to tuples needing to fit onto the page, page headers
2456 : * etc.
2457 : */
2458 700892 : if (ndone == 0 || !starting_with_empty_page)
2459 : {
2460 684574 : npages = heap_multi_insert_pages(heaptuples, ndone, ntuples,
2461 : saveFreeSpace);
2462 684574 : npages_used = 0;
2463 : }
2464 : else
2465 16318 : npages_used++;
2466 :
2467 : /*
2468 : * Find buffer where at least the next tuple will fit. If the page is
2469 : * all-visible, this will also pin the requisite visibility map page.
2470 : *
2471 : * Also pin visibility map page if COPY FREEZE inserts tuples into an
2472 : * empty page. See all_frozen_set below.
2473 : */
2474 700892 : buffer = RelationGetBufferForTuple(relation, heaptuples[ndone]->t_len,
2475 : InvalidBuffer, options, bistate,
2476 : &vmbuffer, NULL,
2477 : npages - npages_used);
2478 700892 : page = BufferGetPage(buffer);
2479 :
2480 700892 : starting_with_empty_page = PageGetMaxOffsetNumber(page) == 0;
2481 :
2482 700892 : if (starting_with_empty_page && (options & HEAP_INSERT_FROZEN))
2483 3322 : all_frozen_set = true;
2484 :
2485 : /* NO EREPORT(ERROR) from here till changes are logged */
2486 700892 : START_CRIT_SECTION();
2487 :
2488 : /*
2489 : * RelationGetBufferForTuple has ensured that the first tuple fits.
2490 : * Put that on the page, and then as many other tuples as fit.
2491 : */
2492 700892 : RelationPutHeapTuple(relation, buffer, heaptuples[ndone], false);
2493 :
2494 : /*
2495 : * For logical decoding we need combo CIDs to properly decode the
2496 : * catalog.
2497 : */
2498 700892 : if (needwal && need_cids)
2499 9568 : log_heap_new_cid(relation, heaptuples[ndone]);
2500 :
2501 2844934 : for (nthispage = 1; ndone + nthispage < ntuples; nthispage++)
2502 : {
2503 2172752 : HeapTuple heaptup = heaptuples[ndone + nthispage];
2504 :
2505 2172752 : if (PageGetHeapFreeSpace(page) < MAXALIGN(heaptup->t_len) + saveFreeSpace)
2506 28710 : break;
2507 :
2508 2144042 : RelationPutHeapTuple(relation, buffer, heaptup, false);
2509 :
2510 : /*
2511 : * For logical decoding we need combo CIDs to properly decode the
2512 : * catalog.
2513 : */
2514 2144042 : if (needwal && need_cids)
2515 9154 : log_heap_new_cid(relation, heaptup);
2516 : }
2517 :
2518 : /*
2519 : * If the page is all visible, need to clear that, unless we're only
2520 : * going to add further frozen rows to it.
2521 : *
2522 : * If we're only adding already frozen rows to a previously empty
2523 : * page, mark it as all-visible.
2524 : */
2525 700892 : if (PageIsAllVisible(page) && !(options & HEAP_INSERT_FROZEN))
2526 : {
2527 5452 : all_visible_cleared = true;
2528 5452 : PageClearAllVisible(page);
2529 5452 : visibilitymap_clear(relation,
2530 : BufferGetBlockNumber(buffer),
2531 : vmbuffer, VISIBILITYMAP_VALID_BITS);
2532 : }
2533 695440 : else if (all_frozen_set)
2534 3322 : PageSetAllVisible(page);
2535 :
2536 : /*
2537 : * XXX Should we set PageSetPrunable on this page ? See heap_insert()
2538 : */
2539 :
2540 700892 : MarkBufferDirty(buffer);
2541 :
2542 : /* XLOG stuff */
2543 700892 : if (needwal)
2544 : {
2545 : XLogRecPtr recptr;
2546 : xl_heap_multi_insert *xlrec;
2547 693250 : uint8 info = XLOG_HEAP2_MULTI_INSERT;
2548 : char *tupledata;
2549 : int totaldatalen;
2550 693250 : char *scratchptr = scratch.data;
2551 : bool init;
2552 693250 : int bufflags = 0;
2553 :
2554 : /*
2555 : * If the page was previously empty, we can reinit the page
2556 : * instead of restoring the whole thing.
2557 : */
2558 693250 : init = starting_with_empty_page;
2559 :
2560 : /* allocate xl_heap_multi_insert struct from the scratch area */
2561 693250 : xlrec = (xl_heap_multi_insert *) scratchptr;
2562 693250 : scratchptr += SizeOfHeapMultiInsert;
2563 :
2564 : /*
2565 : * Allocate offsets array. Unless we're reinitializing the page,
2566 : * in that case the tuples are stored in order starting at
2567 : * FirstOffsetNumber and we don't need to store the offsets
2568 : * explicitly.
2569 : */
2570 693250 : if (!init)
2571 668216 : scratchptr += nthispage * sizeof(OffsetNumber);
2572 :
2573 : /* the rest of the scratch space is used for tuple data */
2574 693250 : tupledata = scratchptr;
2575 :
2576 : /* check that the mutually exclusive flags are not both set */
2577 : Assert(!(all_visible_cleared && all_frozen_set));
2578 :
2579 693250 : xlrec->flags = 0;
2580 693250 : if (all_visible_cleared)
2581 5452 : xlrec->flags = XLH_INSERT_ALL_VISIBLE_CLEARED;
2582 693250 : if (all_frozen_set)
2583 26 : xlrec->flags = XLH_INSERT_ALL_FROZEN_SET;
2584 :
2585 693250 : xlrec->ntuples = nthispage;
2586 :
2587 : /*
2588 : * Write out an xl_multi_insert_tuple and the tuple data itself
2589 : * for each tuple.
2590 : */
2591 3127382 : for (i = 0; i < nthispage; i++)
2592 : {
2593 2434132 : HeapTuple heaptup = heaptuples[ndone + i];
2594 : xl_multi_insert_tuple *tuphdr;
2595 : int datalen;
2596 :
2597 2434132 : if (!init)
2598 1401134 : xlrec->offsets[i] = ItemPointerGetOffsetNumber(&heaptup->t_self);
2599 : /* xl_multi_insert_tuple needs two-byte alignment. */
2600 2434132 : tuphdr = (xl_multi_insert_tuple *) SHORTALIGN(scratchptr);
2601 2434132 : scratchptr = ((char *) tuphdr) + SizeOfMultiInsertTuple;
2602 :
2603 2434132 : tuphdr->t_infomask2 = heaptup->t_data->t_infomask2;
2604 2434132 : tuphdr->t_infomask = heaptup->t_data->t_infomask;
2605 2434132 : tuphdr->t_hoff = heaptup->t_data->t_hoff;
2606 :
2607 : /* write bitmap [+ padding] [+ oid] + data */
2608 2434132 : datalen = heaptup->t_len - SizeofHeapTupleHeader;
2609 2434132 : memcpy(scratchptr,
2610 2434132 : (char *) heaptup->t_data + SizeofHeapTupleHeader,
2611 : datalen);
2612 2434132 : tuphdr->datalen = datalen;
2613 2434132 : scratchptr += datalen;
2614 : }
2615 693250 : totaldatalen = scratchptr - tupledata;
2616 : Assert((scratchptr - scratch.data) < BLCKSZ);
2617 :
2618 693250 : if (need_tuple_data)
2619 144 : xlrec->flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
2620 :
2621 : /*
2622 : * Signal that this is the last xl_heap_multi_insert record
2623 : * emitted by this call to heap_multi_insert(). Needed for logical
2624 : * decoding so it knows when to cleanup temporary data.
2625 : */
2626 693250 : if (ndone + nthispage == ntuples)
2627 671358 : xlrec->flags |= XLH_INSERT_LAST_IN_MULTI;
2628 :
2629 693250 : if (init)
2630 : {
2631 25034 : info |= XLOG_HEAP_INIT_PAGE;
2632 25034 : bufflags |= REGBUF_WILL_INIT;
2633 : }
2634 :
2635 : /*
2636 : * If we're doing logical decoding, include the new tuple data
2637 : * even if we take a full-page image of the page.
2638 : */
2639 693250 : if (need_tuple_data)
2640 144 : bufflags |= REGBUF_KEEP_DATA;
2641 :
2642 693250 : XLogBeginInsert();
2643 693250 : XLogRegisterData(xlrec, tupledata - scratch.data);
2644 693250 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD | bufflags);
2645 :
2646 693250 : XLogRegisterBufData(0, tupledata, totaldatalen);
2647 :
2648 : /* filtering by origin on a row level is much more efficient */
2649 693250 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
2650 :
2651 693250 : recptr = XLogInsert(RM_HEAP2_ID, info);
2652 :
2653 693250 : PageSetLSN(page, recptr);
2654 : }
2655 :
2656 700892 : END_CRIT_SECTION();
2657 :
2658 : /*
2659 : * If we've frozen everything on the page, update the visibilitymap.
2660 : * We're already holding pin on the vmbuffer.
2661 : */
2662 700892 : if (all_frozen_set)
2663 : {
2664 : Assert(PageIsAllVisible(page));
2665 : Assert(visibilitymap_pin_ok(BufferGetBlockNumber(buffer), vmbuffer));
2666 :
2667 : /*
2668 : * It's fine to use InvalidTransactionId here - this is only used
2669 : * when HEAP_INSERT_FROZEN is specified, which intentionally
2670 : * violates visibility rules.
2671 : */
2672 3322 : visibilitymap_set(relation, BufferGetBlockNumber(buffer), buffer,
2673 : InvalidXLogRecPtr, vmbuffer,
2674 : InvalidTransactionId,
2675 : VISIBILITYMAP_ALL_VISIBLE | VISIBILITYMAP_ALL_FROZEN);
2676 : }
2677 :
2678 700892 : UnlockReleaseBuffer(buffer);
2679 700892 : ndone += nthispage;
2680 :
2681 : /*
2682 : * NB: Only release vmbuffer after inserting all tuples - it's fairly
2683 : * likely that we'll insert into subsequent heap pages that are likely
2684 : * to use the same vm page.
2685 : */
2686 : }
2687 :
2688 : /* We're done with inserting all tuples, so release the last vmbuffer. */
2689 672182 : if (vmbuffer != InvalidBuffer)
2690 5656 : ReleaseBuffer(vmbuffer);
2691 :
2692 : /*
2693 : * We're done with the actual inserts. Check for conflicts again, to
2694 : * ensure that all rw-conflicts in to these inserts are detected. Without
2695 : * this final check, a sequential scan of the heap may have locked the
2696 : * table after the "before" check, missing one opportunity to detect the
2697 : * conflict, and then scanned the table before the new tuples were there,
2698 : * missing the other chance to detect the conflict.
2699 : *
2700 : * For heap inserts, we only need to check for table-level SSI locks. Our
2701 : * new tuples can't possibly conflict with existing tuple locks, and heap
2702 : * page locks are only consolidated versions of tuple locks; they do not
2703 : * lock "gaps" as index page locks do. So we don't need to specify a
2704 : * buffer when making the call.
2705 : */
2706 672182 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2707 :
2708 : /*
2709 : * If tuples are cachable, mark them for invalidation from the caches in
2710 : * case we abort. Note it is OK to do this after releasing the buffer,
2711 : * because the heaptuples data structure is all in local memory, not in
2712 : * the shared buffer.
2713 : */
2714 672182 : if (IsCatalogRelation(relation))
2715 : {
2716 2319086 : for (i = 0; i < ntuples; i++)
2717 1649348 : CacheInvalidateHeapTuple(relation, heaptuples[i], NULL);
2718 : }
2719 :
2720 : /* copy t_self fields back to the caller's slots */
2721 3517116 : for (i = 0; i < ntuples; i++)
2722 2844934 : slots[i]->tts_tid = heaptuples[i]->t_self;
2723 :
2724 672182 : pgstat_count_heap_insert(relation, ntuples);
2725 672182 : }
2726 :
2727 : /*
2728 : * simple_heap_insert - insert a tuple
2729 : *
2730 : * Currently, this routine differs from heap_insert only in supplying
2731 : * a default command ID and not allowing access to the speedup options.
2732 : *
2733 : * This should be used rather than using heap_insert directly in most places
2734 : * where we are modifying system catalogs.
2735 : */
2736 : void
2737 1629328 : simple_heap_insert(Relation relation, HeapTuple tup)
2738 : {
2739 1629328 : heap_insert(relation, tup, GetCurrentCommandId(true), 0, NULL);
2740 1629328 : }
2741 :
2742 : /*
2743 : * Given infomask/infomask2, compute the bits that must be saved in the
2744 : * "infobits" field of xl_heap_delete, xl_heap_update, xl_heap_lock,
2745 : * xl_heap_lock_updated WAL records.
2746 : *
2747 : * See fix_infomask_from_infobits.
2748 : */
2749 : static uint8
2750 3834228 : compute_infobits(uint16 infomask, uint16 infomask2)
2751 : {
2752 : return
2753 3834228 : ((infomask & HEAP_XMAX_IS_MULTI) != 0 ? XLHL_XMAX_IS_MULTI : 0) |
2754 3834228 : ((infomask & HEAP_XMAX_LOCK_ONLY) != 0 ? XLHL_XMAX_LOCK_ONLY : 0) |
2755 3834228 : ((infomask & HEAP_XMAX_EXCL_LOCK) != 0 ? XLHL_XMAX_EXCL_LOCK : 0) |
2756 : /* note we ignore HEAP_XMAX_SHR_LOCK here */
2757 7668456 : ((infomask & HEAP_XMAX_KEYSHR_LOCK) != 0 ? XLHL_XMAX_KEYSHR_LOCK : 0) |
2758 : ((infomask2 & HEAP_KEYS_UPDATED) != 0 ?
2759 3834228 : XLHL_KEYS_UPDATED : 0);
2760 : }
2761 :
2762 : /*
2763 : * Given two versions of the same t_infomask for a tuple, compare them and
2764 : * return whether the relevant status for a tuple Xmax has changed. This is
2765 : * used after a buffer lock has been released and reacquired: we want to ensure
2766 : * that the tuple state continues to be the same it was when we previously
2767 : * examined it.
2768 : *
2769 : * Note the Xmax field itself must be compared separately.
2770 : */
2771 : static inline bool
2772 10658 : xmax_infomask_changed(uint16 new_infomask, uint16 old_infomask)
2773 : {
2774 10658 : const uint16 interesting =
2775 : HEAP_XMAX_IS_MULTI | HEAP_XMAX_LOCK_ONLY | HEAP_LOCK_MASK;
2776 :
2777 10658 : if ((new_infomask & interesting) != (old_infomask & interesting))
2778 26 : return true;
2779 :
2780 10632 : return false;
2781 : }
2782 :
2783 : /*
2784 : * heap_delete - delete a tuple
2785 : *
2786 : * See table_tuple_delete() for an explanation of the parameters, except that
2787 : * this routine directly takes a tuple rather than a slot.
2788 : *
2789 : * In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
2790 : * t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
2791 : * only for TM_SelfModified, since we cannot obtain cmax from a combo CID
2792 : * generated by another transaction).
2793 : */
2794 : TM_Result
2795 2943986 : heap_delete(Relation relation, ItemPointer tid,
2796 : CommandId cid, Snapshot crosscheck, bool wait,
2797 : TM_FailureData *tmfd, bool changingPart)
2798 : {
2799 : TM_Result result;
2800 2943986 : TransactionId xid = GetCurrentTransactionId();
2801 : ItemId lp;
2802 : HeapTupleData tp;
2803 : Page page;
2804 : BlockNumber block;
2805 : Buffer buffer;
2806 2943986 : Buffer vmbuffer = InvalidBuffer;
2807 : TransactionId new_xmax;
2808 : uint16 new_infomask,
2809 : new_infomask2;
2810 2943986 : bool have_tuple_lock = false;
2811 : bool iscombo;
2812 2943986 : bool all_visible_cleared = false;
2813 2943986 : HeapTuple old_key_tuple = NULL; /* replica identity of the tuple */
2814 2943986 : bool old_key_copied = false;
2815 :
2816 : Assert(ItemPointerIsValid(tid));
2817 :
2818 : /*
2819 : * Forbid this during a parallel operation, lest it allocate a combo CID.
2820 : * Other workers might need that combo CID for visibility checks, and we
2821 : * have no provision for broadcasting it to them.
2822 : */
2823 2943986 : if (IsInParallelMode())
2824 0 : ereport(ERROR,
2825 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
2826 : errmsg("cannot delete tuples during a parallel operation")));
2827 :
2828 2943986 : block = ItemPointerGetBlockNumber(tid);
2829 2943986 : buffer = ReadBuffer(relation, block);
2830 2943986 : page = BufferGetPage(buffer);
2831 :
2832 : /*
2833 : * Before locking the buffer, pin the visibility map page if it appears to
2834 : * be necessary. Since we haven't got the lock yet, someone else might be
2835 : * in the middle of changing this, so we'll need to recheck after we have
2836 : * the lock.
2837 : */
2838 2943986 : if (PageIsAllVisible(page))
2839 446 : visibilitymap_pin(relation, block, &vmbuffer);
2840 :
2841 2943986 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2842 :
2843 2943986 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
2844 : Assert(ItemIdIsNormal(lp));
2845 :
2846 2943986 : tp.t_tableOid = RelationGetRelid(relation);
2847 2943986 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
2848 2943986 : tp.t_len = ItemIdGetLength(lp);
2849 2943986 : tp.t_self = *tid;
2850 :
2851 2943988 : l1:
2852 :
2853 : /*
2854 : * If we didn't pin the visibility map page and the page has become all
2855 : * visible while we were busy locking the buffer, we'll have to unlock and
2856 : * re-lock, to avoid holding the buffer lock across an I/O. That's a bit
2857 : * unfortunate, but hopefully shouldn't happen often.
2858 : */
2859 2943988 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
2860 : {
2861 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2862 0 : visibilitymap_pin(relation, block, &vmbuffer);
2863 0 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2864 : }
2865 :
2866 2943988 : result = HeapTupleSatisfiesUpdate(&tp, cid, buffer);
2867 :
2868 2943988 : if (result == TM_Invisible)
2869 : {
2870 0 : UnlockReleaseBuffer(buffer);
2871 0 : ereport(ERROR,
2872 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2873 : errmsg("attempted to delete invisible tuple")));
2874 : }
2875 2943988 : else if (result == TM_BeingModified && wait)
2876 : {
2877 : TransactionId xwait;
2878 : uint16 infomask;
2879 :
2880 : /* must copy state data before unlocking buffer */
2881 81110 : xwait = HeapTupleHeaderGetRawXmax(tp.t_data);
2882 81110 : infomask = tp.t_data->t_infomask;
2883 :
2884 : /*
2885 : * Sleep until concurrent transaction ends -- except when there's a
2886 : * single locker and it's our own transaction. Note we don't care
2887 : * which lock mode the locker has, because we need the strongest one.
2888 : *
2889 : * Before sleeping, we need to acquire tuple lock to establish our
2890 : * priority for the tuple (see heap_lock_tuple). LockTuple will
2891 : * release us when we are next-in-line for the tuple.
2892 : *
2893 : * If we are forced to "start over" below, we keep the tuple lock;
2894 : * this arranges that we stay at the head of the line while rechecking
2895 : * tuple state.
2896 : */
2897 81110 : if (infomask & HEAP_XMAX_IS_MULTI)
2898 : {
2899 16 : bool current_is_member = false;
2900 :
2901 16 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
2902 : LockTupleExclusive, ¤t_is_member))
2903 : {
2904 16 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2905 :
2906 : /*
2907 : * Acquire the lock, if necessary (but skip it when we're
2908 : * requesting a lock and already have one; avoids deadlock).
2909 : */
2910 16 : if (!current_is_member)
2911 12 : heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
2912 : LockWaitBlock, &have_tuple_lock);
2913 :
2914 : /* wait for multixact */
2915 16 : MultiXactIdWait((MultiXactId) xwait, MultiXactStatusUpdate, infomask,
2916 : relation, &(tp.t_self), XLTW_Delete,
2917 : NULL);
2918 16 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2919 :
2920 : /*
2921 : * If xwait had just locked the tuple then some other xact
2922 : * could update this tuple before we get to this point. Check
2923 : * for xmax change, and start over if so.
2924 : *
2925 : * We also must start over if we didn't pin the VM page, and
2926 : * the page has become all visible.
2927 : */
2928 32 : if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
2929 32 : xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
2930 16 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
2931 : xwait))
2932 0 : goto l1;
2933 : }
2934 :
2935 : /*
2936 : * You might think the multixact is necessarily done here, but not
2937 : * so: it could have surviving members, namely our own xact or
2938 : * other subxacts of this backend. It is legal for us to delete
2939 : * the tuple in either case, however (the latter case is
2940 : * essentially a situation of upgrading our former shared lock to
2941 : * exclusive). We don't bother changing the on-disk hint bits
2942 : * since we are about to overwrite the xmax altogether.
2943 : */
2944 : }
2945 81094 : else if (!TransactionIdIsCurrentTransactionId(xwait))
2946 : {
2947 : /*
2948 : * Wait for regular transaction to end; but first, acquire tuple
2949 : * lock.
2950 : */
2951 80 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2952 80 : heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
2953 : LockWaitBlock, &have_tuple_lock);
2954 80 : XactLockTableWait(xwait, relation, &(tp.t_self), XLTW_Delete);
2955 72 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2956 :
2957 : /*
2958 : * xwait is done, but if xwait had just locked the tuple then some
2959 : * other xact could update this tuple before we get to this point.
2960 : * Check for xmax change, and start over if so.
2961 : *
2962 : * We also must start over if we didn't pin the VM page, and the
2963 : * page has become all visible.
2964 : */
2965 144 : if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
2966 142 : xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
2967 70 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
2968 : xwait))
2969 2 : goto l1;
2970 :
2971 : /* Otherwise check if it committed or aborted */
2972 70 : UpdateXmaxHintBits(tp.t_data, buffer, xwait);
2973 : }
2974 :
2975 : /*
2976 : * We may overwrite if previous xmax aborted, or if it committed but
2977 : * only locked the tuple without updating it.
2978 : */
2979 162172 : if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
2980 81122 : HEAP_XMAX_IS_LOCKED_ONLY(tp.t_data->t_infomask) ||
2981 50 : HeapTupleHeaderIsOnlyLocked(tp.t_data))
2982 81058 : result = TM_Ok;
2983 42 : else if (!ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
2984 34 : result = TM_Updated;
2985 : else
2986 8 : result = TM_Deleted;
2987 : }
2988 :
2989 : /* sanity check the result HeapTupleSatisfiesUpdate() and the logic above */
2990 : if (result != TM_Ok)
2991 : {
2992 : Assert(result == TM_SelfModified ||
2993 : result == TM_Updated ||
2994 : result == TM_Deleted ||
2995 : result == TM_BeingModified);
2996 : Assert(!(tp.t_data->t_infomask & HEAP_XMAX_INVALID));
2997 : Assert(result != TM_Updated ||
2998 : !ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid));
2999 : }
3000 :
3001 2943978 : if (crosscheck != InvalidSnapshot && result == TM_Ok)
3002 : {
3003 : /* Perform additional check for transaction-snapshot mode RI updates */
3004 2 : if (!HeapTupleSatisfiesVisibility(&tp, crosscheck, buffer))
3005 2 : result = TM_Updated;
3006 : }
3007 :
3008 2943978 : if (result != TM_Ok)
3009 : {
3010 112 : tmfd->ctid = tp.t_data->t_ctid;
3011 112 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
3012 112 : if (result == TM_SelfModified)
3013 42 : tmfd->cmax = HeapTupleHeaderGetCmax(tp.t_data);
3014 : else
3015 70 : tmfd->cmax = InvalidCommandId;
3016 112 : UnlockReleaseBuffer(buffer);
3017 112 : if (have_tuple_lock)
3018 42 : UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
3019 112 : if (vmbuffer != InvalidBuffer)
3020 0 : ReleaseBuffer(vmbuffer);
3021 112 : return result;
3022 : }
3023 :
3024 : /*
3025 : * We're about to do the actual delete -- check for conflict first, to
3026 : * avoid possibly having to roll back work we've just done.
3027 : *
3028 : * This is safe without a recheck as long as there is no possibility of
3029 : * another process scanning the page between this check and the delete
3030 : * being visible to the scan (i.e., an exclusive buffer content lock is
3031 : * continuously held from this point until the tuple delete is visible).
3032 : */
3033 2943866 : CheckForSerializableConflictIn(relation, tid, BufferGetBlockNumber(buffer));
3034 :
3035 : /* replace cid with a combo CID if necessary */
3036 2943838 : HeapTupleHeaderAdjustCmax(tp.t_data, &cid, &iscombo);
3037 :
3038 : /*
3039 : * Compute replica identity tuple before entering the critical section so
3040 : * we don't PANIC upon a memory allocation failure.
3041 : */
3042 2943838 : old_key_tuple = ExtractReplicaIdentity(relation, &tp, true, &old_key_copied);
3043 :
3044 : /*
3045 : * If this is the first possibly-multixact-able operation in the current
3046 : * transaction, set my per-backend OldestMemberMXactId setting. We can be
3047 : * certain that the transaction will never become a member of any older
3048 : * MultiXactIds than that. (We have to do this even if we end up just
3049 : * using our own TransactionId below, since some other backend could
3050 : * incorporate our XID into a MultiXact immediately afterwards.)
3051 : */
3052 2943838 : MultiXactIdSetOldestMember();
3053 :
3054 2943838 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(tp.t_data),
3055 2943838 : tp.t_data->t_infomask, tp.t_data->t_infomask2,
3056 : xid, LockTupleExclusive, true,
3057 : &new_xmax, &new_infomask, &new_infomask2);
3058 :
3059 2943838 : START_CRIT_SECTION();
3060 :
3061 : /*
3062 : * If this transaction commits, the tuple will become DEAD sooner or
3063 : * later. Set flag that this page is a candidate for pruning once our xid
3064 : * falls below the OldestXmin horizon. If the transaction finally aborts,
3065 : * the subsequent page pruning will be a no-op and the hint will be
3066 : * cleared.
3067 : */
3068 2943838 : PageSetPrunable(page, xid);
3069 :
3070 2943838 : if (PageIsAllVisible(page))
3071 : {
3072 446 : all_visible_cleared = true;
3073 446 : PageClearAllVisible(page);
3074 446 : visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
3075 : vmbuffer, VISIBILITYMAP_VALID_BITS);
3076 : }
3077 :
3078 : /* store transaction information of xact deleting the tuple */
3079 2943838 : tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
3080 2943838 : tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
3081 2943838 : tp.t_data->t_infomask |= new_infomask;
3082 2943838 : tp.t_data->t_infomask2 |= new_infomask2;
3083 2943838 : HeapTupleHeaderClearHotUpdated(tp.t_data);
3084 2943838 : HeapTupleHeaderSetXmax(tp.t_data, new_xmax);
3085 2943838 : HeapTupleHeaderSetCmax(tp.t_data, cid, iscombo);
3086 : /* Make sure there is no forward chain link in t_ctid */
3087 2943838 : tp.t_data->t_ctid = tp.t_self;
3088 :
3089 : /* Signal that this is actually a move into another partition */
3090 2943838 : if (changingPart)
3091 964 : HeapTupleHeaderSetMovedPartitions(tp.t_data);
3092 :
3093 2943838 : MarkBufferDirty(buffer);
3094 :
3095 : /*
3096 : * XLOG stuff
3097 : *
3098 : * NB: heap_abort_speculative() uses the same xlog record and replay
3099 : * routines.
3100 : */
3101 2943838 : if (RelationNeedsWAL(relation))
3102 : {
3103 : xl_heap_delete xlrec;
3104 : xl_heap_header xlhdr;
3105 : XLogRecPtr recptr;
3106 :
3107 : /*
3108 : * For logical decode we need combo CIDs to properly decode the
3109 : * catalog
3110 : */
3111 2822620 : if (RelationIsAccessibleInLogicalDecoding(relation))
3112 12310 : log_heap_new_cid(relation, &tp);
3113 :
3114 2822620 : xlrec.flags = 0;
3115 2822620 : if (all_visible_cleared)
3116 446 : xlrec.flags |= XLH_DELETE_ALL_VISIBLE_CLEARED;
3117 2822620 : if (changingPart)
3118 964 : xlrec.flags |= XLH_DELETE_IS_PARTITION_MOVE;
3119 5645240 : xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
3120 2822620 : tp.t_data->t_infomask2);
3121 2822620 : xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
3122 2822620 : xlrec.xmax = new_xmax;
3123 :
3124 2822620 : if (old_key_tuple != NULL)
3125 : {
3126 94028 : if (relation->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
3127 256 : xlrec.flags |= XLH_DELETE_CONTAINS_OLD_TUPLE;
3128 : else
3129 93772 : xlrec.flags |= XLH_DELETE_CONTAINS_OLD_KEY;
3130 : }
3131 :
3132 2822620 : XLogBeginInsert();
3133 2822620 : XLogRegisterData(&xlrec, SizeOfHeapDelete);
3134 :
3135 2822620 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
3136 :
3137 : /*
3138 : * Log replica identity of the deleted tuple if there is one
3139 : */
3140 2822620 : if (old_key_tuple != NULL)
3141 : {
3142 94028 : xlhdr.t_infomask2 = old_key_tuple->t_data->t_infomask2;
3143 94028 : xlhdr.t_infomask = old_key_tuple->t_data->t_infomask;
3144 94028 : xlhdr.t_hoff = old_key_tuple->t_data->t_hoff;
3145 :
3146 94028 : XLogRegisterData(&xlhdr, SizeOfHeapHeader);
3147 94028 : XLogRegisterData((char *) old_key_tuple->t_data
3148 : + SizeofHeapTupleHeader,
3149 94028 : old_key_tuple->t_len
3150 : - SizeofHeapTupleHeader);
3151 : }
3152 :
3153 : /* filtering by origin on a row level is much more efficient */
3154 2822620 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
3155 :
3156 2822620 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
3157 :
3158 2822620 : PageSetLSN(page, recptr);
3159 : }
3160 :
3161 2943838 : END_CRIT_SECTION();
3162 :
3163 2943838 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3164 :
3165 2943838 : if (vmbuffer != InvalidBuffer)
3166 446 : ReleaseBuffer(vmbuffer);
3167 :
3168 : /*
3169 : * If the tuple has toasted out-of-line attributes, we need to delete
3170 : * those items too. We have to do this before releasing the buffer
3171 : * because we need to look at the contents of the tuple, but it's OK to
3172 : * release the content lock on the buffer first.
3173 : */
3174 2943838 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
3175 4806 : relation->rd_rel->relkind != RELKIND_MATVIEW)
3176 : {
3177 : /* toast table entries should never be recursively toasted */
3178 : Assert(!HeapTupleHasExternal(&tp));
3179 : }
3180 2939052 : else if (HeapTupleHasExternal(&tp))
3181 586 : heap_toast_delete(relation, &tp, false);
3182 :
3183 : /*
3184 : * Mark tuple for invalidation from system caches at next command
3185 : * boundary. We have to do this before releasing the buffer because we
3186 : * need to look at the contents of the tuple.
3187 : */
3188 2943838 : CacheInvalidateHeapTuple(relation, &tp, NULL);
3189 :
3190 : /* Now we can release the buffer */
3191 2943838 : ReleaseBuffer(buffer);
3192 :
3193 : /*
3194 : * Release the lmgr tuple lock, if we had it.
3195 : */
3196 2943838 : if (have_tuple_lock)
3197 40 : UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
3198 :
3199 2943838 : pgstat_count_heap_delete(relation);
3200 :
3201 2943838 : if (old_key_tuple != NULL && old_key_copied)
3202 93774 : heap_freetuple(old_key_tuple);
3203 :
3204 2943838 : return TM_Ok;
3205 : }
3206 :
3207 : /*
3208 : * simple_heap_delete - delete a tuple
3209 : *
3210 : * This routine may be used to delete a tuple when concurrent updates of
3211 : * the target tuple are not expected (for example, because we have a lock
3212 : * on the relation associated with the tuple). Any failure is reported
3213 : * via ereport().
3214 : */
3215 : void
3216 1217490 : simple_heap_delete(Relation relation, ItemPointer tid)
3217 : {
3218 : TM_Result result;
3219 : TM_FailureData tmfd;
3220 :
3221 1217490 : result = heap_delete(relation, tid,
3222 : GetCurrentCommandId(true), InvalidSnapshot,
3223 : true /* wait for commit */ ,
3224 : &tmfd, false /* changingPart */ );
3225 1217490 : switch (result)
3226 : {
3227 0 : case TM_SelfModified:
3228 : /* Tuple was already updated in current command? */
3229 0 : elog(ERROR, "tuple already updated by self");
3230 : break;
3231 :
3232 1217490 : case TM_Ok:
3233 : /* done successfully */
3234 1217490 : break;
3235 :
3236 0 : case TM_Updated:
3237 0 : elog(ERROR, "tuple concurrently updated");
3238 : break;
3239 :
3240 0 : case TM_Deleted:
3241 0 : elog(ERROR, "tuple concurrently deleted");
3242 : break;
3243 :
3244 0 : default:
3245 0 : elog(ERROR, "unrecognized heap_delete status: %u", result);
3246 : break;
3247 : }
3248 1217490 : }
3249 :
3250 : /*
3251 : * heap_update - replace a tuple
3252 : *
3253 : * See table_tuple_update() for an explanation of the parameters, except that
3254 : * this routine directly takes a tuple rather than a slot.
3255 : *
3256 : * In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
3257 : * t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
3258 : * only for TM_SelfModified, since we cannot obtain cmax from a combo CID
3259 : * generated by another transaction).
3260 : */
3261 : TM_Result
3262 595790 : heap_update(Relation relation, ItemPointer otid, HeapTuple newtup,
3263 : CommandId cid, Snapshot crosscheck, bool wait,
3264 : TM_FailureData *tmfd, LockTupleMode *lockmode,
3265 : TU_UpdateIndexes *update_indexes)
3266 : {
3267 : TM_Result result;
3268 595790 : TransactionId xid = GetCurrentTransactionId();
3269 : Bitmapset *hot_attrs;
3270 : Bitmapset *sum_attrs;
3271 : Bitmapset *key_attrs;
3272 : Bitmapset *id_attrs;
3273 : Bitmapset *interesting_attrs;
3274 : Bitmapset *modified_attrs;
3275 : ItemId lp;
3276 : HeapTupleData oldtup;
3277 : HeapTuple heaptup;
3278 595790 : HeapTuple old_key_tuple = NULL;
3279 595790 : bool old_key_copied = false;
3280 : Page page;
3281 : BlockNumber block;
3282 : MultiXactStatus mxact_status;
3283 : Buffer buffer,
3284 : newbuf,
3285 595790 : vmbuffer = InvalidBuffer,
3286 595790 : vmbuffer_new = InvalidBuffer;
3287 : bool need_toast;
3288 : Size newtupsize,
3289 : pagefree;
3290 595790 : bool have_tuple_lock = false;
3291 : bool iscombo;
3292 595790 : bool use_hot_update = false;
3293 595790 : bool summarized_update = false;
3294 : bool key_intact;
3295 595790 : bool all_visible_cleared = false;
3296 595790 : bool all_visible_cleared_new = false;
3297 : bool checked_lockers;
3298 : bool locker_remains;
3299 595790 : bool id_has_external = false;
3300 : TransactionId xmax_new_tuple,
3301 : xmax_old_tuple;
3302 : uint16 infomask_old_tuple,
3303 : infomask2_old_tuple,
3304 : infomask_new_tuple,
3305 : infomask2_new_tuple;
3306 :
3307 : Assert(ItemPointerIsValid(otid));
3308 :
3309 : /* Cheap, simplistic check that the tuple matches the rel's rowtype. */
3310 : Assert(HeapTupleHeaderGetNatts(newtup->t_data) <=
3311 : RelationGetNumberOfAttributes(relation));
3312 :
3313 : /*
3314 : * Forbid this during a parallel operation, lest it allocate a combo CID.
3315 : * Other workers might need that combo CID for visibility checks, and we
3316 : * have no provision for broadcasting it to them.
3317 : */
3318 595790 : if (IsInParallelMode())
3319 0 : ereport(ERROR,
3320 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
3321 : errmsg("cannot update tuples during a parallel operation")));
3322 :
3323 : #ifdef USE_ASSERT_CHECKING
3324 : check_lock_if_inplace_updateable_rel(relation, otid, newtup);
3325 : #endif
3326 :
3327 : /*
3328 : * Fetch the list of attributes to be checked for various operations.
3329 : *
3330 : * For HOT considerations, this is wasted effort if we fail to update or
3331 : * have to put the new tuple on a different page. But we must compute the
3332 : * list before obtaining buffer lock --- in the worst case, if we are
3333 : * doing an update on one of the relevant system catalogs, we could
3334 : * deadlock if we try to fetch the list later. In any case, the relcache
3335 : * caches the data so this is usually pretty cheap.
3336 : *
3337 : * We also need columns used by the replica identity and columns that are
3338 : * considered the "key" of rows in the table.
3339 : *
3340 : * Note that we get copies of each bitmap, so we need not worry about
3341 : * relcache flush happening midway through.
3342 : */
3343 595790 : hot_attrs = RelationGetIndexAttrBitmap(relation,
3344 : INDEX_ATTR_BITMAP_HOT_BLOCKING);
3345 595790 : sum_attrs = RelationGetIndexAttrBitmap(relation,
3346 : INDEX_ATTR_BITMAP_SUMMARIZED);
3347 595790 : key_attrs = RelationGetIndexAttrBitmap(relation, INDEX_ATTR_BITMAP_KEY);
3348 595790 : id_attrs = RelationGetIndexAttrBitmap(relation,
3349 : INDEX_ATTR_BITMAP_IDENTITY_KEY);
3350 595790 : interesting_attrs = NULL;
3351 595790 : interesting_attrs = bms_add_members(interesting_attrs, hot_attrs);
3352 595790 : interesting_attrs = bms_add_members(interesting_attrs, sum_attrs);
3353 595790 : interesting_attrs = bms_add_members(interesting_attrs, key_attrs);
3354 595790 : interesting_attrs = bms_add_members(interesting_attrs, id_attrs);
3355 :
3356 595790 : block = ItemPointerGetBlockNumber(otid);
3357 595790 : INJECTION_POINT("heap_update-before-pin");
3358 595790 : buffer = ReadBuffer(relation, block);
3359 595790 : page = BufferGetPage(buffer);
3360 :
3361 : /*
3362 : * Before locking the buffer, pin the visibility map page if it appears to
3363 : * be necessary. Since we haven't got the lock yet, someone else might be
3364 : * in the middle of changing this, so we'll need to recheck after we have
3365 : * the lock.
3366 : */
3367 595790 : if (PageIsAllVisible(page))
3368 3234 : visibilitymap_pin(relation, block, &vmbuffer);
3369 :
3370 595790 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3371 :
3372 595790 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(otid));
3373 :
3374 : /*
3375 : * Usually, a buffer pin and/or snapshot blocks pruning of otid, ensuring
3376 : * we see LP_NORMAL here. When the otid origin is a syscache, we may have
3377 : * neither a pin nor a snapshot. Hence, we may see other LP_ states, each
3378 : * of which indicates concurrent pruning.
3379 : *
3380 : * Failing with TM_Updated would be most accurate. However, unlike other
3381 : * TM_Updated scenarios, we don't know the successor ctid in LP_UNUSED and
3382 : * LP_DEAD cases. While the distinction between TM_Updated and TM_Deleted
3383 : * does matter to SQL statements UPDATE and MERGE, those SQL statements
3384 : * hold a snapshot that ensures LP_NORMAL. Hence, the choice between
3385 : * TM_Updated and TM_Deleted affects only the wording of error messages.
3386 : * Settle on TM_Deleted, for two reasons. First, it avoids complicating
3387 : * the specification of when tmfd->ctid is valid. Second, it creates
3388 : * error log evidence that we took this branch.
3389 : *
3390 : * Since it's possible to see LP_UNUSED at otid, it's also possible to see
3391 : * LP_NORMAL for a tuple that replaced LP_UNUSED. If it's a tuple for an
3392 : * unrelated row, we'll fail with "duplicate key value violates unique".
3393 : * XXX if otid is the live, newer version of the newtup row, we'll discard
3394 : * changes originating in versions of this catalog row after the version
3395 : * the caller got from syscache. See syscache-update-pruned.spec.
3396 : */
3397 595790 : if (!ItemIdIsNormal(lp))
3398 : {
3399 : Assert(RelationSupportsSysCache(RelationGetRelid(relation)));
3400 :
3401 2 : UnlockReleaseBuffer(buffer);
3402 : Assert(!have_tuple_lock);
3403 2 : if (vmbuffer != InvalidBuffer)
3404 2 : ReleaseBuffer(vmbuffer);
3405 2 : tmfd->ctid = *otid;
3406 2 : tmfd->xmax = InvalidTransactionId;
3407 2 : tmfd->cmax = InvalidCommandId;
3408 2 : *update_indexes = TU_None;
3409 :
3410 2 : bms_free(hot_attrs);
3411 2 : bms_free(sum_attrs);
3412 2 : bms_free(key_attrs);
3413 2 : bms_free(id_attrs);
3414 : /* modified_attrs not yet initialized */
3415 2 : bms_free(interesting_attrs);
3416 2 : return TM_Deleted;
3417 : }
3418 :
3419 : /*
3420 : * Fill in enough data in oldtup for HeapDetermineColumnsInfo to work
3421 : * properly.
3422 : */
3423 595788 : oldtup.t_tableOid = RelationGetRelid(relation);
3424 595788 : oldtup.t_data = (HeapTupleHeader) PageGetItem(page, lp);
3425 595788 : oldtup.t_len = ItemIdGetLength(lp);
3426 595788 : oldtup.t_self = *otid;
3427 :
3428 : /* the new tuple is ready, except for this: */
3429 595788 : newtup->t_tableOid = RelationGetRelid(relation);
3430 :
3431 : /*
3432 : * Determine columns modified by the update. Additionally, identify
3433 : * whether any of the unmodified replica identity key attributes in the
3434 : * old tuple is externally stored or not. This is required because for
3435 : * such attributes the flattened value won't be WAL logged as part of the
3436 : * new tuple so we must include it as part of the old_key_tuple. See
3437 : * ExtractReplicaIdentity.
3438 : */
3439 595788 : modified_attrs = HeapDetermineColumnsInfo(relation, interesting_attrs,
3440 : id_attrs, &oldtup,
3441 : newtup, &id_has_external);
3442 :
3443 : /*
3444 : * If we're not updating any "key" column, we can grab a weaker lock type.
3445 : * This allows for more concurrency when we are running simultaneously
3446 : * with foreign key checks.
3447 : *
3448 : * Note that if a column gets detoasted while executing the update, but
3449 : * the value ends up being the same, this test will fail and we will use
3450 : * the stronger lock. This is acceptable; the important case to optimize
3451 : * is updates that don't manipulate key columns, not those that
3452 : * serendipitously arrive at the same key values.
3453 : */
3454 595788 : if (!bms_overlap(modified_attrs, key_attrs))
3455 : {
3456 587540 : *lockmode = LockTupleNoKeyExclusive;
3457 587540 : mxact_status = MultiXactStatusNoKeyUpdate;
3458 587540 : key_intact = true;
3459 :
3460 : /*
3461 : * If this is the first possibly-multixact-able operation in the
3462 : * current transaction, set my per-backend OldestMemberMXactId
3463 : * setting. We can be certain that the transaction will never become a
3464 : * member of any older MultiXactIds than that. (We have to do this
3465 : * even if we end up just using our own TransactionId below, since
3466 : * some other backend could incorporate our XID into a MultiXact
3467 : * immediately afterwards.)
3468 : */
3469 587540 : MultiXactIdSetOldestMember();
3470 : }
3471 : else
3472 : {
3473 8248 : *lockmode = LockTupleExclusive;
3474 8248 : mxact_status = MultiXactStatusUpdate;
3475 8248 : key_intact = false;
3476 : }
3477 :
3478 : /*
3479 : * Note: beyond this point, use oldtup not otid to refer to old tuple.
3480 : * otid may very well point at newtup->t_self, which we will overwrite
3481 : * with the new tuple's location, so there's great risk of confusion if we
3482 : * use otid anymore.
3483 : */
3484 :
3485 595788 : l2:
3486 595790 : checked_lockers = false;
3487 595790 : locker_remains = false;
3488 595790 : result = HeapTupleSatisfiesUpdate(&oldtup, cid, buffer);
3489 :
3490 : /* see below about the "no wait" case */
3491 : Assert(result != TM_BeingModified || wait);
3492 :
3493 595790 : if (result == TM_Invisible)
3494 : {
3495 0 : UnlockReleaseBuffer(buffer);
3496 0 : ereport(ERROR,
3497 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3498 : errmsg("attempted to update invisible tuple")));
3499 : }
3500 595790 : else if (result == TM_BeingModified && wait)
3501 : {
3502 : TransactionId xwait;
3503 : uint16 infomask;
3504 71882 : bool can_continue = false;
3505 :
3506 : /*
3507 : * XXX note that we don't consider the "no wait" case here. This
3508 : * isn't a problem currently because no caller uses that case, but it
3509 : * should be fixed if such a caller is introduced. It wasn't a
3510 : * problem previously because this code would always wait, but now
3511 : * that some tuple locks do not conflict with one of the lock modes we
3512 : * use, it is possible that this case is interesting to handle
3513 : * specially.
3514 : *
3515 : * This may cause failures with third-party code that calls
3516 : * heap_update directly.
3517 : */
3518 :
3519 : /* must copy state data before unlocking buffer */
3520 71882 : xwait = HeapTupleHeaderGetRawXmax(oldtup.t_data);
3521 71882 : infomask = oldtup.t_data->t_infomask;
3522 :
3523 : /*
3524 : * Now we have to do something about the existing locker. If it's a
3525 : * multi, sleep on it; we might be awakened before it is completely
3526 : * gone (or even not sleep at all in some cases); we need to preserve
3527 : * it as locker, unless it is gone completely.
3528 : *
3529 : * If it's not a multi, we need to check for sleeping conditions
3530 : * before actually going to sleep. If the update doesn't conflict
3531 : * with the locks, we just continue without sleeping (but making sure
3532 : * it is preserved).
3533 : *
3534 : * Before sleeping, we need to acquire tuple lock to establish our
3535 : * priority for the tuple (see heap_lock_tuple). LockTuple will
3536 : * release us when we are next-in-line for the tuple. Note we must
3537 : * not acquire the tuple lock until we're sure we're going to sleep;
3538 : * otherwise we're open for race conditions with other transactions
3539 : * holding the tuple lock which sleep on us.
3540 : *
3541 : * If we are forced to "start over" below, we keep the tuple lock;
3542 : * this arranges that we stay at the head of the line while rechecking
3543 : * tuple state.
3544 : */
3545 71882 : if (infomask & HEAP_XMAX_IS_MULTI)
3546 : {
3547 : TransactionId update_xact;
3548 : int remain;
3549 120 : bool current_is_member = false;
3550 :
3551 120 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
3552 : *lockmode, ¤t_is_member))
3553 : {
3554 16 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3555 :
3556 : /*
3557 : * Acquire the lock, if necessary (but skip it when we're
3558 : * requesting a lock and already have one; avoids deadlock).
3559 : */
3560 16 : if (!current_is_member)
3561 0 : heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3562 : LockWaitBlock, &have_tuple_lock);
3563 :
3564 : /* wait for multixact */
3565 16 : MultiXactIdWait((MultiXactId) xwait, mxact_status, infomask,
3566 : relation, &oldtup.t_self, XLTW_Update,
3567 : &remain);
3568 16 : checked_lockers = true;
3569 16 : locker_remains = remain != 0;
3570 16 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3571 :
3572 : /*
3573 : * If xwait had just locked the tuple then some other xact
3574 : * could update this tuple before we get to this point. Check
3575 : * for xmax change, and start over if so.
3576 : */
3577 16 : if (xmax_infomask_changed(oldtup.t_data->t_infomask,
3578 16 : infomask) ||
3579 16 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3580 : xwait))
3581 0 : goto l2;
3582 : }
3583 :
3584 : /*
3585 : * Note that the multixact may not be done by now. It could have
3586 : * surviving members; our own xact or other subxacts of this
3587 : * backend, and also any other concurrent transaction that locked
3588 : * the tuple with LockTupleKeyShare if we only got
3589 : * LockTupleNoKeyExclusive. If this is the case, we have to be
3590 : * careful to mark the updated tuple with the surviving members in
3591 : * Xmax.
3592 : *
3593 : * Note that there could have been another update in the
3594 : * MultiXact. In that case, we need to check whether it committed
3595 : * or aborted. If it aborted we are safe to update it again;
3596 : * otherwise there is an update conflict, and we have to return
3597 : * TableTuple{Deleted, Updated} below.
3598 : *
3599 : * In the LockTupleExclusive case, we still need to preserve the
3600 : * surviving members: those would include the tuple locks we had
3601 : * before this one, which are important to keep in case this
3602 : * subxact aborts.
3603 : */
3604 120 : if (!HEAP_XMAX_IS_LOCKED_ONLY(oldtup.t_data->t_infomask))
3605 16 : update_xact = HeapTupleGetUpdateXid(oldtup.t_data);
3606 : else
3607 104 : update_xact = InvalidTransactionId;
3608 :
3609 : /*
3610 : * There was no UPDATE in the MultiXact; or it aborted. No
3611 : * TransactionIdIsInProgress() call needed here, since we called
3612 : * MultiXactIdWait() above.
3613 : */
3614 136 : if (!TransactionIdIsValid(update_xact) ||
3615 16 : TransactionIdDidAbort(update_xact))
3616 106 : can_continue = true;
3617 : }
3618 71762 : else if (TransactionIdIsCurrentTransactionId(xwait))
3619 : {
3620 : /*
3621 : * The only locker is ourselves; we can avoid grabbing the tuple
3622 : * lock here, but must preserve our locking information.
3623 : */
3624 71578 : checked_lockers = true;
3625 71578 : locker_remains = true;
3626 71578 : can_continue = true;
3627 : }
3628 184 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) && key_intact)
3629 : {
3630 : /*
3631 : * If it's just a key-share locker, and we're not changing the key
3632 : * columns, we don't need to wait for it to end; but we need to
3633 : * preserve it as locker.
3634 : */
3635 58 : checked_lockers = true;
3636 58 : locker_remains = true;
3637 58 : can_continue = true;
3638 : }
3639 : else
3640 : {
3641 : /*
3642 : * Wait for regular transaction to end; but first, acquire tuple
3643 : * lock.
3644 : */
3645 126 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3646 126 : heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3647 : LockWaitBlock, &have_tuple_lock);
3648 126 : XactLockTableWait(xwait, relation, &oldtup.t_self,
3649 : XLTW_Update);
3650 126 : checked_lockers = true;
3651 126 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3652 :
3653 : /*
3654 : * xwait is done, but if xwait had just locked the tuple then some
3655 : * other xact could update this tuple before we get to this point.
3656 : * Check for xmax change, and start over if so.
3657 : */
3658 250 : if (xmax_infomask_changed(oldtup.t_data->t_infomask, infomask) ||
3659 124 : !TransactionIdEquals(xwait,
3660 : HeapTupleHeaderGetRawXmax(oldtup.t_data)))
3661 2 : goto l2;
3662 :
3663 : /* Otherwise check if it committed or aborted */
3664 124 : UpdateXmaxHintBits(oldtup.t_data, buffer, xwait);
3665 124 : if (oldtup.t_data->t_infomask & HEAP_XMAX_INVALID)
3666 30 : can_continue = true;
3667 : }
3668 :
3669 71880 : if (can_continue)
3670 71772 : result = TM_Ok;
3671 108 : else if (!ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid))
3672 98 : result = TM_Updated;
3673 : else
3674 10 : result = TM_Deleted;
3675 : }
3676 :
3677 : /* Sanity check the result HeapTupleSatisfiesUpdate() and the logic above */
3678 : if (result != TM_Ok)
3679 : {
3680 : Assert(result == TM_SelfModified ||
3681 : result == TM_Updated ||
3682 : result == TM_Deleted ||
3683 : result == TM_BeingModified);
3684 : Assert(!(oldtup.t_data->t_infomask & HEAP_XMAX_INVALID));
3685 : Assert(result != TM_Updated ||
3686 : !ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid));
3687 : }
3688 :
3689 595788 : if (crosscheck != InvalidSnapshot && result == TM_Ok)
3690 : {
3691 : /* Perform additional check for transaction-snapshot mode RI updates */
3692 2 : if (!HeapTupleSatisfiesVisibility(&oldtup, crosscheck, buffer))
3693 2 : result = TM_Updated;
3694 : }
3695 :
3696 595788 : if (result != TM_Ok)
3697 : {
3698 304 : tmfd->ctid = oldtup.t_data->t_ctid;
3699 304 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(oldtup.t_data);
3700 304 : if (result == TM_SelfModified)
3701 104 : tmfd->cmax = HeapTupleHeaderGetCmax(oldtup.t_data);
3702 : else
3703 200 : tmfd->cmax = InvalidCommandId;
3704 304 : UnlockReleaseBuffer(buffer);
3705 304 : if (have_tuple_lock)
3706 94 : UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
3707 304 : if (vmbuffer != InvalidBuffer)
3708 0 : ReleaseBuffer(vmbuffer);
3709 304 : *update_indexes = TU_None;
3710 :
3711 304 : bms_free(hot_attrs);
3712 304 : bms_free(sum_attrs);
3713 304 : bms_free(key_attrs);
3714 304 : bms_free(id_attrs);
3715 304 : bms_free(modified_attrs);
3716 304 : bms_free(interesting_attrs);
3717 304 : return result;
3718 : }
3719 :
3720 : /*
3721 : * If we didn't pin the visibility map page and the page has become all
3722 : * visible while we were busy locking the buffer, or during some
3723 : * subsequent window during which we had it unlocked, we'll have to unlock
3724 : * and re-lock, to avoid holding the buffer lock across an I/O. That's a
3725 : * bit unfortunate, especially since we'll now have to recheck whether the
3726 : * tuple has been locked or updated under us, but hopefully it won't
3727 : * happen very often.
3728 : */
3729 595484 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
3730 : {
3731 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3732 0 : visibilitymap_pin(relation, block, &vmbuffer);
3733 0 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3734 0 : goto l2;
3735 : }
3736 :
3737 : /* Fill in transaction status data */
3738 :
3739 : /*
3740 : * If the tuple we're updating is locked, we need to preserve the locking
3741 : * info in the old tuple's Xmax. Prepare a new Xmax value for this.
3742 : */
3743 595484 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3744 595484 : oldtup.t_data->t_infomask,
3745 595484 : oldtup.t_data->t_infomask2,
3746 : xid, *lockmode, true,
3747 : &xmax_old_tuple, &infomask_old_tuple,
3748 : &infomask2_old_tuple);
3749 :
3750 : /*
3751 : * And also prepare an Xmax value for the new copy of the tuple. If there
3752 : * was no xmax previously, or there was one but all lockers are now gone,
3753 : * then use InvalidTransactionId; otherwise, get the xmax from the old
3754 : * tuple. (In rare cases that might also be InvalidTransactionId and yet
3755 : * not have the HEAP_XMAX_INVALID bit set; that's fine.)
3756 : */
3757 667226 : if ((oldtup.t_data->t_infomask & HEAP_XMAX_INVALID) ||
3758 143484 : HEAP_LOCKED_UPGRADED(oldtup.t_data->t_infomask) ||
3759 71638 : (checked_lockers && !locker_remains))
3760 523742 : xmax_new_tuple = InvalidTransactionId;
3761 : else
3762 71742 : xmax_new_tuple = HeapTupleHeaderGetRawXmax(oldtup.t_data);
3763 :
3764 595484 : if (!TransactionIdIsValid(xmax_new_tuple))
3765 : {
3766 523742 : infomask_new_tuple = HEAP_XMAX_INVALID;
3767 523742 : infomask2_new_tuple = 0;
3768 : }
3769 : else
3770 : {
3771 : /*
3772 : * If we found a valid Xmax for the new tuple, then the infomask bits
3773 : * to use on the new tuple depend on what was there on the old one.
3774 : * Note that since we're doing an update, the only possibility is that
3775 : * the lockers had FOR KEY SHARE lock.
3776 : */
3777 71742 : if (oldtup.t_data->t_infomask & HEAP_XMAX_IS_MULTI)
3778 : {
3779 106 : GetMultiXactIdHintBits(xmax_new_tuple, &infomask_new_tuple,
3780 : &infomask2_new_tuple);
3781 : }
3782 : else
3783 : {
3784 71636 : infomask_new_tuple = HEAP_XMAX_KEYSHR_LOCK | HEAP_XMAX_LOCK_ONLY;
3785 71636 : infomask2_new_tuple = 0;
3786 : }
3787 : }
3788 :
3789 : /*
3790 : * Prepare the new tuple with the appropriate initial values of Xmin and
3791 : * Xmax, as well as initial infomask bits as computed above.
3792 : */
3793 595484 : newtup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
3794 595484 : newtup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
3795 595484 : HeapTupleHeaderSetXmin(newtup->t_data, xid);
3796 595484 : HeapTupleHeaderSetCmin(newtup->t_data, cid);
3797 595484 : newtup->t_data->t_infomask |= HEAP_UPDATED | infomask_new_tuple;
3798 595484 : newtup->t_data->t_infomask2 |= infomask2_new_tuple;
3799 595484 : HeapTupleHeaderSetXmax(newtup->t_data, xmax_new_tuple);
3800 :
3801 : /*
3802 : * Replace cid with a combo CID if necessary. Note that we already put
3803 : * the plain cid into the new tuple.
3804 : */
3805 595484 : HeapTupleHeaderAdjustCmax(oldtup.t_data, &cid, &iscombo);
3806 :
3807 : /*
3808 : * If the toaster needs to be activated, OR if the new tuple will not fit
3809 : * on the same page as the old, then we need to release the content lock
3810 : * (but not the pin!) on the old tuple's buffer while we are off doing
3811 : * TOAST and/or table-file-extension work. We must mark the old tuple to
3812 : * show that it's locked, else other processes may try to update it
3813 : * themselves.
3814 : *
3815 : * We need to invoke the toaster if there are already any out-of-line
3816 : * toasted values present, or if the new tuple is over-threshold.
3817 : */
3818 595484 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
3819 0 : relation->rd_rel->relkind != RELKIND_MATVIEW)
3820 : {
3821 : /* toast table entries should never be recursively toasted */
3822 : Assert(!HeapTupleHasExternal(&oldtup));
3823 : Assert(!HeapTupleHasExternal(newtup));
3824 0 : need_toast = false;
3825 : }
3826 : else
3827 1785736 : need_toast = (HeapTupleHasExternal(&oldtup) ||
3828 1190252 : HeapTupleHasExternal(newtup) ||
3829 594720 : newtup->t_len > TOAST_TUPLE_THRESHOLD);
3830 :
3831 595484 : pagefree = PageGetHeapFreeSpace(page);
3832 :
3833 595484 : newtupsize = MAXALIGN(newtup->t_len);
3834 :
3835 595484 : if (need_toast || newtupsize > pagefree)
3836 293418 : {
3837 : TransactionId xmax_lock_old_tuple;
3838 : uint16 infomask_lock_old_tuple,
3839 : infomask2_lock_old_tuple;
3840 293418 : bool cleared_all_frozen = false;
3841 :
3842 : /*
3843 : * To prevent concurrent sessions from updating the tuple, we have to
3844 : * temporarily mark it locked, while we release the page-level lock.
3845 : *
3846 : * To satisfy the rule that any xid potentially appearing in a buffer
3847 : * written out to disk, we unfortunately have to WAL log this
3848 : * temporary modification. We can reuse xl_heap_lock for this
3849 : * purpose. If we crash/error before following through with the
3850 : * actual update, xmax will be of an aborted transaction, allowing
3851 : * other sessions to proceed.
3852 : */
3853 :
3854 : /*
3855 : * Compute xmax / infomask appropriate for locking the tuple. This has
3856 : * to be done separately from the combo that's going to be used for
3857 : * updating, because the potentially created multixact would otherwise
3858 : * be wrong.
3859 : */
3860 293418 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3861 293418 : oldtup.t_data->t_infomask,
3862 293418 : oldtup.t_data->t_infomask2,
3863 : xid, *lockmode, false,
3864 : &xmax_lock_old_tuple, &infomask_lock_old_tuple,
3865 : &infomask2_lock_old_tuple);
3866 :
3867 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(infomask_lock_old_tuple));
3868 :
3869 293418 : START_CRIT_SECTION();
3870 :
3871 : /* Clear obsolete visibility flags ... */
3872 293418 : oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
3873 293418 : oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
3874 293418 : HeapTupleClearHotUpdated(&oldtup);
3875 : /* ... and store info about transaction updating this tuple */
3876 : Assert(TransactionIdIsValid(xmax_lock_old_tuple));
3877 293418 : HeapTupleHeaderSetXmax(oldtup.t_data, xmax_lock_old_tuple);
3878 293418 : oldtup.t_data->t_infomask |= infomask_lock_old_tuple;
3879 293418 : oldtup.t_data->t_infomask2 |= infomask2_lock_old_tuple;
3880 293418 : HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
3881 :
3882 : /* temporarily make it look not-updated, but locked */
3883 293418 : oldtup.t_data->t_ctid = oldtup.t_self;
3884 :
3885 : /*
3886 : * Clear all-frozen bit on visibility map if needed. We could
3887 : * immediately reset ALL_VISIBLE, but given that the WAL logging
3888 : * overhead would be unchanged, that doesn't seem necessarily
3889 : * worthwhile.
3890 : */
3891 295126 : if (PageIsAllVisible(page) &&
3892 1708 : visibilitymap_clear(relation, block, vmbuffer,
3893 : VISIBILITYMAP_ALL_FROZEN))
3894 1402 : cleared_all_frozen = true;
3895 :
3896 293418 : MarkBufferDirty(buffer);
3897 :
3898 293418 : if (RelationNeedsWAL(relation))
3899 : {
3900 : xl_heap_lock xlrec;
3901 : XLogRecPtr recptr;
3902 :
3903 273162 : XLogBeginInsert();
3904 273162 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
3905 :
3906 273162 : xlrec.offnum = ItemPointerGetOffsetNumber(&oldtup.t_self);
3907 273162 : xlrec.xmax = xmax_lock_old_tuple;
3908 546324 : xlrec.infobits_set = compute_infobits(oldtup.t_data->t_infomask,
3909 273162 : oldtup.t_data->t_infomask2);
3910 273162 : xlrec.flags =
3911 273162 : cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
3912 273162 : XLogRegisterData(&xlrec, SizeOfHeapLock);
3913 273162 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
3914 273162 : PageSetLSN(page, recptr);
3915 : }
3916 :
3917 293418 : END_CRIT_SECTION();
3918 :
3919 293418 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3920 :
3921 : /*
3922 : * Let the toaster do its thing, if needed.
3923 : *
3924 : * Note: below this point, heaptup is the data we actually intend to
3925 : * store into the relation; newtup is the caller's original untoasted
3926 : * data.
3927 : */
3928 293418 : if (need_toast)
3929 : {
3930 : /* Note we always use WAL and FSM during updates */
3931 3078 : heaptup = heap_toast_insert_or_update(relation, newtup, &oldtup, 0);
3932 3078 : newtupsize = MAXALIGN(heaptup->t_len);
3933 : }
3934 : else
3935 290340 : heaptup = newtup;
3936 :
3937 : /*
3938 : * Now, do we need a new page for the tuple, or not? This is a bit
3939 : * tricky since someone else could have added tuples to the page while
3940 : * we weren't looking. We have to recheck the available space after
3941 : * reacquiring the buffer lock. But don't bother to do that if the
3942 : * former amount of free space is still not enough; it's unlikely
3943 : * there's more free now than before.
3944 : *
3945 : * What's more, if we need to get a new page, we will need to acquire
3946 : * buffer locks on both old and new pages. To avoid deadlock against
3947 : * some other backend trying to get the same two locks in the other
3948 : * order, we must be consistent about the order we get the locks in.
3949 : * We use the rule "lock the lower-numbered page of the relation
3950 : * first". To implement this, we must do RelationGetBufferForTuple
3951 : * while not holding the lock on the old page, and we must rely on it
3952 : * to get the locks on both pages in the correct order.
3953 : *
3954 : * Another consideration is that we need visibility map page pin(s) if
3955 : * we will have to clear the all-visible flag on either page. If we
3956 : * call RelationGetBufferForTuple, we rely on it to acquire any such
3957 : * pins; but if we don't, we have to handle that here. Hence we need
3958 : * a loop.
3959 : */
3960 : for (;;)
3961 : {
3962 293418 : if (newtupsize > pagefree)
3963 : {
3964 : /* It doesn't fit, must use RelationGetBufferForTuple. */
3965 292490 : newbuf = RelationGetBufferForTuple(relation, heaptup->t_len,
3966 : buffer, 0, NULL,
3967 : &vmbuffer_new, &vmbuffer,
3968 : 0);
3969 : /* We're all done. */
3970 292490 : break;
3971 : }
3972 : /* Acquire VM page pin if needed and we don't have it. */
3973 928 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
3974 0 : visibilitymap_pin(relation, block, &vmbuffer);
3975 : /* Re-acquire the lock on the old tuple's page. */
3976 928 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3977 : /* Re-check using the up-to-date free space */
3978 928 : pagefree = PageGetHeapFreeSpace(page);
3979 928 : if (newtupsize > pagefree ||
3980 928 : (vmbuffer == InvalidBuffer && PageIsAllVisible(page)))
3981 : {
3982 : /*
3983 : * Rats, it doesn't fit anymore, or somebody just now set the
3984 : * all-visible flag. We must now unlock and loop to avoid
3985 : * deadlock. Fortunately, this path should seldom be taken.
3986 : */
3987 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3988 : }
3989 : else
3990 : {
3991 : /* We're all done. */
3992 928 : newbuf = buffer;
3993 928 : break;
3994 : }
3995 : }
3996 : }
3997 : else
3998 : {
3999 : /* No TOAST work needed, and it'll fit on same page */
4000 302066 : newbuf = buffer;
4001 302066 : heaptup = newtup;
4002 : }
4003 :
4004 : /*
4005 : * We're about to do the actual update -- check for conflict first, to
4006 : * avoid possibly having to roll back work we've just done.
4007 : *
4008 : * This is safe without a recheck as long as there is no possibility of
4009 : * another process scanning the pages between this check and the update
4010 : * being visible to the scan (i.e., exclusive buffer content lock(s) are
4011 : * continuously held from this point until the tuple update is visible).
4012 : *
4013 : * For the new tuple the only check needed is at the relation level, but
4014 : * since both tuples are in the same relation and the check for oldtup
4015 : * will include checking the relation level, there is no benefit to a
4016 : * separate check for the new tuple.
4017 : */
4018 595484 : CheckForSerializableConflictIn(relation, &oldtup.t_self,
4019 : BufferGetBlockNumber(buffer));
4020 :
4021 : /*
4022 : * At this point newbuf and buffer are both pinned and locked, and newbuf
4023 : * has enough space for the new tuple. If they are the same buffer, only
4024 : * one pin is held.
4025 : */
4026 :
4027 595460 : if (newbuf == buffer)
4028 : {
4029 : /*
4030 : * Since the new tuple is going into the same page, we might be able
4031 : * to do a HOT update. Check if any of the index columns have been
4032 : * changed.
4033 : */
4034 302970 : if (!bms_overlap(modified_attrs, hot_attrs))
4035 : {
4036 279132 : use_hot_update = true;
4037 :
4038 : /*
4039 : * If none of the columns that are used in hot-blocking indexes
4040 : * were updated, we can apply HOT, but we do still need to check
4041 : * if we need to update the summarizing indexes, and update those
4042 : * indexes if the columns were updated, or we may fail to detect
4043 : * e.g. value bound changes in BRIN minmax indexes.
4044 : */
4045 279132 : if (bms_overlap(modified_attrs, sum_attrs))
4046 3282 : summarized_update = true;
4047 : }
4048 : }
4049 : else
4050 : {
4051 : /* Set a hint that the old page could use prune/defrag */
4052 292490 : PageSetFull(page);
4053 : }
4054 :
4055 : /*
4056 : * Compute replica identity tuple before entering the critical section so
4057 : * we don't PANIC upon a memory allocation failure.
4058 : * ExtractReplicaIdentity() will return NULL if nothing needs to be
4059 : * logged. Pass old key required as true only if the replica identity key
4060 : * columns are modified or it has external data.
4061 : */
4062 595460 : old_key_tuple = ExtractReplicaIdentity(relation, &oldtup,
4063 595460 : bms_overlap(modified_attrs, id_attrs) ||
4064 : id_has_external,
4065 : &old_key_copied);
4066 :
4067 : /* NO EREPORT(ERROR) from here till changes are logged */
4068 595460 : START_CRIT_SECTION();
4069 :
4070 : /*
4071 : * If this transaction commits, the old tuple will become DEAD sooner or
4072 : * later. Set flag that this page is a candidate for pruning once our xid
4073 : * falls below the OldestXmin horizon. If the transaction finally aborts,
4074 : * the subsequent page pruning will be a no-op and the hint will be
4075 : * cleared.
4076 : *
4077 : * XXX Should we set hint on newbuf as well? If the transaction aborts,
4078 : * there would be a prunable tuple in the newbuf; but for now we choose
4079 : * not to optimize for aborts. Note that heap_xlog_update must be kept in
4080 : * sync if this decision changes.
4081 : */
4082 595460 : PageSetPrunable(page, xid);
4083 :
4084 595460 : if (use_hot_update)
4085 : {
4086 : /* Mark the old tuple as HOT-updated */
4087 279132 : HeapTupleSetHotUpdated(&oldtup);
4088 : /* And mark the new tuple as heap-only */
4089 279132 : HeapTupleSetHeapOnly(heaptup);
4090 : /* Mark the caller's copy too, in case different from heaptup */
4091 279132 : HeapTupleSetHeapOnly(newtup);
4092 : }
4093 : else
4094 : {
4095 : /* Make sure tuples are correctly marked as not-HOT */
4096 316328 : HeapTupleClearHotUpdated(&oldtup);
4097 316328 : HeapTupleClearHeapOnly(heaptup);
4098 316328 : HeapTupleClearHeapOnly(newtup);
4099 : }
4100 :
4101 595460 : RelationPutHeapTuple(relation, newbuf, heaptup, false); /* insert new tuple */
4102 :
4103 :
4104 : /* Clear obsolete visibility flags, possibly set by ourselves above... */
4105 595460 : oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
4106 595460 : oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
4107 : /* ... and store info about transaction updating this tuple */
4108 : Assert(TransactionIdIsValid(xmax_old_tuple));
4109 595460 : HeapTupleHeaderSetXmax(oldtup.t_data, xmax_old_tuple);
4110 595460 : oldtup.t_data->t_infomask |= infomask_old_tuple;
4111 595460 : oldtup.t_data->t_infomask2 |= infomask2_old_tuple;
4112 595460 : HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
4113 :
4114 : /* record address of new tuple in t_ctid of old one */
4115 595460 : oldtup.t_data->t_ctid = heaptup->t_self;
4116 :
4117 : /* clear PD_ALL_VISIBLE flags, reset all visibilitymap bits */
4118 595460 : if (PageIsAllVisible(BufferGetPage(buffer)))
4119 : {
4120 3232 : all_visible_cleared = true;
4121 3232 : PageClearAllVisible(BufferGetPage(buffer));
4122 3232 : visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
4123 : vmbuffer, VISIBILITYMAP_VALID_BITS);
4124 : }
4125 595460 : if (newbuf != buffer && PageIsAllVisible(BufferGetPage(newbuf)))
4126 : {
4127 1486 : all_visible_cleared_new = true;
4128 1486 : PageClearAllVisible(BufferGetPage(newbuf));
4129 1486 : visibilitymap_clear(relation, BufferGetBlockNumber(newbuf),
4130 : vmbuffer_new, VISIBILITYMAP_VALID_BITS);
4131 : }
4132 :
4133 595460 : if (newbuf != buffer)
4134 292490 : MarkBufferDirty(newbuf);
4135 595460 : MarkBufferDirty(buffer);
4136 :
4137 : /* XLOG stuff */
4138 595460 : if (RelationNeedsWAL(relation))
4139 : {
4140 : XLogRecPtr recptr;
4141 :
4142 : /*
4143 : * For logical decoding we need combo CIDs to properly decode the
4144 : * catalog.
4145 : */
4146 572742 : if (RelationIsAccessibleInLogicalDecoding(relation))
4147 : {
4148 5346 : log_heap_new_cid(relation, &oldtup);
4149 5346 : log_heap_new_cid(relation, heaptup);
4150 : }
4151 :
4152 572742 : recptr = log_heap_update(relation, buffer,
4153 : newbuf, &oldtup, heaptup,
4154 : old_key_tuple,
4155 : all_visible_cleared,
4156 : all_visible_cleared_new);
4157 572742 : if (newbuf != buffer)
4158 : {
4159 272246 : PageSetLSN(BufferGetPage(newbuf), recptr);
4160 : }
4161 572742 : PageSetLSN(BufferGetPage(buffer), recptr);
4162 : }
4163 :
4164 595460 : END_CRIT_SECTION();
4165 :
4166 595460 : if (newbuf != buffer)
4167 292490 : LockBuffer(newbuf, BUFFER_LOCK_UNLOCK);
4168 595460 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
4169 :
4170 : /*
4171 : * Mark old tuple for invalidation from system caches at next command
4172 : * boundary, and mark the new tuple for invalidation in case we abort. We
4173 : * have to do this before releasing the buffer because oldtup is in the
4174 : * buffer. (heaptup is all in local memory, but it's necessary to process
4175 : * both tuple versions in one call to inval.c so we can avoid redundant
4176 : * sinval messages.)
4177 : */
4178 595460 : CacheInvalidateHeapTuple(relation, &oldtup, heaptup);
4179 :
4180 : /* Now we can release the buffer(s) */
4181 595460 : if (newbuf != buffer)
4182 292490 : ReleaseBuffer(newbuf);
4183 595460 : ReleaseBuffer(buffer);
4184 595460 : if (BufferIsValid(vmbuffer_new))
4185 1486 : ReleaseBuffer(vmbuffer_new);
4186 595460 : if (BufferIsValid(vmbuffer))
4187 3232 : ReleaseBuffer(vmbuffer);
4188 :
4189 : /*
4190 : * Release the lmgr tuple lock, if we had it.
4191 : */
4192 595460 : if (have_tuple_lock)
4193 30 : UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
4194 :
4195 595460 : pgstat_count_heap_update(relation, use_hot_update, newbuf != buffer);
4196 :
4197 : /*
4198 : * If heaptup is a private copy, release it. Don't forget to copy t_self
4199 : * back to the caller's image, too.
4200 : */
4201 595460 : if (heaptup != newtup)
4202 : {
4203 2984 : newtup->t_self = heaptup->t_self;
4204 2984 : heap_freetuple(heaptup);
4205 : }
4206 :
4207 : /*
4208 : * If it is a HOT update, the update may still need to update summarized
4209 : * indexes, lest we fail to update those summaries and get incorrect
4210 : * results (for example, minmax bounds of the block may change with this
4211 : * update).
4212 : */
4213 595460 : if (use_hot_update)
4214 : {
4215 279132 : if (summarized_update)
4216 3282 : *update_indexes = TU_Summarizing;
4217 : else
4218 275850 : *update_indexes = TU_None;
4219 : }
4220 : else
4221 316328 : *update_indexes = TU_All;
4222 :
4223 595460 : if (old_key_tuple != NULL && old_key_copied)
4224 166 : heap_freetuple(old_key_tuple);
4225 :
4226 595460 : bms_free(hot_attrs);
4227 595460 : bms_free(sum_attrs);
4228 595460 : bms_free(key_attrs);
4229 595460 : bms_free(id_attrs);
4230 595460 : bms_free(modified_attrs);
4231 595460 : bms_free(interesting_attrs);
4232 :
4233 595460 : return TM_Ok;
4234 : }
4235 :
4236 : #ifdef USE_ASSERT_CHECKING
4237 : /*
4238 : * Confirm adequate lock held during heap_update(), per rules from
4239 : * README.tuplock section "Locking to write inplace-updated tables".
4240 : */
4241 : static void
4242 : check_lock_if_inplace_updateable_rel(Relation relation,
4243 : ItemPointer otid,
4244 : HeapTuple newtup)
4245 : {
4246 : /* LOCKTAG_TUPLE acceptable for any catalog */
4247 : switch (RelationGetRelid(relation))
4248 : {
4249 : case RelationRelationId:
4250 : case DatabaseRelationId:
4251 : {
4252 : LOCKTAG tuptag;
4253 :
4254 : SET_LOCKTAG_TUPLE(tuptag,
4255 : relation->rd_lockInfo.lockRelId.dbId,
4256 : relation->rd_lockInfo.lockRelId.relId,
4257 : ItemPointerGetBlockNumber(otid),
4258 : ItemPointerGetOffsetNumber(otid));
4259 : if (LockHeldByMe(&tuptag, InplaceUpdateTupleLock, false))
4260 : return;
4261 : }
4262 : break;
4263 : default:
4264 : Assert(!IsInplaceUpdateRelation(relation));
4265 : return;
4266 : }
4267 :
4268 : switch (RelationGetRelid(relation))
4269 : {
4270 : case RelationRelationId:
4271 : {
4272 : /* LOCKTAG_TUPLE or LOCKTAG_RELATION ok */
4273 : Form_pg_class classForm = (Form_pg_class) GETSTRUCT(newtup);
4274 : Oid relid = classForm->oid;
4275 : Oid dbid;
4276 : LOCKTAG tag;
4277 :
4278 : if (IsSharedRelation(relid))
4279 : dbid = InvalidOid;
4280 : else
4281 : dbid = MyDatabaseId;
4282 :
4283 : if (classForm->relkind == RELKIND_INDEX)
4284 : {
4285 : Relation irel = index_open(relid, AccessShareLock);
4286 :
4287 : SET_LOCKTAG_RELATION(tag, dbid, irel->rd_index->indrelid);
4288 : index_close(irel, AccessShareLock);
4289 : }
4290 : else
4291 : SET_LOCKTAG_RELATION(tag, dbid, relid);
4292 :
4293 : if (!LockHeldByMe(&tag, ShareUpdateExclusiveLock, false) &&
4294 : !LockHeldByMe(&tag, ShareRowExclusiveLock, true))
4295 : elog(WARNING,
4296 : "missing lock for relation \"%s\" (OID %u, relkind %c) @ TID (%u,%u)",
4297 : NameStr(classForm->relname),
4298 : relid,
4299 : classForm->relkind,
4300 : ItemPointerGetBlockNumber(otid),
4301 : ItemPointerGetOffsetNumber(otid));
4302 : }
4303 : break;
4304 : case DatabaseRelationId:
4305 : {
4306 : /* LOCKTAG_TUPLE required */
4307 : Form_pg_database dbForm = (Form_pg_database) GETSTRUCT(newtup);
4308 :
4309 : elog(WARNING,
4310 : "missing lock on database \"%s\" (OID %u) @ TID (%u,%u)",
4311 : NameStr(dbForm->datname),
4312 : dbForm->oid,
4313 : ItemPointerGetBlockNumber(otid),
4314 : ItemPointerGetOffsetNumber(otid));
4315 : }
4316 : break;
4317 : }
4318 : }
4319 :
4320 : /*
4321 : * Confirm adequate relation lock held, per rules from README.tuplock section
4322 : * "Locking to write inplace-updated tables".
4323 : */
4324 : static void
4325 : check_inplace_rel_lock(HeapTuple oldtup)
4326 : {
4327 : Form_pg_class classForm = (Form_pg_class) GETSTRUCT(oldtup);
4328 : Oid relid = classForm->oid;
4329 : Oid dbid;
4330 : LOCKTAG tag;
4331 :
4332 : if (IsSharedRelation(relid))
4333 : dbid = InvalidOid;
4334 : else
4335 : dbid = MyDatabaseId;
4336 :
4337 : if (classForm->relkind == RELKIND_INDEX)
4338 : {
4339 : Relation irel = index_open(relid, AccessShareLock);
4340 :
4341 : SET_LOCKTAG_RELATION(tag, dbid, irel->rd_index->indrelid);
4342 : index_close(irel, AccessShareLock);
4343 : }
4344 : else
4345 : SET_LOCKTAG_RELATION(tag, dbid, relid);
4346 :
4347 : if (!LockHeldByMe(&tag, ShareUpdateExclusiveLock, true))
4348 : elog(WARNING,
4349 : "missing lock for relation \"%s\" (OID %u, relkind %c) @ TID (%u,%u)",
4350 : NameStr(classForm->relname),
4351 : relid,
4352 : classForm->relkind,
4353 : ItemPointerGetBlockNumber(&oldtup->t_self),
4354 : ItemPointerGetOffsetNumber(&oldtup->t_self));
4355 : }
4356 : #endif
4357 :
4358 : /*
4359 : * Check if the specified attribute's values are the same. Subroutine for
4360 : * HeapDetermineColumnsInfo.
4361 : */
4362 : static bool
4363 1417750 : heap_attr_equals(TupleDesc tupdesc, int attrnum, Datum value1, Datum value2,
4364 : bool isnull1, bool isnull2)
4365 : {
4366 : /*
4367 : * If one value is NULL and other is not, then they are certainly not
4368 : * equal
4369 : */
4370 1417750 : if (isnull1 != isnull2)
4371 90 : return false;
4372 :
4373 : /*
4374 : * If both are NULL, they can be considered equal.
4375 : */
4376 1417660 : if (isnull1)
4377 9982 : return true;
4378 :
4379 : /*
4380 : * We do simple binary comparison of the two datums. This may be overly
4381 : * strict because there can be multiple binary representations for the
4382 : * same logical value. But we should be OK as long as there are no false
4383 : * positives. Using a type-specific equality operator is messy because
4384 : * there could be multiple notions of equality in different operator
4385 : * classes; furthermore, we cannot safely invoke user-defined functions
4386 : * while holding exclusive buffer lock.
4387 : */
4388 1407678 : if (attrnum <= 0)
4389 : {
4390 : /* The only allowed system columns are OIDs, so do this */
4391 0 : return (DatumGetObjectId(value1) == DatumGetObjectId(value2));
4392 : }
4393 : else
4394 : {
4395 : CompactAttribute *att;
4396 :
4397 : Assert(attrnum <= tupdesc->natts);
4398 1407678 : att = TupleDescCompactAttr(tupdesc, attrnum - 1);
4399 1407678 : return datumIsEqual(value1, value2, att->attbyval, att->attlen);
4400 : }
4401 : }
4402 :
4403 : /*
4404 : * Check which columns are being updated.
4405 : *
4406 : * Given an updated tuple, determine (and return into the output bitmapset),
4407 : * from those listed as interesting, the set of columns that changed.
4408 : *
4409 : * has_external indicates if any of the unmodified attributes (from those
4410 : * listed as interesting) of the old tuple is a member of external_cols and is
4411 : * stored externally.
4412 : */
4413 : static Bitmapset *
4414 595788 : HeapDetermineColumnsInfo(Relation relation,
4415 : Bitmapset *interesting_cols,
4416 : Bitmapset *external_cols,
4417 : HeapTuple oldtup, HeapTuple newtup,
4418 : bool *has_external)
4419 : {
4420 : int attidx;
4421 595788 : Bitmapset *modified = NULL;
4422 595788 : TupleDesc tupdesc = RelationGetDescr(relation);
4423 :
4424 595788 : attidx = -1;
4425 2013538 : while ((attidx = bms_next_member(interesting_cols, attidx)) >= 0)
4426 : {
4427 : /* attidx is zero-based, attrnum is the normal attribute number */
4428 1417750 : AttrNumber attrnum = attidx + FirstLowInvalidHeapAttributeNumber;
4429 : Datum value1,
4430 : value2;
4431 : bool isnull1,
4432 : isnull2;
4433 :
4434 : /*
4435 : * If it's a whole-tuple reference, say "not equal". It's not really
4436 : * worth supporting this case, since it could only succeed after a
4437 : * no-op update, which is hardly a case worth optimizing for.
4438 : */
4439 1417750 : if (attrnum == 0)
4440 : {
4441 0 : modified = bms_add_member(modified, attidx);
4442 1356462 : continue;
4443 : }
4444 :
4445 : /*
4446 : * Likewise, automatically say "not equal" for any system attribute
4447 : * other than tableOID; we cannot expect these to be consistent in a
4448 : * HOT chain, or even to be set correctly yet in the new tuple.
4449 : */
4450 1417750 : if (attrnum < 0)
4451 : {
4452 0 : if (attrnum != TableOidAttributeNumber)
4453 : {
4454 0 : modified = bms_add_member(modified, attidx);
4455 0 : continue;
4456 : }
4457 : }
4458 :
4459 : /*
4460 : * Extract the corresponding values. XXX this is pretty inefficient
4461 : * if there are many indexed columns. Should we do a single
4462 : * heap_deform_tuple call on each tuple, instead? But that doesn't
4463 : * work for system columns ...
4464 : */
4465 1417750 : value1 = heap_getattr(oldtup, attrnum, tupdesc, &isnull1);
4466 1417750 : value2 = heap_getattr(newtup, attrnum, tupdesc, &isnull2);
4467 :
4468 1417750 : if (!heap_attr_equals(tupdesc, attrnum, value1,
4469 : value2, isnull1, isnull2))
4470 : {
4471 53212 : modified = bms_add_member(modified, attidx);
4472 53212 : continue;
4473 : }
4474 :
4475 : /*
4476 : * No need to check attributes that can't be stored externally. Note
4477 : * that system attributes can't be stored externally.
4478 : */
4479 1364538 : if (attrnum < 0 || isnull1 ||
4480 1354556 : TupleDescCompactAttr(tupdesc, attrnum - 1)->attlen != -1)
4481 1303250 : continue;
4482 :
4483 : /*
4484 : * Check if the old tuple's attribute is stored externally and is a
4485 : * member of external_cols.
4486 : */
4487 61298 : if (VARATT_IS_EXTERNAL((struct varlena *) DatumGetPointer(value1)) &&
4488 10 : bms_is_member(attidx, external_cols))
4489 4 : *has_external = true;
4490 : }
4491 :
4492 595788 : return modified;
4493 : }
4494 :
4495 : /*
4496 : * simple_heap_update - replace a tuple
4497 : *
4498 : * This routine may be used to update a tuple when concurrent updates of
4499 : * the target tuple are not expected (for example, because we have a lock
4500 : * on the relation associated with the tuple). Any failure is reported
4501 : * via ereport().
4502 : */
4503 : void
4504 210306 : simple_heap_update(Relation relation, ItemPointer otid, HeapTuple tup,
4505 : TU_UpdateIndexes *update_indexes)
4506 : {
4507 : TM_Result result;
4508 : TM_FailureData tmfd;
4509 : LockTupleMode lockmode;
4510 :
4511 210306 : result = heap_update(relation, otid, tup,
4512 : GetCurrentCommandId(true), InvalidSnapshot,
4513 : true /* wait for commit */ ,
4514 : &tmfd, &lockmode, update_indexes);
4515 210306 : switch (result)
4516 : {
4517 0 : case TM_SelfModified:
4518 : /* Tuple was already updated in current command? */
4519 0 : elog(ERROR, "tuple already updated by self");
4520 : break;
4521 :
4522 210304 : case TM_Ok:
4523 : /* done successfully */
4524 210304 : break;
4525 :
4526 0 : case TM_Updated:
4527 0 : elog(ERROR, "tuple concurrently updated");
4528 : break;
4529 :
4530 2 : case TM_Deleted:
4531 2 : elog(ERROR, "tuple concurrently deleted");
4532 : break;
4533 :
4534 0 : default:
4535 0 : elog(ERROR, "unrecognized heap_update status: %u", result);
4536 : break;
4537 : }
4538 210304 : }
4539 :
4540 :
4541 : /*
4542 : * Return the MultiXactStatus corresponding to the given tuple lock mode.
4543 : */
4544 : static MultiXactStatus
4545 2414 : get_mxact_status_for_lock(LockTupleMode mode, bool is_update)
4546 : {
4547 : int retval;
4548 :
4549 2414 : if (is_update)
4550 192 : retval = tupleLockExtraInfo[mode].updstatus;
4551 : else
4552 2222 : retval = tupleLockExtraInfo[mode].lockstatus;
4553 :
4554 2414 : if (retval == -1)
4555 0 : elog(ERROR, "invalid lock tuple mode %d/%s", mode,
4556 : is_update ? "true" : "false");
4557 :
4558 2414 : return (MultiXactStatus) retval;
4559 : }
4560 :
4561 : /*
4562 : * heap_lock_tuple - lock a tuple in shared or exclusive mode
4563 : *
4564 : * Note that this acquires a buffer pin, which the caller must release.
4565 : *
4566 : * Input parameters:
4567 : * relation: relation containing tuple (caller must hold suitable lock)
4568 : * tid: TID of tuple to lock
4569 : * cid: current command ID (used for visibility test, and stored into
4570 : * tuple's cmax if lock is successful)
4571 : * mode: indicates if shared or exclusive tuple lock is desired
4572 : * wait_policy: what to do if tuple lock is not available
4573 : * follow_updates: if true, follow the update chain to also lock descendant
4574 : * tuples.
4575 : *
4576 : * Output parameters:
4577 : * *tuple: all fields filled in
4578 : * *buffer: set to buffer holding tuple (pinned but not locked at exit)
4579 : * *tmfd: filled in failure cases (see below)
4580 : *
4581 : * Function results are the same as the ones for table_tuple_lock().
4582 : *
4583 : * In the failure cases other than TM_Invisible, the routine fills
4584 : * *tmfd with the tuple's t_ctid, t_xmax (resolving a possible MultiXact,
4585 : * if necessary), and t_cmax (the last only for TM_SelfModified,
4586 : * since we cannot obtain cmax from a combo CID generated by another
4587 : * transaction).
4588 : * See comments for struct TM_FailureData for additional info.
4589 : *
4590 : * See README.tuplock for a thorough explanation of this mechanism.
4591 : */
4592 : TM_Result
4593 169690 : heap_lock_tuple(Relation relation, HeapTuple tuple,
4594 : CommandId cid, LockTupleMode mode, LockWaitPolicy wait_policy,
4595 : bool follow_updates,
4596 : Buffer *buffer, TM_FailureData *tmfd)
4597 : {
4598 : TM_Result result;
4599 169690 : ItemPointer tid = &(tuple->t_self);
4600 : ItemId lp;
4601 : Page page;
4602 169690 : Buffer vmbuffer = InvalidBuffer;
4603 : BlockNumber block;
4604 : TransactionId xid,
4605 : xmax;
4606 : uint16 old_infomask,
4607 : new_infomask,
4608 : new_infomask2;
4609 169690 : bool first_time = true;
4610 169690 : bool skip_tuple_lock = false;
4611 169690 : bool have_tuple_lock = false;
4612 169690 : bool cleared_all_frozen = false;
4613 :
4614 169690 : *buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
4615 169690 : block = ItemPointerGetBlockNumber(tid);
4616 :
4617 : /*
4618 : * Before locking the buffer, pin the visibility map page if it appears to
4619 : * be necessary. Since we haven't got the lock yet, someone else might be
4620 : * in the middle of changing this, so we'll need to recheck after we have
4621 : * the lock.
4622 : */
4623 169690 : if (PageIsAllVisible(BufferGetPage(*buffer)))
4624 3318 : visibilitymap_pin(relation, block, &vmbuffer);
4625 :
4626 169690 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4627 :
4628 169690 : page = BufferGetPage(*buffer);
4629 169690 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
4630 : Assert(ItemIdIsNormal(lp));
4631 :
4632 169690 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
4633 169690 : tuple->t_len = ItemIdGetLength(lp);
4634 169690 : tuple->t_tableOid = RelationGetRelid(relation);
4635 :
4636 169716 : l3:
4637 169716 : result = HeapTupleSatisfiesUpdate(tuple, cid, *buffer);
4638 :
4639 169716 : if (result == TM_Invisible)
4640 : {
4641 : /*
4642 : * This is possible, but only when locking a tuple for ON CONFLICT
4643 : * UPDATE. We return this value here rather than throwing an error in
4644 : * order to give that case the opportunity to throw a more specific
4645 : * error.
4646 : */
4647 24 : result = TM_Invisible;
4648 24 : goto out_locked;
4649 : }
4650 169692 : else if (result == TM_BeingModified ||
4651 154004 : result == TM_Updated ||
4652 : result == TM_Deleted)
4653 : {
4654 : TransactionId xwait;
4655 : uint16 infomask;
4656 : uint16 infomask2;
4657 : bool require_sleep;
4658 : ItemPointerData t_ctid;
4659 :
4660 : /* must copy state data before unlocking buffer */
4661 15690 : xwait = HeapTupleHeaderGetRawXmax(tuple->t_data);
4662 15690 : infomask = tuple->t_data->t_infomask;
4663 15690 : infomask2 = tuple->t_data->t_infomask2;
4664 15690 : ItemPointerCopy(&tuple->t_data->t_ctid, &t_ctid);
4665 :
4666 15690 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
4667 :
4668 : /*
4669 : * If any subtransaction of the current top transaction already holds
4670 : * a lock as strong as or stronger than what we're requesting, we
4671 : * effectively hold the desired lock already. We *must* succeed
4672 : * without trying to take the tuple lock, else we will deadlock
4673 : * against anyone wanting to acquire a stronger lock.
4674 : *
4675 : * Note we only do this the first time we loop on the HTSU result;
4676 : * there is no point in testing in subsequent passes, because
4677 : * evidently our own transaction cannot have acquired a new lock after
4678 : * the first time we checked.
4679 : */
4680 15690 : if (first_time)
4681 : {
4682 15672 : first_time = false;
4683 :
4684 15672 : if (infomask & HEAP_XMAX_IS_MULTI)
4685 : {
4686 : int i;
4687 : int nmembers;
4688 : MultiXactMember *members;
4689 :
4690 : /*
4691 : * We don't need to allow old multixacts here; if that had
4692 : * been the case, HeapTupleSatisfiesUpdate would have returned
4693 : * MayBeUpdated and we wouldn't be here.
4694 : */
4695 : nmembers =
4696 176 : GetMultiXactIdMembers(xwait, &members, false,
4697 176 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
4698 :
4699 526 : for (i = 0; i < nmembers; i++)
4700 : {
4701 : /* only consider members of our own transaction */
4702 378 : if (!TransactionIdIsCurrentTransactionId(members[i].xid))
4703 280 : continue;
4704 :
4705 98 : if (TUPLOCK_from_mxstatus(members[i].status) >= mode)
4706 : {
4707 28 : pfree(members);
4708 28 : result = TM_Ok;
4709 28 : goto out_unlocked;
4710 : }
4711 : else
4712 : {
4713 : /*
4714 : * Disable acquisition of the heavyweight tuple lock.
4715 : * Otherwise, when promoting a weaker lock, we might
4716 : * deadlock with another locker that has acquired the
4717 : * heavyweight tuple lock and is waiting for our
4718 : * transaction to finish.
4719 : *
4720 : * Note that in this case we still need to wait for
4721 : * the multixact if required, to avoid acquiring
4722 : * conflicting locks.
4723 : */
4724 70 : skip_tuple_lock = true;
4725 : }
4726 : }
4727 :
4728 148 : if (members)
4729 148 : pfree(members);
4730 : }
4731 15496 : else if (TransactionIdIsCurrentTransactionId(xwait))
4732 : {
4733 13042 : switch (mode)
4734 : {
4735 324 : case LockTupleKeyShare:
4736 : Assert(HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) ||
4737 : HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4738 : HEAP_XMAX_IS_EXCL_LOCKED(infomask));
4739 324 : result = TM_Ok;
4740 324 : goto out_unlocked;
4741 12 : case LockTupleShare:
4742 24 : if (HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4743 12 : HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4744 : {
4745 0 : result = TM_Ok;
4746 0 : goto out_unlocked;
4747 : }
4748 12 : break;
4749 122 : case LockTupleNoKeyExclusive:
4750 122 : if (HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4751 : {
4752 100 : result = TM_Ok;
4753 100 : goto out_unlocked;
4754 : }
4755 22 : break;
4756 12584 : case LockTupleExclusive:
4757 12584 : if (HEAP_XMAX_IS_EXCL_LOCKED(infomask) &&
4758 2504 : infomask2 & HEAP_KEYS_UPDATED)
4759 : {
4760 2462 : result = TM_Ok;
4761 2462 : goto out_unlocked;
4762 : }
4763 10122 : break;
4764 : }
4765 2472 : }
4766 : }
4767 :
4768 : /*
4769 : * Initially assume that we will have to wait for the locking
4770 : * transaction(s) to finish. We check various cases below in which
4771 : * this can be turned off.
4772 : */
4773 12776 : require_sleep = true;
4774 12776 : if (mode == LockTupleKeyShare)
4775 : {
4776 : /*
4777 : * If we're requesting KeyShare, and there's no update present, we
4778 : * don't need to wait. Even if there is an update, we can still
4779 : * continue if the key hasn't been modified.
4780 : *
4781 : * However, if there are updates, we need to walk the update chain
4782 : * to mark future versions of the row as locked, too. That way,
4783 : * if somebody deletes that future version, we're protected
4784 : * against the key going away. This locking of future versions
4785 : * could block momentarily, if a concurrent transaction is
4786 : * deleting a key; or it could return a value to the effect that
4787 : * the transaction deleting the key has already committed. So we
4788 : * do this before re-locking the buffer; otherwise this would be
4789 : * prone to deadlocks.
4790 : *
4791 : * Note that the TID we're locking was grabbed before we unlocked
4792 : * the buffer. For it to change while we're not looking, the
4793 : * other properties we're testing for below after re-locking the
4794 : * buffer would also change, in which case we would restart this
4795 : * loop above.
4796 : */
4797 1174 : if (!(infomask2 & HEAP_KEYS_UPDATED))
4798 : {
4799 : bool updated;
4800 :
4801 1112 : updated = !HEAP_XMAX_IS_LOCKED_ONLY(infomask);
4802 :
4803 : /*
4804 : * If there are updates, follow the update chain; bail out if
4805 : * that cannot be done.
4806 : */
4807 1112 : if (follow_updates && updated)
4808 : {
4809 : TM_Result res;
4810 :
4811 100 : res = heap_lock_updated_tuple(relation, tuple, &t_ctid,
4812 : GetCurrentTransactionId(),
4813 : mode);
4814 100 : if (res != TM_Ok)
4815 : {
4816 12 : result = res;
4817 : /* recovery code expects to have buffer lock held */
4818 12 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4819 360 : goto failed;
4820 : }
4821 : }
4822 :
4823 1100 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4824 :
4825 : /*
4826 : * Make sure it's still an appropriate lock, else start over.
4827 : * Also, if it wasn't updated before we released the lock, but
4828 : * is updated now, we start over too; the reason is that we
4829 : * now need to follow the update chain to lock the new
4830 : * versions.
4831 : */
4832 1100 : if (!HeapTupleHeaderIsOnlyLocked(tuple->t_data) &&
4833 86 : ((tuple->t_data->t_infomask2 & HEAP_KEYS_UPDATED) ||
4834 86 : !updated))
4835 26 : goto l3;
4836 :
4837 : /* Things look okay, so we can skip sleeping */
4838 1100 : require_sleep = false;
4839 :
4840 : /*
4841 : * Note we allow Xmax to change here; other updaters/lockers
4842 : * could have modified it before we grabbed the buffer lock.
4843 : * However, this is not a problem, because with the recheck we
4844 : * just did we ensure that they still don't conflict with the
4845 : * lock we want.
4846 : */
4847 : }
4848 : }
4849 11602 : else if (mode == LockTupleShare)
4850 : {
4851 : /*
4852 : * If we're requesting Share, we can similarly avoid sleeping if
4853 : * there's no update and no exclusive lock present.
4854 : */
4855 882 : if (HEAP_XMAX_IS_LOCKED_ONLY(infomask) &&
4856 882 : !HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4857 : {
4858 870 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4859 :
4860 : /*
4861 : * Make sure it's still an appropriate lock, else start over.
4862 : * See above about allowing xmax to change.
4863 : */
4864 1740 : if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
4865 870 : HEAP_XMAX_IS_EXCL_LOCKED(tuple->t_data->t_infomask))
4866 0 : goto l3;
4867 870 : require_sleep = false;
4868 : }
4869 : }
4870 10720 : else if (mode == LockTupleNoKeyExclusive)
4871 : {
4872 : /*
4873 : * If we're requesting NoKeyExclusive, we might also be able to
4874 : * avoid sleeping; just ensure that there no conflicting lock
4875 : * already acquired.
4876 : */
4877 310 : if (infomask & HEAP_XMAX_IS_MULTI)
4878 : {
4879 52 : if (!DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
4880 : mode, NULL))
4881 : {
4882 : /*
4883 : * No conflict, but if the xmax changed under us in the
4884 : * meantime, start over.
4885 : */
4886 26 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4887 52 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4888 26 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4889 : xwait))
4890 0 : goto l3;
4891 :
4892 : /* otherwise, we're good */
4893 26 : require_sleep = false;
4894 : }
4895 : }
4896 258 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask))
4897 : {
4898 34 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4899 :
4900 : /* if the xmax changed in the meantime, start over */
4901 68 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4902 34 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4903 : xwait))
4904 0 : goto l3;
4905 : /* otherwise, we're good */
4906 34 : require_sleep = false;
4907 : }
4908 : }
4909 :
4910 : /*
4911 : * As a check independent from those above, we can also avoid sleeping
4912 : * if the current transaction is the sole locker of the tuple. Note
4913 : * that the strength of the lock already held is irrelevant; this is
4914 : * not about recording the lock in Xmax (which will be done regardless
4915 : * of this optimization, below). Also, note that the cases where we
4916 : * hold a lock stronger than we are requesting are already handled
4917 : * above by not doing anything.
4918 : *
4919 : * Note we only deal with the non-multixact case here; MultiXactIdWait
4920 : * is well equipped to deal with this situation on its own.
4921 : */
4922 23416 : if (require_sleep && !(infomask & HEAP_XMAX_IS_MULTI) &&
4923 10652 : TransactionIdIsCurrentTransactionId(xwait))
4924 : {
4925 : /* ... but if the xmax changed in the meantime, start over */
4926 10122 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4927 20244 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4928 10122 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4929 : xwait))
4930 0 : goto l3;
4931 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask));
4932 10122 : require_sleep = false;
4933 : }
4934 :
4935 : /*
4936 : * Time to sleep on the other transaction/multixact, if necessary.
4937 : *
4938 : * If the other transaction is an update/delete that's already
4939 : * committed, then sleeping cannot possibly do any good: if we're
4940 : * required to sleep, get out to raise an error instead.
4941 : *
4942 : * By here, we either have already acquired the buffer exclusive lock,
4943 : * or we must wait for the locking transaction or multixact; so below
4944 : * we ensure that we grab buffer lock after the sleep.
4945 : */
4946 12764 : if (require_sleep && (result == TM_Updated || result == TM_Deleted))
4947 : {
4948 272 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4949 272 : goto failed;
4950 : }
4951 12492 : else if (require_sleep)
4952 : {
4953 : /*
4954 : * Acquire tuple lock to establish our priority for the tuple, or
4955 : * die trying. LockTuple will release us when we are next-in-line
4956 : * for the tuple. We must do this even if we are share-locking,
4957 : * but not if we already have a weaker lock on the tuple.
4958 : *
4959 : * If we are forced to "start over" below, we keep the tuple lock;
4960 : * this arranges that we stay at the head of the line while
4961 : * rechecking tuple state.
4962 : */
4963 340 : if (!skip_tuple_lock &&
4964 308 : !heap_acquire_tuplock(relation, tid, mode, wait_policy,
4965 : &have_tuple_lock))
4966 : {
4967 : /*
4968 : * This can only happen if wait_policy is Skip and the lock
4969 : * couldn't be obtained.
4970 : */
4971 2 : result = TM_WouldBlock;
4972 : /* recovery code expects to have buffer lock held */
4973 2 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4974 2 : goto failed;
4975 : }
4976 :
4977 336 : if (infomask & HEAP_XMAX_IS_MULTI)
4978 : {
4979 80 : MultiXactStatus status = get_mxact_status_for_lock(mode, false);
4980 :
4981 : /* We only ever lock tuples, never update them */
4982 80 : if (status >= MultiXactStatusNoKeyUpdate)
4983 0 : elog(ERROR, "invalid lock mode in heap_lock_tuple");
4984 :
4985 : /* wait for multixact to end, or die trying */
4986 80 : switch (wait_policy)
4987 : {
4988 72 : case LockWaitBlock:
4989 72 : MultiXactIdWait((MultiXactId) xwait, status, infomask,
4990 : relation, &tuple->t_self, XLTW_Lock, NULL);
4991 72 : break;
4992 4 : case LockWaitSkip:
4993 4 : if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
4994 : status, infomask, relation,
4995 : NULL, false))
4996 : {
4997 4 : result = TM_WouldBlock;
4998 : /* recovery code expects to have buffer lock held */
4999 4 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5000 4 : goto failed;
5001 : }
5002 0 : break;
5003 4 : case LockWaitError:
5004 4 : if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
5005 : status, infomask, relation,
5006 : NULL, log_lock_failure))
5007 4 : ereport(ERROR,
5008 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5009 : errmsg("could not obtain lock on row in relation \"%s\"",
5010 : RelationGetRelationName(relation))));
5011 :
5012 0 : break;
5013 : }
5014 :
5015 : /*
5016 : * Of course, the multixact might not be done here: if we're
5017 : * requesting a light lock mode, other transactions with light
5018 : * locks could still be alive, as well as locks owned by our
5019 : * own xact or other subxacts of this backend. We need to
5020 : * preserve the surviving MultiXact members. Note that it
5021 : * isn't absolutely necessary in the latter case, but doing so
5022 : * is simpler.
5023 : */
5024 72 : }
5025 : else
5026 : {
5027 : /* wait for regular transaction to end, or die trying */
5028 256 : switch (wait_policy)
5029 : {
5030 178 : case LockWaitBlock:
5031 178 : XactLockTableWait(xwait, relation, &tuple->t_self,
5032 : XLTW_Lock);
5033 178 : break;
5034 66 : case LockWaitSkip:
5035 66 : if (!ConditionalXactLockTableWait(xwait, false))
5036 : {
5037 66 : result = TM_WouldBlock;
5038 : /* recovery code expects to have buffer lock held */
5039 66 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5040 66 : goto failed;
5041 : }
5042 0 : break;
5043 12 : case LockWaitError:
5044 12 : if (!ConditionalXactLockTableWait(xwait, log_lock_failure))
5045 12 : ereport(ERROR,
5046 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5047 : errmsg("could not obtain lock on row in relation \"%s\"",
5048 : RelationGetRelationName(relation))));
5049 0 : break;
5050 : }
5051 250 : }
5052 :
5053 : /* if there are updates, follow the update chain */
5054 250 : if (follow_updates && !HEAP_XMAX_IS_LOCKED_ONLY(infomask))
5055 : {
5056 : TM_Result res;
5057 :
5058 80 : res = heap_lock_updated_tuple(relation, tuple, &t_ctid,
5059 : GetCurrentTransactionId(),
5060 : mode);
5061 80 : if (res != TM_Ok)
5062 : {
5063 4 : result = res;
5064 : /* recovery code expects to have buffer lock held */
5065 4 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5066 4 : goto failed;
5067 : }
5068 : }
5069 :
5070 246 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5071 :
5072 : /*
5073 : * xwait is done, but if xwait had just locked the tuple then some
5074 : * other xact could update this tuple before we get to this point.
5075 : * Check for xmax change, and start over if so.
5076 : */
5077 470 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
5078 224 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
5079 : xwait))
5080 26 : goto l3;
5081 :
5082 220 : if (!(infomask & HEAP_XMAX_IS_MULTI))
5083 : {
5084 : /*
5085 : * Otherwise check if it committed or aborted. Note we cannot
5086 : * be here if the tuple was only locked by somebody who didn't
5087 : * conflict with us; that would have been handled above. So
5088 : * that transaction must necessarily be gone by now. But
5089 : * don't check for this in the multixact case, because some
5090 : * locker transactions might still be running.
5091 : */
5092 156 : UpdateXmaxHintBits(tuple->t_data, *buffer, xwait);
5093 : }
5094 : }
5095 :
5096 : /* By here, we're certain that we hold buffer exclusive lock again */
5097 :
5098 : /*
5099 : * We may lock if previous xmax aborted, or if it committed but only
5100 : * locked the tuple without updating it; or if we didn't have to wait
5101 : * at all for whatever reason.
5102 : */
5103 12372 : if (!require_sleep ||
5104 380 : (tuple->t_data->t_infomask & HEAP_XMAX_INVALID) ||
5105 288 : HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
5106 128 : HeapTupleHeaderIsOnlyLocked(tuple->t_data))
5107 12256 : result = TM_Ok;
5108 116 : else if (!ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid))
5109 92 : result = TM_Updated;
5110 : else
5111 24 : result = TM_Deleted;
5112 : }
5113 :
5114 154002 : failed:
5115 166734 : if (result != TM_Ok)
5116 : {
5117 : Assert(result == TM_SelfModified || result == TM_Updated ||
5118 : result == TM_Deleted || result == TM_WouldBlock);
5119 :
5120 : /*
5121 : * When locking a tuple under LockWaitSkip semantics and we fail with
5122 : * TM_WouldBlock above, it's possible for concurrent transactions to
5123 : * release the lock and set HEAP_XMAX_INVALID in the meantime. So
5124 : * this assert is slightly different from the equivalent one in
5125 : * heap_delete and heap_update.
5126 : */
5127 : Assert((result == TM_WouldBlock) ||
5128 : !(tuple->t_data->t_infomask & HEAP_XMAX_INVALID));
5129 : Assert(result != TM_Updated ||
5130 : !ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid));
5131 488 : tmfd->ctid = tuple->t_data->t_ctid;
5132 488 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(tuple->t_data);
5133 488 : if (result == TM_SelfModified)
5134 12 : tmfd->cmax = HeapTupleHeaderGetCmax(tuple->t_data);
5135 : else
5136 476 : tmfd->cmax = InvalidCommandId;
5137 488 : goto out_locked;
5138 : }
5139 :
5140 : /*
5141 : * If we didn't pin the visibility map page and the page has become all
5142 : * visible while we were busy locking the buffer, or during some
5143 : * subsequent window during which we had it unlocked, we'll have to unlock
5144 : * and re-lock, to avoid holding the buffer lock across I/O. That's a bit
5145 : * unfortunate, especially since we'll now have to recheck whether the
5146 : * tuple has been locked or updated under us, but hopefully it won't
5147 : * happen very often.
5148 : */
5149 166246 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
5150 : {
5151 0 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
5152 0 : visibilitymap_pin(relation, block, &vmbuffer);
5153 0 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5154 0 : goto l3;
5155 : }
5156 :
5157 166246 : xmax = HeapTupleHeaderGetRawXmax(tuple->t_data);
5158 166246 : old_infomask = tuple->t_data->t_infomask;
5159 :
5160 : /*
5161 : * If this is the first possibly-multixact-able operation in the current
5162 : * transaction, set my per-backend OldestMemberMXactId setting. We can be
5163 : * certain that the transaction will never become a member of any older
5164 : * MultiXactIds than that. (We have to do this even if we end up just
5165 : * using our own TransactionId below, since some other backend could
5166 : * incorporate our XID into a MultiXact immediately afterwards.)
5167 : */
5168 166246 : MultiXactIdSetOldestMember();
5169 :
5170 : /*
5171 : * Compute the new xmax and infomask to store into the tuple. Note we do
5172 : * not modify the tuple just yet, because that would leave it in the wrong
5173 : * state if multixact.c elogs.
5174 : */
5175 166246 : compute_new_xmax_infomask(xmax, old_infomask, tuple->t_data->t_infomask2,
5176 : GetCurrentTransactionId(), mode, false,
5177 : &xid, &new_infomask, &new_infomask2);
5178 :
5179 166246 : START_CRIT_SECTION();
5180 :
5181 : /*
5182 : * Store transaction information of xact locking the tuple.
5183 : *
5184 : * Note: Cmax is meaningless in this context, so don't set it; this avoids
5185 : * possibly generating a useless combo CID. Moreover, if we're locking a
5186 : * previously updated tuple, it's important to preserve the Cmax.
5187 : *
5188 : * Also reset the HOT UPDATE bit, but only if there's no update; otherwise
5189 : * we would break the HOT chain.
5190 : */
5191 166246 : tuple->t_data->t_infomask &= ~HEAP_XMAX_BITS;
5192 166246 : tuple->t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
5193 166246 : tuple->t_data->t_infomask |= new_infomask;
5194 166246 : tuple->t_data->t_infomask2 |= new_infomask2;
5195 166246 : if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
5196 166168 : HeapTupleHeaderClearHotUpdated(tuple->t_data);
5197 166246 : HeapTupleHeaderSetXmax(tuple->t_data, xid);
5198 :
5199 : /*
5200 : * Make sure there is no forward chain link in t_ctid. Note that in the
5201 : * cases where the tuple has been updated, we must not overwrite t_ctid,
5202 : * because it was set by the updater. Moreover, if the tuple has been
5203 : * updated, we need to follow the update chain to lock the new versions of
5204 : * the tuple as well.
5205 : */
5206 166246 : if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
5207 166168 : tuple->t_data->t_ctid = *tid;
5208 :
5209 : /* Clear only the all-frozen bit on visibility map if needed */
5210 169564 : if (PageIsAllVisible(page) &&
5211 3318 : visibilitymap_clear(relation, block, vmbuffer,
5212 : VISIBILITYMAP_ALL_FROZEN))
5213 30 : cleared_all_frozen = true;
5214 :
5215 :
5216 166246 : MarkBufferDirty(*buffer);
5217 :
5218 : /*
5219 : * XLOG stuff. You might think that we don't need an XLOG record because
5220 : * there is no state change worth restoring after a crash. You would be
5221 : * wrong however: we have just written either a TransactionId or a
5222 : * MultiXactId that may never have been seen on disk before, and we need
5223 : * to make sure that there are XLOG entries covering those ID numbers.
5224 : * Else the same IDs might be re-used after a crash, which would be
5225 : * disastrous if this page made it to disk before the crash. Essentially
5226 : * we have to enforce the WAL log-before-data rule even in this case.
5227 : * (Also, in a PITR log-shipping or 2PC environment, we have to have XLOG
5228 : * entries for everything anyway.)
5229 : */
5230 166246 : if (RelationNeedsWAL(relation))
5231 : {
5232 : xl_heap_lock xlrec;
5233 : XLogRecPtr recptr;
5234 :
5235 165558 : XLogBeginInsert();
5236 165558 : XLogRegisterBuffer(0, *buffer, REGBUF_STANDARD);
5237 :
5238 165558 : xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
5239 165558 : xlrec.xmax = xid;
5240 331116 : xlrec.infobits_set = compute_infobits(new_infomask,
5241 165558 : tuple->t_data->t_infomask2);
5242 165558 : xlrec.flags = cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
5243 165558 : XLogRegisterData(&xlrec, SizeOfHeapLock);
5244 :
5245 : /* we don't decode row locks atm, so no need to log the origin */
5246 :
5247 165558 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
5248 :
5249 165558 : PageSetLSN(page, recptr);
5250 : }
5251 :
5252 166246 : END_CRIT_SECTION();
5253 :
5254 166246 : result = TM_Ok;
5255 :
5256 166758 : out_locked:
5257 166758 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
5258 :
5259 169672 : out_unlocked:
5260 169672 : if (BufferIsValid(vmbuffer))
5261 3318 : ReleaseBuffer(vmbuffer);
5262 :
5263 : /*
5264 : * Don't update the visibility map here. Locking a tuple doesn't change
5265 : * visibility info.
5266 : */
5267 :
5268 : /*
5269 : * Now that we have successfully marked the tuple as locked, we can
5270 : * release the lmgr tuple lock, if we had it.
5271 : */
5272 169672 : if (have_tuple_lock)
5273 278 : UnlockTupleTuplock(relation, tid, mode);
5274 :
5275 169672 : return result;
5276 : }
5277 :
5278 : /*
5279 : * Acquire heavyweight lock on the given tuple, in preparation for acquiring
5280 : * its normal, Xmax-based tuple lock.
5281 : *
5282 : * have_tuple_lock is an input and output parameter: on input, it indicates
5283 : * whether the lock has previously been acquired (and this function does
5284 : * nothing in that case). If this function returns success, have_tuple_lock
5285 : * has been flipped to true.
5286 : *
5287 : * Returns false if it was unable to obtain the lock; this can only happen if
5288 : * wait_policy is Skip.
5289 : */
5290 : static bool
5291 526 : heap_acquire_tuplock(Relation relation, ItemPointer tid, LockTupleMode mode,
5292 : LockWaitPolicy wait_policy, bool *have_tuple_lock)
5293 : {
5294 526 : if (*have_tuple_lock)
5295 18 : return true;
5296 :
5297 508 : switch (wait_policy)
5298 : {
5299 426 : case LockWaitBlock:
5300 426 : LockTupleTuplock(relation, tid, mode);
5301 426 : break;
5302 :
5303 68 : case LockWaitSkip:
5304 68 : if (!ConditionalLockTupleTuplock(relation, tid, mode, false))
5305 2 : return false;
5306 66 : break;
5307 :
5308 14 : case LockWaitError:
5309 14 : if (!ConditionalLockTupleTuplock(relation, tid, mode, log_lock_failure))
5310 2 : ereport(ERROR,
5311 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5312 : errmsg("could not obtain lock on row in relation \"%s\"",
5313 : RelationGetRelationName(relation))));
5314 12 : break;
5315 : }
5316 504 : *have_tuple_lock = true;
5317 :
5318 504 : return true;
5319 : }
5320 :
5321 : /*
5322 : * Given an original set of Xmax and infomask, and a transaction (identified by
5323 : * add_to_xmax) acquiring a new lock of some mode, compute the new Xmax and
5324 : * corresponding infomasks to use on the tuple.
5325 : *
5326 : * Note that this might have side effects such as creating a new MultiXactId.
5327 : *
5328 : * Most callers will have called HeapTupleSatisfiesUpdate before this function;
5329 : * that will have set the HEAP_XMAX_INVALID bit if the xmax was a MultiXactId
5330 : * but it was not running anymore. There is a race condition, which is that the
5331 : * MultiXactId may have finished since then, but that uncommon case is handled
5332 : * either here, or within MultiXactIdExpand.
5333 : *
5334 : * There is a similar race condition possible when the old xmax was a regular
5335 : * TransactionId. We test TransactionIdIsInProgress again just to narrow the
5336 : * window, but it's still possible to end up creating an unnecessary
5337 : * MultiXactId. Fortunately this is harmless.
5338 : */
5339 : static void
5340 4206914 : compute_new_xmax_infomask(TransactionId xmax, uint16 old_infomask,
5341 : uint16 old_infomask2, TransactionId add_to_xmax,
5342 : LockTupleMode mode, bool is_update,
5343 : TransactionId *result_xmax, uint16 *result_infomask,
5344 : uint16 *result_infomask2)
5345 : {
5346 : TransactionId new_xmax;
5347 : uint16 new_infomask,
5348 : new_infomask2;
5349 :
5350 : Assert(TransactionIdIsCurrentTransactionId(add_to_xmax));
5351 :
5352 4206914 : l5:
5353 4206914 : new_infomask = 0;
5354 4206914 : new_infomask2 = 0;
5355 4206914 : if (old_infomask & HEAP_XMAX_INVALID)
5356 : {
5357 : /*
5358 : * No previous locker; we just insert our own TransactionId.
5359 : *
5360 : * Note that it's critical that this case be the first one checked,
5361 : * because there are several blocks below that come back to this one
5362 : * to implement certain optimizations; old_infomask might contain
5363 : * other dirty bits in those cases, but we don't really care.
5364 : */
5365 3996842 : if (is_update)
5366 : {
5367 3539130 : new_xmax = add_to_xmax;
5368 3539130 : if (mode == LockTupleExclusive)
5369 3016156 : new_infomask2 |= HEAP_KEYS_UPDATED;
5370 : }
5371 : else
5372 : {
5373 457712 : new_infomask |= HEAP_XMAX_LOCK_ONLY;
5374 457712 : switch (mode)
5375 : {
5376 5022 : case LockTupleKeyShare:
5377 5022 : new_xmax = add_to_xmax;
5378 5022 : new_infomask |= HEAP_XMAX_KEYSHR_LOCK;
5379 5022 : break;
5380 1442 : case LockTupleShare:
5381 1442 : new_xmax = add_to_xmax;
5382 1442 : new_infomask |= HEAP_XMAX_SHR_LOCK;
5383 1442 : break;
5384 260034 : case LockTupleNoKeyExclusive:
5385 260034 : new_xmax = add_to_xmax;
5386 260034 : new_infomask |= HEAP_XMAX_EXCL_LOCK;
5387 260034 : break;
5388 191214 : case LockTupleExclusive:
5389 191214 : new_xmax = add_to_xmax;
5390 191214 : new_infomask |= HEAP_XMAX_EXCL_LOCK;
5391 191214 : new_infomask2 |= HEAP_KEYS_UPDATED;
5392 191214 : break;
5393 0 : default:
5394 0 : new_xmax = InvalidTransactionId; /* silence compiler */
5395 0 : elog(ERROR, "invalid lock mode");
5396 : }
5397 : }
5398 : }
5399 210072 : else if (old_infomask & HEAP_XMAX_IS_MULTI)
5400 : {
5401 : MultiXactStatus new_status;
5402 :
5403 : /*
5404 : * Currently we don't allow XMAX_COMMITTED to be set for multis, so
5405 : * cross-check.
5406 : */
5407 : Assert(!(old_infomask & HEAP_XMAX_COMMITTED));
5408 :
5409 : /*
5410 : * A multixact together with LOCK_ONLY set but neither lock bit set
5411 : * (i.e. a pg_upgraded share locked tuple) cannot possibly be running
5412 : * anymore. This check is critical for databases upgraded by
5413 : * pg_upgrade; both MultiXactIdIsRunning and MultiXactIdExpand assume
5414 : * that such multis are never passed.
5415 : */
5416 250 : if (HEAP_LOCKED_UPGRADED(old_infomask))
5417 : {
5418 0 : old_infomask &= ~HEAP_XMAX_IS_MULTI;
5419 0 : old_infomask |= HEAP_XMAX_INVALID;
5420 0 : goto l5;
5421 : }
5422 :
5423 : /*
5424 : * If the XMAX is already a MultiXactId, then we need to expand it to
5425 : * include add_to_xmax; but if all the members were lockers and are
5426 : * all gone, we can do away with the IS_MULTI bit and just set
5427 : * add_to_xmax as the only locker/updater. If all lockers are gone
5428 : * and we have an updater that aborted, we can also do without a
5429 : * multi.
5430 : *
5431 : * The cost of doing GetMultiXactIdMembers would be paid by
5432 : * MultiXactIdExpand if we weren't to do this, so this check is not
5433 : * incurring extra work anyhow.
5434 : */
5435 250 : if (!MultiXactIdIsRunning(xmax, HEAP_XMAX_IS_LOCKED_ONLY(old_infomask)))
5436 : {
5437 48 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) ||
5438 16 : !TransactionIdDidCommit(MultiXactIdGetUpdateXid(xmax,
5439 : old_infomask)))
5440 : {
5441 : /*
5442 : * Reset these bits and restart; otherwise fall through to
5443 : * create a new multi below.
5444 : */
5445 48 : old_infomask &= ~HEAP_XMAX_IS_MULTI;
5446 48 : old_infomask |= HEAP_XMAX_INVALID;
5447 48 : goto l5;
5448 : }
5449 : }
5450 :
5451 202 : new_status = get_mxact_status_for_lock(mode, is_update);
5452 :
5453 202 : new_xmax = MultiXactIdExpand((MultiXactId) xmax, add_to_xmax,
5454 : new_status);
5455 202 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5456 : }
5457 209822 : else if (old_infomask & HEAP_XMAX_COMMITTED)
5458 : {
5459 : /*
5460 : * It's a committed update, so we need to preserve him as updater of
5461 : * the tuple.
5462 : */
5463 : MultiXactStatus status;
5464 : MultiXactStatus new_status;
5465 :
5466 26 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5467 0 : status = MultiXactStatusUpdate;
5468 : else
5469 26 : status = MultiXactStatusNoKeyUpdate;
5470 :
5471 26 : new_status = get_mxact_status_for_lock(mode, is_update);
5472 :
5473 : /*
5474 : * since it's not running, it's obviously impossible for the old
5475 : * updater to be identical to the current one, so we need not check
5476 : * for that case as we do in the block above.
5477 : */
5478 26 : new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5479 26 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5480 : }
5481 209796 : else if (TransactionIdIsInProgress(xmax))
5482 : {
5483 : /*
5484 : * If the XMAX is a valid, in-progress TransactionId, then we need to
5485 : * create a new MultiXactId that includes both the old locker or
5486 : * updater and our own TransactionId.
5487 : */
5488 : MultiXactStatus new_status;
5489 : MultiXactStatus old_status;
5490 : LockTupleMode old_mode;
5491 :
5492 209778 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
5493 : {
5494 209726 : if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
5495 11244 : old_status = MultiXactStatusForKeyShare;
5496 198482 : else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
5497 862 : old_status = MultiXactStatusForShare;
5498 197620 : else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
5499 : {
5500 197620 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5501 185490 : old_status = MultiXactStatusForUpdate;
5502 : else
5503 12130 : old_status = MultiXactStatusForNoKeyUpdate;
5504 : }
5505 : else
5506 : {
5507 : /*
5508 : * LOCK_ONLY can be present alone only when a page has been
5509 : * upgraded by pg_upgrade. But in that case,
5510 : * TransactionIdIsInProgress() should have returned false. We
5511 : * assume it's no longer locked in this case.
5512 : */
5513 0 : elog(WARNING, "LOCK_ONLY found for Xid in progress %u", xmax);
5514 0 : old_infomask |= HEAP_XMAX_INVALID;
5515 0 : old_infomask &= ~HEAP_XMAX_LOCK_ONLY;
5516 0 : goto l5;
5517 : }
5518 : }
5519 : else
5520 : {
5521 : /* it's an update, but which kind? */
5522 52 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5523 0 : old_status = MultiXactStatusUpdate;
5524 : else
5525 52 : old_status = MultiXactStatusNoKeyUpdate;
5526 : }
5527 :
5528 209778 : old_mode = TUPLOCK_from_mxstatus(old_status);
5529 :
5530 : /*
5531 : * If the lock to be acquired is for the same TransactionId as the
5532 : * existing lock, there's an optimization possible: consider only the
5533 : * strongest of both locks as the only one present, and restart.
5534 : */
5535 209778 : if (xmax == add_to_xmax)
5536 : {
5537 : /*
5538 : * Note that it's not possible for the original tuple to be
5539 : * updated: we wouldn't be here because the tuple would have been
5540 : * invisible and we wouldn't try to update it. As a subtlety,
5541 : * this code can also run when traversing an update chain to lock
5542 : * future versions of a tuple. But we wouldn't be here either,
5543 : * because the add_to_xmax would be different from the original
5544 : * updater.
5545 : */
5546 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
5547 :
5548 : /* acquire the strongest of both */
5549 207738 : if (mode < old_mode)
5550 104340 : mode = old_mode;
5551 : /* mustn't touch is_update */
5552 :
5553 207738 : old_infomask |= HEAP_XMAX_INVALID;
5554 207738 : goto l5;
5555 : }
5556 :
5557 : /* otherwise, just fall back to creating a new multixact */
5558 2040 : new_status = get_mxact_status_for_lock(mode, is_update);
5559 2040 : new_xmax = MultiXactIdCreate(xmax, old_status,
5560 : add_to_xmax, new_status);
5561 2040 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5562 : }
5563 28 : else if (!HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) &&
5564 10 : TransactionIdDidCommit(xmax))
5565 2 : {
5566 : /*
5567 : * It's a committed update, so we gotta preserve him as updater of the
5568 : * tuple.
5569 : */
5570 : MultiXactStatus status;
5571 : MultiXactStatus new_status;
5572 :
5573 2 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5574 0 : status = MultiXactStatusUpdate;
5575 : else
5576 2 : status = MultiXactStatusNoKeyUpdate;
5577 :
5578 2 : new_status = get_mxact_status_for_lock(mode, is_update);
5579 :
5580 : /*
5581 : * since it's not running, it's obviously impossible for the old
5582 : * updater to be identical to the current one, so we need not check
5583 : * for that case as we do in the block above.
5584 : */
5585 2 : new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5586 2 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5587 : }
5588 : else
5589 : {
5590 : /*
5591 : * Can get here iff the locking/updating transaction was running when
5592 : * the infomask was extracted from the tuple, but finished before
5593 : * TransactionIdIsInProgress got to run. Deal with it as if there was
5594 : * no locker at all in the first place.
5595 : */
5596 16 : old_infomask |= HEAP_XMAX_INVALID;
5597 16 : goto l5;
5598 : }
5599 :
5600 3999112 : *result_infomask = new_infomask;
5601 3999112 : *result_infomask2 = new_infomask2;
5602 3999112 : *result_xmax = new_xmax;
5603 3999112 : }
5604 :
5605 : /*
5606 : * Subroutine for heap_lock_updated_tuple_rec.
5607 : *
5608 : * Given a hypothetical multixact status held by the transaction identified
5609 : * with the given xid, does the current transaction need to wait, fail, or can
5610 : * it continue if it wanted to acquire a lock of the given mode? "needwait"
5611 : * is set to true if waiting is necessary; if it can continue, then TM_Ok is
5612 : * returned. If the lock is already held by the current transaction, return
5613 : * TM_SelfModified. In case of a conflict with another transaction, a
5614 : * different HeapTupleSatisfiesUpdate return code is returned.
5615 : *
5616 : * The held status is said to be hypothetical because it might correspond to a
5617 : * lock held by a single Xid, i.e. not a real MultiXactId; we express it this
5618 : * way for simplicity of API.
5619 : */
5620 : static TM_Result
5621 64 : test_lockmode_for_conflict(MultiXactStatus status, TransactionId xid,
5622 : LockTupleMode mode, HeapTuple tup,
5623 : bool *needwait)
5624 : {
5625 : MultiXactStatus wantedstatus;
5626 :
5627 64 : *needwait = false;
5628 64 : wantedstatus = get_mxact_status_for_lock(mode, false);
5629 :
5630 : /*
5631 : * Note: we *must* check TransactionIdIsInProgress before
5632 : * TransactionIdDidAbort/Commit; see comment at top of heapam_visibility.c
5633 : * for an explanation.
5634 : */
5635 64 : if (TransactionIdIsCurrentTransactionId(xid))
5636 : {
5637 : /*
5638 : * The tuple has already been locked by our own transaction. This is
5639 : * very rare but can happen if multiple transactions are trying to
5640 : * lock an ancient version of the same tuple.
5641 : */
5642 0 : return TM_SelfModified;
5643 : }
5644 64 : else if (TransactionIdIsInProgress(xid))
5645 : {
5646 : /*
5647 : * If the locking transaction is running, what we do depends on
5648 : * whether the lock modes conflict: if they do, then we must wait for
5649 : * it to finish; otherwise we can fall through to lock this tuple
5650 : * version without waiting.
5651 : */
5652 32 : if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
5653 32 : LOCKMODE_from_mxstatus(wantedstatus)))
5654 : {
5655 16 : *needwait = true;
5656 : }
5657 :
5658 : /*
5659 : * If we set needwait above, then this value doesn't matter;
5660 : * otherwise, this value signals to caller that it's okay to proceed.
5661 : */
5662 32 : return TM_Ok;
5663 : }
5664 32 : else if (TransactionIdDidAbort(xid))
5665 6 : return TM_Ok;
5666 26 : else if (TransactionIdDidCommit(xid))
5667 : {
5668 : /*
5669 : * The other transaction committed. If it was only a locker, then the
5670 : * lock is completely gone now and we can return success; but if it
5671 : * was an update, then what we do depends on whether the two lock
5672 : * modes conflict. If they conflict, then we must report error to
5673 : * caller. But if they don't, we can fall through to allow the current
5674 : * transaction to lock the tuple.
5675 : *
5676 : * Note: the reason we worry about ISUPDATE here is because as soon as
5677 : * a transaction ends, all its locks are gone and meaningless, and
5678 : * thus we can ignore them; whereas its updates persist. In the
5679 : * TransactionIdIsInProgress case, above, we don't need to check
5680 : * because we know the lock is still "alive" and thus a conflict needs
5681 : * always be checked.
5682 : */
5683 26 : if (!ISUPDATE_from_mxstatus(status))
5684 8 : return TM_Ok;
5685 :
5686 18 : if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
5687 18 : LOCKMODE_from_mxstatus(wantedstatus)))
5688 : {
5689 : /* bummer */
5690 16 : if (!ItemPointerEquals(&tup->t_self, &tup->t_data->t_ctid))
5691 12 : return TM_Updated;
5692 : else
5693 4 : return TM_Deleted;
5694 : }
5695 :
5696 2 : return TM_Ok;
5697 : }
5698 :
5699 : /* Not in progress, not aborted, not committed -- must have crashed */
5700 0 : return TM_Ok;
5701 : }
5702 :
5703 :
5704 : /*
5705 : * Recursive part of heap_lock_updated_tuple
5706 : *
5707 : * Fetch the tuple pointed to by tid in rel, and mark it as locked by the given
5708 : * xid with the given mode; if this tuple is updated, recurse to lock the new
5709 : * version as well.
5710 : */
5711 : static TM_Result
5712 162 : heap_lock_updated_tuple_rec(Relation rel, ItemPointer tid, TransactionId xid,
5713 : LockTupleMode mode)
5714 : {
5715 : TM_Result result;
5716 : ItemPointerData tupid;
5717 : HeapTupleData mytup;
5718 : Buffer buf;
5719 : uint16 new_infomask,
5720 : new_infomask2,
5721 : old_infomask,
5722 : old_infomask2;
5723 : TransactionId xmax,
5724 : new_xmax;
5725 162 : TransactionId priorXmax = InvalidTransactionId;
5726 162 : bool cleared_all_frozen = false;
5727 : bool pinned_desired_page;
5728 162 : Buffer vmbuffer = InvalidBuffer;
5729 : BlockNumber block;
5730 :
5731 162 : ItemPointerCopy(tid, &tupid);
5732 :
5733 : for (;;)
5734 : {
5735 168 : new_infomask = 0;
5736 168 : new_xmax = InvalidTransactionId;
5737 168 : block = ItemPointerGetBlockNumber(&tupid);
5738 168 : ItemPointerCopy(&tupid, &(mytup.t_self));
5739 :
5740 168 : if (!heap_fetch(rel, SnapshotAny, &mytup, &buf, false))
5741 : {
5742 : /*
5743 : * if we fail to find the updated version of the tuple, it's
5744 : * because it was vacuumed/pruned away after its creator
5745 : * transaction aborted. So behave as if we got to the end of the
5746 : * chain, and there's no further tuple to lock: return success to
5747 : * caller.
5748 : */
5749 0 : result = TM_Ok;
5750 0 : goto out_unlocked;
5751 : }
5752 :
5753 168 : l4:
5754 184 : CHECK_FOR_INTERRUPTS();
5755 :
5756 : /*
5757 : * Before locking the buffer, pin the visibility map page if it
5758 : * appears to be necessary. Since we haven't got the lock yet,
5759 : * someone else might be in the middle of changing this, so we'll need
5760 : * to recheck after we have the lock.
5761 : */
5762 184 : if (PageIsAllVisible(BufferGetPage(buf)))
5763 : {
5764 0 : visibilitymap_pin(rel, block, &vmbuffer);
5765 0 : pinned_desired_page = true;
5766 : }
5767 : else
5768 184 : pinned_desired_page = false;
5769 :
5770 184 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
5771 :
5772 : /*
5773 : * If we didn't pin the visibility map page and the page has become
5774 : * all visible while we were busy locking the buffer, we'll have to
5775 : * unlock and re-lock, to avoid holding the buffer lock across I/O.
5776 : * That's a bit unfortunate, but hopefully shouldn't happen often.
5777 : *
5778 : * Note: in some paths through this function, we will reach here
5779 : * holding a pin on a vm page that may or may not be the one matching
5780 : * this page. If this page isn't all-visible, we won't use the vm
5781 : * page, but we hold onto such a pin till the end of the function.
5782 : */
5783 184 : if (!pinned_desired_page && PageIsAllVisible(BufferGetPage(buf)))
5784 : {
5785 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5786 0 : visibilitymap_pin(rel, block, &vmbuffer);
5787 0 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
5788 : }
5789 :
5790 : /*
5791 : * Check the tuple XMIN against prior XMAX, if any. If we reached the
5792 : * end of the chain, we're done, so return success.
5793 : */
5794 190 : if (TransactionIdIsValid(priorXmax) &&
5795 6 : !TransactionIdEquals(HeapTupleHeaderGetXmin(mytup.t_data),
5796 : priorXmax))
5797 : {
5798 0 : result = TM_Ok;
5799 0 : goto out_locked;
5800 : }
5801 :
5802 : /*
5803 : * Also check Xmin: if this tuple was created by an aborted
5804 : * (sub)transaction, then we already locked the last live one in the
5805 : * chain, thus we're done, so return success.
5806 : */
5807 184 : if (TransactionIdDidAbort(HeapTupleHeaderGetXmin(mytup.t_data)))
5808 : {
5809 26 : result = TM_Ok;
5810 26 : goto out_locked;
5811 : }
5812 :
5813 158 : old_infomask = mytup.t_data->t_infomask;
5814 158 : old_infomask2 = mytup.t_data->t_infomask2;
5815 158 : xmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
5816 :
5817 : /*
5818 : * If this tuple version has been updated or locked by some concurrent
5819 : * transaction(s), what we do depends on whether our lock mode
5820 : * conflicts with what those other transactions hold, and also on the
5821 : * status of them.
5822 : */
5823 158 : if (!(old_infomask & HEAP_XMAX_INVALID))
5824 : {
5825 : TransactionId rawxmax;
5826 : bool needwait;
5827 :
5828 60 : rawxmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
5829 60 : if (old_infomask & HEAP_XMAX_IS_MULTI)
5830 : {
5831 : int nmembers;
5832 : int i;
5833 : MultiXactMember *members;
5834 :
5835 : /*
5836 : * We don't need a test for pg_upgrade'd tuples: this is only
5837 : * applied to tuples after the first in an update chain. Said
5838 : * first tuple in the chain may well be locked-in-9.2-and-
5839 : * pg_upgraded, but that one was already locked by our caller,
5840 : * not us; and any subsequent ones cannot be because our
5841 : * caller must necessarily have obtained a snapshot later than
5842 : * the pg_upgrade itself.
5843 : */
5844 : Assert(!HEAP_LOCKED_UPGRADED(mytup.t_data->t_infomask));
5845 :
5846 2 : nmembers = GetMultiXactIdMembers(rawxmax, &members, false,
5847 2 : HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
5848 8 : for (i = 0; i < nmembers; i++)
5849 : {
5850 6 : result = test_lockmode_for_conflict(members[i].status,
5851 6 : members[i].xid,
5852 : mode,
5853 : &mytup,
5854 : &needwait);
5855 :
5856 : /*
5857 : * If the tuple was already locked by ourselves in a
5858 : * previous iteration of this (say heap_lock_tuple was
5859 : * forced to restart the locking loop because of a change
5860 : * in xmax), then we hold the lock already on this tuple
5861 : * version and we don't need to do anything; and this is
5862 : * not an error condition either. We just need to skip
5863 : * this tuple and continue locking the next version in the
5864 : * update chain.
5865 : */
5866 6 : if (result == TM_SelfModified)
5867 : {
5868 0 : pfree(members);
5869 0 : goto next;
5870 : }
5871 :
5872 6 : if (needwait)
5873 : {
5874 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5875 0 : XactLockTableWait(members[i].xid, rel,
5876 : &mytup.t_self,
5877 : XLTW_LockUpdated);
5878 0 : pfree(members);
5879 0 : goto l4;
5880 : }
5881 6 : if (result != TM_Ok)
5882 : {
5883 0 : pfree(members);
5884 0 : goto out_locked;
5885 : }
5886 : }
5887 2 : if (members)
5888 2 : pfree(members);
5889 : }
5890 : else
5891 : {
5892 : MultiXactStatus status;
5893 :
5894 : /*
5895 : * For a non-multi Xmax, we first need to compute the
5896 : * corresponding MultiXactStatus by using the infomask bits.
5897 : */
5898 58 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
5899 : {
5900 20 : if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
5901 20 : status = MultiXactStatusForKeyShare;
5902 0 : else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
5903 0 : status = MultiXactStatusForShare;
5904 0 : else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
5905 : {
5906 0 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5907 0 : status = MultiXactStatusForUpdate;
5908 : else
5909 0 : status = MultiXactStatusForNoKeyUpdate;
5910 : }
5911 : else
5912 : {
5913 : /*
5914 : * LOCK_ONLY present alone (a pg_upgraded tuple marked
5915 : * as share-locked in the old cluster) shouldn't be
5916 : * seen in the middle of an update chain.
5917 : */
5918 0 : elog(ERROR, "invalid lock status in tuple");
5919 : }
5920 : }
5921 : else
5922 : {
5923 : /* it's an update, but which kind? */
5924 38 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5925 28 : status = MultiXactStatusUpdate;
5926 : else
5927 10 : status = MultiXactStatusNoKeyUpdate;
5928 : }
5929 :
5930 58 : result = test_lockmode_for_conflict(status, rawxmax, mode,
5931 : &mytup, &needwait);
5932 :
5933 : /*
5934 : * If the tuple was already locked by ourselves in a previous
5935 : * iteration of this (say heap_lock_tuple was forced to
5936 : * restart the locking loop because of a change in xmax), then
5937 : * we hold the lock already on this tuple version and we don't
5938 : * need to do anything; and this is not an error condition
5939 : * either. We just need to skip this tuple and continue
5940 : * locking the next version in the update chain.
5941 : */
5942 58 : if (result == TM_SelfModified)
5943 0 : goto next;
5944 :
5945 58 : if (needwait)
5946 : {
5947 16 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5948 16 : XactLockTableWait(rawxmax, rel, &mytup.t_self,
5949 : XLTW_LockUpdated);
5950 16 : goto l4;
5951 : }
5952 42 : if (result != TM_Ok)
5953 : {
5954 16 : goto out_locked;
5955 : }
5956 : }
5957 : }
5958 :
5959 : /* compute the new Xmax and infomask values for the tuple ... */
5960 126 : compute_new_xmax_infomask(xmax, old_infomask, mytup.t_data->t_infomask2,
5961 : xid, mode, false,
5962 : &new_xmax, &new_infomask, &new_infomask2);
5963 :
5964 126 : if (PageIsAllVisible(BufferGetPage(buf)) &&
5965 0 : visibilitymap_clear(rel, block, vmbuffer,
5966 : VISIBILITYMAP_ALL_FROZEN))
5967 0 : cleared_all_frozen = true;
5968 :
5969 126 : START_CRIT_SECTION();
5970 :
5971 : /* ... and set them */
5972 126 : HeapTupleHeaderSetXmax(mytup.t_data, new_xmax);
5973 126 : mytup.t_data->t_infomask &= ~HEAP_XMAX_BITS;
5974 126 : mytup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
5975 126 : mytup.t_data->t_infomask |= new_infomask;
5976 126 : mytup.t_data->t_infomask2 |= new_infomask2;
5977 :
5978 126 : MarkBufferDirty(buf);
5979 :
5980 : /* XLOG stuff */
5981 126 : if (RelationNeedsWAL(rel))
5982 : {
5983 : xl_heap_lock_updated xlrec;
5984 : XLogRecPtr recptr;
5985 126 : Page page = BufferGetPage(buf);
5986 :
5987 126 : XLogBeginInsert();
5988 126 : XLogRegisterBuffer(0, buf, REGBUF_STANDARD);
5989 :
5990 126 : xlrec.offnum = ItemPointerGetOffsetNumber(&mytup.t_self);
5991 126 : xlrec.xmax = new_xmax;
5992 126 : xlrec.infobits_set = compute_infobits(new_infomask, new_infomask2);
5993 126 : xlrec.flags =
5994 126 : cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
5995 :
5996 126 : XLogRegisterData(&xlrec, SizeOfHeapLockUpdated);
5997 :
5998 126 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_LOCK_UPDATED);
5999 :
6000 126 : PageSetLSN(page, recptr);
6001 : }
6002 :
6003 126 : END_CRIT_SECTION();
6004 :
6005 126 : next:
6006 : /* if we find the end of update chain, we're done. */
6007 252 : if (mytup.t_data->t_infomask & HEAP_XMAX_INVALID ||
6008 252 : HeapTupleHeaderIndicatesMovedPartitions(mytup.t_data) ||
6009 134 : ItemPointerEquals(&mytup.t_self, &mytup.t_data->t_ctid) ||
6010 8 : HeapTupleHeaderIsOnlyLocked(mytup.t_data))
6011 : {
6012 120 : result = TM_Ok;
6013 120 : goto out_locked;
6014 : }
6015 :
6016 : /* tail recursion */
6017 6 : priorXmax = HeapTupleHeaderGetUpdateXid(mytup.t_data);
6018 6 : ItemPointerCopy(&(mytup.t_data->t_ctid), &tupid);
6019 6 : UnlockReleaseBuffer(buf);
6020 : }
6021 :
6022 : result = TM_Ok;
6023 :
6024 162 : out_locked:
6025 162 : UnlockReleaseBuffer(buf);
6026 :
6027 162 : out_unlocked:
6028 162 : if (vmbuffer != InvalidBuffer)
6029 0 : ReleaseBuffer(vmbuffer);
6030 :
6031 162 : return result;
6032 : }
6033 :
6034 : /*
6035 : * heap_lock_updated_tuple
6036 : * Follow update chain when locking an updated tuple, acquiring locks (row
6037 : * marks) on the updated versions.
6038 : *
6039 : * The initial tuple is assumed to be already locked.
6040 : *
6041 : * This function doesn't check visibility, it just unconditionally marks the
6042 : * tuple(s) as locked. If any tuple in the updated chain is being deleted
6043 : * concurrently (or updated with the key being modified), sleep until the
6044 : * transaction doing it is finished.
6045 : *
6046 : * Note that we don't acquire heavyweight tuple locks on the tuples we walk
6047 : * when we have to wait for other transactions to release them, as opposed to
6048 : * what heap_lock_tuple does. The reason is that having more than one
6049 : * transaction walking the chain is probably uncommon enough that risk of
6050 : * starvation is not likely: one of the preconditions for being here is that
6051 : * the snapshot in use predates the update that created this tuple (because we
6052 : * started at an earlier version of the tuple), but at the same time such a
6053 : * transaction cannot be using repeatable read or serializable isolation
6054 : * levels, because that would lead to a serializability failure.
6055 : */
6056 : static TM_Result
6057 180 : heap_lock_updated_tuple(Relation rel, HeapTuple tuple, ItemPointer ctid,
6058 : TransactionId xid, LockTupleMode mode)
6059 : {
6060 : /*
6061 : * If the tuple has not been updated, or has moved into another partition
6062 : * (effectively a delete) stop here.
6063 : */
6064 180 : if (!HeapTupleHeaderIndicatesMovedPartitions(tuple->t_data) &&
6065 176 : !ItemPointerEquals(&tuple->t_self, ctid))
6066 : {
6067 : /*
6068 : * If this is the first possibly-multixact-able operation in the
6069 : * current transaction, set my per-backend OldestMemberMXactId
6070 : * setting. We can be certain that the transaction will never become a
6071 : * member of any older MultiXactIds than that. (We have to do this
6072 : * even if we end up just using our own TransactionId below, since
6073 : * some other backend could incorporate our XID into a MultiXact
6074 : * immediately afterwards.)
6075 : */
6076 162 : MultiXactIdSetOldestMember();
6077 :
6078 162 : return heap_lock_updated_tuple_rec(rel, ctid, xid, mode);
6079 : }
6080 :
6081 : /* nothing to lock */
6082 18 : return TM_Ok;
6083 : }
6084 :
6085 : /*
6086 : * heap_finish_speculative - mark speculative insertion as successful
6087 : *
6088 : * To successfully finish a speculative insertion we have to clear speculative
6089 : * token from tuple. To do so the t_ctid field, which will contain a
6090 : * speculative token value, is modified in place to point to the tuple itself,
6091 : * which is characteristic of a newly inserted ordinary tuple.
6092 : *
6093 : * NB: It is not ok to commit without either finishing or aborting a
6094 : * speculative insertion. We could treat speculative tuples of committed
6095 : * transactions implicitly as completed, but then we would have to be prepared
6096 : * to deal with speculative tokens on committed tuples. That wouldn't be
6097 : * difficult - no-one looks at the ctid field of a tuple with invalid xmax -
6098 : * but clearing the token at completion isn't very expensive either.
6099 : * An explicit confirmation WAL record also makes logical decoding simpler.
6100 : */
6101 : void
6102 4112 : heap_finish_speculative(Relation relation, ItemPointer tid)
6103 : {
6104 : Buffer buffer;
6105 : Page page;
6106 : OffsetNumber offnum;
6107 4112 : ItemId lp = NULL;
6108 : HeapTupleHeader htup;
6109 :
6110 4112 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
6111 4112 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6112 4112 : page = (Page) BufferGetPage(buffer);
6113 :
6114 4112 : offnum = ItemPointerGetOffsetNumber(tid);
6115 4112 : if (PageGetMaxOffsetNumber(page) >= offnum)
6116 4112 : lp = PageGetItemId(page, offnum);
6117 :
6118 4112 : if (PageGetMaxOffsetNumber(page) < offnum || !ItemIdIsNormal(lp))
6119 0 : elog(ERROR, "invalid lp");
6120 :
6121 4112 : htup = (HeapTupleHeader) PageGetItem(page, lp);
6122 :
6123 : /* NO EREPORT(ERROR) from here till changes are logged */
6124 4112 : START_CRIT_SECTION();
6125 :
6126 : Assert(HeapTupleHeaderIsSpeculative(htup));
6127 :
6128 4112 : MarkBufferDirty(buffer);
6129 :
6130 : /*
6131 : * Replace the speculative insertion token with a real t_ctid, pointing to
6132 : * itself like it does on regular tuples.
6133 : */
6134 4112 : htup->t_ctid = *tid;
6135 :
6136 : /* XLOG stuff */
6137 4112 : if (RelationNeedsWAL(relation))
6138 : {
6139 : xl_heap_confirm xlrec;
6140 : XLogRecPtr recptr;
6141 :
6142 4094 : xlrec.offnum = ItemPointerGetOffsetNumber(tid);
6143 :
6144 4094 : XLogBeginInsert();
6145 :
6146 : /* We want the same filtering on this as on a plain insert */
6147 4094 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
6148 :
6149 4094 : XLogRegisterData(&xlrec, SizeOfHeapConfirm);
6150 4094 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
6151 :
6152 4094 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_CONFIRM);
6153 :
6154 4094 : PageSetLSN(page, recptr);
6155 : }
6156 :
6157 4112 : END_CRIT_SECTION();
6158 :
6159 4112 : UnlockReleaseBuffer(buffer);
6160 4112 : }
6161 :
6162 : /*
6163 : * heap_abort_speculative - kill a speculatively inserted tuple
6164 : *
6165 : * Marks a tuple that was speculatively inserted in the same command as dead,
6166 : * by setting its xmin as invalid. That makes it immediately appear as dead
6167 : * to all transactions, including our own. In particular, it makes
6168 : * HeapTupleSatisfiesDirty() regard the tuple as dead, so that another backend
6169 : * inserting a duplicate key value won't unnecessarily wait for our whole
6170 : * transaction to finish (it'll just wait for our speculative insertion to
6171 : * finish).
6172 : *
6173 : * Killing the tuple prevents "unprincipled deadlocks", which are deadlocks
6174 : * that arise due to a mutual dependency that is not user visible. By
6175 : * definition, unprincipled deadlocks cannot be prevented by the user
6176 : * reordering lock acquisition in client code, because the implementation level
6177 : * lock acquisitions are not under the user's direct control. If speculative
6178 : * inserters did not take this precaution, then under high concurrency they
6179 : * could deadlock with each other, which would not be acceptable.
6180 : *
6181 : * This is somewhat redundant with heap_delete, but we prefer to have a
6182 : * dedicated routine with stripped down requirements. Note that this is also
6183 : * used to delete the TOAST tuples created during speculative insertion.
6184 : *
6185 : * This routine does not affect logical decoding as it only looks at
6186 : * confirmation records.
6187 : */
6188 : void
6189 20 : heap_abort_speculative(Relation relation, ItemPointer tid)
6190 : {
6191 20 : TransactionId xid = GetCurrentTransactionId();
6192 : ItemId lp;
6193 : HeapTupleData tp;
6194 : Page page;
6195 : BlockNumber block;
6196 : Buffer buffer;
6197 :
6198 : Assert(ItemPointerIsValid(tid));
6199 :
6200 20 : block = ItemPointerGetBlockNumber(tid);
6201 20 : buffer = ReadBuffer(relation, block);
6202 20 : page = BufferGetPage(buffer);
6203 :
6204 20 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6205 :
6206 : /*
6207 : * Page can't be all visible, we just inserted into it, and are still
6208 : * running.
6209 : */
6210 : Assert(!PageIsAllVisible(page));
6211 :
6212 20 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
6213 : Assert(ItemIdIsNormal(lp));
6214 :
6215 20 : tp.t_tableOid = RelationGetRelid(relation);
6216 20 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
6217 20 : tp.t_len = ItemIdGetLength(lp);
6218 20 : tp.t_self = *tid;
6219 :
6220 : /*
6221 : * Sanity check that the tuple really is a speculatively inserted tuple,
6222 : * inserted by us.
6223 : */
6224 20 : if (tp.t_data->t_choice.t_heap.t_xmin != xid)
6225 0 : elog(ERROR, "attempted to kill a tuple inserted by another transaction");
6226 20 : if (!(IsToastRelation(relation) || HeapTupleHeaderIsSpeculative(tp.t_data)))
6227 0 : elog(ERROR, "attempted to kill a non-speculative tuple");
6228 : Assert(!HeapTupleHeaderIsHeapOnly(tp.t_data));
6229 :
6230 : /*
6231 : * No need to check for serializable conflicts here. There is never a
6232 : * need for a combo CID, either. No need to extract replica identity, or
6233 : * do anything special with infomask bits.
6234 : */
6235 :
6236 20 : START_CRIT_SECTION();
6237 :
6238 : /*
6239 : * The tuple will become DEAD immediately. Flag that this page is a
6240 : * candidate for pruning by setting xmin to TransactionXmin. While not
6241 : * immediately prunable, it is the oldest xid we can cheaply determine
6242 : * that's safe against wraparound / being older than the table's
6243 : * relfrozenxid. To defend against the unlikely case of a new relation
6244 : * having a newer relfrozenxid than our TransactionXmin, use relfrozenxid
6245 : * if so (vacuum can't subsequently move relfrozenxid to beyond
6246 : * TransactionXmin, so there's no race here).
6247 : */
6248 : Assert(TransactionIdIsValid(TransactionXmin));
6249 : {
6250 20 : TransactionId relfrozenxid = relation->rd_rel->relfrozenxid;
6251 : TransactionId prune_xid;
6252 :
6253 20 : if (TransactionIdPrecedes(TransactionXmin, relfrozenxid))
6254 0 : prune_xid = relfrozenxid;
6255 : else
6256 20 : prune_xid = TransactionXmin;
6257 20 : PageSetPrunable(page, prune_xid);
6258 : }
6259 :
6260 : /* store transaction information of xact deleting the tuple */
6261 20 : tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
6262 20 : tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
6263 :
6264 : /*
6265 : * Set the tuple header xmin to InvalidTransactionId. This makes the
6266 : * tuple immediately invisible everyone. (In particular, to any
6267 : * transactions waiting on the speculative token, woken up later.)
6268 : */
6269 20 : HeapTupleHeaderSetXmin(tp.t_data, InvalidTransactionId);
6270 :
6271 : /* Clear the speculative insertion token too */
6272 20 : tp.t_data->t_ctid = tp.t_self;
6273 :
6274 20 : MarkBufferDirty(buffer);
6275 :
6276 : /*
6277 : * XLOG stuff
6278 : *
6279 : * The WAL records generated here match heap_delete(). The same recovery
6280 : * routines are used.
6281 : */
6282 20 : if (RelationNeedsWAL(relation))
6283 : {
6284 : xl_heap_delete xlrec;
6285 : XLogRecPtr recptr;
6286 :
6287 20 : xlrec.flags = XLH_DELETE_IS_SUPER;
6288 40 : xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
6289 20 : tp.t_data->t_infomask2);
6290 20 : xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
6291 20 : xlrec.xmax = xid;
6292 :
6293 20 : XLogBeginInsert();
6294 20 : XLogRegisterData(&xlrec, SizeOfHeapDelete);
6295 20 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
6296 :
6297 : /* No replica identity & replication origin logged */
6298 :
6299 20 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
6300 :
6301 20 : PageSetLSN(page, recptr);
6302 : }
6303 :
6304 20 : END_CRIT_SECTION();
6305 :
6306 20 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6307 :
6308 20 : if (HeapTupleHasExternal(&tp))
6309 : {
6310 : Assert(!IsToastRelation(relation));
6311 2 : heap_toast_delete(relation, &tp, true);
6312 : }
6313 :
6314 : /*
6315 : * Never need to mark tuple for invalidation, since catalogs don't support
6316 : * speculative insertion
6317 : */
6318 :
6319 : /* Now we can release the buffer */
6320 20 : ReleaseBuffer(buffer);
6321 :
6322 : /* count deletion, as we counted the insertion too */
6323 20 : pgstat_count_heap_delete(relation);
6324 20 : }
6325 :
6326 : /*
6327 : * heap_inplace_lock - protect inplace update from concurrent heap_update()
6328 : *
6329 : * Evaluate whether the tuple's state is compatible with a no-key update.
6330 : * Current transaction rowmarks are fine, as is KEY SHARE from any
6331 : * transaction. If compatible, return true with the buffer exclusive-locked,
6332 : * and the caller must release that by calling
6333 : * heap_inplace_update_and_unlock(), calling heap_inplace_unlock(), or raising
6334 : * an error. Otherwise, call release_callback(arg), wait for blocking
6335 : * transactions to end, and return false.
6336 : *
6337 : * Since this is intended for system catalogs and SERIALIZABLE doesn't cover
6338 : * DDL, this doesn't guarantee any particular predicate locking.
6339 : *
6340 : * One could modify this to return true for tuples with delete in progress,
6341 : * All inplace updaters take a lock that conflicts with DROP. If explicit
6342 : * "DELETE FROM pg_class" is in progress, we'll wait for it like we would an
6343 : * update.
6344 : *
6345 : * Readers of inplace-updated fields expect changes to those fields are
6346 : * durable. For example, vac_truncate_clog() reads datfrozenxid from
6347 : * pg_database tuples via catalog snapshots. A future snapshot must not
6348 : * return a lower datfrozenxid for the same database OID (lower in the
6349 : * FullTransactionIdPrecedes() sense). We achieve that since no update of a
6350 : * tuple can start while we hold a lock on its buffer. In cases like
6351 : * BEGIN;GRANT;CREATE INDEX;COMMIT we're inplace-updating a tuple visible only
6352 : * to this transaction. ROLLBACK then is one case where it's okay to lose
6353 : * inplace updates. (Restoring relhasindex=false on ROLLBACK is fine, since
6354 : * any concurrent CREATE INDEX would have blocked, then inplace-updated the
6355 : * committed tuple.)
6356 : *
6357 : * In principle, we could avoid waiting by overwriting every tuple in the
6358 : * updated tuple chain. Reader expectations permit updating a tuple only if
6359 : * it's aborted, is the tail of the chain, or we already updated the tuple
6360 : * referenced in its t_ctid. Hence, we would need to overwrite the tuples in
6361 : * order from tail to head. That would imply either (a) mutating all tuples
6362 : * in one critical section or (b) accepting a chance of partial completion.
6363 : * Partial completion of a relfrozenxid update would have the weird
6364 : * consequence that the table's next VACUUM could see the table's relfrozenxid
6365 : * move forward between vacuum_get_cutoffs() and finishing.
6366 : */
6367 : bool
6368 269300 : heap_inplace_lock(Relation relation,
6369 : HeapTuple oldtup_ptr, Buffer buffer,
6370 : void (*release_callback) (void *), void *arg)
6371 : {
6372 269300 : HeapTupleData oldtup = *oldtup_ptr; /* minimize diff vs. heap_update() */
6373 : TM_Result result;
6374 : bool ret;
6375 :
6376 : #ifdef USE_ASSERT_CHECKING
6377 : if (RelationGetRelid(relation) == RelationRelationId)
6378 : check_inplace_rel_lock(oldtup_ptr);
6379 : #endif
6380 :
6381 : Assert(BufferIsValid(buffer));
6382 :
6383 : /*
6384 : * Construct shared cache inval if necessary. Because we pass a tuple
6385 : * version without our own inplace changes or inplace changes other
6386 : * sessions complete while we wait for locks, inplace update mustn't
6387 : * change catcache lookup keys. But we aren't bothering with index
6388 : * updates either, so that's true a fortiori. After LockBuffer(), it
6389 : * would be too late, because this might reach a
6390 : * CatalogCacheInitializeCache() that locks "buffer".
6391 : */
6392 269300 : CacheInvalidateHeapTupleInplace(relation, oldtup_ptr, NULL);
6393 :
6394 269300 : LockTuple(relation, &oldtup.t_self, InplaceUpdateTupleLock);
6395 269300 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6396 :
6397 : /*----------
6398 : * Interpret HeapTupleSatisfiesUpdate() like heap_update() does, except:
6399 : *
6400 : * - wait unconditionally
6401 : * - already locked tuple above, since inplace needs that unconditionally
6402 : * - don't recheck header after wait: simpler to defer to next iteration
6403 : * - don't try to continue even if the updater aborts: likewise
6404 : * - no crosscheck
6405 : */
6406 269300 : result = HeapTupleSatisfiesUpdate(&oldtup, GetCurrentCommandId(false),
6407 : buffer);
6408 :
6409 269300 : if (result == TM_Invisible)
6410 : {
6411 : /* no known way this can happen */
6412 0 : ereport(ERROR,
6413 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6414 : errmsg_internal("attempted to overwrite invisible tuple")));
6415 : }
6416 269300 : else if (result == TM_SelfModified)
6417 : {
6418 : /*
6419 : * CREATE INDEX might reach this if an expression is silly enough to
6420 : * call e.g. SELECT ... FROM pg_class FOR SHARE. C code of other SQL
6421 : * statements might get here after a heap_update() of the same row, in
6422 : * the absence of an intervening CommandCounterIncrement().
6423 : */
6424 0 : ereport(ERROR,
6425 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6426 : errmsg("tuple to be updated was already modified by an operation triggered by the current command")));
6427 : }
6428 269300 : else if (result == TM_BeingModified)
6429 : {
6430 : TransactionId xwait;
6431 : uint16 infomask;
6432 :
6433 44 : xwait = HeapTupleHeaderGetRawXmax(oldtup.t_data);
6434 44 : infomask = oldtup.t_data->t_infomask;
6435 :
6436 44 : if (infomask & HEAP_XMAX_IS_MULTI)
6437 : {
6438 10 : LockTupleMode lockmode = LockTupleNoKeyExclusive;
6439 10 : MultiXactStatus mxact_status = MultiXactStatusNoKeyUpdate;
6440 : int remain;
6441 :
6442 10 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
6443 : lockmode, NULL))
6444 : {
6445 4 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6446 4 : release_callback(arg);
6447 4 : ret = false;
6448 4 : MultiXactIdWait((MultiXactId) xwait, mxact_status, infomask,
6449 : relation, &oldtup.t_self, XLTW_Update,
6450 : &remain);
6451 : }
6452 : else
6453 6 : ret = true;
6454 : }
6455 34 : else if (TransactionIdIsCurrentTransactionId(xwait))
6456 2 : ret = true;
6457 32 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask))
6458 2 : ret = true;
6459 : else
6460 : {
6461 30 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6462 30 : release_callback(arg);
6463 30 : ret = false;
6464 30 : XactLockTableWait(xwait, relation, &oldtup.t_self,
6465 : XLTW_Update);
6466 : }
6467 : }
6468 : else
6469 : {
6470 269256 : ret = (result == TM_Ok);
6471 269256 : if (!ret)
6472 : {
6473 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6474 0 : release_callback(arg);
6475 : }
6476 : }
6477 :
6478 : /*
6479 : * GetCatalogSnapshot() relies on invalidation messages to know when to
6480 : * take a new snapshot. COMMIT of xwait is responsible for sending the
6481 : * invalidation. We're not acquiring heavyweight locks sufficient to
6482 : * block if not yet sent, so we must take a new snapshot to ensure a later
6483 : * attempt has a fair chance. While we don't need this if xwait aborted,
6484 : * don't bother optimizing that.
6485 : */
6486 269300 : if (!ret)
6487 : {
6488 34 : UnlockTuple(relation, &oldtup.t_self, InplaceUpdateTupleLock);
6489 34 : ForgetInplace_Inval();
6490 34 : InvalidateCatalogSnapshot();
6491 : }
6492 269300 : return ret;
6493 : }
6494 :
6495 : /*
6496 : * heap_inplace_update_and_unlock - core of systable_inplace_update_finish
6497 : *
6498 : * The tuple cannot change size, and therefore its header fields and null
6499 : * bitmap (if any) don't change either.
6500 : *
6501 : * Since we hold LOCKTAG_TUPLE, no updater has a local copy of this tuple.
6502 : */
6503 : void
6504 156534 : heap_inplace_update_and_unlock(Relation relation,
6505 : HeapTuple oldtup, HeapTuple tuple,
6506 : Buffer buffer)
6507 : {
6508 156534 : HeapTupleHeader htup = oldtup->t_data;
6509 : uint32 oldlen;
6510 : uint32 newlen;
6511 : char *dst;
6512 : char *src;
6513 156534 : int nmsgs = 0;
6514 156534 : SharedInvalidationMessage *invalMessages = NULL;
6515 156534 : bool RelcacheInitFileInval = false;
6516 :
6517 : Assert(ItemPointerEquals(&oldtup->t_self, &tuple->t_self));
6518 156534 : oldlen = oldtup->t_len - htup->t_hoff;
6519 156534 : newlen = tuple->t_len - tuple->t_data->t_hoff;
6520 156534 : if (oldlen != newlen || htup->t_hoff != tuple->t_data->t_hoff)
6521 0 : elog(ERROR, "wrong tuple length");
6522 :
6523 156534 : dst = (char *) htup + htup->t_hoff;
6524 156534 : src = (char *) tuple->t_data + tuple->t_data->t_hoff;
6525 :
6526 : /* Like RecordTransactionCommit(), log only if needed */
6527 156534 : if (XLogStandbyInfoActive())
6528 92604 : nmsgs = inplaceGetInvalidationMessages(&invalMessages,
6529 : &RelcacheInitFileInval);
6530 :
6531 : /*
6532 : * Unlink relcache init files as needed. If unlinking, acquire
6533 : * RelCacheInitLock until after associated invalidations. By doing this
6534 : * in advance, if we checkpoint and then crash between inplace
6535 : * XLogInsert() and inval, we don't rely on StartupXLOG() ->
6536 : * RelationCacheInitFileRemove(). That uses elevel==LOG, so replay would
6537 : * neglect to PANIC on EIO.
6538 : */
6539 156534 : PreInplace_Inval();
6540 :
6541 : /*----------
6542 : * NO EREPORT(ERROR) from here till changes are complete
6543 : *
6544 : * Our buffer lock won't stop a reader having already pinned and checked
6545 : * visibility for this tuple. Hence, we write WAL first, then mutate the
6546 : * buffer. Like in MarkBufferDirtyHint() or RecordTransactionCommit(),
6547 : * checkpoint delay makes that acceptable. With the usual order of
6548 : * changes, a crash after memcpy() and before XLogInsert() could allow
6549 : * datfrozenxid to overtake relfrozenxid:
6550 : *
6551 : * ["D" is a VACUUM (ONLY_DATABASE_STATS)]
6552 : * ["R" is a VACUUM tbl]
6553 : * D: vac_update_datfrozenxid() -> systable_beginscan(pg_class)
6554 : * D: systable_getnext() returns pg_class tuple of tbl
6555 : * R: memcpy() into pg_class tuple of tbl
6556 : * D: raise pg_database.datfrozenxid, XLogInsert(), finish
6557 : * [crash]
6558 : * [recovery restores datfrozenxid w/o relfrozenxid]
6559 : *
6560 : * Like in MarkBufferDirtyHint() subroutine XLogSaveBufferForHint(), copy
6561 : * the buffer to the stack before logging. Here, that facilitates a FPI
6562 : * of the post-mutation block before we accept other sessions seeing it.
6563 : */
6564 : Assert((MyProc->delayChkptFlags & DELAY_CHKPT_START) == 0);
6565 156534 : START_CRIT_SECTION();
6566 156534 : MyProc->delayChkptFlags |= DELAY_CHKPT_START;
6567 :
6568 : /* XLOG stuff */
6569 156534 : if (RelationNeedsWAL(relation))
6570 : {
6571 : xl_heap_inplace xlrec;
6572 : PGAlignedBlock copied_buffer;
6573 156518 : char *origdata = (char *) BufferGetBlock(buffer);
6574 156518 : Page page = BufferGetPage(buffer);
6575 156518 : uint16 lower = ((PageHeader) page)->pd_lower;
6576 156518 : uint16 upper = ((PageHeader) page)->pd_upper;
6577 : uintptr_t dst_offset_in_block;
6578 : RelFileLocator rlocator;
6579 : ForkNumber forkno;
6580 : BlockNumber blkno;
6581 : XLogRecPtr recptr;
6582 :
6583 156518 : xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
6584 156518 : xlrec.dbId = MyDatabaseId;
6585 156518 : xlrec.tsId = MyDatabaseTableSpace;
6586 156518 : xlrec.relcacheInitFileInval = RelcacheInitFileInval;
6587 156518 : xlrec.nmsgs = nmsgs;
6588 :
6589 156518 : XLogBeginInsert();
6590 156518 : XLogRegisterData(&xlrec, MinSizeOfHeapInplace);
6591 156518 : if (nmsgs != 0)
6592 65964 : XLogRegisterData(invalMessages,
6593 : nmsgs * sizeof(SharedInvalidationMessage));
6594 :
6595 : /* register block matching what buffer will look like after changes */
6596 156518 : memcpy(copied_buffer.data, origdata, lower);
6597 156518 : memcpy(copied_buffer.data + upper, origdata + upper, BLCKSZ - upper);
6598 156518 : dst_offset_in_block = dst - origdata;
6599 156518 : memcpy(copied_buffer.data + dst_offset_in_block, src, newlen);
6600 156518 : BufferGetTag(buffer, &rlocator, &forkno, &blkno);
6601 : Assert(forkno == MAIN_FORKNUM);
6602 156518 : XLogRegisterBlock(0, &rlocator, forkno, blkno, copied_buffer.data,
6603 : REGBUF_STANDARD);
6604 156518 : XLogRegisterBufData(0, src, newlen);
6605 :
6606 : /* inplace updates aren't decoded atm, don't log the origin */
6607 :
6608 156518 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_INPLACE);
6609 :
6610 156518 : PageSetLSN(page, recptr);
6611 : }
6612 :
6613 156534 : memcpy(dst, src, newlen);
6614 :
6615 156534 : MarkBufferDirty(buffer);
6616 :
6617 156534 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6618 :
6619 : /*
6620 : * Send invalidations to shared queue. SearchSysCacheLocked1() assumes we
6621 : * do this before UnlockTuple().
6622 : *
6623 : * If we're mutating a tuple visible only to this transaction, there's an
6624 : * equivalent transactional inval from the action that created the tuple,
6625 : * and this inval is superfluous.
6626 : */
6627 156534 : AtInplace_Inval();
6628 :
6629 156534 : MyProc->delayChkptFlags &= ~DELAY_CHKPT_START;
6630 156534 : END_CRIT_SECTION();
6631 156534 : UnlockTuple(relation, &tuple->t_self, InplaceUpdateTupleLock);
6632 :
6633 156534 : AcceptInvalidationMessages(); /* local processing of just-sent inval */
6634 :
6635 : /*
6636 : * Queue a transactional inval. The immediate invalidation we just sent
6637 : * is the only one known to be necessary. To reduce risk from the
6638 : * transition to immediate invalidation, continue sending a transactional
6639 : * invalidation like we've long done. Third-party code might rely on it.
6640 : */
6641 156534 : if (!IsBootstrapProcessingMode())
6642 129894 : CacheInvalidateHeapTuple(relation, tuple, NULL);
6643 156534 : }
6644 :
6645 : /*
6646 : * heap_inplace_unlock - reverse of heap_inplace_lock
6647 : */
6648 : void
6649 112732 : heap_inplace_unlock(Relation relation,
6650 : HeapTuple oldtup, Buffer buffer)
6651 : {
6652 112732 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6653 112732 : UnlockTuple(relation, &oldtup->t_self, InplaceUpdateTupleLock);
6654 112732 : ForgetInplace_Inval();
6655 112732 : }
6656 :
6657 : #define FRM_NOOP 0x0001
6658 : #define FRM_INVALIDATE_XMAX 0x0002
6659 : #define FRM_RETURN_IS_XID 0x0004
6660 : #define FRM_RETURN_IS_MULTI 0x0008
6661 : #define FRM_MARK_COMMITTED 0x0010
6662 :
6663 : /*
6664 : * FreezeMultiXactId
6665 : * Determine what to do during freezing when a tuple is marked by a
6666 : * MultiXactId.
6667 : *
6668 : * "flags" is an output value; it's used to tell caller what to do on return.
6669 : * "pagefrz" is an input/output value, used to manage page level freezing.
6670 : *
6671 : * Possible values that we can set in "flags":
6672 : * FRM_NOOP
6673 : * don't do anything -- keep existing Xmax
6674 : * FRM_INVALIDATE_XMAX
6675 : * mark Xmax as InvalidTransactionId and set XMAX_INVALID flag.
6676 : * FRM_RETURN_IS_XID
6677 : * The Xid return value is a single update Xid to set as xmax.
6678 : * FRM_MARK_COMMITTED
6679 : * Xmax can be marked as HEAP_XMAX_COMMITTED
6680 : * FRM_RETURN_IS_MULTI
6681 : * The return value is a new MultiXactId to set as new Xmax.
6682 : * (caller must obtain proper infomask bits using GetMultiXactIdHintBits)
6683 : *
6684 : * Caller delegates control of page freezing to us. In practice we always
6685 : * force freezing of caller's page unless FRM_NOOP processing is indicated.
6686 : * We help caller ensure that XIDs < FreezeLimit and MXIDs < MultiXactCutoff
6687 : * can never be left behind. We freely choose when and how to process each
6688 : * Multi, without ever violating the cutoff postconditions for freezing.
6689 : *
6690 : * It's useful to remove Multis on a proactive timeline (relative to freezing
6691 : * XIDs) to keep MultiXact member SLRU buffer misses to a minimum. It can also
6692 : * be cheaper in the short run, for us, since we too can avoid SLRU buffer
6693 : * misses through eager processing.
6694 : *
6695 : * NB: Creates a _new_ MultiXactId when FRM_RETURN_IS_MULTI is set, though only
6696 : * when FreezeLimit and/or MultiXactCutoff cutoffs leave us with no choice.
6697 : * This can usually be put off, which is usually enough to avoid it altogether.
6698 : * Allocating new multis during VACUUM should be avoided on general principle;
6699 : * only VACUUM can advance relminmxid, so allocating new Multis here comes with
6700 : * its own special risks.
6701 : *
6702 : * NB: Caller must maintain "no freeze" NewRelfrozenXid/NewRelminMxid trackers
6703 : * using heap_tuple_should_freeze when we haven't forced page-level freezing.
6704 : *
6705 : * NB: Caller should avoid needlessly calling heap_tuple_should_freeze when we
6706 : * have already forced page-level freezing, since that might incur the same
6707 : * SLRU buffer misses that we specifically intended to avoid by freezing.
6708 : */
6709 : static TransactionId
6710 12 : FreezeMultiXactId(MultiXactId multi, uint16 t_infomask,
6711 : const struct VacuumCutoffs *cutoffs, uint16 *flags,
6712 : HeapPageFreeze *pagefrz)
6713 : {
6714 : TransactionId newxmax;
6715 : MultiXactMember *members;
6716 : int nmembers;
6717 : bool need_replace;
6718 : int nnewmembers;
6719 : MultiXactMember *newmembers;
6720 : bool has_lockers;
6721 : TransactionId update_xid;
6722 : bool update_committed;
6723 : TransactionId FreezePageRelfrozenXid;
6724 :
6725 12 : *flags = 0;
6726 :
6727 : /* We should only be called in Multis */
6728 : Assert(t_infomask & HEAP_XMAX_IS_MULTI);
6729 :
6730 24 : if (!MultiXactIdIsValid(multi) ||
6731 12 : HEAP_LOCKED_UPGRADED(t_infomask))
6732 : {
6733 0 : *flags |= FRM_INVALIDATE_XMAX;
6734 0 : pagefrz->freeze_required = true;
6735 0 : return InvalidTransactionId;
6736 : }
6737 12 : else if (MultiXactIdPrecedes(multi, cutoffs->relminmxid))
6738 0 : ereport(ERROR,
6739 : (errcode(ERRCODE_DATA_CORRUPTED),
6740 : errmsg_internal("found multixact %u from before relminmxid %u",
6741 : multi, cutoffs->relminmxid)));
6742 12 : else if (MultiXactIdPrecedes(multi, cutoffs->OldestMxact))
6743 : {
6744 : TransactionId update_xact;
6745 :
6746 : /*
6747 : * This old multi cannot possibly have members still running, but
6748 : * verify just in case. If it was a locker only, it can be removed
6749 : * without any further consideration; but if it contained an update,
6750 : * we might need to preserve it.
6751 : */
6752 8 : if (MultiXactIdIsRunning(multi,
6753 8 : HEAP_XMAX_IS_LOCKED_ONLY(t_infomask)))
6754 0 : ereport(ERROR,
6755 : (errcode(ERRCODE_DATA_CORRUPTED),
6756 : errmsg_internal("multixact %u from before multi freeze cutoff %u found to be still running",
6757 : multi, cutoffs->OldestMxact)));
6758 :
6759 8 : if (HEAP_XMAX_IS_LOCKED_ONLY(t_infomask))
6760 : {
6761 8 : *flags |= FRM_INVALIDATE_XMAX;
6762 8 : pagefrz->freeze_required = true;
6763 8 : return InvalidTransactionId;
6764 : }
6765 :
6766 : /* replace multi with single XID for its updater? */
6767 0 : update_xact = MultiXactIdGetUpdateXid(multi, t_infomask);
6768 0 : if (TransactionIdPrecedes(update_xact, cutoffs->relfrozenxid))
6769 0 : ereport(ERROR,
6770 : (errcode(ERRCODE_DATA_CORRUPTED),
6771 : errmsg_internal("multixact %u contains update XID %u from before relfrozenxid %u",
6772 : multi, update_xact,
6773 : cutoffs->relfrozenxid)));
6774 0 : else if (TransactionIdPrecedes(update_xact, cutoffs->OldestXmin))
6775 : {
6776 : /*
6777 : * Updater XID has to have aborted (otherwise the tuple would have
6778 : * been pruned away instead, since updater XID is < OldestXmin).
6779 : * Just remove xmax.
6780 : */
6781 0 : if (TransactionIdDidCommit(update_xact))
6782 0 : ereport(ERROR,
6783 : (errcode(ERRCODE_DATA_CORRUPTED),
6784 : errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
6785 : multi, update_xact,
6786 : cutoffs->OldestXmin)));
6787 0 : *flags |= FRM_INVALIDATE_XMAX;
6788 0 : pagefrz->freeze_required = true;
6789 0 : return InvalidTransactionId;
6790 : }
6791 :
6792 : /* Have to keep updater XID as new xmax */
6793 0 : *flags |= FRM_RETURN_IS_XID;
6794 0 : pagefrz->freeze_required = true;
6795 0 : return update_xact;
6796 : }
6797 :
6798 : /*
6799 : * Some member(s) of this Multi may be below FreezeLimit xid cutoff, so we
6800 : * need to walk the whole members array to figure out what to do, if
6801 : * anything.
6802 : */
6803 : nmembers =
6804 4 : GetMultiXactIdMembers(multi, &members, false,
6805 4 : HEAP_XMAX_IS_LOCKED_ONLY(t_infomask));
6806 4 : if (nmembers <= 0)
6807 : {
6808 : /* Nothing worth keeping */
6809 0 : *flags |= FRM_INVALIDATE_XMAX;
6810 0 : pagefrz->freeze_required = true;
6811 0 : return InvalidTransactionId;
6812 : }
6813 :
6814 : /*
6815 : * The FRM_NOOP case is the only case where we might need to ratchet back
6816 : * FreezePageRelfrozenXid or FreezePageRelminMxid. It is also the only
6817 : * case where our caller might ratchet back its NoFreezePageRelfrozenXid
6818 : * or NoFreezePageRelminMxid "no freeze" trackers to deal with a multi.
6819 : * FRM_NOOP handling should result in the NewRelfrozenXid/NewRelminMxid
6820 : * trackers managed by VACUUM being ratcheting back by xmax to the degree
6821 : * required to make it safe to leave xmax undisturbed, independent of
6822 : * whether or not page freezing is triggered somewhere else.
6823 : *
6824 : * Our policy is to force freezing in every case other than FRM_NOOP,
6825 : * which obviates the need to maintain either set of trackers, anywhere.
6826 : * Every other case will reliably execute a freeze plan for xmax that
6827 : * either replaces xmax with an XID/MXID >= OldestXmin/OldestMxact, or
6828 : * sets xmax to an InvalidTransactionId XID, rendering xmax fully frozen.
6829 : * (VACUUM's NewRelfrozenXid/NewRelminMxid trackers are initialized with
6830 : * OldestXmin/OldestMxact, so later values never need to be tracked here.)
6831 : */
6832 4 : need_replace = false;
6833 4 : FreezePageRelfrozenXid = pagefrz->FreezePageRelfrozenXid;
6834 8 : for (int i = 0; i < nmembers; i++)
6835 : {
6836 6 : TransactionId xid = members[i].xid;
6837 :
6838 : Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
6839 :
6840 6 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
6841 : {
6842 : /* Can't violate the FreezeLimit postcondition */
6843 2 : need_replace = true;
6844 2 : break;
6845 : }
6846 4 : if (TransactionIdPrecedes(xid, FreezePageRelfrozenXid))
6847 0 : FreezePageRelfrozenXid = xid;
6848 : }
6849 :
6850 : /* Can't violate the MultiXactCutoff postcondition, either */
6851 4 : if (!need_replace)
6852 2 : need_replace = MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff);
6853 :
6854 4 : if (!need_replace)
6855 : {
6856 : /*
6857 : * vacuumlazy.c might ratchet back NewRelminMxid, NewRelfrozenXid, or
6858 : * both together to make it safe to retain this particular multi after
6859 : * freezing its page
6860 : */
6861 2 : *flags |= FRM_NOOP;
6862 2 : pagefrz->FreezePageRelfrozenXid = FreezePageRelfrozenXid;
6863 2 : if (MultiXactIdPrecedes(multi, pagefrz->FreezePageRelminMxid))
6864 0 : pagefrz->FreezePageRelminMxid = multi;
6865 2 : pfree(members);
6866 2 : return multi;
6867 : }
6868 :
6869 : /*
6870 : * Do a more thorough second pass over the multi to figure out which
6871 : * member XIDs actually need to be kept. Checking the precise status of
6872 : * individual members might even show that we don't need to keep anything.
6873 : * That is quite possible even though the Multi must be >= OldestMxact,
6874 : * since our second pass only keeps member XIDs when it's truly necessary;
6875 : * even member XIDs >= OldestXmin often won't be kept by second pass.
6876 : */
6877 2 : nnewmembers = 0;
6878 2 : newmembers = palloc(sizeof(MultiXactMember) * nmembers);
6879 2 : has_lockers = false;
6880 2 : update_xid = InvalidTransactionId;
6881 2 : update_committed = false;
6882 :
6883 : /*
6884 : * Determine whether to keep each member xid, or to ignore it instead
6885 : */
6886 6 : for (int i = 0; i < nmembers; i++)
6887 : {
6888 4 : TransactionId xid = members[i].xid;
6889 4 : MultiXactStatus mstatus = members[i].status;
6890 :
6891 : Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
6892 :
6893 4 : if (!ISUPDATE_from_mxstatus(mstatus))
6894 : {
6895 : /*
6896 : * Locker XID (not updater XID). We only keep lockers that are
6897 : * still running.
6898 : */
6899 8 : if (TransactionIdIsCurrentTransactionId(xid) ||
6900 4 : TransactionIdIsInProgress(xid))
6901 : {
6902 2 : if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
6903 0 : ereport(ERROR,
6904 : (errcode(ERRCODE_DATA_CORRUPTED),
6905 : errmsg_internal("multixact %u contains running locker XID %u from before removable cutoff %u",
6906 : multi, xid,
6907 : cutoffs->OldestXmin)));
6908 2 : newmembers[nnewmembers++] = members[i];
6909 2 : has_lockers = true;
6910 : }
6911 :
6912 4 : continue;
6913 : }
6914 :
6915 : /*
6916 : * Updater XID (not locker XID). Should we keep it?
6917 : *
6918 : * Since the tuple wasn't totally removed when vacuum pruned, the
6919 : * update Xid cannot possibly be older than OldestXmin cutoff unless
6920 : * the updater XID aborted. If the updater transaction is known
6921 : * aborted or crashed then it's okay to ignore it, otherwise not.
6922 : *
6923 : * In any case the Multi should never contain two updaters, whatever
6924 : * their individual commit status. Check for that first, in passing.
6925 : */
6926 0 : if (TransactionIdIsValid(update_xid))
6927 0 : ereport(ERROR,
6928 : (errcode(ERRCODE_DATA_CORRUPTED),
6929 : errmsg_internal("multixact %u has two or more updating members",
6930 : multi),
6931 : errdetail_internal("First updater XID=%u second updater XID=%u.",
6932 : update_xid, xid)));
6933 :
6934 : /*
6935 : * As with all tuple visibility routines, it's critical to test
6936 : * TransactionIdIsInProgress before TransactionIdDidCommit, because of
6937 : * race conditions explained in detail in heapam_visibility.c.
6938 : */
6939 0 : if (TransactionIdIsCurrentTransactionId(xid) ||
6940 0 : TransactionIdIsInProgress(xid))
6941 0 : update_xid = xid;
6942 0 : else if (TransactionIdDidCommit(xid))
6943 : {
6944 : /*
6945 : * The transaction committed, so we can tell caller to set
6946 : * HEAP_XMAX_COMMITTED. (We can only do this because we know the
6947 : * transaction is not running.)
6948 : */
6949 0 : update_committed = true;
6950 0 : update_xid = xid;
6951 : }
6952 : else
6953 : {
6954 : /*
6955 : * Not in progress, not committed -- must be aborted or crashed;
6956 : * we can ignore it.
6957 : */
6958 0 : continue;
6959 : }
6960 :
6961 : /*
6962 : * We determined that updater must be kept -- add it to pending new
6963 : * members list
6964 : */
6965 0 : if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
6966 0 : ereport(ERROR,
6967 : (errcode(ERRCODE_DATA_CORRUPTED),
6968 : errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
6969 : multi, xid, cutoffs->OldestXmin)));
6970 0 : newmembers[nnewmembers++] = members[i];
6971 : }
6972 :
6973 2 : pfree(members);
6974 :
6975 : /*
6976 : * Determine what to do with caller's multi based on information gathered
6977 : * during our second pass
6978 : */
6979 2 : if (nnewmembers == 0)
6980 : {
6981 : /* Nothing worth keeping */
6982 0 : *flags |= FRM_INVALIDATE_XMAX;
6983 0 : newxmax = InvalidTransactionId;
6984 : }
6985 2 : else if (TransactionIdIsValid(update_xid) && !has_lockers)
6986 : {
6987 : /*
6988 : * If there's a single member and it's an update, pass it back alone
6989 : * without creating a new Multi. (XXX we could do this when there's a
6990 : * single remaining locker, too, but that would complicate the API too
6991 : * much; moreover, the case with the single updater is more
6992 : * interesting, because those are longer-lived.)
6993 : */
6994 : Assert(nnewmembers == 1);
6995 0 : *flags |= FRM_RETURN_IS_XID;
6996 0 : if (update_committed)
6997 0 : *flags |= FRM_MARK_COMMITTED;
6998 0 : newxmax = update_xid;
6999 : }
7000 : else
7001 : {
7002 : /*
7003 : * Create a new multixact with the surviving members of the previous
7004 : * one, to set as new Xmax in the tuple
7005 : */
7006 2 : newxmax = MultiXactIdCreateFromMembers(nnewmembers, newmembers);
7007 2 : *flags |= FRM_RETURN_IS_MULTI;
7008 : }
7009 :
7010 2 : pfree(newmembers);
7011 :
7012 2 : pagefrz->freeze_required = true;
7013 2 : return newxmax;
7014 : }
7015 :
7016 : /*
7017 : * heap_prepare_freeze_tuple
7018 : *
7019 : * Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
7020 : * are older than the OldestXmin and/or OldestMxact freeze cutoffs. If so,
7021 : * setup enough state (in the *frz output argument) to enable caller to
7022 : * process this tuple as part of freezing its page, and return true. Return
7023 : * false if nothing can be changed about the tuple right now.
7024 : *
7025 : * Also sets *totally_frozen to true if the tuple will be totally frozen once
7026 : * caller executes returned freeze plan (or if the tuple was already totally
7027 : * frozen by an earlier VACUUM). This indicates that there are no remaining
7028 : * XIDs or MultiXactIds that will need to be processed by a future VACUUM.
7029 : *
7030 : * VACUUM caller must assemble HeapTupleFreeze freeze plan entries for every
7031 : * tuple that we returned true for, and then execute freezing. Caller must
7032 : * initialize pagefrz fields for page as a whole before first call here for
7033 : * each heap page.
7034 : *
7035 : * VACUUM caller decides on whether or not to freeze the page as a whole.
7036 : * We'll often prepare freeze plans for a page that caller just discards.
7037 : * However, VACUUM doesn't always get to make a choice; it must freeze when
7038 : * pagefrz.freeze_required is set, to ensure that any XIDs < FreezeLimit (and
7039 : * MXIDs < MultiXactCutoff) can never be left behind. We help to make sure
7040 : * that VACUUM always follows that rule.
7041 : *
7042 : * We sometimes force freezing of xmax MultiXactId values long before it is
7043 : * strictly necessary to do so just to ensure the FreezeLimit postcondition.
7044 : * It's worth processing MultiXactIds proactively when it is cheap to do so,
7045 : * and it's convenient to make that happen by piggy-backing it on the "force
7046 : * freezing" mechanism. Conversely, we sometimes delay freezing MultiXactIds
7047 : * because it is expensive right now (though only when it's still possible to
7048 : * do so without violating the FreezeLimit/MultiXactCutoff postcondition).
7049 : *
7050 : * It is assumed that the caller has checked the tuple with
7051 : * HeapTupleSatisfiesVacuum() and determined that it is not HEAPTUPLE_DEAD
7052 : * (else we should be removing the tuple, not freezing it).
7053 : *
7054 : * NB: This function has side effects: it might allocate a new MultiXactId.
7055 : * It will be set as tuple's new xmax when our *frz output is processed within
7056 : * heap_execute_freeze_tuple later on. If the tuple is in a shared buffer
7057 : * then caller had better have an exclusive lock on it already.
7058 : */
7059 : bool
7060 32113240 : heap_prepare_freeze_tuple(HeapTupleHeader tuple,
7061 : const struct VacuumCutoffs *cutoffs,
7062 : HeapPageFreeze *pagefrz,
7063 : HeapTupleFreeze *frz, bool *totally_frozen)
7064 : {
7065 32113240 : bool xmin_already_frozen = false,
7066 32113240 : xmax_already_frozen = false;
7067 32113240 : bool freeze_xmin = false,
7068 32113240 : replace_xvac = false,
7069 32113240 : replace_xmax = false,
7070 32113240 : freeze_xmax = false;
7071 : TransactionId xid;
7072 :
7073 32113240 : frz->xmax = HeapTupleHeaderGetRawXmax(tuple);
7074 32113240 : frz->t_infomask2 = tuple->t_infomask2;
7075 32113240 : frz->t_infomask = tuple->t_infomask;
7076 32113240 : frz->frzflags = 0;
7077 32113240 : frz->checkflags = 0;
7078 :
7079 : /*
7080 : * Process xmin, while keeping track of whether it's already frozen, or
7081 : * will become frozen iff our freeze plan is executed by caller (could be
7082 : * neither).
7083 : */
7084 32113240 : xid = HeapTupleHeaderGetXmin(tuple);
7085 32113240 : if (!TransactionIdIsNormal(xid))
7086 26455174 : xmin_already_frozen = true;
7087 : else
7088 : {
7089 5658066 : if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
7090 0 : ereport(ERROR,
7091 : (errcode(ERRCODE_DATA_CORRUPTED),
7092 : errmsg_internal("found xmin %u from before relfrozenxid %u",
7093 : xid, cutoffs->relfrozenxid)));
7094 :
7095 : /* Will set freeze_xmin flags in freeze plan below */
7096 5658066 : freeze_xmin = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
7097 :
7098 : /* Verify that xmin committed if and when freeze plan is executed */
7099 5658066 : if (freeze_xmin)
7100 4583760 : frz->checkflags |= HEAP_FREEZE_CHECK_XMIN_COMMITTED;
7101 : }
7102 :
7103 : /*
7104 : * Old-style VACUUM FULL is gone, but we have to process xvac for as long
7105 : * as we support having MOVED_OFF/MOVED_IN tuples in the database
7106 : */
7107 32113240 : xid = HeapTupleHeaderGetXvac(tuple);
7108 32113240 : if (TransactionIdIsNormal(xid))
7109 : {
7110 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7111 : Assert(TransactionIdPrecedes(xid, cutoffs->OldestXmin));
7112 :
7113 : /*
7114 : * For Xvac, we always freeze proactively. This allows totally_frozen
7115 : * tracking to ignore xvac.
7116 : */
7117 0 : replace_xvac = pagefrz->freeze_required = true;
7118 :
7119 : /* Will set replace_xvac flags in freeze plan below */
7120 : }
7121 :
7122 : /* Now process xmax */
7123 32113240 : xid = frz->xmax;
7124 32113240 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7125 : {
7126 : /* Raw xmax is a MultiXactId */
7127 : TransactionId newxmax;
7128 : uint16 flags;
7129 :
7130 : /*
7131 : * We will either remove xmax completely (in the "freeze_xmax" path),
7132 : * process xmax by replacing it (in the "replace_xmax" path), or
7133 : * perform no-op xmax processing. The only constraint is that the
7134 : * FreezeLimit/MultiXactCutoff postcondition must never be violated.
7135 : */
7136 12 : newxmax = FreezeMultiXactId(xid, tuple->t_infomask, cutoffs,
7137 : &flags, pagefrz);
7138 :
7139 12 : if (flags & FRM_NOOP)
7140 : {
7141 : /*
7142 : * xmax is a MultiXactId, and nothing about it changes for now.
7143 : * This is the only case where 'freeze_required' won't have been
7144 : * set for us by FreezeMultiXactId, as well as the only case where
7145 : * neither freeze_xmax nor replace_xmax are set (given a multi).
7146 : *
7147 : * This is a no-op, but the call to FreezeMultiXactId might have
7148 : * ratcheted back NewRelfrozenXid and/or NewRelminMxid trackers
7149 : * for us (the "freeze page" variants, specifically). That'll
7150 : * make it safe for our caller to freeze the page later on, while
7151 : * leaving this particular xmax undisturbed.
7152 : *
7153 : * FreezeMultiXactId is _not_ responsible for the "no freeze"
7154 : * NewRelfrozenXid/NewRelminMxid trackers, though -- that's our
7155 : * job. A call to heap_tuple_should_freeze for this same tuple
7156 : * will take place below if 'freeze_required' isn't set already.
7157 : * (This repeats work from FreezeMultiXactId, but allows "no
7158 : * freeze" tracker maintenance to happen in only one place.)
7159 : */
7160 : Assert(!MultiXactIdPrecedes(newxmax, cutoffs->MultiXactCutoff));
7161 : Assert(MultiXactIdIsValid(newxmax) && xid == newxmax);
7162 : }
7163 10 : else if (flags & FRM_RETURN_IS_XID)
7164 : {
7165 : /*
7166 : * xmax will become an updater Xid (original MultiXact's updater
7167 : * member Xid will be carried forward as a simple Xid in Xmax).
7168 : */
7169 : Assert(!TransactionIdPrecedes(newxmax, cutoffs->OldestXmin));
7170 :
7171 : /*
7172 : * NB -- some of these transformations are only valid because we
7173 : * know the return Xid is a tuple updater (i.e. not merely a
7174 : * locker.) Also note that the only reason we don't explicitly
7175 : * worry about HEAP_KEYS_UPDATED is because it lives in
7176 : * t_infomask2 rather than t_infomask.
7177 : */
7178 0 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7179 0 : frz->xmax = newxmax;
7180 0 : if (flags & FRM_MARK_COMMITTED)
7181 0 : frz->t_infomask |= HEAP_XMAX_COMMITTED;
7182 0 : replace_xmax = true;
7183 : }
7184 10 : else if (flags & FRM_RETURN_IS_MULTI)
7185 : {
7186 : uint16 newbits;
7187 : uint16 newbits2;
7188 :
7189 : /*
7190 : * xmax is an old MultiXactId that we have to replace with a new
7191 : * MultiXactId, to carry forward two or more original member XIDs.
7192 : */
7193 : Assert(!MultiXactIdPrecedes(newxmax, cutoffs->OldestMxact));
7194 :
7195 : /*
7196 : * We can't use GetMultiXactIdHintBits directly on the new multi
7197 : * here; that routine initializes the masks to all zeroes, which
7198 : * would lose other bits we need. Doing it this way ensures all
7199 : * unrelated bits remain untouched.
7200 : */
7201 2 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7202 2 : frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
7203 2 : GetMultiXactIdHintBits(newxmax, &newbits, &newbits2);
7204 2 : frz->t_infomask |= newbits;
7205 2 : frz->t_infomask2 |= newbits2;
7206 2 : frz->xmax = newxmax;
7207 2 : replace_xmax = true;
7208 : }
7209 : else
7210 : {
7211 : /*
7212 : * Freeze plan for tuple "freezes xmax" in the strictest sense:
7213 : * it'll leave nothing in xmax (neither an Xid nor a MultiXactId).
7214 : */
7215 : Assert(flags & FRM_INVALIDATE_XMAX);
7216 : Assert(!TransactionIdIsValid(newxmax));
7217 :
7218 : /* Will set freeze_xmax flags in freeze plan below */
7219 8 : freeze_xmax = true;
7220 : }
7221 :
7222 : /* MultiXactId processing forces freezing (barring FRM_NOOP case) */
7223 : Assert(pagefrz->freeze_required || (!freeze_xmax && !replace_xmax));
7224 : }
7225 32113228 : else if (TransactionIdIsNormal(xid))
7226 : {
7227 : /* Raw xmax is normal XID */
7228 11842528 : if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
7229 0 : ereport(ERROR,
7230 : (errcode(ERRCODE_DATA_CORRUPTED),
7231 : errmsg_internal("found xmax %u from before relfrozenxid %u",
7232 : xid, cutoffs->relfrozenxid)));
7233 :
7234 : /* Will set freeze_xmax flags in freeze plan below */
7235 11842528 : freeze_xmax = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
7236 :
7237 : /*
7238 : * Verify that xmax aborted if and when freeze plan is executed,
7239 : * provided it's from an update. (A lock-only xmax can be removed
7240 : * independent of this, since the lock is released at xact end.)
7241 : */
7242 11842528 : if (freeze_xmax && !HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask))
7243 1188 : frz->checkflags |= HEAP_FREEZE_CHECK_XMAX_ABORTED;
7244 : }
7245 20270700 : else if (!TransactionIdIsValid(xid))
7246 : {
7247 : /* Raw xmax is InvalidTransactionId XID */
7248 : Assert((tuple->t_infomask & HEAP_XMAX_IS_MULTI) == 0);
7249 20270700 : xmax_already_frozen = true;
7250 : }
7251 : else
7252 0 : ereport(ERROR,
7253 : (errcode(ERRCODE_DATA_CORRUPTED),
7254 : errmsg_internal("found raw xmax %u (infomask 0x%04x) not invalid and not multi",
7255 : xid, tuple->t_infomask)));
7256 :
7257 32113240 : if (freeze_xmin)
7258 : {
7259 : Assert(!xmin_already_frozen);
7260 :
7261 4583760 : frz->t_infomask |= HEAP_XMIN_FROZEN;
7262 : }
7263 32113240 : if (replace_xvac)
7264 : {
7265 : /*
7266 : * If a MOVED_OFF tuple is not dead, the xvac transaction must have
7267 : * failed; whereas a non-dead MOVED_IN tuple must mean the xvac
7268 : * transaction succeeded.
7269 : */
7270 : Assert(pagefrz->freeze_required);
7271 0 : if (tuple->t_infomask & HEAP_MOVED_OFF)
7272 0 : frz->frzflags |= XLH_INVALID_XVAC;
7273 : else
7274 0 : frz->frzflags |= XLH_FREEZE_XVAC;
7275 : }
7276 : if (replace_xmax)
7277 : {
7278 : Assert(!xmax_already_frozen && !freeze_xmax);
7279 : Assert(pagefrz->freeze_required);
7280 :
7281 : /* Already set replace_xmax flags in freeze plan earlier */
7282 : }
7283 32113240 : if (freeze_xmax)
7284 : {
7285 : Assert(!xmax_already_frozen && !replace_xmax);
7286 :
7287 3126 : frz->xmax = InvalidTransactionId;
7288 :
7289 : /*
7290 : * The tuple might be marked either XMAX_INVALID or XMAX_COMMITTED +
7291 : * LOCKED. Normalize to INVALID just to be sure no one gets confused.
7292 : * Also get rid of the HEAP_KEYS_UPDATED bit.
7293 : */
7294 3126 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7295 3126 : frz->t_infomask |= HEAP_XMAX_INVALID;
7296 3126 : frz->t_infomask2 &= ~HEAP_HOT_UPDATED;
7297 3126 : frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
7298 : }
7299 :
7300 : /*
7301 : * Determine if this tuple is already totally frozen, or will become
7302 : * totally frozen (provided caller executes freeze plans for the page)
7303 : */
7304 63149048 : *totally_frozen = ((freeze_xmin || xmin_already_frozen) &&
7305 31035808 : (freeze_xmax || xmax_already_frozen));
7306 :
7307 32113240 : if (!pagefrz->freeze_required && !(xmin_already_frozen &&
7308 : xmax_already_frozen))
7309 : {
7310 : /*
7311 : * So far no previous tuple from the page made freezing mandatory.
7312 : * Does this tuple force caller to freeze the entire page?
7313 : */
7314 14581744 : pagefrz->freeze_required =
7315 14581744 : heap_tuple_should_freeze(tuple, cutoffs,
7316 : &pagefrz->NoFreezePageRelfrozenXid,
7317 : &pagefrz->NoFreezePageRelminMxid);
7318 : }
7319 :
7320 : /* Tell caller if this tuple has a usable freeze plan set in *frz */
7321 32113240 : return freeze_xmin || replace_xvac || replace_xmax || freeze_xmax;
7322 : }
7323 :
7324 : /*
7325 : * Perform xmin/xmax XID status sanity checks before actually executing freeze
7326 : * plans.
7327 : *
7328 : * heap_prepare_freeze_tuple doesn't perform these checks directly because
7329 : * pg_xact lookups are relatively expensive. They shouldn't be repeated by
7330 : * successive VACUUMs that each decide against freezing the same page.
7331 : */
7332 : void
7333 40176 : heap_pre_freeze_checks(Buffer buffer,
7334 : HeapTupleFreeze *tuples, int ntuples)
7335 : {
7336 40176 : Page page = BufferGetPage(buffer);
7337 :
7338 1774622 : for (int i = 0; i < ntuples; i++)
7339 : {
7340 1734446 : HeapTupleFreeze *frz = tuples + i;
7341 1734446 : ItemId itemid = PageGetItemId(page, frz->offset);
7342 : HeapTupleHeader htup;
7343 :
7344 1734446 : htup = (HeapTupleHeader) PageGetItem(page, itemid);
7345 :
7346 : /* Deliberately avoid relying on tuple hint bits here */
7347 1734446 : if (frz->checkflags & HEAP_FREEZE_CHECK_XMIN_COMMITTED)
7348 : {
7349 1734444 : TransactionId xmin = HeapTupleHeaderGetRawXmin(htup);
7350 :
7351 : Assert(!HeapTupleHeaderXminFrozen(htup));
7352 1734444 : if (unlikely(!TransactionIdDidCommit(xmin)))
7353 0 : ereport(ERROR,
7354 : (errcode(ERRCODE_DATA_CORRUPTED),
7355 : errmsg_internal("uncommitted xmin %u needs to be frozen",
7356 : xmin)));
7357 : }
7358 :
7359 : /*
7360 : * TransactionIdDidAbort won't work reliably in the presence of XIDs
7361 : * left behind by transactions that were in progress during a crash,
7362 : * so we can only check that xmax didn't commit
7363 : */
7364 1734446 : if (frz->checkflags & HEAP_FREEZE_CHECK_XMAX_ABORTED)
7365 : {
7366 94 : TransactionId xmax = HeapTupleHeaderGetRawXmax(htup);
7367 :
7368 : Assert(TransactionIdIsNormal(xmax));
7369 94 : if (unlikely(TransactionIdDidCommit(xmax)))
7370 0 : ereport(ERROR,
7371 : (errcode(ERRCODE_DATA_CORRUPTED),
7372 : errmsg_internal("cannot freeze committed xmax %u",
7373 : xmax)));
7374 : }
7375 : }
7376 40176 : }
7377 :
7378 : /*
7379 : * Helper which executes freezing of one or more heap tuples on a page on
7380 : * behalf of caller. Caller passes an array of tuple plans from
7381 : * heap_prepare_freeze_tuple. Caller must set 'offset' in each plan for us.
7382 : * Must be called in a critical section that also marks the buffer dirty and,
7383 : * if needed, emits WAL.
7384 : */
7385 : void
7386 40176 : heap_freeze_prepared_tuples(Buffer buffer, HeapTupleFreeze *tuples, int ntuples)
7387 : {
7388 40176 : Page page = BufferGetPage(buffer);
7389 :
7390 1774622 : for (int i = 0; i < ntuples; i++)
7391 : {
7392 1734446 : HeapTupleFreeze *frz = tuples + i;
7393 1734446 : ItemId itemid = PageGetItemId(page, frz->offset);
7394 : HeapTupleHeader htup;
7395 :
7396 1734446 : htup = (HeapTupleHeader) PageGetItem(page, itemid);
7397 1734446 : heap_execute_freeze_tuple(htup, frz);
7398 : }
7399 40176 : }
7400 :
7401 : /*
7402 : * heap_freeze_tuple
7403 : * Freeze tuple in place, without WAL logging.
7404 : *
7405 : * Useful for callers like CLUSTER that perform their own WAL logging.
7406 : */
7407 : bool
7408 748098 : heap_freeze_tuple(HeapTupleHeader tuple,
7409 : TransactionId relfrozenxid, TransactionId relminmxid,
7410 : TransactionId FreezeLimit, TransactionId MultiXactCutoff)
7411 : {
7412 : HeapTupleFreeze frz;
7413 : bool do_freeze;
7414 : bool totally_frozen;
7415 : struct VacuumCutoffs cutoffs;
7416 : HeapPageFreeze pagefrz;
7417 :
7418 748098 : cutoffs.relfrozenxid = relfrozenxid;
7419 748098 : cutoffs.relminmxid = relminmxid;
7420 748098 : cutoffs.OldestXmin = FreezeLimit;
7421 748098 : cutoffs.OldestMxact = MultiXactCutoff;
7422 748098 : cutoffs.FreezeLimit = FreezeLimit;
7423 748098 : cutoffs.MultiXactCutoff = MultiXactCutoff;
7424 :
7425 748098 : pagefrz.freeze_required = true;
7426 748098 : pagefrz.FreezePageRelfrozenXid = FreezeLimit;
7427 748098 : pagefrz.FreezePageRelminMxid = MultiXactCutoff;
7428 748098 : pagefrz.NoFreezePageRelfrozenXid = FreezeLimit;
7429 748098 : pagefrz.NoFreezePageRelminMxid = MultiXactCutoff;
7430 :
7431 748098 : do_freeze = heap_prepare_freeze_tuple(tuple, &cutoffs,
7432 : &pagefrz, &frz, &totally_frozen);
7433 :
7434 : /*
7435 : * Note that because this is not a WAL-logged operation, we don't need to
7436 : * fill in the offset in the freeze record.
7437 : */
7438 :
7439 748098 : if (do_freeze)
7440 531988 : heap_execute_freeze_tuple(tuple, &frz);
7441 748098 : return do_freeze;
7442 : }
7443 :
7444 : /*
7445 : * For a given MultiXactId, return the hint bits that should be set in the
7446 : * tuple's infomask.
7447 : *
7448 : * Normally this should be called for a multixact that was just created, and
7449 : * so is on our local cache, so the GetMembers call is fast.
7450 : */
7451 : static void
7452 2378 : GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
7453 : uint16 *new_infomask2)
7454 : {
7455 : int nmembers;
7456 : MultiXactMember *members;
7457 : int i;
7458 2378 : uint16 bits = HEAP_XMAX_IS_MULTI;
7459 2378 : uint16 bits2 = 0;
7460 2378 : bool has_update = false;
7461 2378 : LockTupleMode strongest = LockTupleKeyShare;
7462 :
7463 : /*
7464 : * We only use this in multis we just created, so they cannot be values
7465 : * pre-pg_upgrade.
7466 : */
7467 2378 : nmembers = GetMultiXactIdMembers(multi, &members, false, false);
7468 :
7469 7270 : for (i = 0; i < nmembers; i++)
7470 : {
7471 : LockTupleMode mode;
7472 :
7473 : /*
7474 : * Remember the strongest lock mode held by any member of the
7475 : * multixact.
7476 : */
7477 4892 : mode = TUPLOCK_from_mxstatus(members[i].status);
7478 4892 : if (mode > strongest)
7479 1318 : strongest = mode;
7480 :
7481 : /* See what other bits we need */
7482 4892 : switch (members[i].status)
7483 : {
7484 4510 : case MultiXactStatusForKeyShare:
7485 : case MultiXactStatusForShare:
7486 : case MultiXactStatusForNoKeyUpdate:
7487 4510 : break;
7488 :
7489 104 : case MultiXactStatusForUpdate:
7490 104 : bits2 |= HEAP_KEYS_UPDATED;
7491 104 : break;
7492 :
7493 258 : case MultiXactStatusNoKeyUpdate:
7494 258 : has_update = true;
7495 258 : break;
7496 :
7497 20 : case MultiXactStatusUpdate:
7498 20 : bits2 |= HEAP_KEYS_UPDATED;
7499 20 : has_update = true;
7500 20 : break;
7501 : }
7502 4892 : }
7503 :
7504 2378 : if (strongest == LockTupleExclusive ||
7505 : strongest == LockTupleNoKeyExclusive)
7506 438 : bits |= HEAP_XMAX_EXCL_LOCK;
7507 1940 : else if (strongest == LockTupleShare)
7508 874 : bits |= HEAP_XMAX_SHR_LOCK;
7509 1066 : else if (strongest == LockTupleKeyShare)
7510 1066 : bits |= HEAP_XMAX_KEYSHR_LOCK;
7511 :
7512 2378 : if (!has_update)
7513 2100 : bits |= HEAP_XMAX_LOCK_ONLY;
7514 :
7515 2378 : if (nmembers > 0)
7516 2378 : pfree(members);
7517 :
7518 2378 : *new_infomask = bits;
7519 2378 : *new_infomask2 = bits2;
7520 2378 : }
7521 :
7522 : /*
7523 : * MultiXactIdGetUpdateXid
7524 : *
7525 : * Given a multixact Xmax and corresponding infomask, which does not have the
7526 : * HEAP_XMAX_LOCK_ONLY bit set, obtain and return the Xid of the updating
7527 : * transaction.
7528 : *
7529 : * Caller is expected to check the status of the updating transaction, if
7530 : * necessary.
7531 : */
7532 : static TransactionId
7533 1094 : MultiXactIdGetUpdateXid(TransactionId xmax, uint16 t_infomask)
7534 : {
7535 1094 : TransactionId update_xact = InvalidTransactionId;
7536 : MultiXactMember *members;
7537 : int nmembers;
7538 :
7539 : Assert(!(t_infomask & HEAP_XMAX_LOCK_ONLY));
7540 : Assert(t_infomask & HEAP_XMAX_IS_MULTI);
7541 :
7542 : /*
7543 : * Since we know the LOCK_ONLY bit is not set, this cannot be a multi from
7544 : * pre-pg_upgrade.
7545 : */
7546 1094 : nmembers = GetMultiXactIdMembers(xmax, &members, false, false);
7547 :
7548 1094 : if (nmembers > 0)
7549 : {
7550 : int i;
7551 :
7552 2890 : for (i = 0; i < nmembers; i++)
7553 : {
7554 : /* Ignore lockers */
7555 2890 : if (!ISUPDATE_from_mxstatus(members[i].status))
7556 1796 : continue;
7557 :
7558 : /* there can be at most one updater */
7559 : Assert(update_xact == InvalidTransactionId);
7560 1094 : update_xact = members[i].xid;
7561 : #ifndef USE_ASSERT_CHECKING
7562 :
7563 : /*
7564 : * in an assert-enabled build, walk the whole array to ensure
7565 : * there's no other updater.
7566 : */
7567 1094 : break;
7568 : #endif
7569 : }
7570 :
7571 1094 : pfree(members);
7572 : }
7573 :
7574 1094 : return update_xact;
7575 : }
7576 :
7577 : /*
7578 : * HeapTupleGetUpdateXid
7579 : * As above, but use a HeapTupleHeader
7580 : *
7581 : * See also HeapTupleHeaderGetUpdateXid, which can be used without previously
7582 : * checking the hint bits.
7583 : */
7584 : TransactionId
7585 1078 : HeapTupleGetUpdateXid(const HeapTupleHeaderData *tup)
7586 : {
7587 1078 : return MultiXactIdGetUpdateXid(HeapTupleHeaderGetRawXmax(tup),
7588 1078 : tup->t_infomask);
7589 : }
7590 :
7591 : /*
7592 : * Does the given multixact conflict with the current transaction grabbing a
7593 : * tuple lock of the given strength?
7594 : *
7595 : * The passed infomask pairs up with the given multixact in the tuple header.
7596 : *
7597 : * If current_is_member is not NULL, it is set to 'true' if the current
7598 : * transaction is a member of the given multixact.
7599 : */
7600 : static bool
7601 198 : DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
7602 : LockTupleMode lockmode, bool *current_is_member)
7603 : {
7604 : int nmembers;
7605 : MultiXactMember *members;
7606 198 : bool result = false;
7607 198 : LOCKMODE wanted = tupleLockExtraInfo[lockmode].hwlock;
7608 :
7609 198 : if (HEAP_LOCKED_UPGRADED(infomask))
7610 0 : return false;
7611 :
7612 198 : nmembers = GetMultiXactIdMembers(multi, &members, false,
7613 198 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
7614 198 : if (nmembers >= 0)
7615 : {
7616 : int i;
7617 :
7618 618 : for (i = 0; i < nmembers; i++)
7619 : {
7620 : TransactionId memxid;
7621 : LOCKMODE memlockmode;
7622 :
7623 434 : if (result && (current_is_member == NULL || *current_is_member))
7624 : break;
7625 :
7626 420 : memlockmode = LOCKMODE_from_mxstatus(members[i].status);
7627 :
7628 : /* ignore members from current xact (but track their presence) */
7629 420 : memxid = members[i].xid;
7630 420 : if (TransactionIdIsCurrentTransactionId(memxid))
7631 : {
7632 184 : if (current_is_member != NULL)
7633 156 : *current_is_member = true;
7634 184 : continue;
7635 : }
7636 236 : else if (result)
7637 16 : continue;
7638 :
7639 : /* ignore members that don't conflict with the lock we want */
7640 220 : if (!DoLockModesConflict(memlockmode, wanted))
7641 142 : continue;
7642 :
7643 78 : if (ISUPDATE_from_mxstatus(members[i].status))
7644 : {
7645 : /* ignore aborted updaters */
7646 34 : if (TransactionIdDidAbort(memxid))
7647 2 : continue;
7648 : }
7649 : else
7650 : {
7651 : /* ignore lockers-only that are no longer in progress */
7652 44 : if (!TransactionIdIsInProgress(memxid))
7653 14 : continue;
7654 : }
7655 :
7656 : /*
7657 : * Whatever remains are either live lockers that conflict with our
7658 : * wanted lock, and updaters that are not aborted. Those conflict
7659 : * with what we want. Set up to return true, but keep going to
7660 : * look for the current transaction among the multixact members,
7661 : * if needed.
7662 : */
7663 62 : result = true;
7664 : }
7665 198 : pfree(members);
7666 : }
7667 :
7668 198 : return result;
7669 : }
7670 :
7671 : /*
7672 : * Do_MultiXactIdWait
7673 : * Actual implementation for the two functions below.
7674 : *
7675 : * 'multi', 'status' and 'infomask' indicate what to sleep on (the status is
7676 : * needed to ensure we only sleep on conflicting members, and the infomask is
7677 : * used to optimize multixact access in case it's a lock-only multi); 'nowait'
7678 : * indicates whether to use conditional lock acquisition, to allow callers to
7679 : * fail if lock is unavailable. 'rel', 'ctid' and 'oper' are used to set up
7680 : * context information for error messages. 'remaining', if not NULL, receives
7681 : * the number of members that are still running, including any (non-aborted)
7682 : * subtransactions of our own transaction. 'logLockFailure' indicates whether
7683 : * to log details when a lock acquisition fails with 'nowait' enabled.
7684 : *
7685 : * We do this by sleeping on each member using XactLockTableWait. Any
7686 : * members that belong to the current backend are *not* waited for, however;
7687 : * this would not merely be useless but would lead to Assert failure inside
7688 : * XactLockTableWait. By the time this returns, it is certain that all
7689 : * transactions *of other backends* that were members of the MultiXactId
7690 : * that conflict with the requested status are dead (and no new ones can have
7691 : * been added, since it is not legal to add members to an existing
7692 : * MultiXactId).
7693 : *
7694 : * But by the time we finish sleeping, someone else may have changed the Xmax
7695 : * of the containing tuple, so the caller needs to iterate on us somehow.
7696 : *
7697 : * Note that in case we return false, the number of remaining members is
7698 : * not to be trusted.
7699 : */
7700 : static bool
7701 116 : Do_MultiXactIdWait(MultiXactId multi, MultiXactStatus status,
7702 : uint16 infomask, bool nowait,
7703 : Relation rel, ItemPointer ctid, XLTW_Oper oper,
7704 : int *remaining, bool logLockFailure)
7705 : {
7706 116 : bool result = true;
7707 : MultiXactMember *members;
7708 : int nmembers;
7709 116 : int remain = 0;
7710 :
7711 : /* for pre-pg_upgrade tuples, no need to sleep at all */
7712 116 : nmembers = HEAP_LOCKED_UPGRADED(infomask) ? -1 :
7713 116 : GetMultiXactIdMembers(multi, &members, false,
7714 116 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
7715 :
7716 116 : if (nmembers >= 0)
7717 : {
7718 : int i;
7719 :
7720 374 : for (i = 0; i < nmembers; i++)
7721 : {
7722 266 : TransactionId memxid = members[i].xid;
7723 266 : MultiXactStatus memstatus = members[i].status;
7724 :
7725 266 : if (TransactionIdIsCurrentTransactionId(memxid))
7726 : {
7727 48 : remain++;
7728 48 : continue;
7729 : }
7730 :
7731 218 : if (!DoLockModesConflict(LOCKMODE_from_mxstatus(memstatus),
7732 218 : LOCKMODE_from_mxstatus(status)))
7733 : {
7734 44 : if (remaining && TransactionIdIsInProgress(memxid))
7735 16 : remain++;
7736 44 : continue;
7737 : }
7738 :
7739 : /*
7740 : * This member conflicts with our multi, so we have to sleep (or
7741 : * return failure, if asked to avoid waiting.)
7742 : *
7743 : * Note that we don't set up an error context callback ourselves,
7744 : * but instead we pass the info down to XactLockTableWait. This
7745 : * might seem a bit wasteful because the context is set up and
7746 : * tore down for each member of the multixact, but in reality it
7747 : * should be barely noticeable, and it avoids duplicate code.
7748 : */
7749 174 : if (nowait)
7750 : {
7751 8 : result = ConditionalXactLockTableWait(memxid, logLockFailure);
7752 8 : if (!result)
7753 8 : break;
7754 : }
7755 : else
7756 166 : XactLockTableWait(memxid, rel, ctid, oper);
7757 : }
7758 :
7759 116 : pfree(members);
7760 : }
7761 :
7762 116 : if (remaining)
7763 20 : *remaining = remain;
7764 :
7765 116 : return result;
7766 : }
7767 :
7768 : /*
7769 : * MultiXactIdWait
7770 : * Sleep on a MultiXactId.
7771 : *
7772 : * By the time we finish sleeping, someone else may have changed the Xmax
7773 : * of the containing tuple, so the caller needs to iterate on us somehow.
7774 : *
7775 : * We return (in *remaining, if not NULL) the number of members that are still
7776 : * running, including any (non-aborted) subtransactions of our own transaction.
7777 : */
7778 : static void
7779 108 : MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
7780 : Relation rel, ItemPointer ctid, XLTW_Oper oper,
7781 : int *remaining)
7782 : {
7783 108 : (void) Do_MultiXactIdWait(multi, status, infomask, false,
7784 : rel, ctid, oper, remaining, false);
7785 108 : }
7786 :
7787 : /*
7788 : * ConditionalMultiXactIdWait
7789 : * As above, but only lock if we can get the lock without blocking.
7790 : *
7791 : * By the time we finish sleeping, someone else may have changed the Xmax
7792 : * of the containing tuple, so the caller needs to iterate on us somehow.
7793 : *
7794 : * If the multixact is now all gone, return true. Returns false if some
7795 : * transactions might still be running.
7796 : *
7797 : * We return (in *remaining, if not NULL) the number of members that are still
7798 : * running, including any (non-aborted) subtransactions of our own transaction.
7799 : */
7800 : static bool
7801 8 : ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status,
7802 : uint16 infomask, Relation rel, int *remaining,
7803 : bool logLockFailure)
7804 : {
7805 8 : return Do_MultiXactIdWait(multi, status, infomask, true,
7806 : rel, NULL, XLTW_None, remaining, logLockFailure);
7807 : }
7808 :
7809 : /*
7810 : * heap_tuple_needs_eventual_freeze
7811 : *
7812 : * Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
7813 : * will eventually require freezing (if tuple isn't removed by pruning first).
7814 : */
7815 : bool
7816 218420 : heap_tuple_needs_eventual_freeze(HeapTupleHeader tuple)
7817 : {
7818 : TransactionId xid;
7819 :
7820 : /*
7821 : * If xmin is a normal transaction ID, this tuple is definitely not
7822 : * frozen.
7823 : */
7824 218420 : xid = HeapTupleHeaderGetXmin(tuple);
7825 218420 : if (TransactionIdIsNormal(xid))
7826 5350 : return true;
7827 :
7828 : /*
7829 : * If xmax is a valid xact or multixact, this tuple is also not frozen.
7830 : */
7831 213070 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7832 : {
7833 : MultiXactId multi;
7834 :
7835 0 : multi = HeapTupleHeaderGetRawXmax(tuple);
7836 0 : if (MultiXactIdIsValid(multi))
7837 0 : return true;
7838 : }
7839 : else
7840 : {
7841 213070 : xid = HeapTupleHeaderGetRawXmax(tuple);
7842 213070 : if (TransactionIdIsNormal(xid))
7843 10 : return true;
7844 : }
7845 :
7846 213060 : if (tuple->t_infomask & HEAP_MOVED)
7847 : {
7848 0 : xid = HeapTupleHeaderGetXvac(tuple);
7849 0 : if (TransactionIdIsNormal(xid))
7850 0 : return true;
7851 : }
7852 :
7853 213060 : return false;
7854 : }
7855 :
7856 : /*
7857 : * heap_tuple_should_freeze
7858 : *
7859 : * Return value indicates if heap_prepare_freeze_tuple sibling function would
7860 : * (or should) force freezing of the heap page that contains caller's tuple.
7861 : * Tuple header XIDs/MXIDs < FreezeLimit/MultiXactCutoff trigger freezing.
7862 : * This includes (xmin, xmax, xvac) fields, as well as MultiXact member XIDs.
7863 : *
7864 : * The *NoFreezePageRelfrozenXid and *NoFreezePageRelminMxid input/output
7865 : * arguments help VACUUM track the oldest extant XID/MXID remaining in rel.
7866 : * Our working assumption is that caller won't decide to freeze this tuple.
7867 : * It's up to caller to only ratchet back its own top-level trackers after the
7868 : * point that it fully commits to not freezing the tuple/page in question.
7869 : */
7870 : bool
7871 14582804 : heap_tuple_should_freeze(HeapTupleHeader tuple,
7872 : const struct VacuumCutoffs *cutoffs,
7873 : TransactionId *NoFreezePageRelfrozenXid,
7874 : MultiXactId *NoFreezePageRelminMxid)
7875 : {
7876 : TransactionId xid;
7877 : MultiXactId multi;
7878 14582804 : bool freeze = false;
7879 :
7880 : /* First deal with xmin */
7881 14582804 : xid = HeapTupleHeaderGetXmin(tuple);
7882 14582804 : if (TransactionIdIsNormal(xid))
7883 : {
7884 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7885 3394362 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7886 41604 : *NoFreezePageRelfrozenXid = xid;
7887 3394362 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7888 38588 : freeze = true;
7889 : }
7890 :
7891 : /* Now deal with xmax */
7892 14582804 : xid = InvalidTransactionId;
7893 14582804 : multi = InvalidMultiXactId;
7894 14582804 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7895 4 : multi = HeapTupleHeaderGetRawXmax(tuple);
7896 : else
7897 14582800 : xid = HeapTupleHeaderGetRawXmax(tuple);
7898 :
7899 14582804 : if (TransactionIdIsNormal(xid))
7900 : {
7901 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7902 : /* xmax is a non-permanent XID */
7903 11677492 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7904 2 : *NoFreezePageRelfrozenXid = xid;
7905 11677492 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7906 8 : freeze = true;
7907 : }
7908 2905312 : else if (!MultiXactIdIsValid(multi))
7909 : {
7910 : /* xmax is a permanent XID or invalid MultiXactId/XID */
7911 : }
7912 4 : else if (HEAP_LOCKED_UPGRADED(tuple->t_infomask))
7913 : {
7914 : /* xmax is a pg_upgrade'd MultiXact, which can't have updater XID */
7915 0 : if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
7916 0 : *NoFreezePageRelminMxid = multi;
7917 : /* heap_prepare_freeze_tuple always freezes pg_upgrade'd xmax */
7918 0 : freeze = true;
7919 : }
7920 : else
7921 : {
7922 : /* xmax is a MultiXactId that may have an updater XID */
7923 : MultiXactMember *members;
7924 : int nmembers;
7925 :
7926 : Assert(MultiXactIdPrecedesOrEquals(cutoffs->relminmxid, multi));
7927 4 : if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
7928 4 : *NoFreezePageRelminMxid = multi;
7929 4 : if (MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff))
7930 4 : freeze = true;
7931 :
7932 : /* need to check whether any member of the mxact is old */
7933 4 : nmembers = GetMultiXactIdMembers(multi, &members, false,
7934 4 : HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask));
7935 :
7936 10 : for (int i = 0; i < nmembers; i++)
7937 : {
7938 6 : xid = members[i].xid;
7939 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7940 6 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7941 0 : *NoFreezePageRelfrozenXid = xid;
7942 6 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7943 0 : freeze = true;
7944 : }
7945 4 : if (nmembers > 0)
7946 2 : pfree(members);
7947 : }
7948 :
7949 14582804 : if (tuple->t_infomask & HEAP_MOVED)
7950 : {
7951 0 : xid = HeapTupleHeaderGetXvac(tuple);
7952 0 : if (TransactionIdIsNormal(xid))
7953 : {
7954 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7955 0 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7956 0 : *NoFreezePageRelfrozenXid = xid;
7957 : /* heap_prepare_freeze_tuple forces xvac freezing */
7958 0 : freeze = true;
7959 : }
7960 : }
7961 :
7962 14582804 : return freeze;
7963 : }
7964 :
7965 : /*
7966 : * Maintain snapshotConflictHorizon for caller by ratcheting forward its value
7967 : * using any committed XIDs contained in 'tuple', an obsolescent heap tuple
7968 : * that caller is in the process of physically removing, e.g. via HOT pruning
7969 : * or index deletion.
7970 : *
7971 : * Caller must initialize its value to InvalidTransactionId, which is
7972 : * generally interpreted as "definitely no need for a recovery conflict".
7973 : * Final value must reflect all heap tuples that caller will physically remove
7974 : * (or remove TID references to) via its ongoing pruning/deletion operation.
7975 : * ResolveRecoveryConflictWithSnapshot() is passed the final value (taken from
7976 : * caller's WAL record) by REDO routine when it replays caller's operation.
7977 : */
7978 : void
7979 3121238 : HeapTupleHeaderAdvanceConflictHorizon(HeapTupleHeader tuple,
7980 : TransactionId *snapshotConflictHorizon)
7981 : {
7982 3121238 : TransactionId xmin = HeapTupleHeaderGetXmin(tuple);
7983 3121238 : TransactionId xmax = HeapTupleHeaderGetUpdateXid(tuple);
7984 3121238 : TransactionId xvac = HeapTupleHeaderGetXvac(tuple);
7985 :
7986 3121238 : if (tuple->t_infomask & HEAP_MOVED)
7987 : {
7988 0 : if (TransactionIdPrecedes(*snapshotConflictHorizon, xvac))
7989 0 : *snapshotConflictHorizon = xvac;
7990 : }
7991 :
7992 : /*
7993 : * Ignore tuples inserted by an aborted transaction or if the tuple was
7994 : * updated/deleted by the inserting transaction.
7995 : *
7996 : * Look for a committed hint bit, or if no xmin bit is set, check clog.
7997 : */
7998 3121238 : if (HeapTupleHeaderXminCommitted(tuple) ||
7999 206982 : (!HeapTupleHeaderXminInvalid(tuple) && TransactionIdDidCommit(xmin)))
8000 : {
8001 5597392 : if (xmax != xmin &&
8002 2626334 : TransactionIdFollows(xmax, *snapshotConflictHorizon))
8003 187156 : *snapshotConflictHorizon = xmax;
8004 : }
8005 3121238 : }
8006 :
8007 : #ifdef USE_PREFETCH
8008 : /*
8009 : * Helper function for heap_index_delete_tuples. Issues prefetch requests for
8010 : * prefetch_count buffers. The prefetch_state keeps track of all the buffers
8011 : * we can prefetch, and which have already been prefetched; each call to this
8012 : * function picks up where the previous call left off.
8013 : *
8014 : * Note: we expect the deltids array to be sorted in an order that groups TIDs
8015 : * by heap block, with all TIDs for each block appearing together in exactly
8016 : * one group.
8017 : */
8018 : static void
8019 37380 : index_delete_prefetch_buffer(Relation rel,
8020 : IndexDeletePrefetchState *prefetch_state,
8021 : int prefetch_count)
8022 : {
8023 37380 : BlockNumber cur_hblkno = prefetch_state->cur_hblkno;
8024 37380 : int count = 0;
8025 : int i;
8026 37380 : int ndeltids = prefetch_state->ndeltids;
8027 37380 : TM_IndexDelete *deltids = prefetch_state->deltids;
8028 :
8029 1306732 : for (i = prefetch_state->next_item;
8030 1277020 : i < ndeltids && count < prefetch_count;
8031 1269352 : i++)
8032 : {
8033 1269352 : ItemPointer htid = &deltids[i].tid;
8034 :
8035 2527540 : if (cur_hblkno == InvalidBlockNumber ||
8036 1258188 : ItemPointerGetBlockNumber(htid) != cur_hblkno)
8037 : {
8038 33744 : cur_hblkno = ItemPointerGetBlockNumber(htid);
8039 33744 : PrefetchBuffer(rel, MAIN_FORKNUM, cur_hblkno);
8040 33744 : count++;
8041 : }
8042 : }
8043 :
8044 : /*
8045 : * Save the prefetch position so that next time we can continue from that
8046 : * position.
8047 : */
8048 37380 : prefetch_state->next_item = i;
8049 37380 : prefetch_state->cur_hblkno = cur_hblkno;
8050 37380 : }
8051 : #endif
8052 :
8053 : /*
8054 : * Helper function for heap_index_delete_tuples. Checks for index corruption
8055 : * involving an invalid TID in index AM caller's index page.
8056 : *
8057 : * This is an ideal place for these checks. The index AM must hold a buffer
8058 : * lock on the index page containing the TIDs we examine here, so we don't
8059 : * have to worry about concurrent VACUUMs at all. We can be sure that the
8060 : * index is corrupt when htid points directly to an LP_UNUSED item or
8061 : * heap-only tuple, which is not the case during standard index scans.
8062 : */
8063 : static inline void
8064 1067118 : index_delete_check_htid(TM_IndexDeleteOp *delstate,
8065 : Page page, OffsetNumber maxoff,
8066 : ItemPointer htid, TM_IndexStatus *istatus)
8067 : {
8068 1067118 : OffsetNumber indexpagehoffnum = ItemPointerGetOffsetNumber(htid);
8069 : ItemId iid;
8070 :
8071 : Assert(OffsetNumberIsValid(istatus->idxoffnum));
8072 :
8073 1067118 : if (unlikely(indexpagehoffnum > maxoff))
8074 0 : ereport(ERROR,
8075 : (errcode(ERRCODE_INDEX_CORRUPTED),
8076 : 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\"",
8077 : ItemPointerGetBlockNumber(htid),
8078 : indexpagehoffnum,
8079 : istatus->idxoffnum, delstate->iblknum,
8080 : RelationGetRelationName(delstate->irel))));
8081 :
8082 1067118 : iid = PageGetItemId(page, indexpagehoffnum);
8083 1067118 : if (unlikely(!ItemIdIsUsed(iid)))
8084 0 : ereport(ERROR,
8085 : (errcode(ERRCODE_INDEX_CORRUPTED),
8086 : errmsg_internal("heap tid from index tuple (%u,%u) points to unused heap page item at offset %u of block %u in index \"%s\"",
8087 : ItemPointerGetBlockNumber(htid),
8088 : indexpagehoffnum,
8089 : istatus->idxoffnum, delstate->iblknum,
8090 : RelationGetRelationName(delstate->irel))));
8091 :
8092 1067118 : if (ItemIdHasStorage(iid))
8093 : {
8094 : HeapTupleHeader htup;
8095 :
8096 : Assert(ItemIdIsNormal(iid));
8097 626020 : htup = (HeapTupleHeader) PageGetItem(page, iid);
8098 :
8099 626020 : if (unlikely(HeapTupleHeaderIsHeapOnly(htup)))
8100 0 : ereport(ERROR,
8101 : (errcode(ERRCODE_INDEX_CORRUPTED),
8102 : errmsg_internal("heap tid from index tuple (%u,%u) points to heap-only tuple at offset %u of block %u in index \"%s\"",
8103 : ItemPointerGetBlockNumber(htid),
8104 : indexpagehoffnum,
8105 : istatus->idxoffnum, delstate->iblknum,
8106 : RelationGetRelationName(delstate->irel))));
8107 : }
8108 1067118 : }
8109 :
8110 : /*
8111 : * heapam implementation of tableam's index_delete_tuples interface.
8112 : *
8113 : * This helper function is called by index AMs during index tuple deletion.
8114 : * See tableam header comments for an explanation of the interface implemented
8115 : * here and a general theory of operation. Note that each call here is either
8116 : * a simple index deletion call, or a bottom-up index deletion call.
8117 : *
8118 : * It's possible for this to generate a fair amount of I/O, since we may be
8119 : * deleting hundreds of tuples from a single index block. To amortize that
8120 : * cost to some degree, this uses prefetching and combines repeat accesses to
8121 : * the same heap block.
8122 : */
8123 : TransactionId
8124 11164 : heap_index_delete_tuples(Relation rel, TM_IndexDeleteOp *delstate)
8125 : {
8126 : /* Initial assumption is that earlier pruning took care of conflict */
8127 11164 : TransactionId snapshotConflictHorizon = InvalidTransactionId;
8128 11164 : BlockNumber blkno = InvalidBlockNumber;
8129 11164 : Buffer buf = InvalidBuffer;
8130 11164 : Page page = NULL;
8131 11164 : OffsetNumber maxoff = InvalidOffsetNumber;
8132 : TransactionId priorXmax;
8133 : #ifdef USE_PREFETCH
8134 : IndexDeletePrefetchState prefetch_state;
8135 : int prefetch_distance;
8136 : #endif
8137 : SnapshotData SnapshotNonVacuumable;
8138 11164 : int finalndeltids = 0,
8139 11164 : nblocksaccessed = 0;
8140 :
8141 : /* State that's only used in bottom-up index deletion case */
8142 11164 : int nblocksfavorable = 0;
8143 11164 : int curtargetfreespace = delstate->bottomupfreespace,
8144 11164 : lastfreespace = 0,
8145 11164 : actualfreespace = 0;
8146 11164 : bool bottomup_final_block = false;
8147 :
8148 11164 : InitNonVacuumableSnapshot(SnapshotNonVacuumable, GlobalVisTestFor(rel));
8149 :
8150 : /* Sort caller's deltids array by TID for further processing */
8151 11164 : index_delete_sort(delstate);
8152 :
8153 : /*
8154 : * Bottom-up case: resort deltids array in an order attuned to where the
8155 : * greatest number of promising TIDs are to be found, and determine how
8156 : * many blocks from the start of sorted array should be considered
8157 : * favorable. This will also shrink the deltids array in order to
8158 : * eliminate completely unfavorable blocks up front.
8159 : */
8160 11164 : if (delstate->bottomup)
8161 3828 : nblocksfavorable = bottomup_sort_and_shrink(delstate);
8162 :
8163 : #ifdef USE_PREFETCH
8164 : /* Initialize prefetch state. */
8165 11164 : prefetch_state.cur_hblkno = InvalidBlockNumber;
8166 11164 : prefetch_state.next_item = 0;
8167 11164 : prefetch_state.ndeltids = delstate->ndeltids;
8168 11164 : prefetch_state.deltids = delstate->deltids;
8169 :
8170 : /*
8171 : * Determine the prefetch distance that we will attempt to maintain.
8172 : *
8173 : * Since the caller holds a buffer lock somewhere in rel, we'd better make
8174 : * sure that isn't a catalog relation before we call code that does
8175 : * syscache lookups, to avoid risk of deadlock.
8176 : */
8177 11164 : if (IsCatalogRelation(rel))
8178 7968 : prefetch_distance = maintenance_io_concurrency;
8179 : else
8180 : prefetch_distance =
8181 3196 : get_tablespace_maintenance_io_concurrency(rel->rd_rel->reltablespace);
8182 :
8183 : /* Cap initial prefetch distance for bottom-up deletion caller */
8184 11164 : if (delstate->bottomup)
8185 : {
8186 : Assert(nblocksfavorable >= 1);
8187 : Assert(nblocksfavorable <= BOTTOMUP_MAX_NBLOCKS);
8188 3828 : prefetch_distance = Min(prefetch_distance, nblocksfavorable);
8189 : }
8190 :
8191 : /* Start prefetching. */
8192 11164 : index_delete_prefetch_buffer(rel, &prefetch_state, prefetch_distance);
8193 : #endif
8194 :
8195 : /* Iterate over deltids, determine which to delete, check their horizon */
8196 : Assert(delstate->ndeltids > 0);
8197 1078282 : for (int i = 0; i < delstate->ndeltids; i++)
8198 : {
8199 1070946 : TM_IndexDelete *ideltid = &delstate->deltids[i];
8200 1070946 : TM_IndexStatus *istatus = delstate->status + ideltid->id;
8201 1070946 : ItemPointer htid = &ideltid->tid;
8202 : OffsetNumber offnum;
8203 :
8204 : /*
8205 : * Read buffer, and perform required extra steps each time a new block
8206 : * is encountered. Avoid refetching if it's the same block as the one
8207 : * from the last htid.
8208 : */
8209 2130728 : if (blkno == InvalidBlockNumber ||
8210 1059782 : ItemPointerGetBlockNumber(htid) != blkno)
8211 : {
8212 : /*
8213 : * Consider giving up early for bottom-up index deletion caller
8214 : * first. (Only prefetch next-next block afterwards, when it
8215 : * becomes clear that we're at least going to access the next
8216 : * block in line.)
8217 : *
8218 : * Sometimes the first block frees so much space for bottom-up
8219 : * caller that the deletion process can end without accessing any
8220 : * more blocks. It is usually necessary to access 2 or 3 blocks
8221 : * per bottom-up deletion operation, though.
8222 : */
8223 30044 : if (delstate->bottomup)
8224 : {
8225 : /*
8226 : * We often allow caller to delete a few additional items
8227 : * whose entries we reached after the point that space target
8228 : * from caller was satisfied. The cost of accessing the page
8229 : * was already paid at that point, so it made sense to finish
8230 : * it off. When that happened, we finalize everything here
8231 : * (by finishing off the whole bottom-up deletion operation
8232 : * without needlessly paying the cost of accessing any more
8233 : * blocks).
8234 : */
8235 8254 : if (bottomup_final_block)
8236 272 : break;
8237 :
8238 : /*
8239 : * Give up when we didn't enable our caller to free any
8240 : * additional space as a result of processing the page that we
8241 : * just finished up with. This rule is the main way in which
8242 : * we keep the cost of bottom-up deletion under control.
8243 : */
8244 7982 : if (nblocksaccessed >= 1 && actualfreespace == lastfreespace)
8245 3556 : break;
8246 4426 : lastfreespace = actualfreespace; /* for next time */
8247 :
8248 : /*
8249 : * Deletion operation (which is bottom-up) will definitely
8250 : * access the next block in line. Prepare for that now.
8251 : *
8252 : * Decay target free space so that we don't hang on for too
8253 : * long with a marginal case. (Space target is only truly
8254 : * helpful when it allows us to recognize that we don't need
8255 : * to access more than 1 or 2 blocks to satisfy caller due to
8256 : * agreeable workload characteristics.)
8257 : *
8258 : * We are a bit more patient when we encounter contiguous
8259 : * blocks, though: these are treated as favorable blocks. The
8260 : * decay process is only applied when the next block in line
8261 : * is not a favorable/contiguous block. This is not an
8262 : * exception to the general rule; we still insist on finding
8263 : * at least one deletable item per block accessed. See
8264 : * bottomup_nblocksfavorable() for full details of the theory
8265 : * behind favorable blocks and heap block locality in general.
8266 : *
8267 : * Note: The first block in line is always treated as a
8268 : * favorable block, so the earliest possible point that the
8269 : * decay can be applied is just before we access the second
8270 : * block in line. The Assert() verifies this for us.
8271 : */
8272 : Assert(nblocksaccessed > 0 || nblocksfavorable > 0);
8273 4426 : if (nblocksfavorable > 0)
8274 4220 : nblocksfavorable--;
8275 : else
8276 206 : curtargetfreespace /= 2;
8277 : }
8278 :
8279 : /* release old buffer */
8280 26216 : if (BufferIsValid(buf))
8281 15052 : UnlockReleaseBuffer(buf);
8282 :
8283 26216 : blkno = ItemPointerGetBlockNumber(htid);
8284 26216 : buf = ReadBuffer(rel, blkno);
8285 26216 : nblocksaccessed++;
8286 : Assert(!delstate->bottomup ||
8287 : nblocksaccessed <= BOTTOMUP_MAX_NBLOCKS);
8288 :
8289 : #ifdef USE_PREFETCH
8290 :
8291 : /*
8292 : * To maintain the prefetch distance, prefetch one more page for
8293 : * each page we read.
8294 : */
8295 26216 : index_delete_prefetch_buffer(rel, &prefetch_state, 1);
8296 : #endif
8297 :
8298 26216 : LockBuffer(buf, BUFFER_LOCK_SHARE);
8299 :
8300 26216 : page = BufferGetPage(buf);
8301 26216 : maxoff = PageGetMaxOffsetNumber(page);
8302 : }
8303 :
8304 : /*
8305 : * In passing, detect index corruption involving an index page with a
8306 : * TID that points to a location in the heap that couldn't possibly be
8307 : * correct. We only do this with actual TIDs from caller's index page
8308 : * (not items reached by traversing through a HOT chain).
8309 : */
8310 1067118 : index_delete_check_htid(delstate, page, maxoff, htid, istatus);
8311 :
8312 1067118 : if (istatus->knowndeletable)
8313 : Assert(!delstate->bottomup && !istatus->promising);
8314 : else
8315 : {
8316 801756 : ItemPointerData tmp = *htid;
8317 : HeapTupleData heapTuple;
8318 :
8319 : /* Are any tuples from this HOT chain non-vacuumable? */
8320 801756 : if (heap_hot_search_buffer(&tmp, rel, buf, &SnapshotNonVacuumable,
8321 : &heapTuple, NULL, true))
8322 476150 : continue; /* can't delete entry */
8323 :
8324 : /* Caller will delete, since whole HOT chain is vacuumable */
8325 325606 : istatus->knowndeletable = true;
8326 :
8327 : /* Maintain index free space info for bottom-up deletion case */
8328 325606 : if (delstate->bottomup)
8329 : {
8330 : Assert(istatus->freespace > 0);
8331 14928 : actualfreespace += istatus->freespace;
8332 14928 : if (actualfreespace >= curtargetfreespace)
8333 4780 : bottomup_final_block = true;
8334 : }
8335 : }
8336 :
8337 : /*
8338 : * Maintain snapshotConflictHorizon value for deletion operation as a
8339 : * whole by advancing current value using heap tuple headers. This is
8340 : * loosely based on the logic for pruning a HOT chain.
8341 : */
8342 590968 : offnum = ItemPointerGetOffsetNumber(htid);
8343 590968 : priorXmax = InvalidTransactionId; /* cannot check first XMIN */
8344 : for (;;)
8345 41258 : {
8346 : ItemId lp;
8347 : HeapTupleHeader htup;
8348 :
8349 : /* Sanity check (pure paranoia) */
8350 632226 : if (offnum < FirstOffsetNumber)
8351 0 : break;
8352 :
8353 : /*
8354 : * An offset past the end of page's line pointer array is possible
8355 : * when the array was truncated
8356 : */
8357 632226 : if (offnum > maxoff)
8358 0 : break;
8359 :
8360 632226 : lp = PageGetItemId(page, offnum);
8361 632226 : if (ItemIdIsRedirected(lp))
8362 : {
8363 18334 : offnum = ItemIdGetRedirect(lp);
8364 18334 : continue;
8365 : }
8366 :
8367 : /*
8368 : * We'll often encounter LP_DEAD line pointers (especially with an
8369 : * entry marked knowndeletable by our caller up front). No heap
8370 : * tuple headers get examined for an htid that leads us to an
8371 : * LP_DEAD item. This is okay because the earlier pruning
8372 : * operation that made the line pointer LP_DEAD in the first place
8373 : * must have considered the original tuple header as part of
8374 : * generating its own snapshotConflictHorizon value.
8375 : *
8376 : * Relying on XLOG_HEAP2_PRUNE_VACUUM_SCAN records like this is
8377 : * the same strategy that index vacuuming uses in all cases. Index
8378 : * VACUUM WAL records don't even have a snapshotConflictHorizon
8379 : * field of their own for this reason.
8380 : */
8381 613892 : if (!ItemIdIsNormal(lp))
8382 396520 : break;
8383 :
8384 217372 : htup = (HeapTupleHeader) PageGetItem(page, lp);
8385 :
8386 : /*
8387 : * Check the tuple XMIN against prior XMAX, if any
8388 : */
8389 240296 : if (TransactionIdIsValid(priorXmax) &&
8390 22924 : !TransactionIdEquals(HeapTupleHeaderGetXmin(htup), priorXmax))
8391 0 : break;
8392 :
8393 217372 : HeapTupleHeaderAdvanceConflictHorizon(htup,
8394 : &snapshotConflictHorizon);
8395 :
8396 : /*
8397 : * If the tuple is not HOT-updated, then we are at the end of this
8398 : * HOT-chain. No need to visit later tuples from the same update
8399 : * chain (they get their own index entries) -- just move on to
8400 : * next htid from index AM caller.
8401 : */
8402 217372 : if (!HeapTupleHeaderIsHotUpdated(htup))
8403 194448 : break;
8404 :
8405 : /* Advance to next HOT chain member */
8406 : Assert(ItemPointerGetBlockNumber(&htup->t_ctid) == blkno);
8407 22924 : offnum = ItemPointerGetOffsetNumber(&htup->t_ctid);
8408 22924 : priorXmax = HeapTupleHeaderGetUpdateXid(htup);
8409 : }
8410 :
8411 : /* Enable further/final shrinking of deltids for caller */
8412 590968 : finalndeltids = i + 1;
8413 : }
8414 :
8415 11164 : UnlockReleaseBuffer(buf);
8416 :
8417 : /*
8418 : * Shrink deltids array to exclude non-deletable entries at the end. This
8419 : * is not just a minor optimization. Final deltids array size might be
8420 : * zero for a bottom-up caller. Index AM is explicitly allowed to rely on
8421 : * ndeltids being zero in all cases with zero total deletable entries.
8422 : */
8423 : Assert(finalndeltids > 0 || delstate->bottomup);
8424 11164 : delstate->ndeltids = finalndeltids;
8425 :
8426 11164 : return snapshotConflictHorizon;
8427 : }
8428 :
8429 : /*
8430 : * Specialized inlineable comparison function for index_delete_sort()
8431 : */
8432 : static inline int
8433 25151366 : index_delete_sort_cmp(TM_IndexDelete *deltid1, TM_IndexDelete *deltid2)
8434 : {
8435 25151366 : ItemPointer tid1 = &deltid1->tid;
8436 25151366 : ItemPointer tid2 = &deltid2->tid;
8437 :
8438 : {
8439 25151366 : BlockNumber blk1 = ItemPointerGetBlockNumber(tid1);
8440 25151366 : BlockNumber blk2 = ItemPointerGetBlockNumber(tid2);
8441 :
8442 25151366 : if (blk1 != blk2)
8443 10295590 : return (blk1 < blk2) ? -1 : 1;
8444 : }
8445 : {
8446 14855776 : OffsetNumber pos1 = ItemPointerGetOffsetNumber(tid1);
8447 14855776 : OffsetNumber pos2 = ItemPointerGetOffsetNumber(tid2);
8448 :
8449 14855776 : if (pos1 != pos2)
8450 14855776 : return (pos1 < pos2) ? -1 : 1;
8451 : }
8452 :
8453 : Assert(false);
8454 :
8455 0 : return 0;
8456 : }
8457 :
8458 : /*
8459 : * Sort deltids array from delstate by TID. This prepares it for further
8460 : * processing by heap_index_delete_tuples().
8461 : *
8462 : * This operation becomes a noticeable consumer of CPU cycles with some
8463 : * workloads, so we go to the trouble of specialization/micro optimization.
8464 : * We use shellsort for this because it's easy to specialize, compiles to
8465 : * relatively few instructions, and is adaptive to presorted inputs/subsets
8466 : * (which are typical here).
8467 : */
8468 : static void
8469 11164 : index_delete_sort(TM_IndexDeleteOp *delstate)
8470 : {
8471 11164 : TM_IndexDelete *deltids = delstate->deltids;
8472 11164 : int ndeltids = delstate->ndeltids;
8473 :
8474 : /*
8475 : * Shellsort gap sequence (taken from Sedgewick-Incerpi paper).
8476 : *
8477 : * This implementation is fast with array sizes up to ~4500. This covers
8478 : * all supported BLCKSZ values.
8479 : */
8480 11164 : const int gaps[9] = {1968, 861, 336, 112, 48, 21, 7, 3, 1};
8481 :
8482 : /* Think carefully before changing anything here -- keep swaps cheap */
8483 : StaticAssertDecl(sizeof(TM_IndexDelete) <= 8,
8484 : "element size exceeds 8 bytes");
8485 :
8486 111640 : for (int g = 0; g < lengthof(gaps); g++)
8487 : {
8488 15053840 : for (int hi = gaps[g], i = hi; i < ndeltids; i++)
8489 : {
8490 14953364 : TM_IndexDelete d = deltids[i];
8491 14953364 : int j = i;
8492 :
8493 25877794 : while (j >= hi && index_delete_sort_cmp(&deltids[j - hi], &d) >= 0)
8494 : {
8495 10924430 : deltids[j] = deltids[j - hi];
8496 10924430 : j -= hi;
8497 : }
8498 14953364 : deltids[j] = d;
8499 : }
8500 : }
8501 11164 : }
8502 :
8503 : /*
8504 : * Returns how many blocks should be considered favorable/contiguous for a
8505 : * bottom-up index deletion pass. This is a number of heap blocks that starts
8506 : * from and includes the first block in line.
8507 : *
8508 : * There is always at least one favorable block during bottom-up index
8509 : * deletion. In the worst case (i.e. with totally random heap blocks) the
8510 : * first block in line (the only favorable block) can be thought of as a
8511 : * degenerate array of contiguous blocks that consists of a single block.
8512 : * heap_index_delete_tuples() will expect this.
8513 : *
8514 : * Caller passes blockgroups, a description of the final order that deltids
8515 : * will be sorted in for heap_index_delete_tuples() bottom-up index deletion
8516 : * processing. Note that deltids need not actually be sorted just yet (caller
8517 : * only passes deltids to us so that we can interpret blockgroups).
8518 : *
8519 : * You might guess that the existence of contiguous blocks cannot matter much,
8520 : * since in general the main factor that determines which blocks we visit is
8521 : * the number of promising TIDs, which is a fixed hint from the index AM.
8522 : * We're not really targeting the general case, though -- the actual goal is
8523 : * to adapt our behavior to a wide variety of naturally occurring conditions.
8524 : * The effects of most of the heuristics we apply are only noticeable in the
8525 : * aggregate, over time and across many _related_ bottom-up index deletion
8526 : * passes.
8527 : *
8528 : * Deeming certain blocks favorable allows heapam to recognize and adapt to
8529 : * workloads where heap blocks visited during bottom-up index deletion can be
8530 : * accessed contiguously, in the sense that each newly visited block is the
8531 : * neighbor of the block that bottom-up deletion just finished processing (or
8532 : * close enough to it). It will likely be cheaper to access more favorable
8533 : * blocks sooner rather than later (e.g. in this pass, not across a series of
8534 : * related bottom-up passes). Either way it is probably only a matter of time
8535 : * (or a matter of further correlated version churn) before all blocks that
8536 : * appear together as a single large batch of favorable blocks get accessed by
8537 : * _some_ bottom-up pass. Large batches of favorable blocks tend to either
8538 : * appear almost constantly or not even once (it all depends on per-index
8539 : * workload characteristics).
8540 : *
8541 : * Note that the blockgroups sort order applies a power-of-two bucketing
8542 : * scheme that creates opportunities for contiguous groups of blocks to get
8543 : * batched together, at least with workloads that are naturally amenable to
8544 : * being driven by heap block locality. This doesn't just enhance the spatial
8545 : * locality of bottom-up heap block processing in the obvious way. It also
8546 : * enables temporal locality of access, since sorting by heap block number
8547 : * naturally tends to make the bottom-up processing order deterministic.
8548 : *
8549 : * Consider the following example to get a sense of how temporal locality
8550 : * might matter: There is a heap relation with several indexes, each of which
8551 : * is low to medium cardinality. It is subject to constant non-HOT updates.
8552 : * The updates are skewed (in one part of the primary key, perhaps). None of
8553 : * the indexes are logically modified by the UPDATE statements (if they were
8554 : * then bottom-up index deletion would not be triggered in the first place).
8555 : * Naturally, each new round of index tuples (for each heap tuple that gets a
8556 : * heap_update() call) will have the same heap TID in each and every index.
8557 : * Since these indexes are low cardinality and never get logically modified,
8558 : * heapam processing during bottom-up deletion passes will access heap blocks
8559 : * in approximately sequential order. Temporal locality of access occurs due
8560 : * to bottom-up deletion passes behaving very similarly across each of the
8561 : * indexes at any given moment. This keeps the number of buffer misses needed
8562 : * to visit heap blocks to a minimum.
8563 : */
8564 : static int
8565 3828 : bottomup_nblocksfavorable(IndexDeleteCounts *blockgroups, int nblockgroups,
8566 : TM_IndexDelete *deltids)
8567 : {
8568 3828 : int64 lastblock = -1;
8569 3828 : int nblocksfavorable = 0;
8570 :
8571 : Assert(nblockgroups >= 1);
8572 : Assert(nblockgroups <= BOTTOMUP_MAX_NBLOCKS);
8573 :
8574 : /*
8575 : * We tolerate heap blocks that will be accessed only slightly out of
8576 : * physical order. Small blips occur when a pair of almost-contiguous
8577 : * blocks happen to fall into different buckets (perhaps due only to a
8578 : * small difference in npromisingtids that the bucketing scheme didn't
8579 : * quite manage to ignore). We effectively ignore these blips by applying
8580 : * a small tolerance. The precise tolerance we use is a little arbitrary,
8581 : * but it works well enough in practice.
8582 : */
8583 12104 : for (int b = 0; b < nblockgroups; b++)
8584 : {
8585 11574 : IndexDeleteCounts *group = blockgroups + b;
8586 11574 : TM_IndexDelete *firstdtid = deltids + group->ifirsttid;
8587 11574 : BlockNumber block = ItemPointerGetBlockNumber(&firstdtid->tid);
8588 :
8589 11574 : if (lastblock != -1 &&
8590 7746 : ((int64) block < lastblock - BOTTOMUP_TOLERANCE_NBLOCKS ||
8591 6540 : (int64) block > lastblock + BOTTOMUP_TOLERANCE_NBLOCKS))
8592 : break;
8593 :
8594 8276 : nblocksfavorable++;
8595 8276 : lastblock = block;
8596 : }
8597 :
8598 : /* Always indicate that there is at least 1 favorable block */
8599 : Assert(nblocksfavorable >= 1);
8600 :
8601 3828 : return nblocksfavorable;
8602 : }
8603 :
8604 : /*
8605 : * qsort comparison function for bottomup_sort_and_shrink()
8606 : */
8607 : static int
8608 384564 : bottomup_sort_and_shrink_cmp(const void *arg1, const void *arg2)
8609 : {
8610 384564 : const IndexDeleteCounts *group1 = (const IndexDeleteCounts *) arg1;
8611 384564 : const IndexDeleteCounts *group2 = (const IndexDeleteCounts *) arg2;
8612 :
8613 : /*
8614 : * Most significant field is npromisingtids (which we invert the order of
8615 : * so as to sort in desc order).
8616 : *
8617 : * Caller should have already normalized npromisingtids fields into
8618 : * power-of-two values (buckets).
8619 : */
8620 384564 : if (group1->npromisingtids > group2->npromisingtids)
8621 18830 : return -1;
8622 365734 : if (group1->npromisingtids < group2->npromisingtids)
8623 21698 : return 1;
8624 :
8625 : /*
8626 : * Tiebreak: desc ntids sort order.
8627 : *
8628 : * We cannot expect power-of-two values for ntids fields. We should
8629 : * behave as if they were already rounded up for us instead.
8630 : */
8631 344036 : if (group1->ntids != group2->ntids)
8632 : {
8633 248074 : uint32 ntids1 = pg_nextpower2_32((uint32) group1->ntids);
8634 248074 : uint32 ntids2 = pg_nextpower2_32((uint32) group2->ntids);
8635 :
8636 248074 : if (ntids1 > ntids2)
8637 35230 : return -1;
8638 212844 : if (ntids1 < ntids2)
8639 45282 : return 1;
8640 : }
8641 :
8642 : /*
8643 : * Tiebreak: asc offset-into-deltids-for-block (offset to first TID for
8644 : * block in deltids array) order.
8645 : *
8646 : * This is equivalent to sorting in ascending heap block number order
8647 : * (among otherwise equal subsets of the array). This approach allows us
8648 : * to avoid accessing the out-of-line TID. (We rely on the assumption
8649 : * that the deltids array was sorted in ascending heap TID order when
8650 : * these offsets to the first TID from each heap block group were formed.)
8651 : */
8652 263524 : if (group1->ifirsttid > group2->ifirsttid)
8653 129048 : return 1;
8654 134476 : if (group1->ifirsttid < group2->ifirsttid)
8655 134476 : return -1;
8656 :
8657 0 : pg_unreachable();
8658 :
8659 : return 0;
8660 : }
8661 :
8662 : /*
8663 : * heap_index_delete_tuples() helper function for bottom-up deletion callers.
8664 : *
8665 : * Sorts deltids array in the order needed for useful processing by bottom-up
8666 : * deletion. The array should already be sorted in TID order when we're
8667 : * called. The sort process groups heap TIDs from deltids into heap block
8668 : * groupings. Earlier/more-promising groups/blocks are usually those that are
8669 : * known to have the most "promising" TIDs.
8670 : *
8671 : * Sets new size of deltids array (ndeltids) in state. deltids will only have
8672 : * TIDs from the BOTTOMUP_MAX_NBLOCKS most promising heap blocks when we
8673 : * return. This often means that deltids will be shrunk to a small fraction
8674 : * of its original size (we eliminate many heap blocks from consideration for
8675 : * caller up front).
8676 : *
8677 : * Returns the number of "favorable" blocks. See bottomup_nblocksfavorable()
8678 : * for a definition and full details.
8679 : */
8680 : static int
8681 3828 : bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate)
8682 : {
8683 : IndexDeleteCounts *blockgroups;
8684 : TM_IndexDelete *reordereddeltids;
8685 3828 : BlockNumber curblock = InvalidBlockNumber;
8686 3828 : int nblockgroups = 0;
8687 3828 : int ncopied = 0;
8688 3828 : int nblocksfavorable = 0;
8689 :
8690 : Assert(delstate->bottomup);
8691 : Assert(delstate->ndeltids > 0);
8692 :
8693 : /* Calculate per-heap-block count of TIDs */
8694 3828 : blockgroups = palloc(sizeof(IndexDeleteCounts) * delstate->ndeltids);
8695 1857900 : for (int i = 0; i < delstate->ndeltids; i++)
8696 : {
8697 1854072 : TM_IndexDelete *ideltid = &delstate->deltids[i];
8698 1854072 : TM_IndexStatus *istatus = delstate->status + ideltid->id;
8699 1854072 : ItemPointer htid = &ideltid->tid;
8700 1854072 : bool promising = istatus->promising;
8701 :
8702 1854072 : if (curblock != ItemPointerGetBlockNumber(htid))
8703 : {
8704 : /* New block group */
8705 74754 : nblockgroups++;
8706 :
8707 : Assert(curblock < ItemPointerGetBlockNumber(htid) ||
8708 : !BlockNumberIsValid(curblock));
8709 :
8710 74754 : curblock = ItemPointerGetBlockNumber(htid);
8711 74754 : blockgroups[nblockgroups - 1].ifirsttid = i;
8712 74754 : blockgroups[nblockgroups - 1].ntids = 1;
8713 74754 : blockgroups[nblockgroups - 1].npromisingtids = 0;
8714 : }
8715 : else
8716 : {
8717 1779318 : blockgroups[nblockgroups - 1].ntids++;
8718 : }
8719 :
8720 1854072 : if (promising)
8721 235858 : blockgroups[nblockgroups - 1].npromisingtids++;
8722 : }
8723 :
8724 : /*
8725 : * We're about ready to sort block groups to determine the optimal order
8726 : * for visiting heap blocks. But before we do, round the number of
8727 : * promising tuples for each block group up to the next power-of-two,
8728 : * unless it is very low (less than 4), in which case we round up to 4.
8729 : * npromisingtids is far too noisy to trust when choosing between a pair
8730 : * of block groups that both have very low values.
8731 : *
8732 : * This scheme divides heap blocks/block groups into buckets. Each bucket
8733 : * contains blocks that have _approximately_ the same number of promising
8734 : * TIDs as each other. The goal is to ignore relatively small differences
8735 : * in the total number of promising entries, so that the whole process can
8736 : * give a little weight to heapam factors (like heap block locality)
8737 : * instead. This isn't a trade-off, really -- we have nothing to lose. It
8738 : * would be foolish to interpret small differences in npromisingtids
8739 : * values as anything more than noise.
8740 : *
8741 : * We tiebreak on nhtids when sorting block group subsets that have the
8742 : * same npromisingtids, but this has the same issues as npromisingtids,
8743 : * and so nhtids is subject to the same power-of-two bucketing scheme. The
8744 : * only reason that we don't fix nhtids in the same way here too is that
8745 : * we'll need accurate nhtids values after the sort. We handle nhtids
8746 : * bucketization dynamically instead (in the sort comparator).
8747 : *
8748 : * See bottomup_nblocksfavorable() for a full explanation of when and how
8749 : * heap locality/favorable blocks can significantly influence when and how
8750 : * heap blocks are accessed.
8751 : */
8752 78582 : for (int b = 0; b < nblockgroups; b++)
8753 : {
8754 74754 : IndexDeleteCounts *group = blockgroups + b;
8755 :
8756 : /* Better off falling back on nhtids with low npromisingtids */
8757 74754 : if (group->npromisingtids <= 4)
8758 63854 : group->npromisingtids = 4;
8759 : else
8760 10900 : group->npromisingtids =
8761 10900 : pg_nextpower2_32((uint32) group->npromisingtids);
8762 : }
8763 :
8764 : /* Sort groups and rearrange caller's deltids array */
8765 3828 : qsort(blockgroups, nblockgroups, sizeof(IndexDeleteCounts),
8766 : bottomup_sort_and_shrink_cmp);
8767 3828 : reordereddeltids = palloc(delstate->ndeltids * sizeof(TM_IndexDelete));
8768 :
8769 3828 : nblockgroups = Min(BOTTOMUP_MAX_NBLOCKS, nblockgroups);
8770 : /* Determine number of favorable blocks at the start of final deltids */
8771 3828 : nblocksfavorable = bottomup_nblocksfavorable(blockgroups, nblockgroups,
8772 : delstate->deltids);
8773 :
8774 25574 : for (int b = 0; b < nblockgroups; b++)
8775 : {
8776 21746 : IndexDeleteCounts *group = blockgroups + b;
8777 21746 : TM_IndexDelete *firstdtid = delstate->deltids + group->ifirsttid;
8778 :
8779 21746 : memcpy(reordereddeltids + ncopied, firstdtid,
8780 21746 : sizeof(TM_IndexDelete) * group->ntids);
8781 21746 : ncopied += group->ntids;
8782 : }
8783 :
8784 : /* Copy final grouped and sorted TIDs back into start of caller's array */
8785 3828 : memcpy(delstate->deltids, reordereddeltids,
8786 : sizeof(TM_IndexDelete) * ncopied);
8787 3828 : delstate->ndeltids = ncopied;
8788 :
8789 3828 : pfree(reordereddeltids);
8790 3828 : pfree(blockgroups);
8791 :
8792 3828 : return nblocksfavorable;
8793 : }
8794 :
8795 : /*
8796 : * Perform XLogInsert for a heap-visible operation. 'block' is the block
8797 : * being marked all-visible, and vm_buffer is the buffer containing the
8798 : * corresponding visibility map block. Both should have already been modified
8799 : * and dirtied.
8800 : *
8801 : * snapshotConflictHorizon comes from the largest xmin on the page being
8802 : * marked all-visible. REDO routine uses it to generate recovery conflicts.
8803 : *
8804 : * If checksums or wal_log_hints are enabled, we may also generate a full-page
8805 : * image of heap_buffer. Otherwise, we optimize away the FPI (by specifying
8806 : * REGBUF_NO_IMAGE for the heap buffer), in which case the caller should *not*
8807 : * update the heap page's LSN.
8808 : */
8809 : XLogRecPtr
8810 83652 : log_heap_visible(Relation rel, Buffer heap_buffer, Buffer vm_buffer,
8811 : TransactionId snapshotConflictHorizon, uint8 vmflags)
8812 : {
8813 : xl_heap_visible xlrec;
8814 : XLogRecPtr recptr;
8815 : uint8 flags;
8816 :
8817 : Assert(BufferIsValid(heap_buffer));
8818 : Assert(BufferIsValid(vm_buffer));
8819 :
8820 83652 : xlrec.snapshotConflictHorizon = snapshotConflictHorizon;
8821 83652 : xlrec.flags = vmflags;
8822 83652 : if (RelationIsAccessibleInLogicalDecoding(rel))
8823 266 : xlrec.flags |= VISIBILITYMAP_XLOG_CATALOG_REL;
8824 83652 : XLogBeginInsert();
8825 83652 : XLogRegisterData(&xlrec, SizeOfHeapVisible);
8826 :
8827 83652 : XLogRegisterBuffer(0, vm_buffer, 0);
8828 :
8829 83652 : flags = REGBUF_STANDARD;
8830 83652 : if (!XLogHintBitIsNeeded())
8831 5854 : flags |= REGBUF_NO_IMAGE;
8832 83652 : XLogRegisterBuffer(1, heap_buffer, flags);
8833 :
8834 83652 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_VISIBLE);
8835 :
8836 83652 : return recptr;
8837 : }
8838 :
8839 : /*
8840 : * Perform XLogInsert for a heap-update operation. Caller must already
8841 : * have modified the buffer(s) and marked them dirty.
8842 : */
8843 : static XLogRecPtr
8844 572742 : log_heap_update(Relation reln, Buffer oldbuf,
8845 : Buffer newbuf, HeapTuple oldtup, HeapTuple newtup,
8846 : HeapTuple old_key_tuple,
8847 : bool all_visible_cleared, bool new_all_visible_cleared)
8848 : {
8849 : xl_heap_update xlrec;
8850 : xl_heap_header xlhdr;
8851 : xl_heap_header xlhdr_idx;
8852 : uint8 info;
8853 : uint16 prefix_suffix[2];
8854 572742 : uint16 prefixlen = 0,
8855 572742 : suffixlen = 0;
8856 : XLogRecPtr recptr;
8857 572742 : Page page = BufferGetPage(newbuf);
8858 572742 : bool need_tuple_data = RelationIsLogicallyLogged(reln);
8859 : bool init;
8860 : int bufflags;
8861 :
8862 : /* Caller should not call me on a non-WAL-logged relation */
8863 : Assert(RelationNeedsWAL(reln));
8864 :
8865 572742 : XLogBeginInsert();
8866 :
8867 572742 : if (HeapTupleIsHeapOnly(newtup))
8868 276768 : info = XLOG_HEAP_HOT_UPDATE;
8869 : else
8870 295974 : info = XLOG_HEAP_UPDATE;
8871 :
8872 : /*
8873 : * If the old and new tuple are on the same page, we only need to log the
8874 : * parts of the new tuple that were changed. That saves on the amount of
8875 : * WAL we need to write. Currently, we just count any unchanged bytes in
8876 : * the beginning and end of the tuple. That's quick to check, and
8877 : * perfectly covers the common case that only one field is updated.
8878 : *
8879 : * We could do this even if the old and new tuple are on different pages,
8880 : * but only if we don't make a full-page image of the old page, which is
8881 : * difficult to know in advance. Also, if the old tuple is corrupt for
8882 : * some reason, it would allow the corruption to propagate the new page,
8883 : * so it seems best to avoid. Under the general assumption that most
8884 : * updates tend to create the new tuple version on the same page, there
8885 : * isn't much to be gained by doing this across pages anyway.
8886 : *
8887 : * Skip this if we're taking a full-page image of the new page, as we
8888 : * don't include the new tuple in the WAL record in that case. Also
8889 : * disable if wal_level='logical', as logical decoding needs to be able to
8890 : * read the new tuple in whole from the WAL record alone.
8891 : */
8892 572742 : if (oldbuf == newbuf && !need_tuple_data &&
8893 276612 : !XLogCheckBufferNeedsBackup(newbuf))
8894 : {
8895 275696 : char *oldp = (char *) oldtup->t_data + oldtup->t_data->t_hoff;
8896 275696 : char *newp = (char *) newtup->t_data + newtup->t_data->t_hoff;
8897 275696 : int oldlen = oldtup->t_len - oldtup->t_data->t_hoff;
8898 275696 : int newlen = newtup->t_len - newtup->t_data->t_hoff;
8899 :
8900 : /* Check for common prefix between old and new tuple */
8901 22655448 : for (prefixlen = 0; prefixlen < Min(oldlen, newlen); prefixlen++)
8902 : {
8903 22606490 : if (newp[prefixlen] != oldp[prefixlen])
8904 226738 : break;
8905 : }
8906 :
8907 : /*
8908 : * Storing the length of the prefix takes 2 bytes, so we need to save
8909 : * at least 3 bytes or there's no point.
8910 : */
8911 275696 : if (prefixlen < 3)
8912 44148 : prefixlen = 0;
8913 :
8914 : /* Same for suffix */
8915 8867970 : for (suffixlen = 0; suffixlen < Min(oldlen, newlen) - prefixlen; suffixlen++)
8916 : {
8917 8818514 : if (newp[newlen - suffixlen - 1] != oldp[oldlen - suffixlen - 1])
8918 226240 : break;
8919 : }
8920 275696 : if (suffixlen < 3)
8921 67674 : suffixlen = 0;
8922 : }
8923 :
8924 : /* Prepare main WAL data chain */
8925 572742 : xlrec.flags = 0;
8926 572742 : if (all_visible_cleared)
8927 3230 : xlrec.flags |= XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED;
8928 572742 : if (new_all_visible_cleared)
8929 1486 : xlrec.flags |= XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED;
8930 572742 : if (prefixlen > 0)
8931 231548 : xlrec.flags |= XLH_UPDATE_PREFIX_FROM_OLD;
8932 572742 : if (suffixlen > 0)
8933 208022 : xlrec.flags |= XLH_UPDATE_SUFFIX_FROM_OLD;
8934 572742 : if (need_tuple_data)
8935 : {
8936 94044 : xlrec.flags |= XLH_UPDATE_CONTAINS_NEW_TUPLE;
8937 94044 : if (old_key_tuple)
8938 : {
8939 286 : if (reln->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
8940 126 : xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_TUPLE;
8941 : else
8942 160 : xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_KEY;
8943 : }
8944 : }
8945 :
8946 : /* If new tuple is the single and first tuple on page... */
8947 579276 : if (ItemPointerGetOffsetNumber(&(newtup->t_self)) == FirstOffsetNumber &&
8948 6534 : PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
8949 : {
8950 6258 : info |= XLOG_HEAP_INIT_PAGE;
8951 6258 : init = true;
8952 : }
8953 : else
8954 566484 : init = false;
8955 :
8956 : /* Prepare WAL data for the old page */
8957 572742 : xlrec.old_offnum = ItemPointerGetOffsetNumber(&oldtup->t_self);
8958 572742 : xlrec.old_xmax = HeapTupleHeaderGetRawXmax(oldtup->t_data);
8959 1145484 : xlrec.old_infobits_set = compute_infobits(oldtup->t_data->t_infomask,
8960 572742 : oldtup->t_data->t_infomask2);
8961 :
8962 : /* Prepare WAL data for the new page */
8963 572742 : xlrec.new_offnum = ItemPointerGetOffsetNumber(&newtup->t_self);
8964 572742 : xlrec.new_xmax = HeapTupleHeaderGetRawXmax(newtup->t_data);
8965 :
8966 572742 : bufflags = REGBUF_STANDARD;
8967 572742 : if (init)
8968 6258 : bufflags |= REGBUF_WILL_INIT;
8969 572742 : if (need_tuple_data)
8970 94044 : bufflags |= REGBUF_KEEP_DATA;
8971 :
8972 572742 : XLogRegisterBuffer(0, newbuf, bufflags);
8973 572742 : if (oldbuf != newbuf)
8974 272246 : XLogRegisterBuffer(1, oldbuf, REGBUF_STANDARD);
8975 :
8976 572742 : XLogRegisterData(&xlrec, SizeOfHeapUpdate);
8977 :
8978 : /*
8979 : * Prepare WAL data for the new tuple.
8980 : */
8981 572742 : if (prefixlen > 0 || suffixlen > 0)
8982 : {
8983 274788 : if (prefixlen > 0 && suffixlen > 0)
8984 : {
8985 164782 : prefix_suffix[0] = prefixlen;
8986 164782 : prefix_suffix[1] = suffixlen;
8987 164782 : XLogRegisterBufData(0, &prefix_suffix, sizeof(uint16) * 2);
8988 : }
8989 110006 : else if (prefixlen > 0)
8990 : {
8991 66766 : XLogRegisterBufData(0, &prefixlen, sizeof(uint16));
8992 : }
8993 : else
8994 : {
8995 43240 : XLogRegisterBufData(0, &suffixlen, sizeof(uint16));
8996 : }
8997 : }
8998 :
8999 572742 : xlhdr.t_infomask2 = newtup->t_data->t_infomask2;
9000 572742 : xlhdr.t_infomask = newtup->t_data->t_infomask;
9001 572742 : xlhdr.t_hoff = newtup->t_data->t_hoff;
9002 : Assert(SizeofHeapTupleHeader + prefixlen + suffixlen <= newtup->t_len);
9003 :
9004 : /*
9005 : * PG73FORMAT: write bitmap [+ padding] [+ oid] + data
9006 : *
9007 : * The 'data' doesn't include the common prefix or suffix.
9008 : */
9009 572742 : XLogRegisterBufData(0, &xlhdr, SizeOfHeapHeader);
9010 572742 : if (prefixlen == 0)
9011 : {
9012 341194 : XLogRegisterBufData(0,
9013 341194 : (char *) newtup->t_data + SizeofHeapTupleHeader,
9014 341194 : newtup->t_len - SizeofHeapTupleHeader - suffixlen);
9015 : }
9016 : else
9017 : {
9018 : /*
9019 : * Have to write the null bitmap and data after the common prefix as
9020 : * two separate rdata entries.
9021 : */
9022 : /* bitmap [+ padding] [+ oid] */
9023 231548 : if (newtup->t_data->t_hoff - SizeofHeapTupleHeader > 0)
9024 : {
9025 231548 : XLogRegisterBufData(0,
9026 231548 : (char *) newtup->t_data + SizeofHeapTupleHeader,
9027 231548 : newtup->t_data->t_hoff - SizeofHeapTupleHeader);
9028 : }
9029 :
9030 : /* data after common prefix */
9031 231548 : XLogRegisterBufData(0,
9032 231548 : (char *) newtup->t_data + newtup->t_data->t_hoff + prefixlen,
9033 231548 : newtup->t_len - newtup->t_data->t_hoff - prefixlen - suffixlen);
9034 : }
9035 :
9036 : /* We need to log a tuple identity */
9037 572742 : if (need_tuple_data && old_key_tuple)
9038 : {
9039 : /* don't really need this, but its more comfy to decode */
9040 286 : xlhdr_idx.t_infomask2 = old_key_tuple->t_data->t_infomask2;
9041 286 : xlhdr_idx.t_infomask = old_key_tuple->t_data->t_infomask;
9042 286 : xlhdr_idx.t_hoff = old_key_tuple->t_data->t_hoff;
9043 :
9044 286 : XLogRegisterData(&xlhdr_idx, SizeOfHeapHeader);
9045 :
9046 : /* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
9047 286 : XLogRegisterData((char *) old_key_tuple->t_data + SizeofHeapTupleHeader,
9048 286 : old_key_tuple->t_len - SizeofHeapTupleHeader);
9049 : }
9050 :
9051 : /* filtering by origin on a row level is much more efficient */
9052 572742 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
9053 :
9054 572742 : recptr = XLogInsert(RM_HEAP_ID, info);
9055 :
9056 572742 : return recptr;
9057 : }
9058 :
9059 : /*
9060 : * Perform XLogInsert of an XLOG_HEAP2_NEW_CID record
9061 : *
9062 : * This is only used in wal_level >= WAL_LEVEL_LOGICAL, and only for catalog
9063 : * tuples.
9064 : */
9065 : static XLogRecPtr
9066 48200 : log_heap_new_cid(Relation relation, HeapTuple tup)
9067 : {
9068 : xl_heap_new_cid xlrec;
9069 :
9070 : XLogRecPtr recptr;
9071 48200 : HeapTupleHeader hdr = tup->t_data;
9072 :
9073 : Assert(ItemPointerIsValid(&tup->t_self));
9074 : Assert(tup->t_tableOid != InvalidOid);
9075 :
9076 48200 : xlrec.top_xid = GetTopTransactionId();
9077 48200 : xlrec.target_locator = relation->rd_locator;
9078 48200 : xlrec.target_tid = tup->t_self;
9079 :
9080 : /*
9081 : * If the tuple got inserted & deleted in the same TX we definitely have a
9082 : * combo CID, set cmin and cmax.
9083 : */
9084 48200 : if (hdr->t_infomask & HEAP_COMBOCID)
9085 : {
9086 : Assert(!(hdr->t_infomask & HEAP_XMAX_INVALID));
9087 : Assert(!HeapTupleHeaderXminInvalid(hdr));
9088 4036 : xlrec.cmin = HeapTupleHeaderGetCmin(hdr);
9089 4036 : xlrec.cmax = HeapTupleHeaderGetCmax(hdr);
9090 4036 : xlrec.combocid = HeapTupleHeaderGetRawCommandId(hdr);
9091 : }
9092 : /* No combo CID, so only cmin or cmax can be set by this TX */
9093 : else
9094 : {
9095 : /*
9096 : * Tuple inserted.
9097 : *
9098 : * We need to check for LOCK ONLY because multixacts might be
9099 : * transferred to the new tuple in case of FOR KEY SHARE updates in
9100 : * which case there will be an xmax, although the tuple just got
9101 : * inserted.
9102 : */
9103 57786 : if (hdr->t_infomask & HEAP_XMAX_INVALID ||
9104 13622 : HEAP_XMAX_IS_LOCKED_ONLY(hdr->t_infomask))
9105 : {
9106 30544 : xlrec.cmin = HeapTupleHeaderGetRawCommandId(hdr);
9107 30544 : xlrec.cmax = InvalidCommandId;
9108 : }
9109 : /* Tuple from a different tx updated or deleted. */
9110 : else
9111 : {
9112 13620 : xlrec.cmin = InvalidCommandId;
9113 13620 : xlrec.cmax = HeapTupleHeaderGetRawCommandId(hdr);
9114 : }
9115 44164 : xlrec.combocid = InvalidCommandId;
9116 : }
9117 :
9118 : /*
9119 : * Note that we don't need to register the buffer here, because this
9120 : * operation does not modify the page. The insert/update/delete that
9121 : * called us certainly did, but that's WAL-logged separately.
9122 : */
9123 48200 : XLogBeginInsert();
9124 48200 : XLogRegisterData(&xlrec, SizeOfHeapNewCid);
9125 :
9126 : /* will be looked at irrespective of origin */
9127 :
9128 48200 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_NEW_CID);
9129 :
9130 48200 : return recptr;
9131 : }
9132 :
9133 : /*
9134 : * Build a heap tuple representing the configured REPLICA IDENTITY to represent
9135 : * the old tuple in an UPDATE or DELETE.
9136 : *
9137 : * Returns NULL if there's no need to log an identity or if there's no suitable
9138 : * key defined.
9139 : *
9140 : * Pass key_required true if any replica identity columns changed value, or if
9141 : * any of them have any external data. Delete must always pass true.
9142 : *
9143 : * *copy is set to true if the returned tuple is a modified copy rather than
9144 : * the same tuple that was passed in.
9145 : */
9146 : static HeapTuple
9147 3539298 : ExtractReplicaIdentity(Relation relation, HeapTuple tp, bool key_required,
9148 : bool *copy)
9149 : {
9150 3539298 : TupleDesc desc = RelationGetDescr(relation);
9151 3539298 : char replident = relation->rd_rel->relreplident;
9152 : Bitmapset *idattrs;
9153 : HeapTuple key_tuple;
9154 : bool nulls[MaxHeapAttributeNumber];
9155 : Datum values[MaxHeapAttributeNumber];
9156 :
9157 3539298 : *copy = false;
9158 :
9159 3539298 : if (!RelationIsLogicallyLogged(relation))
9160 3338722 : return NULL;
9161 :
9162 200576 : if (replident == REPLICA_IDENTITY_NOTHING)
9163 462 : return NULL;
9164 :
9165 200114 : if (replident == REPLICA_IDENTITY_FULL)
9166 : {
9167 : /*
9168 : * When logging the entire old tuple, it very well could contain
9169 : * toasted columns. If so, force them to be inlined.
9170 : */
9171 382 : if (HeapTupleHasExternal(tp))
9172 : {
9173 8 : *copy = true;
9174 8 : tp = toast_flatten_tuple(tp, desc);
9175 : }
9176 382 : return tp;
9177 : }
9178 :
9179 : /* if the key isn't required and we're only logging the key, we're done */
9180 199732 : if (!key_required)
9181 93758 : return NULL;
9182 :
9183 : /* find out the replica identity columns */
9184 105974 : idattrs = RelationGetIndexAttrBitmap(relation,
9185 : INDEX_ATTR_BITMAP_IDENTITY_KEY);
9186 :
9187 : /*
9188 : * If there's no defined replica identity columns, treat as !key_required.
9189 : * (This case should not be reachable from heap_update, since that should
9190 : * calculate key_required accurately. But heap_delete just passes
9191 : * constant true for key_required, so we can hit this case in deletes.)
9192 : */
9193 105974 : if (bms_is_empty(idattrs))
9194 12042 : return NULL;
9195 :
9196 : /*
9197 : * Construct a new tuple containing only the replica identity columns,
9198 : * with nulls elsewhere. While we're at it, assert that the replica
9199 : * identity columns aren't null.
9200 : */
9201 93932 : heap_deform_tuple(tp, desc, values, nulls);
9202 :
9203 301782 : for (int i = 0; i < desc->natts; i++)
9204 : {
9205 207850 : if (bms_is_member(i + 1 - FirstLowInvalidHeapAttributeNumber,
9206 : idattrs))
9207 : Assert(!nulls[i]);
9208 : else
9209 113894 : nulls[i] = true;
9210 : }
9211 :
9212 93932 : key_tuple = heap_form_tuple(desc, values, nulls);
9213 93932 : *copy = true;
9214 :
9215 93932 : bms_free(idattrs);
9216 :
9217 : /*
9218 : * If the tuple, which by here only contains indexed columns, still has
9219 : * toasted columns, force them to be inlined. This is somewhat unlikely
9220 : * since there's limits on the size of indexed columns, so we don't
9221 : * duplicate toast_flatten_tuple()s functionality in the above loop over
9222 : * the indexed columns, even if it would be more efficient.
9223 : */
9224 93932 : if (HeapTupleHasExternal(key_tuple))
9225 : {
9226 8 : HeapTuple oldtup = key_tuple;
9227 :
9228 8 : key_tuple = toast_flatten_tuple(oldtup, desc);
9229 8 : heap_freetuple(oldtup);
9230 : }
9231 :
9232 93932 : return key_tuple;
9233 : }
9234 :
9235 : /*
9236 : * HeapCheckForSerializableConflictOut
9237 : * We are reading a tuple. If it's not visible, there may be a
9238 : * rw-conflict out with the inserter. Otherwise, if it is visible to us
9239 : * but has been deleted, there may be a rw-conflict out with the deleter.
9240 : *
9241 : * We will determine the top level xid of the writing transaction with which
9242 : * we may be in conflict, and ask CheckForSerializableConflictOut() to check
9243 : * for overlap with our own transaction.
9244 : *
9245 : * This function should be called just about anywhere in heapam.c where a
9246 : * tuple has been read. The caller must hold at least a shared lock on the
9247 : * buffer, because this function might set hint bits on the tuple. There is
9248 : * currently no known reason to call this function from an index AM.
9249 : */
9250 : void
9251 60098946 : HeapCheckForSerializableConflictOut(bool visible, Relation relation,
9252 : HeapTuple tuple, Buffer buffer,
9253 : Snapshot snapshot)
9254 : {
9255 : TransactionId xid;
9256 : HTSV_Result htsvResult;
9257 :
9258 60098946 : if (!CheckForSerializableConflictOutNeeded(relation, snapshot))
9259 60048276 : return;
9260 :
9261 : /*
9262 : * Check to see whether the tuple has been written to by a concurrent
9263 : * transaction, either to create it not visible to us, or to delete it
9264 : * while it is visible to us. The "visible" bool indicates whether the
9265 : * tuple is visible to us, while HeapTupleSatisfiesVacuum checks what else
9266 : * is going on with it.
9267 : *
9268 : * In the event of a concurrently inserted tuple that also happens to have
9269 : * been concurrently updated (by a separate transaction), the xmin of the
9270 : * tuple will be used -- not the updater's xid.
9271 : */
9272 50670 : htsvResult = HeapTupleSatisfiesVacuum(tuple, TransactionXmin, buffer);
9273 50670 : switch (htsvResult)
9274 : {
9275 49066 : case HEAPTUPLE_LIVE:
9276 49066 : if (visible)
9277 49040 : return;
9278 26 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9279 26 : break;
9280 704 : case HEAPTUPLE_RECENTLY_DEAD:
9281 : case HEAPTUPLE_DELETE_IN_PROGRESS:
9282 704 : if (visible)
9283 562 : xid = HeapTupleHeaderGetUpdateXid(tuple->t_data);
9284 : else
9285 142 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9286 :
9287 704 : if (TransactionIdPrecedes(xid, TransactionXmin))
9288 : {
9289 : /* This is like the HEAPTUPLE_DEAD case */
9290 : Assert(!visible);
9291 124 : return;
9292 : }
9293 580 : break;
9294 652 : case HEAPTUPLE_INSERT_IN_PROGRESS:
9295 652 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9296 652 : break;
9297 248 : case HEAPTUPLE_DEAD:
9298 : Assert(!visible);
9299 248 : return;
9300 0 : default:
9301 :
9302 : /*
9303 : * The only way to get to this default clause is if a new value is
9304 : * added to the enum type without adding it to this switch
9305 : * statement. That's a bug, so elog.
9306 : */
9307 0 : elog(ERROR, "unrecognized return value from HeapTupleSatisfiesVacuum: %u", htsvResult);
9308 :
9309 : /*
9310 : * In spite of having all enum values covered and calling elog on
9311 : * this default, some compilers think this is a code path which
9312 : * allows xid to be used below without initialization. Silence
9313 : * that warning.
9314 : */
9315 : xid = InvalidTransactionId;
9316 : }
9317 :
9318 : Assert(TransactionIdIsValid(xid));
9319 : Assert(TransactionIdFollowsOrEquals(xid, TransactionXmin));
9320 :
9321 : /*
9322 : * Find top level xid. Bail out if xid is too early to be a conflict, or
9323 : * if it's our own xid.
9324 : */
9325 1258 : if (TransactionIdEquals(xid, GetTopTransactionIdIfAny()))
9326 124 : return;
9327 1134 : xid = SubTransGetTopmostTransaction(xid);
9328 1134 : if (TransactionIdPrecedes(xid, TransactionXmin))
9329 0 : return;
9330 :
9331 1134 : CheckForSerializableConflictOut(relation, xid, snapshot);
9332 : }
|